Economia: Spring 2005 (Economia: Journal of the Latin American & Caribbean Economic Association)

2005

Volume 5 Number 2

Latin American and Caribbean Economic Associationwww.lacea.org

Brookings Institution PressWashington, D.C.

www.brookings.edu

Volume

5 Num

ber2Brookings /LACEA

Cover design by Rogue Element

2005Spring

Journal of the Latin American and Caribbean Economic Association

Lora and Olivera on Electoral

Consequences of the Washington

Consensus

Suominen and Estevadeordalon Rules of Origin in Trade Agreements

Majnoni and Powell on Basel II

and Bank Capital Requirements

Echeverry, Ibáñez, Moya, and Hillón on Reforming Public

Transport in Bogotá

Kaplan, Martínez González,and Robertson on Wages after

Displacement

econ

omia econom

ía Spring

E D I T O R

Andrés Velasco

LATIN AMERICAN AND CARIBBEANECONOMIC ASSOCIATION

BROOKINGS INSTITUTION PRESSWashington, D.C.

2005

Volume 5 Number 2 economíaSpring


Articles in this publication were developed by the authors for the biannual Economía meetings.In all cases the papers are the product of the authors’ thinking alone and do not imply endorse-ment by the staff members, officers, or trustees of the Brookings Institution or of LACEA, or ofthose institutions with which the authors are affiliated.

Copyright © 2005

LATIN AMERICAN AND CARIBBEAN ECONOMIC ASSOCIATION

www.lacea.org

Published by

BROOKINGS INSTITUTION PRESS

1775 Massachusetts Avenue, N.W., Washington, DC 20036

www.brookings.edu

ISSN 1529-7470

ISBN 0-8157-2079-3

For information on subscriptions, standing orders, and individual copies, contact Brookings Insti-tution Press, P.O. Box 465, Hanover, PA 17331-0465. Or call 866-698-0010. E-mail [email protected]. Visit Brookings online at http://bookstore.brookings.edu.

Brookings periodicals are available online through Online Computer Library Center (contact the OCLC subscriptions department at 800/848-5878, ext. 6251) and Project Muse (http://muse.jhu.edu).

Authorization to photocopy items for internal or personal use or the internal or personal use of specific clientsis granted by the Brookings Institution for libraries and other users registered with the Copyright ClearanceCenter Transactional Reporting Service, provided that the basic fee is paid to the Copyright Clearance Center,222 Rosewood Drive, Danvers, MA 01923. For more information, please contact CCC at 508/750-8400. Thisauthorization does not extend to other kinds of copying, such as copying for general distribution, or for creatingnew collective works, or for sale. Specific written permission for such copying must be obtained from the Per-missions Department, Brookings Institution Press, 1775 Massachusetts Avenue, N.W., Washington, DC 20036,fax 202/797-6195; e-mail [email protected].

Editor’s Summary vii

EDUARDO LORA and MAURICIO OLIVERA

The Electoral Consequences of the Washington Consensus 1Comments by Sebastián Galiani and Ernesto Dal Bó 46

ANTONI ESTEVADEORDAL and KATI SUOMINEN

Rules of Origin in Preferential Trading Arrangements:Is All Well with the Spaghetti Bowl in the Americas? 63Comments by Pablo Sanguinetti and Alberto Trejos 93

GIOVANNI MAJNONI and ANDREW POWELL

Reforming Bank Capital Requirements:Implications of Basel II for Latin American Countries 105Comments by Patricia Correa and Philip Brock 141

JUAN CARLOS ECHEVERRY, ANA MARÍA IBÁÑEZ, ANDRÉS MOYA, and LUIS CARLOS HILLÓN

The Economics of TransMilenio, a Mass Transit System for Bogotá 151Comments by Mauricio Cárdenas and Andrés Gómez-Lobo 189

DAVID S. KAPLAN, GABRIEL MARTÍNEZ GONZÁLEZ, and RAYMOND ROBERTSON

What Happens to Wages after Displacement? 197Comments by Naércio Menezes-Filho and Omar Arias 235

2005

Volume 5 Number 2 economíaSpring


LATIN AMERICAN AND CARIBBEAN ECONOMIC ASSOCIATION

The Latin American and Caribbean Economic Association (LACEA), orAsociación de Economía de América Latina y el Caribe, is an internationalassociation of economists with common research interests in LatinAmerica. It was formed in 1992 to facilitate the exchange of ideas amongeconomists and policymakers.

Membership in LACEA is open to all individuals or institutionsprofessionally concerned with the study of Latin American and Caribbeaneconomies. For membership information, please visit the LACEA websiteat www.lacea.org and click on “join LACEA.”

O F F I C E R S

PresidentMariano Tommasi, Universidad de San Andrés, Argentina

Vice PresidentAndrés Velasco, Harvard University and Universidad de Chile

Past PresidentsSebastián Edwards, University of California–Los Angeles

Guillermo Calvo, Inter-American Development Bank and University of Maryland

Nora Lustig, Universidad de las Américas, Puebla

Albert Fishlow, Columbia University

SecretaryAriel Fiszbein, World Bank

TreasurerSergio Schmukler, World Bank

E X E C U T I V E C O M M I T T E E

Nancy Birdsall, Center for Global Development

François Bourguignon, World Bank

Raquel Fernández, New York University

Francisco H. G. Ferreira, Pontifícia Universidade Católica, Rio de Janeiro, and

World Bank

Nora Lustig, Universidad de las Américas, Puebla

José A. Ocampo, United Nations

Guillermo Perry, World Bank

Carola Pessino, Universidad Torcuato Di Tella, Argentina

Carmen Reinhart, University of Maryland

Andrés Rodríquez-Clare, Inter-American Development Bank

Jaime Saavedra, World Bank

Cristina T. Terra, Fundação Getúlio Vargas

ECONOMIA

EditorAndrés Velasco, Harvard University and Universidad de Chile

Editorial AssociateJennifer Hoover

Managing EditorMagdalena Balcells

Editorial BoardRafael Di Tella, Harvard University

Eduardo Engel, Yale University

Francisco H. G. Ferreira, Pontifícia Universidade Católica, Rio de Janeiro, and

World Bank

Carmen Pagés, World Bank

Roberto Rigobón, Massachusetts Institute of Technology

Andrés Rodríguez-Clare, Inter-American Development Bank

Roberto Steiner, International Monetary Fund

Miguel Urquiola, Columbia University

Sebastián Edwards (ex officio), University of California–Los Angeles

Mariano Tommasi (ex officio), Universidad de San Andrés, Argentina

ECONOMIA PANEL FOR VOLUME 5

Mauricio Cárdenas, Fedesarrollo-Colombia

Gerardo Esquivel Hernández, Colegio de México

Ronald Fischer, Universidad de Chile

Ilan Goldfajn, Pontifícia Universidade Católica, Rio de Janeiro

Eduardo Levy Yeyati, Universidad Torcuato Di Tella, Argentina

María Soledad Martínez Pería, World Bank

Francisco Rodríguez, Instituto de Estudios Superiores de Administración,

Venezuela

Nouriel Roubini, New York University

Jaime Saavedra, World Bank

Ernesto Schagrodsky, Universidad Torcuato Di Tella, Argentina

AUTHORS AND DISCUSSANTS

Omar Arias, World Bank

Philip Brock, University of Washington

Mauricio Cárdenas, Fedesarrollo-Colombia

Patricia Correa, Super Intendencia de Bancos de Colombia

Ernesto Dal Bó, University of California–Berkeley

Juan Carlos Echeverry, Universidad de los Andes

Antoni Estevadeordal, Inter-American Development Bank

Sebastián Galiani, Universidad de San Andrés

Andrés Gómez-Lobo, Universidad de Chile

Luis Carlos Hillón, Universidad de los Andes

Ana María Ibáñez, Universidad de los Andes

David S. Kaplan, Instituto Tecnológico Autónomo de México

Eduardo Lora, Inter-American Development Bank

Giovanni Majnoni, World Bank

Gabriel Martínez González, Inter-American Conference on Social Security

Naércio Menezes-Filho, University of São Paulo

Andrés Moya, Universidad de los Andes

Mauricio Olivera, Inter-American Development Bank

Andrew Powell, Universidad Torcuato Di Tella

Raymond Robertson, Macalester College

Pablo Sanguinetti, Universidad Torcuato Di Tella

Kati Suominen, Inter-American Development Bank

Alberto Trejos, Instituto Centroamericano de Administracion de Empresas INCAE

Editor’s Summary

Nearly two decades after a wave of policy changes swept through LatinAmerica, economic reforms continue to be the focus of much dis-cussion. Critics claim that the promarket reforms have failed to

deliver economic growth, and that the time has come to try something else.Advocates claim that the reforms were never given a fair chance—too lit-tle was done, often too late. Complete the reform process, they claim, andgrowth will come.

Both sides do agree on one point: Latin America seems to be sufferingfrom reform fatigue, and another wave of reforms is unlikely to happen anytime soon. Certainly not in countries led by left-leaning populists, such asArgentina’s Néstor Kirchner or Venezuela’s Hugo Chávez. The reformmomentum has even stalled in countries led by promarket conservatives—Mexico under Vicente Fox and Colombia under Alvaro Uribe are twoexamples. If such reforms are now unpopular in many quarters, did thepoliticians who initially adopted them bear a political cost? Was the Wash-ington Consensus electorally bad for friends of Washington? That is thequestion studied by Eduardo Lora and Mauricio Olivera in the lead articleof this, the tenth issue of Economía.

Lora and Olivera analyze the outcome of sixty-six presidential elec-tions and eighty-one parliamentary elections in seventeen Latin Americancountries from 1985 to 2002. Their general conclusion is striking: reform-ing parties and politicians were rewarded electorally only when reformsinvolved macroeconomic stabilization and a sharp reduction in inflation;otherwise, their reforming zeal cost them dearly at the polls. Economic out-comes do matter for electoral outcomes. Lora and Olivera find that theincumbent’s party is rewarded in presidential elections for reductions in theinflation rate and in legislative elections for increases in the growth rate.Changes in unemployment and income distribution, however, do not appearto influence voters’ behavior.

vii

A N D R É S V E L A S C O

What is even more surprising is that, at the polls, policies matter irre-spective of their results—that is, their effects on growth or inflation. Elec-torates seem not to like reform policies of the kind applied in LatinAmerica in the 1990s. In a regression with electoral outcomes on the righthand side, reform indexes have a negative and significant effect, even whenthe authors control for changes in inflation and growth. The point esti-mate of the effect of policies on electoral results implies that the incum-bent’s party typically lost 15 percent of its vote in presidential elections onaccount of the average amount of promarket reforms introduced duringits term. More aggressive reformers (say, those reforming one standarddeviation above the mean) sacrificed 27 percent of their vote on account ofpromarket reforms. Statistically, this seems to be a very robust result forpresidential elections.1

Moreover, lying about one’s true intentions does not seem to be a goodelectoral strategy. Several Latin American politicians—including Fujimoriin Peru, Menem in Argentina, and more recently Gutiérrez in Ecuador—first ran as opponents of the Washington Consensus, then followed ortho-dox policies. The paper shows that a candidate that said one thing on taxpolicy and then did another was, on average, punished more severely atthe polls. Campaign promises do not seem to matter for the effect of otherpolicies on voting behavior.

These results raise two kinds of questions. For academics, the issue iswhy inputs (policies) matter and not just outputs. Is it ideology, pure andsimple? Or is it that because outcomes represent an extremely noisy signalof politicians’ competence, the choice of policies conveys some informa-tion that voters find useful? For policymakers, the question is political:what has to change in Latin America before ambitious reforms become fea-sible again? Are all large-scale reforms out of the question, or only thosethat carry the Washington Consensus label? Both sets of questions remainvery much open.

The unpopularity of the reforms does not mean, however, that policy isfrozen everywhere. Trade is one area in which reform has not stopped dead

viii E C O N O M I A , Spring 2005

1. The total effect of reforms on electoral outcomes is the sum of two effects: a directeffect that runs from policies to votes and an indirect effect that runs from policies to eco-nomic outcomes to votes. The first is typically large and negative, while the second is pos-itive insofar as the reforms lowered inflation and stimulated growth. The figures givencorrespond to the total effect—that is, after the positive indirect effects have been taken intoaccount. The direct negative effects are much larger. See the paper for details.

on its tracks. Liberalization agreements, of both the bilateral and regionalkind, continue to be signed, though at a less frenzied pace than a decadeago. Nearly fifty deals have been forged in the Americas since 1990s. Butthis veritable “spaghetti bowl” of overlapping and sometimes contradictoryagreements has costs as well as benefits. One cost, studied by Antoni Este-vadeordal and Kati Suominen in the second paper of this issue, results fromthe rules of origin applied.

At heart, the matter is simple: if Brazil gives Paraguay preferentialaccess to its market, Brazilian policymakers want to make sure that the newimports entering Brazil are, in fact, made in Paraguay and not in a thirdcountry attempting to benefit from Paraguay’s preferential status. But whatsounds simple in theory becomes devilishly complicated in practice. What,precisely, is a Paraguayan good? Goods often have imported inputs, and infew or no items is 100 percent of value added likely to originate inParaguay. Where, then, should a country draw the line?

Rules of origin attempt to settle the issue, but in doing so they face manypitfalls. If the set of rules is too stringent, then the bulk of Paraguayangoods may be left out of the Brazilian market. Indeed, such rules can beused as protectionist devices that effectively undercut trade preferences andcontradict the avowed liberalizing intent of free trade agreements. Anotherproblem is that rules of origin are almost inevitably complex (the paperidentifies several kinds, each with its own subcategories). Applying themcan be very costly, especially for the poorer economies in the region.

Estevadeordal and Suominen offer three main conclusions. First, puttingstringent rules of origin into an agreement makes it more politically feasi-ble, since the rules can be used as a tool to pay off protectionist interests.Second, there is evidence that restrictive rules of origin undercut the liber-alizing potential of free trade agreements. NAFTA is a particularly egre-gious example of this, with many Mexican goods subjected to rules thatverge on ludicrous. It is unfortunate, therefore—and this is the third con-clusion of the paper—that the NAFTA model of rules of origin is increas-ingly being used in other agreements in the region. This does not bodewell for free trade in the Americas.

Not all is lost, however. NAFTA-type rules are at least precise, and theyleave less room for arbitrary application than do other types of rules oforigin used in earlier agreements. Moreover, the growing homogeneity ofrules that follow the NAFTA model simplifies the life of customs officialsand lowers transaction costs. Last, and most important, the NAFTA rules of

Andrés Velasco ix

origin have what trade experts call lenient facilitation devices. In English,this means that the rules themselves include ways to reduce their restric-tiveness. A key aspect is diagonal cumulation, which allows countries tiedby the same set of origin rules to use products that originate in any part ofthe common rules-of-origin zone as if they originated in the exportingcountry. Therefore, argue the authors, the rules of origin in a future FreeTrade Area of the Americas—if one ever materializes—should not be allthat restrictive. One can only hope they are right.

Financial regulation is another area in which policy is changing, as aresult of both internal needs and international changes in standards. The1988 Basel Accord on bank capital—the so-called Basel I agreement—isnow in place throughout the region, and discussion has shifted to whetherand how Latin American countries should apply Basel II. It is widelyaccepted that bank capital ought to be regulated, but how to do so remainsopen to debate. The simple approach of Basel I divides assets into verybroad risk categories and establishes an 8 percent minimum capital require-ment for risky assets. The potential for arbitraging one’s way around thissimple rule has grown, however, as risk management becomes moresophisticated. In response, Basel II goes well beyond simple quantitativerequirements, proposing two basic approaches: the standardized approach,which uses external credit rating agencies together with a table that mapsthose ratings directly into capital requirements; and the internal ratings-based (IRB) approach, in which the banks themselves estimate their cus-tomers’ default probability (without relying on external rating agencies) andthen use a particular formula to determine capital requirements as a func-tion of the estimated default probability.

The third paper in this issue, by Giovanni Majnoni and Andrew Powell,focuses on a key aspect of Basel II application. Many emerging marketsdo not have many (or any) external rating agencies, so the standardizedapproach may not be applicable. The internal ratings-based approach, inturn, is complex, and its application and supervision may stretch limitedsupervisory resources. Majnoni and Powell suggest a simplification of theIRB approach that could be used as a transition arrangement. In theircentralized ratings-based (CRB) approach, banks would rate their clients,but the regulator would determine the rating scale and the way in whichthe banks’ ratings map into default probabilities. Using a centralized scalewould facilitate comparison across banks and greatly ease the monitor-ing of banks’ ratings. Those requirements would also be easier to monitor,

x E C O N O M I A , Spring 2005

since the regulator would determine how banks’ ratings feed into capitalrequirements.

The hard part of the approach is deciding what kinds of standards theregulator should apply, since what works in rich countries may not workin emerging economies. Basel II’s IRB approach suggests a formula for cal-culating a bank’s capital requirement as a function of three basic variables:default probability, exposure at default, and loss given default. A regulatormight then ask a bank to hold provisions and capital to cover a specifiedpercentage of the distribution of losses to ensure the continued solvencyof the bank except in highly extraordinary circumstances. The calibration ofthe Basel II IRB formula uses a value at risk of 99.9 percent with a hori-zon of one year—that is, a bank is only expected to use up its capital inone year with a probability of 0.1 percent, or once every 1,000 years.

Majnoni and Powell employ a bootstrapping technique to calculate lossdistribution functions for Argentina, Brazil, and Mexico, using data on loanperformance from public credit registries. They then use these functions toestimate the size of expected and unexpected losses of an average-sizedbank with a loan portfolio randomly drawn from the universe of loanswithin the financial system. Their results show that these three countrieshave significantly higher default probabilities than Group of Ten (G10)countries. As a result, both current practice under Basel I and the sug-gested standards under Basel II may be inadequate. To achieve a 99 percentlevel of protection, capital requirements would need to be close to 15 per-cent, which is significantly higher than the 8 percent level recommendedin Basel I. Even higher levels would be required to achieve 99.9 percentprotection, as intended in Basel II. They also find that Basel II’s IRBapproach would result in levels of 90–95 percent protection rather than the99.9 percent goal. This is not surprising, since the IRB was calibrated forthe safer economies of G10 countries.

If bank regulation needs modernizing in Latin America, public trans-port does too. The spectacle of streets packed with old buses spewing blacksmoke is all too common in many cities of the region, from Mexico Cityto Quito and from São Paulo to Santiago. Poor public transport inducesmore private cars to enter the streets, worsening congestion and pollution.

If you think that this is a textbook case of the state not doing the job ofproviding public services, think again. Bus systems are private in manycities in Latin America, and that does not seem to solve the problem. AsJuan Carlos Echeverry, Ana María Ibáñez, Andrés Moya, and Luis Carlos

Andrés Velasco xi

Hillón document in their paper in this volume, the market for urban busesis ripe with market failures: unclear definition of property rights on thecurbside and on the road; cartelization that results in fares set above thecompetitive equilibrium levels; misalignment between the incentives of busdrivers and owners, in a typical principal-agent conflict; and congestion andpollution externalities. In many developing countries, these market fail-ures are exacerbated by weak regulation and enforcement. The result oftenis too many buses each carrying too few passengers in unsafe conditions,clogging the streets and soiling the air as they move (or fail to move) along.

One city in Latin America to have tackled the problem head-on isBogotá, Colombia. Its so-called TransMilenio system is now being imitatedin Quito and Santiago, among others, as well as several cities in Colom-bia. Echeverry, Ibáñez, Moya, and Hillón explain the logic behind the newsystem and analyze is effects. The key elements of the new system are asfollows: (i) a hybrid public-private system, with concession contracts forprivate service providers; (ii) competition “for the road” (rather than “onthe road”) in a tendering process, with fare-setting based on long-terminvestment recovery; (iii) remuneration based on kilometers traveled ratherthan passengers transported, so as to prevent drivers from fighting over pas-sengers on the street; (iv) separation between the transportation service andthe fare collection process; and (v) exclusive road and curb-side service inmetro-like stations.

Congestion, pollution, traffic accidents, travel times, and waiting timesall fell dramatically along the corridors where TransMilenio was first put towork. The system was initially hailed as the solution of Bogotá’s serioustransport problems. Not all results were unambiguously positive, however,as the paper makes clear. Increased ridership resulted in jammed busesand rising waiting times. Moreover, the full system covering the entirecity is not expected to be operating until 2015. This gradual transition didnot help: older buses were displaced to secondary streets, where trafficand pollution increased.

A cost-benefit analysis of the system as is, with approximately 25 per-cent of the routes in operation, reveals welfare gains for users of the newroutes, but an overall negative effect stemming primarily from increasesin travel time for passengers using the traditional transport system. Sincecongestion costs are highly nonlinear, the welfare losses from heightenedcongestion in unserved corridors more than offset the benefits from Trans-Milenio, even though those benefits are sizeable. The authors conclude by

xii E C O N O M I A , Spring 2005

arguing that the adoption of a new public transport system must be cou-pled with improved regulation of all other public transport providers, so asto avoid the problem that arose in Bogotá.

What happens to workers’ wages and employment prospects once theyare displaced from their current jobs—for instance, by trade reform? If theyare likely to be re-hired quickly at comparable wages, then no policyresponse is called for; but if some wage losses are large and lasting, thentargeted help for displaced workers may be called for. David Kaplan,Gabriel Martínez, and Raymond Robertson study the issue for the case ofMexico, using an administrative data set that allows them to follow indi-vidual workers over a period of thirty-two quarters in four regions that varysignificantly in labor market conditions. They focus on the differences ininstitutions, inequality, and labor market conditions that may explain dif-ferences in wage behavior after displacement.

One striking result is the heterogeneity of worker experiences, whichrange from large wage losses to many instances of gains after displacement.This is consistent with earlier results for other countries, but it cannot beattributed to differences in institutions (rates of unionization) or inequal-ity, which are quite similar across Mexico. Rather, Kaplan, Martínez, andRobertson argue that labor market conditions, which vary quite a bit acrosstime and regions within Mexico, explain the heterogeneity of experiences.In good times and in the most economically active regions, postdisplace-ment wages are generally higher than they were in the previous jobs. How-ever, workers who are fired during times of high unemployment and inless economically active regions experience lasting effects on wages. If anypublic assistance is to be disbursed, Kaplan, Martínez, and Robertsonargue, it should go to these workers.

All papers but one included in this issue were presented at the panelmeeting held in San José, Costa Rica, in October 2004. The local hosts, andparticularly Juan Rafael Vargas, provided much help. As usual, associateeditors of Economía, members of the 2004 panel, and outside discussantsand referees have done an outstanding job. Thanks is due to them all.

Andrés Velasco xiii

The Electoral Consequences of theWashington Consensus

No country in Latin America escaped the dictums of the WashingtonConsensus. From Brazil under left-leaning Fernando Henrique Car-doso to Mexico under ultra-orthodox economist Ernesto Zedillo

and Peru under Alberto Fujimori’s yoke, macroeconomic imbalances werebrought under control, barriers to international trade were lifted, and state-owned enterprises were privatized. Whether this one-size-fits-all prescrip-tion was imposed from outside or adopted at will by the governments electedon the promise of improving the lot of their peoples may be a matter ofdebate. But all sides seem to agree on one point: the results did not meetthe expectations created both by outsiders and by those in power.

Up to the mid-1980s only two countries in Latin America had adopteda package of policies similar to what became to be known as the Wash-ington Consensus at the turn of the decade. Those two were undemocraticChile and impoverished Bolivia, by then among the most politically andeconomically unstable countries, if not in the world, then certainly in LatinAmerica. Extreme cases, extreme policies: that was a common interpreta-tion of the two experiences. Less common was the expectation that thosepolicies were about to be adopted by virtually every Latin American coun-try in the next few years, both those in which democracy had been the rule

1

E D U A R D O L O R AM A U R I C I O O L I V E R A

Lora and Olivera are with the Inter-American Development Bank.We are grateful for the valuable research assistance of Carlos Andrés Gómez. We also

thank Benito Arrunada, Mauricio Cárdenas, Stephen Kay, Ugo Panizza, Andres Rodríguez-Clare, Mariano Tommasi, Jessica Wallack, and seminar participants at ISNIE-UniversityPompeu Fabra, LACEA-PEG, Econnet-IADB, and the Economia panel meeting for com-ments and suggestions. We are especially grateful to Rafael Di Tella, Sebastian Galiani, andErnesto Dal Bó for their detailed and very useful comments and suggestions. We would likealso to thank Sebastian Saiegh for allowing us to use his data on political coalitions.

for decades, like Colombia, Costa Rica, and Venezuela, and those wherethe third wave of democratization was just arriving, such as Argentina,Brazil, and Uruguay.

The years of high expectations, both about democratization and aboutWashington Consensus–type policies, are over. Latin Americans are stillconvinced democrats, but enthusiasm has waned. Three out of every fourLatin Americans see democracy as the best form of government—or rather,as the least bad, since 68 percent think that democracy is not function-ing well in their countries. Latin Americans are even more sceptical aboutthe benefits of promarket economic policies. Only one out of four LatinAmericans considers privatization to have been beneficial for his or her coun-try and barely 16 percent think that the market economy is doing a good job.1

Malaise is getting the upper hand in a number of places. Electricity andwater privatizations were blocked in Arequipa (Peru) and Cochabamba(Bolivia), following violent clashes between vociferous opponents and thepolice. An ambitious project to attract foreign direct investment toBolivia’s gas sector was derailed by the Indian communities. While theseevents may be dismissed as isolated expressions of popular feeling, a newcrop of presidents from Néstor Kirchner in Argentina to Lucio Gutiérrezin Ecuador and Tabaré Vásquez in Uruguay has won clear majorities inpopular elections after campaigning against the excesses of market-oriented policies.

In an attempt to establish whether this malaise is justified or not, econ-omists have devoted substantial effort to assessing the economic and socialconsequences of the Washington Consensus policies. The dominant viewseems to be that they have had positive effects on economic growth andincome levels, though there is intense debate over the size of those effects,over whether they are transient or permanent, and over the importance ofeach of the components of the Washington Consensus. The dominant viewalso holds that the effects have been muted by lack of regulatory and insti-tutional support for the liberalization efforts, though the specific forms ofregulation and institutions necessary for that purpose are far from clear.Even more intense is the debate over the social and distributional effects offiscal stabilization and promarket reforms, which are the two main pillars ofthe Washington Consensus.2

2 E C O N O M I A , Spring 2005

1. Opinion data come from the 2003 issue of Latinobarómetro, a public opinion surveyconducted by the Corporación Latinobarómetro, Santiago, Chile.

2. These debates are surveyed in Lora and Panizza (2002); Kuczynski and Williamson(2003); and Lora, Panizza and Quispe-Agnoli (2004).

However, the future of these policies will depend not so much on theirefficacy but on whether they receive the support of the electorate. On this,the state of knowledge is much more scant and fragmentary, as will be seenbelow. This paper attempts to help fill that vacuum by evaluating througheconometric methods the electoral consequences of the Washington Con-sensus. Although our approach is backward looking, it sheds considerablelight on the future. Our study shows that the electorate cares not only aboutthe outcomes of the policies (maybe about only some outcomes and notothers), but also about the policies themselves, irrespective of whether theyproduce good or bad (observable) outcomes. In addition, the electorateseems to care about whether the policies adopted by a government are in line with the ideology of the incumbent’s party and with preelectoralpromises. Furthermore, in presidential regimes voters cast separate votesfor the executive and the legislature, and outcomes and policies affect eachvote differently. The presidential vote is more volatile and more suscepti-ble to economic outcomes and policies, but votes for the legislature are notcompletely immune: policies in which the legislature clearly plays a role,such as privatizations, tend to have electoral consequences. These resultsprovide a nuanced landscape for the future of Washington Consensus poli-cies, where neither bold backslashes nor aggressive promarket reformsshould be expected in the future. Not only is the time of high expectationsover; perhaps the time for deep reforms is also past.

In the next section of the paper we present a short survey of the literatureassessing the electoral consequences of the Washington Consensus policiesand derive our empirical hypotheses. On that basis, we then discuss the the-oretical and econometric approaches that support the empirical analysis.In subsequent sections we describe the data, present the econometric find-ings, and discuss our conclusions.

A note on terminology is in order before proceeding. “Neo-liberal,”“market-oriented,” “orthodox,” and a variety of other labels have beenattached to the set of economic policies in vogue since the early nineties inLatin America and elsewhere. We use these terms interchangeably, but notloosely: for the sake of clarity and brevity, this paper deals with the ten poli-cies summarized in the classic article by Williamson that made the term“Washington Consensus” famous.3 We assume that all those labels refer tothat same set of policies (as detailed below in the section titled “Data”).

Eduardo Lora and Mauricio Olivera 3

3. Williamson (1990a).

Review of the Empirical Literature and Some Testable Hypotheses

The most straightforward view of the response of the electorate to eco-nomic policies is based on the “economic voting” argument: people basetheir electoral decisions on cost-benefit calculations. If the policies bringnet benefits to them, they cast their votes to support the government, or theparty, administering those policies; if the policies bring losses to them,they lend their support to the candidate, or the party, opposing them. Eco-nomic voting is usually assumed to be retrospective: voters observe pastperformance and assume that past trends will persist into the future if thegovernment or the party remains in power. If those trends are deemedacceptable, given a set of standards or expectations voters decide to reelectthe incumbent, or his party if the option of reelection does not exist. There-fore, in retrospective economic voting policies play no direct role, sincevoters decide entirely on the base of past outcomes.4

Considerable evidence from advanced industrial democracies supports theview that past economic performance influences people’s voting decisionsand their support for governments.5 An important empirical finding from thisliterature is that voters base their decisions on aggregate (or “sociotropic”)economic outcomes such as growth, inflation, and unemployment, ratherthan on individual (or “pocketbook”) outcomes. Most of the empirical liter-ature on developed countries comes from single-country analyses, basedeither on time-series electoral outcomes or public opinion polls. The eco-nomic voting hypothesis is more robust for public opinion polls than foractual electoral outcomes.6 Empirical studies of electoral behavior in theUnited States using state-level data lend support to the simple economicvoting hypothesis, in the sense that voters are able to evaluate their state’seconomic performance relative to that of the national economy. Further-more, they (irrationally) reward state governors for economic fluctuationsthat are unrelated to gubernatorial actions, which implies that they have lim-ited ability to filter aggregate economic information.7 The ability of voters


4. Stokes (2001b, pp. 1–18) provides a concise review and discussion of the theoreticalunderpinnings of retrospective economic voting.

5. Based on the seminal work by Downs (1957); among the initial papers on economicvoting in the United States are Kramer (1971); Meltzer and Vellrath (1975); and Arcelus andMeltzer (1975).

6. Lewis-Beck (1988) is a salient example of the early empirical literature based on opin-ion polls in European countries. For a review of this literature, see Stokes (2001b, pp. 2–8).

7. Wolfers (2002).

to gather and update information is a central issue in the theoretical andempirical literature on economic voting.8 Although some evidence pointsout to the presence of prospective as well as retrospective behavior, uncer-tainty about the workings of the economy and the relatively high cost ofgathering and processing the information necessary to forecast outcomesis consistent with the importance of retrospective voting in the empiricalfindings.9

Empirical support for the economic voting hypothesis in Latin Americahas been uncovered by Karen Remmer, Michael Coppedge, Kenneth Robertsand Erik Wibbels, and Susan Stokes.10 A concise summary of these findingsis presented in table 1. Based on data for twenty-one competitive electionsbetween 1982 and 1990, Remmer has found that conditions of economiccrisis undermine support for incumbents and provoke high levels of elec-toral volatility.11 The magnitude of the electoral change is found to be asso-ciated with the depth of the crisis during the campaign period, with variationsin exchange rates, GDP, and inflation highly correlated with variousindicators of electoral outcomes. Her results also suggest that the effectof economic conditions on electoral instability are mediated by the struc-ture of the party system (insulating two-party systems from the volatilityexperienced by more fragmented systems). However, as Stokes pointsout, these results are anomalous given the predictions of normal economicvoting, as she “finds that incumbent parties suffered larger losses at thepolls when inflation went down (significant) and when GDP rose (notsignificant).”12

In a subsequent paper, Remmer presents new estimates on the influenceof inflation and growth on the incumbent vote in presidential elections.13

Her new database covers forty-nine elections for seven countries between1983 and 1999. Her results indicate that after controlling for the advantageof incumbency as well as major differences in the structure of party sys-tems, electoral outcomes are strongly influenced, in the direction expected,by macroeconomic performance in the year before the election. That is, infla-tion is found to be negatively correlated with electoral support, whereas


8. For a review of this debate, see Duch and Stevenson (2004); and Keech (1995).9. On prospective behavior, see, for instance, Lewis-Beck (1988).10. Remmer (1991, 2003); Coppedge (2001); Roberts and Wibbels (1999); Stokes (2001b).11. Remmer (1991).12. Stokes (2001b, p. 27).13. Remmer (2003).

growth is positively correlated with it. Furthermore, inflation is significantin all the regressions presented, while growth is more significant for theelections held in the 1990s than for those in the 1980s, indicating that thesensitivity of the electorate to economic performance has increased ratherthan waned over time.

Coppedge’s empirical work focuses on the impact of changes in inflationon legislative vote shares. His dependent variable consists of 132 changes inlegislative vote shares for major parties in eleven countries from 1978 to1995. His only indicator of economic performance is the change in (thelog of) inflation from the last year of the previous government to the lastyear of the current government. By interacting this variable with appropri-ate dummies, Coppedge finds that changes (whether increases or decreases)in inflation affect electoral support for the incumbents’ parties in theexpected way, while only increases in inflation improve the vote share ofthe opposition parties. However, these results apply only to parties “with afluid base,” that is, parties that do not count on a strong party identification.When there is such identification, voters are reluctant to question theirparty identification on the basis of macroeconomic outcomes.

Roberts and Wibbels consider economic voting as a possible explana-tion of electoral volatility in Latin America. Their database includes fifty-eight congressional elections and forty-three presidential elections in sixteenLatin American countries during the 1980s and 1990s. Their results showthat economic performance has an effect on electoral stability. Economicgrowth stabilizes partisan support in legislative elections, whereas sharp


T A B L E 1 . Summary of Empirical Findings on Economic Voting in Latin America

Election type (number of Estimation

Study Dependent variable countries) Period method Main results

Remmer (1991)

Remmer (2003)

Roberts and Wibbels (1999)

Coppedge (2001)Stokes (2001b)

Source: Authors’ calculations.a. Not significant.

Electoral volatility

Vote shares

Electoral volatility

Vote sharesProbability of a

security-orientedcandidate beingelected

Inflation −; GDPgrowth +a

Inflation −; GDPgrowth +

Inflation −a; GDPgrowth +

InflationInflation −; GDP

growth +

Presidential (12)

Presidential (8)

Legislative andpresidential (16)

Legislative (11)Presidential (15)

1982–90

1983–99

1980–97

1978–951982–95

Pooled OLS

Pooled OLS

Pooled OLS

Pooled OLSProbit

changes in the rate of inflation from one administration to the next, whetherpositive or negative, produce the opposite effect. Short-term inflation influ-ences support for incumbent presidents, but growth changes have only aweak effect on the vote for incumbents, “which suggests that voters aremore inclined to hold them directly accountable for monetary stabilitythan economic growth.” Although electoral volatility is influenced by eco-nomic performance, it is also related to the institutional characteristics ofpolitical regimes and party systems, and to the structure and organizationof class cleavages.14

In her study of “neoliberalism by surprise,” Stokes uses data fromtwenty-three elections in the 1980s and 1990s in order to assess how theelectorate judges incumbents who, having campaigned for stability-orientedor protectionist policies, once in office switch to market-oriented ones. Shefinds that for both, “switchers” and “non-switchers,” economic growthand inflation affect their vote share in the expected ways. Furthermore,voters are more sensitive to economic outcomes in the case of “switchers,”although this result is not statistically significant (more on these resultsbelow).15

These empirical studies taken together lend support to the retrospectiveeconomic voting argument in both presidential and legislative elections.They make clear that voting decisions are also influenced by political, insti-tutional, and structural factors and that some of these factors may influencethe severity with which voters judge economic outcomes. Therefore, basedon these studies, two testable propositions are derived:

1. Electoral support for the incumbent’s party is higher, the better theaggregate economic outcomes during his or her administration.

2. The sensitivity of electoral support to economic outcomes depends onthe institutional characteristics of the political regime and the party system.

As mentioned, in normal economic voting only past outcomes influencepeople’s views. However, as in all six of the Stokes case studies on mar-ket reforms in new democracies, people sometimes react to economicdeterioration by supporting the government more strongly; and conversely,they sometimes respond to economic improvements with pessimism andopposition.16 Normal economic voting is not the only pattern, especially inthe process of deep economic reform. If there are good reasons to believe


14. Roberts and Wibbels (1999), quote from p. 584.15. Stokes (2001a).16. Stokes (2001a).

that past circumstances are not good indicators of the future, informationother than past economic outcomes may influence people’s electoral deci-sions. For instance, voters may recognize that past circumstances wereaffected by factors beyond the government’s control and exonerate theincumbent from the responsibility for past declines in their welfare. Vot-ers may then forecast their future welfare as a function of government pol-icy, rather than as an extrapolation of the past. This sounds simpler thanit is, of course, because future government policies are unknown andbecause the relationship between policies and outcomes is diffuse. Peo-ple’s expectations of future policies may be formed on the basis of thepolicies adopted or announced by the incumbent or on the basis of hisparty’s ideology. These policy expectations may then be translated intoexpected outcomes through a set of beliefs and hypotheses about their pos-sible consequences.

It is often implicitly assumed that people’s (average) beliefs conformto the actual functioning of the real world. If that is so, assessing theeffects of economic policies would help explain voters’ electoral decisions.Economists have devoted considerable effort to evaluating the impact ofWashington Consensus policies on economic growth, income distribu-tion, employment levels, and a host of other variables.17 However, therehas been no comparable effort to examine whether these results are con-sistent with how the electorate responds to those policies. The only studyon the subject, by Carlos Gervasoni, has found positive correlationsbetween several indicators of heterodox (that is, anti-neoliberal) policiesand losses in the vote shares of the parties of the incumbents who adoptedthose policies.18 The variable with the largest and most significant effect ismoney supply growth. Import protection indicators are also significant,whereas fiscal deficit and the share of the state in GDP are not significant.These results suggest that Washington Consensus policies do not entailelectoral costs and may even produce electoral benefits, probably becausethey bring positive economic effects. It is suggestive that the most signif-icant policy variable is the money supply, because it is well known thatinflation is, ultimately, a monetary phenomenon, and as mentioned, empiri-cal evidence suggests that inflation is a key economic outcome influencingelectoral decisions.


17. For surveys of the literature, see Inter-American Development Bank (2003, chap. 5);Kuczynski and Williamson (2003); and Lora and Panizza (2002).

18. See Gervasoni (1997), citing a 1995 study.

However, it is a great leap of faith to assume that people’s beliefs con-form to the actual consequences of policies. In mapping policies on out-comes, ideology and leaders’ opinions may be more important for mostpeople than their limited understanding of how policies work their influencethrough the social and economic structures to affect production, employ-ment, or income distribution. Evidence on how those factors influenceelectoral responses to economic policies is very scant. However, in-depthcase studies on Argentina and Venezuela by Javier Corrales clearly showthat the reaction of the electorate to the adoption of neoliberal economicpolicies in the 1990s was mediated by the party structure and other insti-tutional factors.19 The cohesion and tactics of the Partido Peronista helpexplain the electorate’s support of the neoliberal reforms in Argentina inthe early 1990s, as well as their demise a decade later. Venezuela’s AcciónDemocrática lacked that cohesion, and its reforms were soon rejected bythe electorate.

If voters care about policies and not only about past outcomes, the pol-icy announcements of presidential candidates will be a key source of infor-mation. However, campaign promises are often poor predictors of actualpolicy: according to Stokes, of the thirty-three Latin American govern-ments that adopted promarket reforms between 1982 and 1995, only abouthalf (seventeen) hinted during their campaigns that such reforms weregoing to be implemented.20 This raises several empirical issues. First, dopolicy announcements in fact influence electoral decisions? Empirical evi-dence from the United States and other advanced industrialized economiesshows that they do: people seem to base their opinions in part on campaignannouncements, and voters punish ambiguous campaigns.21 Of course,some promises may resonate more than others, depending on, among otherthings, economic circumstances. For thirty-eight Latin American electionsin the 1980s and 1990s, Stokes finds that stability-oriented candidates (asopposed to market-oriented ones) stand a better chance of being elected,the lower the rates of GDP growth and inflation.22

A second empirical issue is whether deviating from campaign promisescarries electoral costs for the incumbent. Although deviations may inprinciple be costly, they may produce a positive payoff if they signal the


19. Corrales (2002).20. Stokes (2001a).21. For a brief review of this topic, see Stokes (2001a, pp. 4–5).22. Stokes (2001a, pp. 93–97).

incumbent’s commitment to achieving highly desirable economic out-comes at the expense of more immediate partisan support.23 According toStokes, deviating from campaign promises does carry electoral costs,although only weakly.24 However, since her estimates control for eco-nomic outcomes, this result implies that policy switches may still have apositive electoral payoff if the new policies bring substantial economicimprovement. Neoliberalism by surprise may still be a good politicalstrategy.25

A common theme in the literature on economic voting is the conditionalnature of voters’ responses to economic outcomes and policies. As men-tioned, the severity of their judgment depends on their attachment to theparty in power, the structure of the party system, and other institutionalconsiderations. It also depends, although weakly, on whether the policiesadopted by the incumbent are in line with his campaign pronouncements.An additional variation on this theme holds that the electorate is better pre-pared to support untested policies, even if they may cause short-term duressor if they run counter to established beliefs, when economic conditions havedeteriorated.26 However, once conditions improve or simply stabilize, tol-erance subsides and support for further reforms wanes. Therefore, whileuncertainty is welcome at the outset of the reform process, certainty is thekey factor for its consolidation. Based on case studies of Peru and Argentina,Kurt Weyland offers persuasive evidence that the public was supportive tothe reform process while there was a perception of acute economic crisis.27

Even though the reformers were reelected, support for their economic pro-grams was already diminishing. Corrales endorses this view in his analysisof the reform process in Argentina and Venezuela, although he acknowl-edges that in the latter case support for reform was never very strong.28

Therefore, the literature on economic voting suggests that policies, notonly outcomes, may influence electoral decisions. As with outcomes, voters’position with respect to policies may be mediated by a host of factors,including ideological considerations, policy pronouncements during the


23. For a theoretical approach, see Cukierman and Tommasi (1998).24. Stokes (2001a, p. 95).25. Cukierman and Tommasi (1998); Navia and Velasco (2003).26. This behavioral hypothesis is based on seminal work by Thaler and others (1997),

Kahneman and Tversky (1979), and Tversky and Kahneman (1991), who find that people aremore prone, even eager, to assume risks after experiencing losses.

27. Weyland (2002).28. Corrales (2002).

electoral campaign, and the state of the economy at the time of elections.This gives rise to the following additional testable propositions:

3. Electoral support for the incumbent’s party depends on the economicpolicies adopted. Policies may carry electoral costs even when they delivergood economic outcomes.

4. The electorate’s tolerance of unpopular policies depends on the ide-ology of the incumbent’s party, his or her campaign statements, and theinitial state of the economy.

Empirical Approach

None of the empirical literature just reviewed offers a full-fledged theo-retical model of electoral behavior, and we have no intention of providingone. However, the series of hypotheses arising from that literature can beorganized in a simple framework such that the persistence of the vote forthe incumbent’s party is a function of a vector of economic outcomes anda vector of policies (both relative to their past values):

where Vt and Vt −1 are the share of the vote for the incumbent’s party at theend and the beginning, respectively, of its term in office; Xt and Xt −1

are the economic outcomes at the time of each election; and Pt and Pt −1

are the policies at those two moments. A is the set of other parameters thatmay influence the stability of the vote for the party in office, and ut is anerror term. β and γ are our parameters of interest. In this simple frame-work, hypothesis 1 states that β is positive for economic outcomes that aredesirable, such as growth, or negative for undesirable ones, such as infla-tion or unemployment (and assumes that γ is zero, since it ignores theinfluence of policies). Hypothesis 2 postulates that β is a function of somefeatures of the political system, such as party fragmentation or the ideo-logical polarization of the party system. The stronger these features, thehigher the electorate’s response to the economic outcomes. Hypothesis 3,which postulates that the electorate cares about the choice of policies,implies that γ is not zero but probably negative if the policies are marketoriented. Finally, hypothesis 4 states that some aspects of the politicaland economic context when the incumbent’s party was initially electedmay affect the way the electorate judges the adoption of policies. This

V

VA

X

X

P

Put

t

t

t

t

t

t

− − −

= ∗

∗

∗1 1 1

β γ

,


hypothesis can be incorporated in our framework by assuming that γ is afunction of those factors. More specifically, γ will be smaller (in absolutevalue) when the policies adopted were those announced by the incumbentduring his election campaign, when they are in line with his party’s ideol-ogy, or when the economy started from a situation of crisis. Although ourframework is general enough to test further hypotheses, due to sample sizelimitations and for the sake of parsimony and tractability, we restrict itsapplication to the hypotheses identified in the literature review.

Our economic voting framework is relevant both for presidential andfor legislative elections. An important feature of presidential systems is theseparation of powers between the legislative and the executive, aimed atimposing checks and balances in order to discipline parties and make themaccountable.29 Since checks and balances force the two powers to agree onpolicies, voters should be expected to pass judgment on the performanceof the incumbent’s party in both branches on the basis of economic out-comes and policy decisions. Of course, we should expect that the influenceof each policy on presidential vis-à-vis legislative elections will depend onwhether such policy is controlled exclusively by the executive or not. Whilelegislatures have very little influence on monetary, exchange rate, and tariffpolicies in most Latin American countries, they do have a strong (even over-riding) influence on tax policies, privatization decisions, and the regulationof financial, capital, and labor markets. As Brian Crisp and Gregg Johnsonshow, contrary to widespread belief, Latin American legislatures make useof their powers to influence the timing and depth of promarket reforms.30

And according to Roberts and Wibbels, the electorate holds each branch ofpower more accountable for some outcomes than for others.31 When assess-ing the role of the legislature in policy decisions in Latin America, it is impor-tant to keep in mind that the incumbent’s party (or the coalition of partiesbacking the incumbent) usually holds the majority in that body (see below).

To estimate the relevant parameters, the previous expression can bewritten in logs as

where d log(Vt) corresponds to the change in (the log of) the share of votesfor the incumbent party between t, the time when its performance is eval-

d V F d X d pt t t tlog log log log ,( ) = + ( ) + ∗ ( ) + ∗ ( ) +α ψ β γ ε


29. Persson, Roland, and Tabellini (1997).30. Crisp and Johnson (2003).31. Roberts and Wibbels (1999).

uated, and t − 1, when it was elected for office; d log(Xt) and d log(Pt) arethe changes in (log measures of the) outcomes and policies, respectively;εt is equivalent to log(ut); and α + ψ log(F ) is equal to log(A), with α as aconstant parameter and F as a set of political control variables.

We estimate separate models for presidential and legislative electionswith panel data for seventeen countries starting from the mid-1980sdescribed below. Potential problems of heteroscedasticity and endogene-ity need to be addressed in this type of specification. The former may arisefrom country or party heterogeneity and is dealt with by the use of Whiterobust standard errors. The endogeneity problem stems from potential omittedvariables, since differentiating countries solely by the economic and policy-related variables included in sets X and P may not capture all the sources ofheterogeneity.32 This is partly dealt with by the inclusion as controls of a setof political variables (represented by F). However, other country-relatedfactors might bias the estimations if they are correlated with the explanatoryvariables. To take care of this problem, we run all the regressions with coun-try fixed effects (although, admittedly, our sample size is too small to getprecise estimation of these effects).33 The fixed effects estimator is

where C is the set of country dummies.

Data and Sources

Table 2 presents the structure of our database, and table 3 shows correlationsbetween the more relevant variables. The database includes a total of sixty-six presidential elections and eighty-one legislative elections in seventeen

d V F d X d p Ct t t t tlog log log log ,( ) = + ( ) + ∗ ( ) + ∗ ( ) + +α ψ β γ λ ε


32. We assume that the two other sources of endogeneity—reverse causality and measure-ment error—are not latent in our model. Reverse causality is not a concern, since votersevaluate the incumbent’s behavior after policies and outcomes are known. Measurement errorproblems may be present, depending on the actual process of expectations formation. However,ample empirical evidence provides support for the hypotheses of retrospective voting, whichfor our framework implies that expectations are formed on the basis of past outcomes only.

33. All the regressions were also run without fixed effects: while virtually all the conclu-sions are the same, in these regressions, some of the explanatory variables (especially thosemeasuring promarket policies) show higher levels of significance. We have also run the re-gressions including a common time trend, or including five-year period fixed effects, with-out any important divergence from the results presented below. Results are available uponrequest from the authors.

TA

BL

E 2

.St

ruct

ure

of th

e Da

ta S

et

Num

ber o

f elec

tions

(per

iod)

Num

ber o

f par

ties t

hat h

eld .

..Fr

agm

enta

tiona

Coun

tryPr

esid

entia

lLe

gisla

tive

Pres

iden

cyLa

rges

t sha

re in

the l

egisl

atur

eM

ean

Min

imum

Max

imum

Pola

rizat

ion

inde

x

Arge

ntin

a3

(198

9–99

)8

(198

5–99

)2

32.

772.

303.

060.

23Bo

livia

4 (1

985–

97)

4 (1

985–

97)

33

4.06

3.42

5.08

0.52

Braz

il3

(198

9–98

)4

(198

6–98

)2

26.

602.

768.

270.

25Ch

ileb

3 (1

989–

99)

4 (1

989–

2001

)2

24.

904.

844.

990.

16Co

lom

bia

5 (1

986–

02)

5 (1

986–

98)

21

2.66

2.21

3.09

0.16

Cost

a Rica

5 (1

986–

02)

5 (1

986–

2002

)2

22.

312.

212.

560.

42Do

min

ican

Repu

blic

5 (1

986–

2000

)4

(198

6–20

00)

33

2.48

2.18

2.88

0.55

Ecua

dor

4 (1

988–

98)

7 (1

986–

98)

42

6.05

4.29

7.56

0.36

El Sa

lvad

or4

(198

4–99

)6

(198

5–20

00)

22

2.68

2.41

3.06

0.39

Guat

emal

a4

(198

5–99

)5

(198

5–97

)4

43.

312.

354.

440.

24Ho

ndur

as5

(198

5–20

01)

5 (1

985–

2001

)2

22.

182.

002.

580.

42M

exico

3 (1

988–

2000

)6

(198

5–20

00)

21

2.38

1.85

2.82

0.32

Nica

ragu

a3

(199

0–20

01)

3 (1

990–

2001

)1

22.

052.

052.

050.

58Pe

ru4

(198

5–20

00)

4 (1

985–

2000

)3

43.

802.

505.

830.

51Pa

ragu

ay4

(198

9–20

03)

4 (1

989–

2003

)1

12.

211.

882.

540.

40Ur

ugua

y3

(198

4–99

)3

(198

4–99

)2

23.

192.

923.

320.

42Ve

nezu

ela

4 (1

988–

2000

)4

(198

8–20

00)

41

3.92

2.34

5.79

0.30

Tota

l or a

vera

ge66

812.

42.

23.

392.

624.

110.

37

Sour

ce: P

ayne

and

othe

rs (2

002)

, com

plem

ente

d w

ith th

e Pol

itica

l Dat

abas

e of t

he A

mer

icas (

Orga

niza

tion

of A

mer

ican

Stat

es an

d Ge

orge

tow

n Un

iver

sity)

.a.

Effe

ctiv

e num

ber o

f par

ties i

n th

e leg

islat

ure.

b. In

Chi

le, t

he ef

fect

ive n

umbe

r of p

artie

s diff

ers f

rom

the n

umbe

r of c

oalit

ions

(Con

certa

ción

and

Alia

nza p

or C

hile

), w

hich

are c

lose

to 2

in ef

fect

ive t

erm

s and

of w

hich

onl

y Con

certa

ción

has h

eld

the p

resid

ency

.

TA

BL

E 3

.Co

rrel

atio

ns

Infla

tion

(loss

In

stitu

tiona

l Vo

tes

Grow

th

of p

urch

asin

g Un

empl

oym

ent

Gini

inde

xM

acro

inde

x St

ruct

ural

inde

x in

dex

(sha

re)

Frag

men

tatio

nPo

lariz

atio

nPr

omise

sId

eolo

gy(lo

g, ch

ange

)po

wer

, cha

nge)

(cha

nge)

(cha

nge)

(log,

chan

ge)

(log,

chan

ge)

(log,

chan

ge)

Pres

iden

tial e

lectio

nsVo

tes (

shar

e)1.

00Fr

agm

enta

tion

−0.3

21.

00Po

lariz

atio

n−0

.20

−0.0

91.

00Pr

omise

s−0

.16

−0.0

5−0

.32

1.00

Ideo

logy

−0.1

60.

03−0

.45

0.37

1.00

Grow

th (l

og, c

hang

e)0.

21−0

.02

−0.0

60.

250.

141.

00In

flatio

n (lo

ss o

f pu

rcha

sing

pow

er, c

hang

e)−0

.24

−0.5

50.

330.

02−0

.16

−0.4

11.

00Un

empl

oym

ent (

chan

ge)

−0.0

20.

13−0

.46

−0.1

60.

28−0

.27

−0.4

81.

00Gi

ni in

dex (

chan

ge)

−0.2

2−0

.16

−0.1

90.

210.

090.

230.

13−0

.09

1.00

Mac

ro in

dex

(log,

chan

ge)

0.14

−0.2

5−0

.24

0.27

0.11

0.25

0.12

−0.2

70.

001.

00St

ruct

ural

inde

x (lo

g, ch

ange

)−0

.24

−0.0

5−0

.27

0.08

−0.0

50.

24−0

.26

0.33

0.37

−0.0

21.

00In

stitu

tiona

l Ind

ex

(log,

chan

ge)

−0.0

7−0

.18

0.09

−0.1

50.

000.

01−0

.10

0.24

−0.1

5−0

.03

0.44

1.00

(con

tinue

d)

TA

BL

E 3

.Co

rrel

atio

ns(c

ontin

ued)

Infla

tion

(loss

In

stitu

tiona

l Vo

tes

Grow

th

of p

urch

asin

g Un

empl

oym

ent

Gini

inde

xM

acro

inde

x St

ruct

ural

inde

x in

dex

(sha

re)

Frag

men

tatio

nPo

lariz

atio

nPr

omise

sId

eolo

gy(lo

g, ch

ange

)po

wer

, cha

nge)

(cha

nge)

(cha

nge)

(log,

chan

ge)

(log,

chan

ge)

(log,

chan

ge)

Legi

slativ

e elec

tions

Vote

s (sh

are)

1.00

Frag

men

tatio

n0.

111.

00Po

lariz

atio

n−0

.43

−0.1

61.

00Pr

omise

s−0

.19

0.32

−0.4

01.

00Id

eolo

gy0.

09−0

.20

−0.3

80.

361.

00Gr

owth

(log

, cha

nge)

0.19

0.33

−0.0

30.

23−0

.05

1.00

Infla

tion

(loss

of

purc

hasin

g po

wer

, cha

nge)

−0.1

9−0

.38

0.34

−0.3

2−0

.24

−0.3

01.

00Un

empl

oym

ent

(cha

nge)

0.09

−0.1

8−0

.47

0.04

0.43

−0.3

8−0

.34

1.00

Gini

inde

x (c

hang

e)−0

.29

−0.1

7−0

.24

0.14

0.07

−0.0

10.

170.

181.

00M

acro

inde

x (lo

g, ch

ange

)0.

080.

31−0

.13

0.26

0.08

0.83

−0.6

3−0

.10

−0.0

81.

00St

ruct

ural

inde

x (lo

g, ch

ange

)−0

.61

−0.0

7−0

.10

0.15

−0.3

4−0

.08

0.20

0.01

0.51

−0.1

51.

00In

stitu

tiona

l Ind

ex

(log,

chan

ge)

−0.2

60.

120.

29−0

.20

−0.2

1−0

.15

0.23

−0.4

7−0

.20

−0.2

90.

321.

00

Sour

ce:

Auth

ors’

calcu

latio

ns.

Latin American countries over the period 1985–2002. Party alternationwas moderate over this period: the average number of parties that heldpower or had a majority was 2.4 for presidential elections and 2.2 for legis-lative elections (with a maximum of 4 and a minimum of 1). However, theeffective number of parties (also known as political fragmentation) washigher: 3.4 on average, with a maximum of 8.3 in Brazil and 7.6 inEcuador.34 Except for Mexico during the 1980s and Paraguay at the end onthat decade, none of the seventeen countries showed party fragmentationbelow 2, implying a generally healthy level of political competition. Theideological polarization of the political system was low during the period,as measured by a polarization index that computes the (weighted average)distance between the ideological positions of the parties on a scale from0 to 1; parties are classified as extreme left, center left, center right, orextreme right. When all the parties have the same ideological position, theindex takes the value 0, when half of them (measured by the number ofvotes) are extreme left and the other half are extreme right, the index takesthe value 1.35 The average value of the index in our data set was 0.37, witha maximum of 0.58 for Nicaragua and a minimum of 0.16 for Chile andColombia.

Dependent Variable

Our dependent variable is the change in the share (in logs) of votes36 forthe incumbent’s party in presidential elections, and for the majority party


34. The effective number of parties is calculated using the Laako-Taagepera index, de-fined as the inverse of the sum of the squares of the shares (measured by the number of seats)of all the parties in the legislature; Payne and others (2002).

35. More precisely, the index is calculated in two steps. First, the average position ofthe electorate on a left-right scale (APLR) is calculated as a weighted average of the partypositions on a scale from −1 to +1, where the weights are the shares of the votes:

APLR = −1*(% votes obtained by parties on the extreme left) − 0.5*(% votes for partieson the center left) + 0.5*(% votes for parties on the center right) + 1*(% votes for parties onthe extreme right).

In the second step, the polarization index (IP) is calculated as a weighed deviation fromthe APLR:

IP = −1 − APLR * (% votes left) + −0.5 − APLR * (% votes center left) + 0.5 −APLR * (% votes center right) + 1 − APLR *(% votes right).

A minimum of 0 is reached when all the votes are in one ideological bloc; and a max-imum of 1, when half of the votes are at each extreme. Ideological orientations are takenfrom Coppedge (1997) and the World Bank’s Database of Political Institutions, 2002(www.worldbank.org/research/bios/pkeefer.htm).

36. The share of votes comes from Payne and others (2002).

in the legislature in legislative elections. Since we use logs for both thedependent and (when possible) the independent variables, the estimatedcoefficients can be interpreted as elasticities.

Some calculations were necessary in order to compute the share of votes,especially for presidential elections, when party coalitions or party dissolu-tions had taken place before and after the elections, as well as to be able toaccount for new independent parties. These calculations treat coalitions asregular parties. The vote for the coalition party in the election previous to itscreation is simply computed as the sum of the votes of the joining parties.When parties break up, the same procedure is used for the following elec-tions. Table 4, which presents summary statistics for the most importantvariables, shows that the share of votes varies from 0 to 64 percent forpresidential elections and 62 percent for legislative elections, with meansof 35 percent and 36 percent, respectively.

Political Variables

The political variables used as independent variables attempt to measurekey dimensions of the party system and the political environment. Follow-ing the literature review, they are to be included in the regressions both asindependent controls and/or interacted with the variables measuring eco-nomic outcomes. Fragmentation (or the effective number of parties) andpolarization, already described, are the two basic dimensions of the partysystem. In addition, we use a dummy for divided governments (when thepresident’s party is not the largest party in the legislature).37

We also use several variables intended to measure the electorate’s expec-tations about the future orientation of economic policies. The first, named“promarket promises,” measures to what extent the positions adopted byincumbents during their election campaigns were promarket; it is a re-scaled version of a variable computed by Stokes.38 The second, named“right-oriented ideology,” a measure taken from the World Bank’s Database


37. Taken from Payne and others (2002). Divided government is not frequent in LatinAmerica, in contrast to the United States; Alesina, Londregan, and Rosenthal (1993);Alesina and Rosenthal (1995, 1996); Fiorina (1992). The only cases in our data set aremainly concentrated in Brazil and Ecuador (six), with one in the Dominican Republic. Morerecently, the PRI lost its monopoly power in Mexico.

38. Based on an ordinal variable computed by Stokes (2001a, p. 3) that classifies forty pres-idential pre-electoral campaigns according to the importance assigned by the candidates toissues of economic security vis-à-vis economic efficiency, the promises variable takes valueson a scale from 0 to 1, where higher values indicate more efficiency-oriented messages.

TA

BL

E 4

.Su

mm

ary

Stat

isti

cs

Pres

iden

tial e

lectio

nsLe

gisla

tive e

lectio

ns

Num

ber o

f St

anda

rd

Num

ber o

fSt

anda

rd

Varia

ble

obse

rvat

ions

Mea

nM

edia

nde

viatio

nM

inim

umM

axim

umob

serv

atio

nsM

ean

Med

ian

devia

tion

Min

imum

Max

imum

Depe

nden

tVo

tes (

shar

es)

480.

350.

360.

170.

000.

6472

0.36

0.38

0.14

0.00

0.62

Polit

ical

Frag

men

tatio

n52

3.33

2.70

1.62

1.85

8.27

723.

372.

801.

541.

858.

27Po

lariz

atio

n (0

–1)

510.

380.

330.

220.

030.

9267

0.34

0.30

0.19

0.03

0.91

Divi

ded

gove

rnm

ent (

dum

my)

470.

130.

000.

340.

001.

0072

0.01

0.00

0.12

0.00

1.00

Rule

of l

aw16

−0.2

3−0

.44

0.50

−0.8

11.

1916

−0.2

3−0

.44

0.50

−0.8

11.

19Pr

omise

s (0–

1)33

0.55

0.50

0.26

0.25

1.00

250.

560.

500.

270.

251.

00Id

eolo

gy (0

–1)

420.

600.

500.

330.

001.

0068

0.55

0.33

0.32

0.00

1.00

Switc

h in

dex w

ith id

eolo

gy (−

1, +

1)39

0.00

0.00

0.08

−0.1

10.

4839

0.00

0.00

0.08

−0.1

10.

48Sw

itch

inde

x with

pro

mise

s (−1

, +1)

320.

050.

010.

12−0

.06

0.24

Outc

ome

Infla

tion

(log,

chan

ge)

49−0

.09

−0.0

30.

24−0

.74

0.32

71−0

.07

−0.0

40.

20−0

.71

0.40

Grow

th (l

og, c

hang

e)52

−0.0

1−0

.01

0.08

−0.2

50.

1472

−0.0

2−0

.02

0.06

−0.1

90.

15Cr

isis

520.

020.

010.

040.

000.

1572

−0.0

20.

000.

03−0

.14

0.00

Unem

ploy

men

t (lo

g, ch

ange

)44

0.00

0.00

0.03

−0.0

80.

0963

0.00

0.00

0.03

−0.0

60.

07Gi

ni in

dex (

log,

chan

ge)

49−0

.01

−0.0

10.

03−0

.10

0.10

690.

00−0

.01

0.03

−0.1

00.

10W

ashi

ngto

n Co

nsen

sus

Mac

ro in

dex (

log,

chan

ge)

510.

120.

060.

40−1

.00

1.70

700.

080.

040.

34−0

.99

1.70

Stru

ctur

al fi

scal

bal

ance

(rat

io to

GDP

, cha

nge)

440.

000.

000.

04−0

.07

0.12

630.

000.

000.

03−0

.07

0.11

Real

exch

ange

rate

(det

rend

ed in

logs

, cha

nge)

510.

00−0

.04

0.26

−0.6

10.

6770

−0.0

1−0

.01

0.22

−0.6

10.

67So

cial e

xpen

ditu

res (

shar

e of G

DP, c

hang

e)50

0.01

0.00

0.02

−0.0

20.

0569

0.01

0.00

0.02

−0.0

30.

06St

ruct

ural

refo

rms i

ndex

Stru

ctur

al re

form

inde

x (lo

g, ch

ange

)49

0.23

0.19

0.21

−0.0

90.

7868

0.17

0.12

0.19

−0.0

90.

78Tr

ade i

ndex

(log

, cha

nge)

490.

180.

060.

30−0

.32

1.15

680.

150.

040.

29−0

.24

1.15

Fina

ncia

l ind

ex (l

og, c

hang

e)51

0.30

0.03

0.44

−0.1

51.

3870

0.23

0.03

0.35

−0.1

51.

38Ta

x ind

ex (l

og, c

hang

e)52

0.10

0.06

0.23

−0.2

90.

8172

0.10

0.03

0.23

−0.2

90.

71Pr

ivat

izatio

ns in

dex (

chan

ge)

520.

090.

000.

17−0

.03

0.97

720.

060.

000.

15−0

.01

0.97

Inst

itutio

nal i

ndex

510.

270.

190.

35−0

.22

1.26

710.

190.

100.

28−0

.51

0.93

of Political Institutions and a study by Coppedge, classifies parties on a leftto right scale according to their economic ideology.39 In order to test theStokes hypothesis on the electoral effects of switching, we have createdtwo types of switch indexes, one measuring the deviation between the amountof promarket reforms implemented by the administration (see below for thedescription of this variable) and the promarket promises during the cam-paign, and the other measuring the deviation between the reforms and themeasure of right-oriented ideology of the party.40 Note that only the latteris applicable to legislative elections.

Economic Outcomes

Following the empirical literature on economic voting, we focus on inflationand growth as the two main economic outcomes of interest, but we also testother variables, such as unemployment and income concentration. We mea-sure inflation as the average annual loss of purchasing power of a currencyunit, rather than as the increase in the price index, since this reduces theextreme observation problem that arises with the cases of high or hyper-inflation. We apply the formula 1 − (1/(1 + π)), where π is the price increaseduring the last year of the administration. Economic growth is measuredas the rate of annual change (in logs) in GDP. In addition to inflation andgrowth, we test for the influence of two other outcomes: the unemploymentrate and the Gini coefficient of distribution of per capita household income.41

Policy Variables

As mentioned in the introduction, we define the Washington Consensus inaccordance with the list of policies included in Williamson.42 Since thosepolicies cover a variety of areas, from fiscal to institutional, we use the


39. World Bank, Database of Political Institutions, 2002 (www.worldbank.org/research/bios/pkeefer.htm); Coppedge (1997).

40. The switch indexes range from −1 to +1. In the first type, −1 indicates that havingadopted the most pro-efficiency stance during the campaign, the candidate does not imple-ment any promarket reform once in office; +1 indicates that having adopted the most pro-security position in the campaign, the candidate once in office becomes the most aggressivepromarket reformer. The formula is then SI = [change in reforms − (PROMISES − medianPROMISES)]. In the other type of switch index, the variable PROMISES is replaced by ourmeasure of party ideology.

41. Prices and GDP are taken from International Monetary Fund, World Economic Out-look (online). Unemployment is from ECLAC (various years). Gini coefficients for incomesare from Deininger and Squire (1998).

42. Williamson (1990a).

following—admittedly somewhat arbitrary—classification (numbers inparentheses refer to Williamson’s list):

Macroeconomic policies: fiscal discipline (1), public expenditure onsocial services and infrastructure (2), and competitive exchange rates (5).

Structural reforms: tax reform (flat, low, and effective tax rates) (3),interest rate liberalization (4), trade liberalization (6), liberalization offoreign direct investment inflows (7), and privatization (8).

Institutional reforms: deregulation of entry and exit (9) and protectionof property rights (10).

The most important distinction is that between macroeconomic poli-cies and structural reforms, the latter referring to sectoral or microeconomicpolicies that affect the functioning of specific markets (imports, credit,infrastructure services, and so forth). The inclusion of public expenditureon social services and infrastructure as a macroeconomic policy is arbi-trary but justifiable for the sake of simplicity. Institutional reforms includeprotection of property rights, a policy that is usually seen not as a core ele-ment of the Washington Consensus (as a matter of fact, it was added byWilliamson as an afterthought) but rather as a key element of what ana-lysts starting with Moisés Naim have referred to as second generationreforms.43 However, these also include regulatory institutions, moderniza-tion of the state apparatus (especially for the provision of social services),and reform of the judiciary sector, none of which are considered here.

We use quantitative indicators to measure eight of the ten policies thatconstitute the Washington Consensus, as well as composite indexes formacroeconomic policies and structural reforms. We do not have quantita-tive indicators for foreign direct investment policies or deregulation ofentry and exit. Therefore, these policies are not included in our reformindexes. A brief description of the policy indicators follows (furtherdetails are in the footnotes):

—Fiscal discipline is measured by the fiscal balance of the centralgovernment, adjusted by the endogenous influence of the economic cycleand changes in the terms of trade on fiscal revenues. The purpose ofthese adjustments is to isolate the exogenous or policy component of thefiscal balance, which is a better measure of fiscal discipline than theobserved fiscal balance.44 Fiscal balance, fiscal revenue, and GDP data


43. Naím (1994).44. Specifically, we subtract from the fiscal balance of the central government the rev-

enue that is associated with either the economic cycle or the terms of trade cycle (obtainedapplying standard Hodrick-Prescott filters).

used in this calculation are from the World Bank and terms of trade dataare from the Economic Commission for Latin America and the Caribbean(ECLAC).45

—Public expenditure in social services includes only education andhealth expenditures, based on data from ECLAC and complemented withdata from World Bank.46

—The measure of competitive exchange rates is the log distance betweenthe observed real exchange rate and its trend, computed with a standardHodrick-Prescott filter.47

—Tax reform is taken from previous work by Lora, who constructs acomposite index of the levels and effectiveness of corporate, personal, andvalue added taxes.48

—Interest rate liberalization is measured by Lora’s index of financialliberalization, which includes information on interest rate freedom, reserverequirements, and quality of regulation and supervision of the financialsector.

—Trade liberalization is also taken from Lora, who uses an index thatcombines import tariff averages and dispersion.

—Privatization is measured by Lora’s index of the cumulated value ofthe sales of state-owned firms to the private sector, as a share of the GDP.

—Protection of property rights is a combined measure of the risk ofexpropriation and the risk of repudiation of government contracts, on ascale from 0 to 1 (the higher the index, the lower the risk).49

—The composite index of macroeconomic policies is a simple averageof the indicators of its three components scaled from 0 to 1, where 0 cor-responds to the lowest observation and 1 to the highest observation for thewhole period and set of countries in the sample.

—The composite index for structural reforms is calculated as the sim-ple average of the indexes for tax reform, financial liberalization, trade lib-eralization, and privatization (each of which is also calculated on a scalefrom 0 to 1).50


45. World Bank, World Development Indicators (online); ECLAC (various years).46. World Bank, World Development Indicators (online); ECLAC (various years).47. Real exchange rate data are from IMF, World Economic Outlook (online).48. Tax reform, interest rate liberalization, and privatization are all from Lora (2001).49. Taken from the International Country Risk Guide, 2004 (www.icrgonline.com).50. Note that this composite index is not identical to the total reform index computed

by Lora (2001), since the latter includes labor reform, which is not among the WashingtonConsensus policies.

All variables are measured as changes between the previous electionyear and the current election year. Since taking the current election year issomewhat arbitrary, we checked the robustness of our main results by alsousing the year prior to the election year.51

Econometric Results

Before discussing the hypotheses in detail, it is helpful to convey the thrustof our findings. The regression summarized in table 5 indicates that theelectorate is highly sensitive to one economic outcome—inflation—andstrongly rejects the adoption of promarket policies. Our estimates implythat the typical reduction in the rate of inflation, from say 20 percent to 8 percent during a president’s tenure, boosts the vote for his party by 21 percent.52 However, if that same incumbent also introduces the averageamount of promarket reform, the resultant party losses account for 23 per-cent of the vote. Put a different way, the adoption of the standard Wash-ington Consensus package brings positive electoral payoffs only whenimplemented in a period of high inflation. Thus, if the same dose of pro-market reform is adopted as part of a package that reduces inflation from100 percent to 8 percent, the net electoral effect is a handsome 82 percentincrease in vote share.

Admittedly, our basic regression overstates the negative effect of thepromarket policies because those policies may help to reduce the rate ofinflation and increase the rate of growth.53 Taken to the extreme, this argu-ment would imply that the total effect of the adoption of promarket poli-cies would be the sum of the direct effect captured in the coefficient of theregression in table 5 and the indirect effects of the changes in the rates ofinflation and growth. Based on this calculation (see tables 6A and 6B), thetotal effect does appear to be substantially milder: −0.97 instead of −1.57.Nevertheless, it would still be substantial, as it would imply that the typi-


51. These results, which are not included in this version of the paper, are available uponrequest from the authors.

52. This reduction corresponds to the average value of our measure of the change ofinflation.

53. The regression includes several other control variables that may also affect the vote(see notes to table 5).


T A B L E 5 . Impact of Economic Outcomes and Washington Consensus Policies in Presidential Elections, 1985–2002: Country Fixed Effects Results

Independent variablesa Dependent variable: Change in vote shareb

Economic outcomesInflation (change in loss of purchasing power) −2.030 (2.09)*Growth (change in growth rate, log) −1.016 (0.74)Washington Consensus reformsStructural reforms index (log, change) −1.569 (2.98)***Constant 0.627 (0.85)Summary statisticNumber of observations 37Number of countries 17R2 0.80Country fixed effects Yes

Source: Authors’ calculations.*Significant at 10 percent; ***significant at 1 percent.a. The regression also includes as control variables measures of divided government, polarization, and fragmentation (see text for

definitions and method of calculation).b. The dependent variable is the change in the log of the vote share of the incumbent president’s party. Robust t statistics are in

parentheses.

cal reformist government must still sacrifice 15 percent of the vote for thesake of the reforms.54 However, this calculation most likely overestimatesthe effects of the reforms on growth and inflation, as we have not isolatedthe influence of other factors on these two variables. Therefore, the centralconclusion is that even if we grant that promarket reforms have strong benefi-cial effects on growth and inflation, their electoral cost is far from negligible.

Apart from promarket reforms, the other Washington Consensus poli-cies do not affect the electorate’s behavior. Likewise, we find no robustevidence that any economic outcomes other than inflation affect the votein presidential elections. We do find that these results are affected by somefeatures of the political system. In legislative elections the results are lessstraightforward, as they are strongly mediated by several contextual andpolitical variables.

Do Outcomes Matter?

We start our empirical analysis by testing the simplest version of the eco-nomic voting model, in which voters update their opinion on the incum-bent’s party based entirely on the changes observed since the last election

54. Note that the total effect would be reduced only slightly (to −0.84) if the indirecteffect through growth, which has the wrong sign, is not included.

in the key economic variables, Xt. As mentioned, we include as additionalcontrols a set of political variables (represented by F below) that may affectthe stability of the vote share, namely, our measures of political fragmenta-tion, polarization, and divided government (lagged to reduce endogeneityand better capture the political environment prevailing during the adminis-tration).55 Since other country-specific factors may also have an influence onthe persistence of the vote for the incumbent’s party, we attempt to isolatethem by using fixed effects. We start with

d V F d X Ct t t tlog log log .( ) = + ( ) + ∗ ( ) + +−α ψ β λ ε1


55. Divided government is a dummy equal to 1 when the party with the greatest repre-sentation in the legislature is not the incumbent’s party (this is unusual in Latin America:in our database, it occurs in only seven instances). In regressions not shown, a dummy formidterm elections was also included in legislative elections. It was never significant and itdid not affect any of the results.

T A B L E 6 A . Rough Estimate of Total Effect of Promarket Reforms on the Presidential Vote (Elasticities)

Effect Inflation Growth Total

Effect of reforms on inflation or growtha −0.361 0.133Effect of inflation or growth on the vote −2.030 −1.016Indirect effect of reforms on the vote (via inflation or growth) 0.733 −0.135 0.598Direct effectb −1.569Total effect (indirect + direct) −0.971

Source: Authors’ calculations.a. See table 6B for regressions.b. As estimated in table 5.

T A B L E 6 B . Estimates for Inflation and Growth Used in Table 6A

Inflationa (1) Growthb (2)

Structural reforms index (log, change) −0.361 (1.88)* 0.133 (2.10)**Constant −0.103 (0.48) −0.063 (0.78)Summary statisticNumber of observations 49 49Number of countries 17 17R2 0.37 0.26Country fixed effects Yes Yes

Source: Authors’ calculations.*Significant at 10 percent; **significant at 5 percent.a. Dependent variable is the independent variable from table 5; that is, the change in the inflation rate, where inflation is measured

as the annual loss of purchasing power of the currency. Robust t statistics are in parentheses.b. Dependent variable is the independent variable from table 5; that is, the change in the growth rate, in logs. Robust t statistics are

in parentheses.

TA

BL

E 7

.Im

pact

of E

cono

mic

Out

com

es in

Pre

side

ntia

l Ele

ctio

ns, 1

985–

2002

: Cou

ntry

Fix

ed E

ffec

ts R

esul

tsa

Depe

nden

t var

iabl

e: Ch

ange

in lo

g of

vote

shar

e of i

ncum

bent

pre

siden

t’s p

arty

Inde

pend

ent v

aria

ble

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Polit

ical

Frag

men

tatio

n (la

gged

)0.

588

0.58

80.

121

0.58

90.

669

−0.3

64−0

.546

0.02

3(1

.31)

(1.1

9)(0

.18)

(1.1

4)(1

.19)

(0.4

5)(0

.59)

(0.0

4)Po

lariz

atio

n (la

gged

)0.

938

0.93

90.

700

0.93

81.

192

3.63

63.

150

0.50

9(0

.80)

(0.7

9)(0

.73)

(0.7

8)(0

.70)

(1.1

1)(1

.44)

(0.6

4)Di

vide

d go

vern

men

t (du

mm

y, la

gged

)−0

.345

−0.4

42−0

.539

−0.4

51−0

.537

−0.5

840.

629

0.17

1−0

.269

(1.0

5)(0

.82)

(1.1

1)(0

.86)

(1.0

9)(1

.05)

(1.9

8)*

(0.4

8)(0

.76)

Econ

omic

outc

omes

Infla

tion

(cha

nge i

n lo

ss o

f pur

chas

ing

pow

er)

−1.1

27−1

.924

−1.6

74(1

.80)

*(2

.08)

*(1

.91)

*Gr

owth

(cha

nge i

n gr

owth

rate

, log

)−0

.034

−2.6

83−2

.828

(0.0

3)(1

.45)

(1.7

0)Gi

ni in

dex (

chan

ge)

0.62

4−2

.256

(0.2

3)(0

.70)

Unem

ploy

men

t rat

e (ch

ange

)2.

510

−4.3

07(1

.06)

(1.2

4)Co

nsta

nt−0

.804

−0.4

53−1

.028

−0.7

43−1

.030

−1.1

57−0

.824

−0.8

06−0

.786

(1.7

3)*

(1.4

1)(2

.13)

**(1

.30)

(1.9

9)*

(1.8

2)*

(1.0

1)(0

.99)

(1.5

6)Su

mm

ary s

tatis

ticNu

mbe

r of o

bser

vatio

ns43

4343

4343

4137

3743

Num

ber o

f cou

ntrie

s17

1717

1717

1715

1517

R20.

470.

480.

480.

60.

480.

480.

560.

740.

66Co

untry

fixe

d ef

fect

sYe

sYe

sYe

sYe

sYe

sYe

sYe

sYe

sYe

s

Sour

ce:

Auth

ors’

calcu

latio

ns.

*Sig

nific

ant a

t 10

perc

ent;

**sig

nific

ant a

t 5 p

erce

nt.

a. Se

e tex

t for

des

crip

tions

and

met

hod

of co

nstru

ctio

n of

inde

pend

ent v

aria

bles

. Rob

ust t

stat

istics

are i

n pa

rent

hese

s.

Table 7 lends some support to this simple version of the economic votinghypothesis: in all the regressions, changes in inflation have the expected signand have a significant impact on the presidential vote. However, changes ingrowth rates are seldom significant and when included in a regression withinflation show the wrong sign. Results for unemployment and inequalityare similarly weak. When all four economic variables are included in thesame regression, inflation remains the only significant variable. In leg-islative elections (see table 8) growth is the only one that is sometimessignificant—but it is not when all economic variables are included in thesame regression. Therefore, inflation and growth seem to matter for the lead-ing party or parties, but through different channels. The size of the co-efficients suggests that the incumbent loses 1–2 percent of his vote foreach (additional) 1 percent of (annual) loss in the purchasing power of thecurrency in the last year of his administration (with respect to the loss inthe year prior to his administration). Likewise, the largest party in the leg-islature increases its share of seats by about 1 percent for each (additional)1 percent of economic growth in the year before the election (with respectto the year immediately before the previous election). Neither changes inthe unemployment rate nor income distribution changes appear to have aclear effect on electoral behavior.

These conclusions must now be qualified in accordance with our sec-ond hypothesis, namely, that the electorate’s response to the economicoutcomes, β, depends on several features of the political system, F (someof which, as tables 7 and 8 show, also have a direct influence on voters’behavior):

Replacing β in the previous equations gives (with fixed effects)

Note that in the interaction terms we use the values of F at the earli-est period of our sample, F0, in order to reduce endogeneity. However,we use the values of F at the beginning of each electoral cycle, Ft −1, todirectly control for these variables, since the inclusion of country fixed effects

d V F d X

F

t t tlog log log

log

( ) = + ∗ ( ) + ∗ ( )+ ∗ ( )

−α ψ ν

µ

1

0 ∗∗ ( ) + +d X Ct tlog .λ ε

β ν µ= + ∗ ( )log .F0


TA

BL

E 8

.Im

pact

of E

cono

mic

Out

com

es in

Leg

isla

tive

Ele

ctio

ns, 1

985–

2002

: Cou

ntry

Fix

ed E

ffec

ts R

esul

tsa

Depe

nden

t var

iabl

e: Ch

ange

in lo

g of

vote

shar

e of p

arty

with

mos

t sea

ts in

legi

slatu

re

Inde

pend

ent v

aria

ble

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Polit

ical

Frag

men

tatio

n (la

gged

)0.

868

0.89

40.

572

0.50

70.

915

1.05

60.

705

0.52

9(3

.86)

***

(4.0

1)**

*(2

.99)

***

(3.2

6)**

*(4

.23)

***

(4.4

2)**

*(2

.87)

***

(2.8

5)**

*Po

lariz

atio

n (la

gged

)−0

.068

0.24

00.

157

0.12

50.

304

1.11

60.

853

0.12

0(0

.21)

(0.7

0)(0

.53)

(0.4

9)(0

.78)

(2.2

6)**

(2.0

8)**

(0.4

6)Di

vide

d go

vern

men

t (du

mm

y, la

gged

)0.

007

−0.2

790.

014

−0.0

11−0

.052

0.01

70.

029

−0.0

01−0

.051

(0.0

5)(1

.52)

(0.0

9)(0

.08)

(0.4

1)(0

.12)

(0.1

9)(0

.00)

(0.4

0)Ec

onom

ic ou

tcom

esIn

flatio

n (c

hang

e in

loss

of p

urch

asin

g po

wer

)−0

.086

−0.0

70.

053

(0.3

8)(0

.26)

(0.2

3)Gr

owth

(cha

nge i

n gr

owth

rate

, log

)0.

861

−0.0

640.

913

(1.7

7)*

(0.1

0)(1

.64)

Gini

inde

x (ch

ange

)1.

322

−0.0

43(0

.79)

(0.0

3)Un

empl

oym

ent (

chan

ge)

−1.5

29−1

.070

(0.8

6)(0

.59)

Cons

tant

−0.9

110.

204

−0.9

98−0

.649

−0.5

26−1

.027

−1.3

56−0

.937

−0.5

42(3

.33)

***

(1.1

9)(3

.31)

***

(2.7

7)**

*(2

.26)

**(3

.39)

***

(3.6

9)**

*(2

.75)

***

(2.3

3)**

Sum

mar

y sta

tistic

Num

ber o

f obs

erva

tions

7474

7471

7172

6562

71Nu

mbe

r of c

ount

ries

1717

1717

1717

1515

17R2

0.53

0.30

0.53

0.40

0.43

0.55

0.60

0.47

0.43

Coun

try fi

xed

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Sour

ce:

Auth

ors’

calcu

latio

ns.

*Sig

nific

ant a

t 10

perc

ent;

**sig

nific

ant a

t 5 p

erce

nt; *

**sig

nific

ant a

t 1 p

erce

nt.

a. Se

e tex

t for

des

crip

tions

and

met

hod

of co

nstru

ctio

n of

inde

pend

ent v

aria

bles

. Rob

ust t

stat

istics

are i

n pa

rent

hese

s.


precludes the use of time-invariant F0. None of the results reported below issensitive to whether we include the set of F variables as direct controls.

Tables 9 and 10 suggest that the electorate’s response to the economicoutcomes is indeed affected by the structure of the political system, and inthe expected manner. In presidential elections (table 9), the more frag-mented the party system, the more harshly the electorate punishes theincumbent’s party for an increase in the inflation rate.56 The intuitionbehind this result is that in more fragmented party systems there is morecompetition for votes, and probably also more information available tothe voters and a wider choice of policy proposals, all of which enhancethe response of the electorate to changes in the economic situation. Oneshould expect this response to be stronger in presidential than in legisla-tive elections, given the winner-take-all nature of the former. A dividedgovernment affects the response of the electorate to inflation in a similarway. However, due to the small number of cases of divided government,we do not attach much relevance to this result.57 In contrast to party frag-mentation, the degree of polarization does not seem to have any significantinfluence on the electorate’s response to the economic outcomes in presi-dential elections.

In legislative elections (table 10), the opposite is the case: while the inter-action terms between economic outcomes and fragmentation are not signif-icant, the interaction with ideological polarization is significant for inflationand for growth. This implies that the more distanced the economic policyplatforms of the parties, the stronger the swings of the electorate in responseto changes in the macroeconomic outcomes. From regression 5, when thedegree of polarization is high (0.53), each percentage point of extra growthbrings an increase of about 1 percent in the vote for the largest party in thelegislature, while this elasticity becomes negative (−0.4) when the degree ofpolarization is low (0.15). Our results indicate that the legislative vote is alsosensitive to inflation outcomes, depending on the degree of ideologicalpolarization of the party system, with implied elasticities of −0.3 whenpolarization is high and 0.38 when polarization is low (regression 2).

To summarize, our results suggest that economic outcomes do matter inpresidential as well as in legislative elections, though in different ways.The executive is held more accountable for increases in inflation, and

56. However, this result does not hold in a similar regression without fixed effects (resultsavailable from the authors upon request).

57. Furthermore, similar regressions for growth show implausibly high coefficients forthe interaction term (GROWTH*DIVIDED GOVERNMENT).

TA

BL

E 9

.Im

pact

of E

cono

mic

Out

com

es In

tera

cted

wit

h Po

litic

al F

eatu

res i

n Pr

esid

enti

al E

lect

ions

, 198

5–20

02: C

ount

ry F

ixed

Eff

ects

Res

ults

a

Depe

nden

t var

iabl

e: Ch

ange

in lo

g of

vote

shar

e of i

ncum

bent

pre

siden

t’s p

arty

Inde

pend

ent v

aria

ble

(1)

(2)

(3)

(4)

(5)

(6)

Polit

ical

Frag

men

tatio

n (la

gged

)1.

316

−0.1

870.

692

0.73

20.

667

−0.1

41(2

.27)

**(0

.24)

(1.5

3)(1

.15)

(1.3

5)(0

.19)

Pola

rizat

ion

(lagg

ed)

0.42

60.

054

−0.2

051.

002

1.04

1.51

8(0

.77)

(0.0

8)(0

.40)

(0.8

3)(0

.78)

(1.1

1)Di

vide

d go

vern

men

t (du

mm

y, la

gged

)−0

.5−0

.19

−0.3

73−0

.492

−0.6

64−1

.028

(1.4

6)(0

.40)

(0.9

2)(0

.91)

(1.0

0)(1

.59)

Econ

omic

outc

omes

and

inte

ract

ions

Infla

tion

(cha

nge i

n lo

ss o

f pur

chas

ing

pow

er)

3.15

3−3

.288

−0.4

02(2

.66)

**(1

.86)

*(1

.89)

*In

flatio

n*in

itial

frag

men

tatio

n−3

.351

(3.5

0)**

*In

flatio

n*in

itial

pol

ariza

tion

5.03

9(1

.62)

Infla

tion*

divi

ded

gove

rnm

ent

−4.8

68(1

9.97

)***

Grow

th (c

hang

e in

grow

th ra

te, l

og)

−1.7

17−1

.198

1.03

2(0

.48)

(0.4

9)(0

.95)

Grow

th*i

nitia

l fra

gmen

tatio

n1.

751

(0.4

0)Gr

owth

*ini

tial p

olar

izatio

n2.

849

(0.4

0)

Grow

th*d

ivid

ed g

over

nmen

t−3

6.07

6(1

.80)

*Co

nsta

nt−1

.545

−0.4

74−0

.938

−1.1

91−1

.142

−0.4

27(2

.45)

**(0

.52)

(2.1

0)**

(1.9

5)*

(2.0

3)*

(0.6

7)Su

mm

ary s

tatis

ticNu

mbe

r of o

bser

vatio

ns43

4343

4343

43Nu

mbe

r of c

ount

ries

1717

1717

1717

R20.

800.

700.

870.

490.

490.

59Co

untry

fixe

d ef

fect

sYe

sYe

sYe

sYe

sYe

sYe

s

Sour

ce:

Auth

ors’

calcu

latio

ns.

*Sig

nific

ant a

t 10

perc

ent;

**sig

nific

ant a

t 5 p

erce

nt; *

**sig

nific

ant a

t 1 p

erce

nt.

a. Se

e tex

t for

des

crip

tions

and

met

hod

of co

nstru

ctio

n of

inde

pend

ent v

aria

bles

. Rob

ust t

stat

istics

are i

n pa

rent

hese

s.

TA

BL

E 1

0.

Impa

ct o

f Eco

nom

ic O

utco

mes

Inte

ract

ed w

ith

Polit

ical

Fea

ture

s in

Legi

slat

ive

Elec

tion

s, 1

985–

2002

: Cou

ntry

Fix

ed E

ffec

ts R

esul

tsa

Depe

nden

t var

iabl

e: Ch

ange

in lo

g of

vote

shar

e of p

arty

with

mos

t sea

ts in

legi

slatu

re

Inde

pend

ent v

aria

ble

(1)

(2)

(3)

(4)

(5)

(6)

(7)

Polit

ical

Frag

men

tatio

n (la

gged

)0.

606

0.56

60.

565

0.52

30.

566

0.46

80.

557

(3.2

5)**

*(2

.77)

***

(2.9

6)**

*(3

.32)

***

(3.7

5)**

*(2

.56)

**(2

.82)

***

Pola

rizat

ion

(lagg

ed)

0.13

50.

205

0.13

20.

124

0.07

80.

066

0.12

7(0

.45)

(0.7

6)(0

.52)

(0.4

8)(0

.35)

(0.2

5)(0

.57)

Divi

ded

gove

rnm

ent (

dum

my,

lagg

ed)

0.00

70.

069

−0.0

39−0

.038

0.00

2−0

.158

0.04

4(0

.05)

(0.5

0)(0

.31)

(0.2

9)(0

.02)

(1.0

5)(0

.31)

Econ

omic

outc

omes

and

inte

ract

ions

Infla

tion

(cha

nge i

n lo

ss o

f pur

chas

ing

pow

er)

0.56

00.

646

0.32

20.

543

(0.6

1)(2

.16)

**(1

.14)

(1.6

8)*

Infla

tion*

initi

al fr

agm

enta

tion

−0.6

80(0

.66)

Infla

tion*

initi

al p

olar

izatio

n−1

.781

−1.2

63(3

.03)

***

(1.7

7)*

Infla

tion*

divi

ded

gove

rnm

ent

−0.4

95(1

.37)

Grow

th (c

hang

e in

grow

th ra

te, l

og)

−0.0

35−1

.024

2.67

8−0

.429

(0.0

3)(1

.54)

(1.5

4)(0

.57)

Grow

th*i

nitia

l fra

gmen

tatio

n0.

981

(0.6

9)Gr

owth

*ini

tial p

olar

izatio

n3.

781

2.54

8(3

.40)

***

(1.8

0)*

Grow

th*d

ivid

ed g

over

nmen

t−2

.072

(1.1

5)Co

nsta

nt−0

.687

−0.6

99−0

.630

−0.5

53−0

.631

−0.3

40−0

.647

(3.0

0)**

*(2

.89)

***

(2.5

3)**

(2.3

2)**

(2.8

4)**

*(1

.26)

(2.6

6)**

Sum

mar

y sta

tistic

Num

ber o

f obs

erva

tions

7171

7171

7171

71Nu

mbe

r of c

ount

ries

1717

1717

1717

17R2

0.41

0.46

0.45

0.44

0.48

0.45

0.51

Coun

try fi

xed

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Sour

ce:

Auth

ors’

calcu

latio

ns.

*Sig

nific

ant a

t 10

perc

ent;

**sig

nific

ant a

t 5 p

erce

nt; *

**sig

nific

ant a

t 1 p

erce

nt.

a. Se

e tex

t for

des

crip

tions

and

met

hod

of co

nstru

ctio

n of

inde

pend

ent v

aria

bles

. Rob

ust t

stat

istics

are i

n pa

rent

hese

s.

more so in highly fragmented party environments. The largest party in thelegislature (which usually is the incumbent’s) is rewarded when economicgrowth improves, and this reaction seems to increase with the degree ofideological polarization.58 Party polarization even makes the legislativevote sensitive to changes in the inflation rate.59 Our results thus lend sup-port to hypotheses 1 and 2 above.

Do Policies Matter?

The next step is to establish whether the electorate cares about policies,and not only about outcomes. For parsimony, and given our limited sam-ple sizes, we ignore the influence that the features of the political systemmay have on voters’ sensitivity to the economic outcomes. We also ignoreother factors that may affect how the electorate feels about the adoption ofcertain policies and focus on the direct electoral effects of the policiesthemselves, as captured in γ:

The first four regressions in table 11 assess the influence on presiden-tial elections of the set of macroeconomic policy indicators defined earlier.The only indicator that shows some significance is the structural fiscal bal-ance, which appears with a negative sign in regression 2, implying that theelectorate reacts against fiscal restraint (however, the coefficient impliesthat this effect is very small). Note that inflation always keeps the rightsign and remains significant in this particular regression, although it losesits significance in some others. Therefore, although the electorate seems towant price stability, it does not reward—and may even punish—an incum-bent for some of the macroeconomic policies that may be needed to achievethose outcomes, such as stronger fiscal balance.

The electorate is more emphatically opposed to some of the promarketreforms, according to regressions 5 to 9. The coefficients for the total indexof reforms and for trade liberalization policies are highly significant, withelasticities of −1.57 and −0.84, respectively. Regression 5 is the basis forthe analysis in tables 3 and 4, where we show that the total electoral pay-

d V F d X d P Ct t t t tlog log log log .( ) = + ( ) + ∗ ( ) + ∗ ( ) + +−α ψ β γ λ ε1


58. None of our main conclusions, in either this or the following sections, is alteredwhen the regressions are run for the share of votes of the incumbent’s party (results avail-able from the authors upon request).

59. All these results persist when the set of political control variables is excluded fromthe regressions.

off of the reforms remains strongly negative, even if we take into accountthe full indirect effects implied in the correlations between the changes inthe reform index and the changes in inflation and growth. As mentioned, thepoint estimate of the direct effect implies that the incumbent’s party typ-ically lost 23 percent of its vote in presidential elections on account ofthe average amount of promarket reforms introduced during its term (or15 percent if we take into account our rough estimate of indirect effects).More aggressive reformers—say, those reforming 1 standard deviationabove the mean—would sacrifice 40 percent of their vote on account of allthe promarket reforms (or 27 percent with the indirect effects). As theremainder of the paper shows, the negative electoral payoff of the adop-tion of promarket reforms is a remarkably robust result.

Regression 10 evaluates the effect of the protection of property rightsand finds that it does not influence the behavior of the electorate. Regres-sion 11 is an attempt to summarize the influence of all the WashingtonConsensus policies, using the composite indexes for the macroeconomicand structural policies, along with the index of property rights. This regres-sion indicates that while the electorate does not hold strong views onmacroeconomic or property rights policies, it does on promarket policies.Finally, the last two regressions in table 11 test the robustness of the pol-icy variables that were found to be significant in previous regressions,namely, the fiscal balance, the total reform index, and the trade liberaliza-tion index. Only the total reform index is robust to the inclusion of theother variables.

In summary, this evidence lends support to the hypothesis that the elec-torate rewards the incumbent’s party for good macroeconomic results—inflation, in particular—but punishes it for the adoption of the promarketpolicies endorsed by the Washington Consensus.

Table 12 presents a similar set of regressions for legislative elections.Those that test the significance of the macroeconomic policy indicators areconsistent with the conclusion that the electorate does not care about thesepolicies. However, in regression 3 the real exchange rate is significant at10 percent with a positive sign, suggesting that the electorate favors moredepreciated exchange rates.60 The set of regressions dealing with the vari-ous indicators of promarket reforms suggests that they do not carry electoralcosts in legislative elections. Since some of these policies fall under thecontrol of the executive, this result is not surprising. However, as we showbelow, privatizations, which are strongly influenced by the legislature, do

60. However, this result does not hold in a similar regression without fixed effects.


TA

BL

E 1

1.

Impa

ct o

f Eco

nom

ic O

utco

mes

and

Was

hing

ton

Cons

ensu

s Pol

icie

s in

Pres

iden

tial

Ele

ctio

ns, 1

985–

2002

: Cou

ntry

Fix

ed E

ffec

ts R

esul

tsa

Depe

nden

t var

iabl

e: Ch

ange

in lo

g of

vote

shar

e of i

ncum

bent

pre

siden

t’s p

arty

Inde

pend

ent v

aria

ble

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

(13)

Polit

ical

Frag

men

tatio

n (la

gged

)0.

122

1.08

70.

263

0.27

6−1

.436

−0.9

45−0

.043

0.09

9−0

.118

−0.0

52−1

.722

−1.3

47−1

.026

(0.1

6)(1

.21)

(0.3

6)(0

.36)

(1.5

5)(0

.97)

(0.0

5)(0

.12)

(0.1

5)(0

.06)

(2.1

1)*

(1.9

3)*

(0.8

4)Po

lariz

atio

n (la

gged

)0.

341

0.93

40.

365

0.39

12.

298

1.83

40.

165

0.35

0.60

80.

159

2.11

43.

357

4.45

6(0

.42)

(0.9

5)(0

.48)

(0.4

8)(1

.72)

(1.2

1)(0

.23)

(0.4

2)(0

.61)

(0.2

1)(2

.17)

**(2

.48)

**(2

.60)

**Di

vide

d go

vern

men

t (du

mm

y, la

gged

)−0

.19

−0.6

95−0

.396

−0.2

110.

103

−0.1

99−0

.177

−0.2

58−0

.401

−0.2

09−0

.107

−0.1

94−0

.404

(0.5

1)(1

.99)

*(1

.07)

(0.6

1)(0

.51)

(0.5

7)(0

.53)

(0.6

9)(0

.87)

(0.6

1)(0

.46)

(0.6

9)(0

.65)

Econ

omic

outc

omes

Infla

tion

(cha

nge i

n lo

ss o

f pur

chas

ing

pow

er)

−1.5

44−1

.82

−1.7

93−1

.558

−2.0

3−1

.736

−2.3

81−1

.674

−1.8

99−2

.424

−3.1

95−2

.735

−2.1

59(1

.61)

(2.1

5)*

(1.5

4)(1

.35)

(2.0

9)*

(1.8

1)*

(2.0

0)*

(1.5

2)(1

.86)

*(2

.10)

*(3

.16)

***

(3.8

2)**

*(3

.11)

**Gr

owth

(cha

nge i

n gr

owth

rate

, log

)−2

.861

−2.7

88−2

.14

−2.9

98−1

.016

−1.2

56−3

.447

−2.6

52−2

.54

−3.0

5−1

.793

−2.2

11−2

.003

(1.7

5)*

(1.5

9)(1

.47)

(1.6

1)(0

.74)

(0.9

9)(1

.78)

*(1

.55)

(1.6

3)(1

.82)

*(1

.32)

(1.2

2)(1

.17)

Was

hing

ton

Cons

ensu

sM

acro

econ

omic

refo

rms i

ndex

(log

, cha

nge)

−0.4

630.

353

(1.7

2)(1

.43)

Stru

ctur

al fi

scal

bal

ance

(rat

io to

GDP

, cha

nge)

−0.0

66−0

.007

−0.0

12(2

.20)

**(0

.41)

(0.3

8)Re

al ex

chan

ge ra

te (d

etre

nded

in lo

gs, c

hang

e)0.

403

(0.8

8)So

cial e

xpen

ditu

res (

shar

e of G

DP, c

hang

e)−6

.79

(1.3

1)St

ruct

ural

refo

rms i

ndex

(log

, cha

nge)

−1.5

69−1

.938

−1.8

25(2

.98)

***

(3.9

5)**

*(4

.23)

***

Trad

e ref

orm

inde

x (lo

g, ch

ange

)−0

.844

−0.8

45(2

.24)

**(1

.82)

Fina

ncia

l ref

orm

inde

x (lo

g, ch

ange

)−0

.073

(0.3

7)Pr

ivat

izatio

ns in

dex (

chan

ge)

0.06

4(0

.09)

Tax i

ndex

(log

, cha

nge)

−0.7

76−0

.877

(1.5

0)(1

.43)

Inst

itutio

nal r

efor

ms i

ndex

(log

, cha

nge)

−0.2

32−0

.049

(0.7

3)(0

.20)

Cons

tant

−0.7

92−1

.837

−0.9

33−0

.85

0.62

7−0

.061

−0.7

59−0

.826

−0.6

47−0

.654

0.80

50.

197

−0.6

2(1

.18)

(1.8

3)*

(1.5

6)(1

.47)

(0.8

5)(0

.08)

(0.9

2)(1

.31)

(1.0

3)(0

.79)

(1.0

7)(0

.29)

(0.6

2)Su

mm

ary s

tatis

ticNu

mbe

r of o

bser

vatio

ns40

3340

3937

3739

4040

3936

3131

Num

ber o

f cou

ntrie

s15

1517

1617

1717

1717

1615

1515

R20.

670.

790.

640.

640.

800.

760.

700.

630.

690.

710.

890.

910.

91Co

untry

fixe

d ef

fect

sYe

sYe

sYe

sYe

sYe

sYe

sYe

sYe

sYe

sYe

sYe

sYe

sYe

s

Sour

ce:

Auth

ors’

calcu

latio

ns.

*Sig

nific

ant a

t 10

perc

ent;

**sig

nific

ant a

t 5 p

erce

nt; *

**sig

nific

ant a

t 1 p

erce

nt.

a. Se

e tex

t for

des

crip

tions

and

met

hod

of co

nstru

ctio

n of

inde

pend

ent v

aria

bles

. Rob

ust t

stat

istics

are i

n pa

rent

hese

s.

TA

BL

E 1

2.

Impa

ct o

f Eco

nom

ic O

utco

mes

and

Was

hing

ton

Cons

ensu

s Pol

icie

s in

Legi

slat

ive

Elec

tion

s, 1

985–

2002

: Cou

ntry

Fix

ed E

ffec

ts R

esul

tsa

Depe

nden

t var

iabl

e: Ch

ange

in lo

g of

vote

shar

e of p

arty

with

mos

t sea

ts in

legi

slatu

re

Inde

pend

ent v

aria

ble

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

(13)

Polit

ical

Frag

men

tatio

n (la

gged

)0.

588

0.61

80.

590.

593

0.52

40.

533

0.59

50.

615

0.50

70.

609

0.52

40.

551

0.57

2(3

.24)

***

(3.0

1)**

*(3

.65)

***

(3.1

9)**

*(2

.60)

**(2

.43)

**(3

.03)

***

(3.1

5)**

*(2

.51)

**(3

.19)

***

(2.6

4)**

(3.0

0)**

*(2

.65)

**Po

lariz

atio

n (la

gged

)0.

138

0.17

40.

110.

142

0.26

0.29

20.

314

0.17

30.

070.

279

0.43

80.

224

0.27

6(0

.54)

(0.5

8)(0

.48)

(0.5

4)(1

.10)

(1.3

0)(1

.27)

(0.6

6)(0

.26)

(1.0

3)(2

.17)

**(1

.04)

(1.3

5)Di

vide

d go

vern

men

t (du

mm

y, la

gged

)−0

.038

−0.0

2−0

.034

−0.0

24−0

.043

−0.0

38−0

.021

−0.0

24−0

.061

−0.0

35−0

.033

−0.0

34−0

.018

(0.2

9)(0

.12)

(0.2

7)(0

.17)

(0.3

2)(0

.29)

(0.1

6)(0

.18)

(0.4

8)(0

.26)

(0.2

3)(0

.25)

(0.1

3)Ec

onom

ic ou

tcom

esIn

flatio

n (c

hang

e in

loss

of p

urch

asin

g po

wer

)0.

077

0.08

60.

048

0.08

90.

007

0.02

5−0

.023

0.05

10.

018

0.02

5−0

.071

0.00

60.

004

(0.3

3)(0

.33)

(0.2

1)(0

.36)

(0.0

3)(0

.10)

(0.0

9)(0

.22)

(0.0

7)(0

.10)

(0.2

8)(0

.03)

(0.0

2)Gr

owth

(cha

nge i

n gr

owth

rate

, log

)0.

948

1.03

70.

916

1.03

40.

786

0.71

20.

599

1.08

61.

084

0.70

30.

133

0.69

30.

653

(1.6

0)(1

.50)

(1.6

9)*

(1.7

8)*

(1.2

6)(1

.16)

(1.0

5)(1

.84)

*(1

.89)

*−1

.16

(0.2

3)(1

.21)

(1.1

3)W

ashi

ngto

n Co

nsen

sus

Mac

roec

onom

ic re

form

s ind

ex (l

og, c

hang

e)0.

080.

167

(0.7

4)(1

.12)

Stru

ctur

al fi

scal

bal

ance

(rat

io to

GDP

, cha

nge)

−0.0

02(0

.20)

Real

exch

ange

rate

(det

rend

ed in

logs

, cha

nge)

0.28

10.

327

0.30

5(1

.78)

*(1

.84)

*(1

.78)

*

Socia

l exp

endi

ture

s (sh

are o

f GDP

, cha

nge)

0.28

3(0

.13)

Stru

ctur

al re

form

s ind

ex (l

og, c

hang

e)−0

.134

−0.2

96−0

.103

(0.7

1)(1

.44)

(0.5

8)Tr

ade r

efor

m in

dex (

log,

chan

ge)

−0.0

83−0

.087

(0.7

3)(0

.71)

Fina

ncia

l ref

orm

inde

x (lo

g, ch

ange

)0.

057

(0.6

6)Pr

ivat

izatio

ns in

dex (

chan

ge)

−0.3

18−0

.23

(1.0

3)(0

.78)

Tax i

ndex

(log

, cha

nge)

−0.1

31(0

.83)

Inst

itutio

nal r

efor

ms i

ndex

(log

, cha

nge)

0.19

10.

298

(1.1

2)−1

.6Co

nsta

nt−0

.634

−0.6

83−0

.615

−0.6

5−0

.583

−0.6

15−0

.718

−0.6

63−0

.518

−0.7

41−0

.7−0

.592

−0.6

38(2

.69)

***

(2.2

4)**

(2.8

6)**

*(2

.60)

**(2

.27)

**(2

.23)

**(3

.06)

***

(2.6

9)**

*(2

.01)

**(2

.90)

***

(2.4

5)**

(2.2

3)**

(2.2

6)**

Sum

mar

y sta

tistic

Num

ber o

f obs

erva

tions

6861

6867

6565

6769

6968

6364

64Nu

mbe

r of c

ount

ries

1515

1716

1717

1717

1716

1515

15R2

0.45

0.41

0.49

0.41

0.48

0.48

0.46

0.47

0.46

0.47

0.53

0.52

0.53

Coun

try fi

xed

effe

cts

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Sour

ce:

Auth

ors’

calcu

latio

ns.

*Sig

nific

ant a

t 10

perc

ent;

**sig

nific

ant a

t 5 p

erce

nt; *

**sig

nific

ant a

t 1 p

erce

nt.

a. Se

e tex

t for

des

crip

tions

and

met

hod

of co

nstru

ctio

n of

inde

pend

ent v

aria

bles

. Rob

ust t

stat

istics

are i

n pa

rent

hese

s.

have electoral implications in some political contexts. As in the previousset of regressions, policies concerning property rights do not have signifi-cant effects on the behavior of voters. The regressions that include the threesummary indexes confirm that none of them is significant. The two finalregressions indicate that the real exchange rate index remains weakly signifi-cant when other policy variables are included. Therefore, evidence on the con-sequences of the Washington Consensus policies in legislative elections is notrobust. Somewhat surprisingly, the policy indicator that turns out to bemore robust is outside the direct influence of the legislature.

The main conclusion that emerges from the empirical evidence pre-sented so far is that the electorate is in favor of some economic outcomesas well as some economic policies. Inflation and the advancement of somepromarket reforms are key reasons for withdrawing support from the incum-bent’s party in presidential elections. (For legislative elections, the evi-dence so far is very scant, regarding both outcomes and policies). It is veryunlikely that the negative payoffs of promarket reforms in presidentialelections would be countered by their positive effects on inflation, growth,or other economic or social outcomes, because the electorate does notseem to be very sensitive to these variables. It is only fair to conclude thatthe electorate dislikes promarket policies, irrespective of their results.However, these conclusions require some additional testing, because theresponse of the electorate may depend on political, institutional, and eco-nomic circumstances, as stated in hypothesis 4.

Does Context Matter?

The sensitivity of the electorate to Washington Consensus policies may beinfluenced by a host of contextual variables, such as the ideology of theincumbent’s party, the incumbent’s promises during the election cam-paign, and whether the economy was in crisis at the time of the previouselections.61 As mentioned above, to treat this hypothesis we endogenizethe coefficient γ as follows:

Replacing γ in the previous equations gives (with fixed effects)

γ ρ τ ζ= + ∗ + ∗PROMISES IDEOLOGY CRISIS.


61. Crisis is measured as the (log) distance between GDP and its trend when GDP isbelow its trend, and 0 otherwise.

Tables 13 and 14 present the relevant results from this specification.62

The context in which reforms take place does not seem to affect the elec-torate’s sensitivity to those reforms. As shown in table 13, the only excep-tion occurs when the tax reform index is interacted with our “switch” index,measured with respect to promises.63 The negative coefficient in regression3 implies that the adoption of measures that make the tax system more neu-tral and effective leads to vote gains when the incumbent has campaignedon the adoption of promarket policies, but brings losses when the incum-bent has argued against them on campaign but switched once in power. Inlegislative elections, contextual factors seem to play an important role forprivatizations. In regression 1 of table 14 the coefficient for the privatiza-tions variable is negative and significant and the coefficient for the inter-action term (IDEOLOGY*PRIVATIZATIONS) is positive and significant.The values of the coefficients suggest that while privatizations do carryelectoral costs, these are reduced by about a third when the largest party inthe legislature is market oriented. Regression 4 includes two interactionterms found significant in previous regressions, namely, (IDEOLOGY*PRIVATIZATIONS) and (GROWTH*POLARIZATION). It finds that bothremain strongly significant. These results confirm the importance of ideol-ogy in legislative elections. It is revealing that the influence of ideology isdetected in connection with privatizations, because this is the area of reformin which the legislature plays the most important role and on which publicopinion is strongest.

Conclusion

This paper has assessed the electoral consequences of Washington Con-sensus policies in Latin America on the basis of testable hypotheses derivedfrom econometric and case studies on the subject. The results lend quali-fied support for our main four hypotheses, as follows.

d V F d X d Pt t t tlog log log log( ) = + ( ) + ∗ ( ) + ∗ ( )−α ψ β ρ1 ++ ∗

∗ ( ) + ∗ ∗ ( ) + +

τ

ζ λ

PROMISES

CRISISd P d P Ct tlog log εε t .


62. A more complete set of results is available from the authors upon request.63. The switch indexes are defined above. We also tested a switch index measured with

respect to the ideology of the party, and those same indexes in absolute values (which mea-sure whether the incumbent has lied or not, regardless of the direction of the switch). Noneof these alternative measures was found to be significant.


T A B L E 1 3 . Impact of Washington Consensus Policies Interacted with ContextualFeatures in Presidential Elections, 1985–2002: Country Fixed Effects Resultsa

Dependent variable: Change in log of vote share of incumbent president’s party

Independent variable (1) (2) (3) (4)

PoliticalFragmentation (lagged) −0.377 0.762 −0.192 0.808

(0.65) (1.70) (0.38) (1.75)*Polarization (lagged) 1.124 −0.156 1.143 −0.063

(1.33) (0.30) (1.34) (0.13)Divided government (dummy, lagged) 0.000 −0.317 0.000 −0.382

(0.00) (0.91) (0.00) (1.19)Economic outcomesInflation (change in loss of purchasing power) −0.696 −0.517 −0.701 −0.622

(1.57) (1.34) (1.85)* (1.50)Growth (change in growth rate, log) −0.524 −0.897 −0.319 −0.909

(0.49) (1.17) (0.37) (1.18)Washington ConsensusTax index (log, change) −1.564 0.283 −0.394 −0.061

(1.92)* (0.36) (1.65) (0.11)Promises*tax reforms index 1.424

(1.62)Ideology*tax reforms index −0.154

(0.65)Promises switch index*tax reforms index −1.658

(2.72)**Crisis*tax reforms index −6.489

(0.48)Constant −0.190 −1.057 −0.371 −1.115

(0.32) (2.27)** (0.67) (2.31)**Summary statisticNumber of observations 27 38 26 38Number of countries 14 17 14 17R2 0.67 0.64 0.72 0.64Country fixed effects Yes Yes Yes Yes

Source: Authors’ calculations.*Significant at 10 percent; **significant at 5 percent.a. See text for descriptions and method of construction of independent variables. Robust t statistics are in parentheses.

1. Electoral support for the incumbent’s party is higher, the better theaggregate economic outcomes during his or her administration. The incum-bent’s party is rewarded in presidential elections for reductions in the rateof inflation; and in legislative elections, for increases in the rate of growth(although the latter result is not robust in this first hypothesis). Neitherchanges in unemployment nor changes in income distribution appear toinfluence voters’ behavior.

2. The sensitivity of electoral support to economic outcomes dependson the institutional characteristics of the political regime and the party sys-tem. We find that in presidential elections, the more fragmented the partysystem, the higher the payoff of inflation rate decreases. There is also someevidence that in presidential elections a divided government increases thepayoff of the inflation rate decreases or increases in the rate of economicgrowth (however, this finding is based on a very small sample). In legislative


T A B L E 1 4 . Impact of Washington Consensus Policies Interacted with ContextualFeatures in Legislative Elections, 1985–2002: Country Fixed Effects Resultsa

Dependent variable: Change in log of vote share of party with most seats in legislature

Independent variable (1) (2) (3) (4)

PoliticalFragmentation (lagged) 0.568 0.657 0.658 0.59

(2.80)*** (3.36)*** (3.25)*** (2.90)***Polarization (lagged) −0.045 0.185 0.098 −0.003

(0.16) −0.7 (0.35) −0.01Divided government (dummy, lagged) −0.104 −0.085 0.042 −0.018

(0.79) −0.62 (0.28) −0.12Economic outcomesInflation (change in loss of purchasing power) 0.278 0.089 0.251 0.291

(1.19) −0.4 (0.94) −1.32Growth (change in growth rate, log) 1.168 1.198 0.638 −0.539

(1.99)* (2.05)** (0.94) −0.64Washington ConsensusPrivatizations index (change) −2.076 −0.333 −0.583 −2.068

(2.97)*** −0.79 (1.61) (3.01)***Ideology*privatization reforms index 0.680 0.67

(2.45)** (2.48)**Promises switch index*privatizations index −0.046

−0.19Crisis*privatization reforms index 15.173

(1.61)Growth*polarization 3.546

(3.12)***Constant −0.525 −0.684 −0.760 −0.626

(2.04)** (2.34)** (2.90)*** (2.43)**Summary statisticNumber of observations 67 67 69 67Number of countries 14 17 14 17R2 0.54 0.5 0.5 0.58Country fixed effects Yes Yes Yes Yes

Source: Authors’ calculations.*Significant at 10 percent; **significant at 5 percent; ***significant at 1 percent.a. See text for descriptions and method of construction of independent variables. Robust t statistics are in parentheses.

elections, there is strong evidence that party polarization enhances theelectoral payoff of higher rates of growth.

3. Electoral support for the incumbent’s party depends on the eco-nomic policies adopted. Policies may carry electoral costs even when theydeliver good economic outcomes. We find strong evidence that voters carenot only about economic outcomes but also, in some cases, about policies.While the electorate seems to be blind to macroeconomic policies, it isantagonistic to promarket policies beyond their effects on growth or infla-tion. Promarket reforms in general carry very large electoral costs for theincumbent’s party in presidential elections. If the context of these reformsis not taken into consideration, the evidence of adverse payoffs in legisla-tive elections is weak.

4. The electorate’s tolerance of unpopular policies depends on the ide-ology of the incumbent’s party, his or her campaign statements, and theinitial state of the economy. Ideology does influence the reaction of theelectorate in legislative elections, according to our results. While the elec-torate dislikes privatization measures, it is more tolerant of them when thelargest party in the legislature has a promarket ideology. In presidentialelections, some evidence suggests that the electorate punishes the incum-bent for the adoption of tax reforms when they run counter to his or hercampaign statements.

In synthesis, adopting the Washington Consensus was costly to thereformers, although these costs were mitigated in some circumstances.The parties in power were able to harvest juicy electoral dividends onlywhen the government pursued ambitious stabilization policies in high-inflation economies. These findings seem to fit well the salient facts of thelast two decades, where a few incumbents were favored by the electoratefor their success in taming inflation, but little electoral recognition wasgiven to those who advanced the other macroeconomic and structural poli-cies deemed necessary to accelerate growth and ensure stability. It mightbe tempting to conclude that the days of economic orthodoxy are numbered.However, it is unclear that reversing the reforms will produce electoral ben-efits. To date, the experience of reversals is limited to a few countries, andit is too soon to assess their political payoffs.

The strongest conclusion of this paper—that promarket reforms carrylarge electoral costs, irrespective of their macroeconomic effects—maynot surprise political scientists, but it certainly will surprise many econo-mists: why should the electorate reject policies that improve aggregateeconomic outcomes and welfare? Although this paper does not address


this question, some results (not reported) suggest that many of the simplerhypotheses proposed to answer it are at best incomplete. It has been widelyargued that such rejection is due to the social and distributional effects ofthe reforms, but we have not found any evidence either that voting deci-sions are directly affected by social or distributional outcomes or that theelectorate’s response to the reforms is influenced by them.64 It has alsobeen argued that frustration with the reforms is due to their weak economicimpact in countries that lack the institutional support needed to harvest thebenefits of market liberalization.65 Again, we find no evidence in supportof this view. Relatedly, several authors have suggested that opposition topromarket policies is stronger when those who make the liberalizationdecisions or benefit from them are perceived to be corrupt.66 However, wedo not find that any measure of perceived corruption helps explain theelectorate’s response to the adoption of promarket reforms.67

Many other hypotheses beyond those that we have been able to testare possible. Based on psychological theory and experimentation, SergioPernice and Federico Sturzenegger have argued that public opposition toa successful reform process can be explained by universal cognitivebiases—confirmatory bias and self-serving bias—if the principles of thereform are at odds with their beliefs and self-serving view of the world.68

And Sanjay Jain and Sharun Mukand have developed a theoretical modelto explain why successful reforms may run aground: if the reform processtilts the political balance in a way that makes the redistribution of the ben-efits less likely, public opinion may turn against the continuation of thereform process.69 Why Latin Americans reject promarket reforms at thepolls remains an open question.


64. For a summary of such arguments, see Lora and Panizza (2002); and Lora, Panizzaand Quispe-Agnoli (2004).

65. See Lora and Panizza (2002), on the basis of public opinion data.66. Di Tella and MacCulloch (2004) have uncovered empirical evidence consistent

with this hypothesis.67. For instance, when we include the interaction between our measure of reform and

a measure of control of corruption (taken from International Country Risk Guide, 2004;www.icrgonline.com) in the basic regression from table 5, the coefficient is positive but notsignificant.

68. Pernice and Sturzenegger (2003).69. Jain and Mukand (2003).

46

Comments

Sebastián Galiani: This paper by Lora and Olivera is very interesting.It investigates whether the Latin American governments that adoptedmarket-oriented reforms during the late 1980s and early 1990s were re-warded with votes by the electorate. Certainly, this is an interesting andmuch-debated question. It is also a very hard one. There are manysources of complexity.

A virtue of the paper is that the authors keep the analysis at an explo-rative level. Identifying the impact of the reforms implemented during the1990s on the voting behavior of the electorate is an intrinsically delicatematter, since there is no good way to assess how the same electorate wouldhave voted had the reforms not been implemented (in precisely the waythey were). Another serious difficulty is the lack of a more detailed theoryto help disentangle the effects under consideration from the data.

The authors rely on cross-country panel data to attempt to control for fac-tors that could potentially impact the effects they are interested in. Clearly,this is the best strategy for the data sets they use. They analyze both pres-idential and legislative elections, with separate empirical models.

A first question, which is very important for their analysis, is whetherthe electorate cares about policies per se, or only in relation to the effectsthey have on outcomes. This would determine whether the effect of eco-nomic reforms on voting behavior is a structural parameter or not. Theauthors assume that voters do care about policies per se, and that they voteon the basis of policies in the recent past.

Of course, the reforms of the 1990s were not implemented in a vacuum.They could be expected to affect variables such as growth and inflationand were also likely to hit unemployment, although transitorily. All ofthese variables are likely to influence voters, and thus should be includedas controls in the regression models. The authors do so.1 What is more,

1. They also include other time-varying political control variables. The authors reportthat the results are robust to the inclusion of year effects, but it is not clear why they do notjust report those results instead, since they encompass the ones presented here.

they also rightly compute the total effect (in addition to the direct effect)of the reforms on voting, using ancillary models that estimate the effectsof the promarket reforms on these macroeconomic outcomes.

In presidential elections, the authors find that the incumbent’s party isrewarded for reducing inflation, but once this is controlled for, the elec-torate seems to oppose market-oriented reforms.2 Even when the totaleffect of these reforms is computed, it appears that reforming parties paida price for adopting them. Given the relative importance of inflation amongthe explanatory variables, it would have been useful to check extensivelythe robustness of this result. Unfortunately, there are episodes of very highinflation that might be driving the results.

In any event, one question naturally arises: why did so many govern-ments adopt promarket reforms? A common argument today is that struc-tural reforms were supposed to deliver more growth than they did. However,even if this were the case, growth is found to have little effect on presiden-tial elections. Thus, if one sticks to the estimated model, one needs to lookelsewhere for the answer. Perhaps these reforms were also an essential com-ponent of the stabilization programs adopted by the countries that reducedinflation during this period, in a way not captured by the ancillary modelsreported in the paper? This seems to be the case for Argentina.3

Another important issue is how the reforms affect voting, once the maineconomic outcomes are controlled for. It is true that voting may well reflectthe taste of the electorate for these policies. But it could also capture otherthings. It is likely to depend on the way these policies were implemented.It might perhaps capture distributive effects. For example, voters might nothave fixed ideas about trade liberalization in general. Maybe, in decidinghow to vote, an individual only considers his own economic situation,which could have been affected by this particular policy reform and is alsoaffected by the macroeconomic performance of his country.

Consider privatization. Firms improved substantially after privatization.And consumers, in general, also benefited.4 But not everyone gained: dis-placed workers lost earnings and employment security, even in the long run.5


2. They find that the results for legislative elections are different. This is troublesome,because there is no theory to help interpret it. Why should one expect inflation to be moreinfluential in presidential elections, for example, and growth to be more influential in leg-islative elections?

3. Galiani, Heymann, and Tommasi (2003).4. See, for example, Galiani and others (2005); Galiani, Gertler, and Schargrodsky (2005).5. See, for example, Galiani and Sturzenegger (2005).

Although the average benefits outweigh the costs, privatization might havea small positive effect on the welfare of each consumer but a large costamong the small group of displaced workers. The former might not decidehow to vote on the basis of the outcome of privatization, but the latter wouldcertainly do so. This is just a speculative counterexample. These issues needto be explored further. A promising research strategy would be to exploitpanel data on voting at the smallest electoral district level, where the pro-portions of potential winners and losers from each reform can be identifiedusing census data.

Ernesto Dal Bó: The 1990s saw several governments in Latin Americaimplement various combinations of the reforms commonly associated withthe Washington Consensus. One important question concerns the effects ofthese reforms. A vast body of work in several different applied literatureshas dealt with, for example, the impact of trade liberalization, deregula-tion, and privatization. As the reform process seems to be losing momen-tum across the continent, a second important question has emerged: whatare the constraints on the reforms’ sustainability?

This paper by Lora and Olivera relates to the second, which is a crucialquestion if one believes that market-friendly reforms are, under some guise,beneficial. In a democratic government, reforms must appeal to the elec-torate, or a large enough portion of it. This, in turn, is a precondition forpoliticians to have a stake in implementing such reforms. Lora and Olivera’scentral claim is that apart from macroeconomic stabilization programs, theeffect of the Washington Consensus policies tends to be to drive votes away:privatization, tax reform, trade and financial liberalization generally costvotes.1 If this is true and more reforms are desirable, the challenge appearssubstantial: why expect political entrepreneurs to provide policies that losevotes? This question is crucial because we would like both to see policiesthat can bring Latin America prosperity and to preserve democracy.

The authors study variations in the vote shares of the party associatedwith the incumbent president or the dominant party in the legislature.


1. Williamson (1990b) includes ten priorities in his synthesis on “what Washingtonmeans by policy reform”: fiscal discipline; tax reform; the liberalization of interest rates,foreign investment, and trade; a competitive exchange rate; privatization; deregulation; theredirection of public expenditure toward social areas such as health, education, and infra-structure; and the protection of property rights. In practice, some reforms were emphasizedover others. As I argue below, there are reasons to believe this bias may have had importantconsequences.

The explanatory factors are economic outcomes (such as inflation andgrowth), political controls (such as political polarization or party frag-mentation), and policies in place (such as privatizations, trade opening,macroeconomic stabilization, or a composite index of structural reforms).Coefficients on inflation tend to be negative and sometimes significant;those on growth tend to be quite unstable, but appear positive and sig-nificant in specifications related to the party dominating the legislature;and those on policies associated with the Washington Consensus tend tobe negative. For example, the indexes of both structural reforms and tradeliberalization appear to lose votes to presidents (see table 11), while pri-vatizations tend to cost dominant legislative parties votes (see table 14).The authors interpret these results as evidence that macroeconomic stabi-lization programs that reduce inflation pay off quite handsomely in termsof votes and growth tends to pay off as well, while policies associatedwith the Washington Consensus are, in themselves, vote losers. The factthat these policy variables have negative coefficients in the presence ofeconomic outcome variables is interpreted as evidence that the publichas an intrinsic negative attitude toward such policies. In other words,voters disapprove of market-friendly reforms per se, regardless of theirimpact on economic performance. This is the second important claim inthe paper. Calculations are provided to show that although reforms mayyield votes by virtue of their beneficial effects on economic performance,they lose votes overall because Latin Americans dislike market-friendlyreforms per se.

Two Observations

The problems of estimating the impact of a policy (and hence how it willaffect voters’ opinions) are well known. Here, I argue that the potentialbenefits of policy reform might be underestimated, in terms of both theireconomic and their electoral impacts. For instance, consider a few coun-tries set on various growth trajectories. Assume now that, foreseeing col-lision, policy reform is implemented more aggressively when countriesget to very bad stages of their growth trajectories—that is, reforms areendogenous. It may be possible for the reforms to improve growth out-comes relative to what the countries would have experienced without thereforms, while estimates indicate that reforms are associated with less,rather than more, growth. The key problem is that we lack the counterfac-tual trajectory for each country, corresponding to less or no reform. Whenestimates of the reforms’ impact on economic outcomes are used in com-


bination with estimates of the outcomes’ impact on votes, the electoralpayoff from the reforms will tend to be underestimated.

A second possible problem could arise when reformist governments areelected right after an acute crisis. Suppose, now, that the old administra-tion, which is seen to be responsible for the crisis, is heavily punished atthe polls. This is likely to get the reformer-to-be government elected witha very large share of votes. Once reforms have been put in place and thecrisis resolved, vote shares may well regress to the mean corresponding tonormal times. Successful reformers might then be reelected because vot-ers approve of their work, but their reelection vote shares may be lowerthan when they were first elected. In this case, estimates will show reformsas destroying votes even when they really do not; in the absence of thereforms, the incumbent reformers would have been hit even harder at thepolls. True, inflation reduction appears to continue to earn votes, accord-ing to the authors’ results. Thus if the bias I mention is affecting both thereform and the outcome coefficients, it is not enough to alter the sign ofthe inflation coefficient. However, it is still possible that the electoralimpacts of both reforms and inflation are biased downward.

The Specification

The basic specification driving the analysis in this paper is reported in col-umn 1 of table 5 and in table 11. This specification includes both economicoutcomes and policy indexes as explanatory variables of variation in voteshares. One problem with this specification is that it could be unfair to thepolicies. After all, if the policies are improving economic outcomes andthese outcomes are included as regressors, the coefficient on the policy vari-able will never capture some of the policy’s electoral rewards. These will beattributed to the improved economic outcomes. Also, to the extent that thepolicies affect the economic outcomes, the regressors are not independent.

A better approach might be to estimate the impact of policies on all rel-evant outcomes, and then estimate the impact of those outcomes on votes.A crucial question, of course, is how to identify all the relevant outcomesthat the policies might affect. Moreover, this strategy does not attempt todetermine whether voters disapprove of reforms for reasons other thantheir economic consequences. The authors are highly concerned about thissecond issue, and the inclusion of policy variables next to economic out-come variables is meant to capture any intrinsic value that the public mayplace on policies.


To get around the problem of evaluating the electoral impact of policiesin a specification that includes outcomes, the authors attempt to separatethe effect of policies due to the intrinsic value voters may perceive fromthe effect due to their impact on economic outcomes. In order to do this,the authors regress both growth and inflation on the index of structuralreforms, which yields significant estimates of how much reforms reducedinflation and fostered growth (reported in table 6); call this effect A. Giventhat improvements in inflation and growth are estimated to yield votes(column 1 of table 5)—call this effect B—the authors compute the elec-toral gains from reform through improved performance by multiplyingeffects A and B. Then they add the resulting (positive) estimate to the negative coefficient on policies themselves in the basic regression fromcolumn 1 of table 5. They conclude that the net effect that WashingtonConsensus policies have on votes is negative: the direct “intrinsic value”effect is larger than the “outcome improvement” effect.

On Interpretation

The authors emphasize that the negative coefficient on the structural reformsindex represents an intrinsic distaste on the part of voters for market-orientedpolicies. One could argue against this interpretation. The reason relatesto the question of defining the complete set of relevant outcomes. To beconcrete, suppose that reforms reduce inflation and improve growth butalso severely damage the environment, causing air quality to drop tounhealthy levels. Now, suppose one runs the authors’ regression with the change in the share of votes as the dependent variable and inflation,growth, and a reform policy index as independent ones, without includ-ing an air quality variable. If people punish the government for thedecline in air quality, using the authors’ interpretation for this regressionone would conclude that the public dislikes market-friendly reforms. Inthis case, that conclusion is wrong. People may not care in the least aboutpolicy labels but care strongly about air quality, which has been affectedby the policies.

Therefore, I believe that the conclusion that Latin American people dislikemarket-friendly reforms per se is premature. A much larger set of controls forvarious relevant outcomes should be explored in the future. I discuss belowsome possible reasons why voters may be unhappy with reforms irrespectiveof their ideological profile. One potential candidate is the deterioration in thesocial atmosphere that some countries have experienced. As reforms wentdeeper in Argentina, for example, there were alarming increases in crimes


against property.2 A relatively safe society just a few decades ago, its “socialair” has now become a lot less easy to breathe; even middle-class people nowdiscuss defensive tactics against “express kidnapping.”

Why Are Latin Americans Unhappy with Reforms?

For the reasons explained above, I view the evidence that reforms cost votesas indicative rather than definitive. In particular, I am skeptical that reformsmay cost votes because of their labels. Even if they do, I would still argue thatwhat is now being interpreted as ideological motivation may be capturingother concerns. However, when these findings are considered next to the avail-able survey evidence indicating that Latin Americans are dissatisfied withsome structural reforms—especially privatization—a rather negative pictureemerges.3 This is puzzling. The reforms are backed by reasonable theory, andmoreover, several of them have had good effects. Most economists wouldhave expected the electorate to endorse the reform process strongly and visi-bly. Why do we find ourselves analyzing evidence that seems to point theother way?

As the profession expected, a number of reforms have had measurablebenefits. I will not attempt a systematic overview of the impact of struc-tural reforms here. An arbitrary selection of the literature suggests thateven privatizations, which are strongly opposed in opinion polls, have hada number of benefits. Some such studies have appeared in this journal.David McKenzie and Dilip Mookherjee, for example, review evidence onthe effects of privatizations from studies applying to four Latin Americancountries: Argentina, Bolivia, Mexico, and Nicaragua.4 The emerging pic-ture shows improvements in labor efficiency and access to services, with-out generalized negative effects on the income distribution or poverty.There is evidence that privatizations have increased firm efficiency.5

Sebastian Galiani, Paul Gertler and Ernesto Schargrodsky show that inArgentina the privatization of water led to improvements in service and


2. Between 1991 and 1998, the years during which most reforms were introduced and per capita GDP increased by about 40 percent, crimes against property rose by around71 percent.

3. See the polls conducted by Latinobarómetro. See also Lora, Panizza and Quispe-Agnoli (2004) on reform fatigue.

4. McKenzie and Mookherjee (2003).5. See, for instance, Boubakri and Cosset (1998) for a study of developing countries that

includes some Latin American cases.

thus to reductions in child mortality.6 The privatization of power genera-tion in Argentina led to a large increase in generating capacity, which inturn contributed to lower energy prices. The privatization of energy distri-bution, in particular, when coupled with regulation through price caps, hadbeneficial effects on productivity.7

Given the need for electoral support in order for the reform agenda toprogress, it seems important to ask why, if reforms have a number of ben-efits, one does not see stronger support for them. There are several possi-ble reasons.

One answer is given by Lora and Olivera: voters are strongly ideologi-cal. There is little reformers can do about this. If ideology is a fixed tastethat goes against economic convenience, then maximizing social happi-ness may require abstaining from further reforms.

It is possible that reforms constitute an investment, and as such entail costsduring the first few years, whereas benefits will tend to accrue later on. Ifvoters are not aware of this temporal pattern, they may be quick to punishreformers when experiencing the costs. Voters may not realize that judged interms of net present value contribution, the reformers deserve to be rewarded.

Moreover, reforms may create very diffuse benefits and highly concen-trated costs, and voters may react more strongly in the face of large varia-tions in payoffs. For example, privatizations tend to increase consumersurplus but generate layoffs. Suppose that one such reform creates a smallbenefit for 80 percent of the population, so that this fraction of the popu-lation is now likely to reelect the reformer with a 60 percent chance whenhe faces a challenger (assume no abstentions nor a third candidate). Thisgives the reformer a 48 percent vote share from the “winners.” But theremaining 20 percent of the population is badly hurt by the reform, so only5 percent of this group will reelect the reformer. Thus the overall voteshare for the reformer is only 49 percent of the vote, which is insufficientto obtain reelection even though a large majority gained from the reform.

Economists can get biased estimates when they lack the appropriatecounterfactual, and so can voters. Therefore, voters may not associatereforms with improved economic outcomes, even when they have beenbeneficial. This outcome is even more likely if voters compare their gov-ernments’ performance to that of neighboring nations, and it is the coun-tries on the worst trajectories that implement more reforms.


6. Galiani, Gertler, and Schargrodsky (2005).7. Estache and Rossi (2005).

It may be the case that reforms are made possible by one of two things(or both): a large fiscal crisis or a substantial amount of corruption.8 Eitherof these two factors could facilitate reform by softening the resistance ofvested interests. Traditional economics emphasizes the role of compensat-ing transfers to facilitate Pareto-improving moves. These transfers canbe hard to implement openly. Corruption could then facilitate reform,because it allows reformers to pass compensating transfers to “losers” orplayers with veto power over the reform process. But if crisis and corrup-tion facilitate reform, voters may associate reforms with these two traits,which in themselves have a number of negative consequences that maycost votes. In this case, it is not reform that costs votes. The electoral con-sequences of the facilitating factors are confused with those of reforms.9

Voters may also be motivated by spite and considerations of fairness.Even when reforms create a substantial consumer surplus, individuals maybe negative about them as a result of the fortunes that some key playershave made in the process.

Another interpretation is that voters are cruel pragmatists. They electleaders who they think will be pragmatic enough to effect the trans-fers that would make reform happen. Examples of successful reformerswho were later accused of corruption are Fujimori in Peru and Menem inArgentina. Once these leaders have delivered the reforms, why wouldrational voters keep them around? The social benefits from their (alleged)corruption have been realized, and there would be only costs to be reapedfurther down the line.

Voters’ assessments of reform may be affected by the economic cycle.The debate on the negative opinions of reform is relatively new. As such, itmay be marked by the fact that since about 1998 a few Latin Americancountries have encountered new macroeconomic problems and less abun-dant foreign capital. It would be interesting to see if opinions remain nega-tive during a future wave of capital inflows and macroeconomic recovery.

It is possible that voters may associate reforms (rightly or spuriously)with a number of outcomes that economists have not paid enough attentionto when discussing policy reform. Here, I discuss only one such possibility:reforms may be correlated with a deterioration in the “social atmosphere.”


8. On fiscal crisis as a catalyst, see, for instance, Drazen and Grilli (1993).9. In their conclusion, Lora and Olivera note in passing that they did not find any con-

nection between perceptions of corruption and electoral results. It would be worthwhile toexplore this connection further, under alternative specifications.

By this I mean that while reforms were being implemented, perceptions ofsecurity, social cohesion, and trust in others may have suffered.

The Social Constraint to Policy Analysis

I begin with a simple conjecture: policy reform indexes may be positivelycorrelated with data on crimes against property and kidnapping with extor-tion. (This may not be the fault of the reforms; the issue also requiresinvestigation.) These phenomena, like air quality in my earlier example,are essentially missing in the specification used by Lora and Olivera. If myconjecture is true, it may imply that voters do not care about reforms froman ideological standpoint, but associate them with decreased security andenjoyment of public spaces and social life.

Note that unemployment figures, inequality, or simple poverty mea-sures may not capture all the relevant variation when it comes to explain-ing how social issues affect electoral results. Indicators of marginality, theimportance of the shadow economy, the average spell of unemployment,and the concentration of longer spells at certain skill levels may be moretelling. In this connection, a pertinent question is to what extent reformschange not the levels, but the profiles of employment.10 For instance, ifreform pushes some workers to a different sector, it may impose on thema large loss of industry-specific human capital. Some of these workers maydecide to drop out of the labor force and not even look for a job. Measuredunemployment will remain the same, while marginality increases.

Although the original body of Washington Consensus recommenda-tions did include elements such as public expenditures in social areas andthe protection of property rights, these do not seem to have been the mainfocus of the reform process. There are various possible reasons for this.The most energetic reforms may have been those fueled by business inter-ests. Also, it may have been easier to pursue reforms that entailed reducingthe scope of public intervention than those that would make interventionmore sophisticated. Further, in the absence of clear theoretical predictionsof how reforms may impact welfare when property rights are not perfectlyenforceable, the protection of property may not have been seen as a vitalcomponent of the reform agenda.


10. Galiani and Hopenhayn (2003) show that after a decade of reforms in Argentina,taking into account reincidence, the risk of unemployment rose significantly and was dis-tributed unequally in the labor force.

A reassessment of the implementation constraints to policy reform couldbe productive. Political economics has done a lot to clarify how politicalconstraints may shape the process of endogenous policy formation. Ananalogous analysis of how social constraints may limit the set of desirableand implementable policies is still lacking.

My study with Pedro Dal Bó takes a step in this direction, arguing thatsocial conflict may impose a social constraint on policy analysis.11 Thereason is that in a world where property rights cannot be perfectly pro-tected, certain policies could unleash a “social backlash” of forced redistri-bution. We examine a simple, small open economy model in the presenceof one critical distortion: individuals may engage in appropriation activi-ties; for example, they may become members of a guerrilla organizationthat extorts payments from nonmembers, or simply become criminals. Thesize of this “appropriation sector” is governed by a balance between theopportunity costs and the returns to appropriation activities.

The key assumption in the model is that appropriation activities aremore labor intensive than the overall economy. (The model uses the labels“capital” and “labor,” but these could be interpreted to mean, respectively,skilled labor and unskilled labor; we would then say that appropriation isrelatively intensive in unskilled labor.) Shocks or policies that hurt labor-intensive activities (or that favor capital-intensive ones) drive wages downrelative to the mass or disputable wealth in the economy. This lowers themain opportunity cost of appropriation activities relative to their returns,so appropriation expands. An example is trade liberalization that increasesthe perceived price of capital-intensive goods relative to labor-intensivegoods.12 Similar effects could follow from deregulation and foreign invest-ment that increase productivity in capital-intensive industries more than inlabor-intensive ones. These forces clearly have the potential to generateefficiency gains and make society wealthier. However, by hurting the rela-tive value of labor in a world where appropriation is relatively labor inten-


11. Dal Bó and Dal Bó (2004).12. Under the skilled versus unskilled labor interpretation, this increases the wage gap

across skills, increasing inequality, and it drives appropriation activities up. Thus the modelcan account for simultaneous increases in GDP per capita, inequality, skill premiums, andcrime, as experienced by Argentina in the 1990s. One prediction of the model is that not allforms of inequality will affect crime, so simple measures of inequality may not always beneatly associated with crime. Appropriation activities are only affected by forms of inequal-ity that affect the structure of remuneration across factors (or skill levels). Otherwise, thebalance between the costs and returns to appropriation may remain unaltered.

sive, such forces may unleash an appropriation backlash that leaves allagents in the economy worse off.

We tend to consider a number of reforms convenient because we expecttheir efficiency gains will make societies more prosperous. However,when the protection of property rights is imperfect, we are in a second-best world. The welfare implications of reforms may then depend on newconsiderations, and the set of reforms that are desirable and sustainablewill change. It may pay to reassess the Washington Consensus from thisperspective.

Conclusion

Lora and Olivera have produced an interesting, thought-provoking piece.The problem they tackle belongs to a set of complicated issues on whichit is hard to derive definitive conclusions. More work is needed. Their evi-dence suggests that market-friendly reforms may not have produced hand-some electoral payoffs beyond those secured in the macroeconomicstabilization phase. This is consistent with survey evidence on public atti-tudes toward reforms and with the sentiment one perceives in casual inter-actions on the continent.

One challenge that lies ahead is to reconcile the evidence of importantbenefits from some reforms—notably privatizations—with the fact thatmany Latin Americans seem to reject the idea that they have benefitedfrom the reforms. I have here proposed a few potential approaches,including (1) examining how opinions on reforms are affected by factorsranging from the economic cycle to the perceived fairness of their imple-mentation, (2) analyzing how corruption—a much vilified trait—mayhave been instrumental in pushing reforms forward, and (3) investigatingthe role of a social constraint on the implementation of reforms. In partic-ular, a reassessment of the reform process in terms of its impact on socialpeace and criminal activities would be profitable. This should entail morereflection on what constitutes desirable and sustainable policy in a worldwhere the protection of life and property can only be imperfect.


References

Alesina, Alberto, and Howard Rosenthal. 1995. Partisan Politics, Divided Govern-ment, and the Economy. Cambridge University Press.

———. 1996. “A Theory of Divided Government.” Econometrica 64(6): 1311–41.Alesina, Alberto, John Londregan, and Howard Rosenthal. 1993. “A Model of

the Political Economy of the United States.” American Political ScienceReview 87(1): 12–33.

Arcelus, Francisco, and Allan H. Meltzer. 1975. “The Effect of Aggregate Eco-nomic Variables on Congressional Elections.” American Political ScienceReview 69(4): 1232–39.

Boubakri, Narjess, and Jean Claude Cosset. 1998. “The Financial and OperatingPerformance of Newly Privatized Firms: Evidence from Developing Coun-tries.” Journal of Finance 53(3): 1081–110.

Coppedge, Michael. 1997. “A Classification of Latin American Political Parties.”Working Paper 244. University of Notre Dame (November).

———. 2001. “Political Darwinism in Latin America’s Lost Decade.” In Politi-cal Parties and Democracy, edited by Larry Diamond and Richard Gunther.John Hopkins University Press.

Corrales, Javier. 2002. Presidents without Parties: The Politics of EconomicReform in Argentina and Venezuela in the 1990s. Pennsylvania State Univer-sity Press.

Crisp, Brian F., and Gregg B. Johnson. 2003. “Mandates, Powers and Policies.”American Journal of Political Science 47(1): 128–42.

Cukierman, Alex, and Mariano Tommasi. 1998. “When Does It Take a Nixon toGo to China?” American Economic Review 88(1): 180–97.

Dal Bó, Ernesto, and Pedro Dal Bó. 2004. “Workers, Warriors, and Criminals:Social Conflict in General Equilibrium.” University of California, Berkeley,mimeo.

Deininger Klaus, and Lyn Squire. 1998. “New Ways of Looking at Old Issues:Inequality and Growth.” Journal of Development Economics 57(2): 259–87.

Di Tella, Rafael, and Robert MacCulloch. 2004. “Why Doesn’t Capitalism Flowto Poor Countries?” Harvard Business School and Princeton University,mimeo.

Downs, Anthony. 1957. An Economic Theory of Democracy. New York: Harper.Drazen, Allan, and Vittorio Grilli. 1993. “The Benefit of Crises for Economic

Reforms.” American Economic Review 83(3): 598–607.Duch, Raymond M., and Randy Stevenson. 2004. “The Economy: Do They Get It

Right?” University of Houston and Rice University, mimeo (April).Economic Commission for Latin America and the Caribbean (ECLAC). Various

years. Panorama preliminar de las economías de América Latina y El Caribe.Santiago, Chile: United Nations.


Estache, Antonio, and Martín Rossi. 2005. “Do Regulation and Ownership Drivethe Efficiency of Electricity Distribution? Evidence from Latin America.” Eco-nomics Letters 86(2): 253–57.

Fiorina, Morris P. 1992. “An Era of Divided Government.” Political ScienceQuarterly 107(3): 387–410.

Galiani, Sebastián, Paul Gertler, and Ernesto Schargrodsky. 2005. “Water forLife: The Impact of the Privatization of Water Supply on Child Mortality.”Journal of Political Economy 113(February): 83–120.

Galiani, Sebastián, Daniel Heymann, and Mariano Tommasi. 2003. “Great Expec-tations and Hard Times: The Argentine Convertibility Plan.” Economia 3(2):109–60.

Galiani, Sebastián, and Hugo Hopenhayn. 2003. “Duration and Risk of Unem-ployment in Argentina.” Journal of Development Economics 71(1): 199–212.

Galiani, Sebastián, and Federico Sturzenegger. 2005. “Structural Reform and Dis-placed Workers Earnings.” Universidad Torcuato Di Tella, mimeo.

Galiani, Sebastián, and others. 2005. “The Benefits and Costs of Privatization inArgentina: A Microeconomic Analysis.” In Privatization in Latin America:Myth and Realities, edited by A. Chong and F. Lopez-De-Silanes. StanfordUniversity Press (forthcoming).

Gervasoni, Carlos. 1997. “La sostenibilidad electoral de los programas de estabi-lización y reforma estructural: los casos de Argentina y Peru.” UniversidadCatólica Argentina, Universidad Torcuato Di Tella, mimeo.

Inter-American Development Bank. 2003. Good Jobs Wanted. Economic andSocial Progress in Latin America series. Washington.

Jain, Sanjay and Mukand, Sharun. 2003. “Redistributive Promises and the Adop-tion of Economic Reform.” American Economic Review 93(1): 256–64.

Kahneman, Daniel, and Amos Tversky. 1979. “Prospect Theory: An Analysis ofDecision under Risk.” Econometrica 47(2): 263–91.

Keech, W. 1995. Economic Politics: The Costs of Democracy. Cambridge Univer-sity Press.

Kramer, G. H. 1971. “Short-term Fluctuations in U.S. Voting Behavior, 1896–1964.”American Political Science Review 65(March): 131–43.

Kuczynski, Pedro-Pablo, and John Williamson, eds. 2003. After the WashingtonConsensus. Restarting Growth and Reform in Latin America. Washington:Institute for International Economics (March).

Lewis-Beck, Michael S. 1988. Economics and Elections. University of MichiganPress.

Lora, Eduardo. 2001. “Structural Reforms in Latin America: What Has BeenReformed and How to Measure It.” Research Department Working Paper 462.Washington: Inter-American Development Bank.

Lora, Eduardo, and Ugo Panizza. 2002. “Structural Reforms under Scrutiny.”Research Department Working Paper 470. Washington: Inter-American Devel-opment Bank.


Lora, Eduardo, Ugo Panizza, and Myriam Quispe-Agnoli. 2004. “Reform Fatigue:Symptoms, Reasons, Implications.” Federal Reserve Bank of Atlanta Eco-nomic Review (2nd quarter): 1–28.

McKenzie, David, and Dilip Mookherjee. 2003. “The Distributive Impact of Pri-vatization in Latin America: An Overview of Evidence from Four Countries.”Economía 3(2): 161–233.

Meltzer, Allan H., and Marc Vellrath. 1975. “The Effect of Economic Policies onVotes for the Presidency: Some Evidence for Recent Elections.” Journal ofLaw and Economics 18(3): 781–89.

Naím, Moisés. 1994. “Latin America: The Second Stage Reform.” Journal ofDemocracy 5(4): 32–48.

Navia, Patricio, and Andrés Velasco. 2003. “The Politics of Second GenerationReforms.” In After the Washington Consensus: Restarting Growth and Reformin Latin America, edited by Pedro-Pablo Kuczynski and John Williamson.Washington: Institute for International Economics (March).

Payne, J. Mark, and others. 2002. Democracies in Development: Politics andReform in Latin America. Washington: Inter-American Development Bank.

Pernice, Sergio, and Federico Sturzenegger. 2003. “Culture and Social Resistenceto Reforms: A Theory about the Endogeneity of Public Beliefs with an Appli-cation to the Case of Argentina.” CEMA and Universidad Torcuato Di Tella,unpublished.

Persson, Torsten, Gerard Roland, and Guido Tabellini. 1997. “Separation of Pow-ers and Political Accountability.” Quarterly Journal of Economics 112(4):1163–202.

Remmer, Karen L. 1991. “The Political Impact of Economic Crisis in Latin Amer-ica in the 1980s.” American Political Science Review 85(3): 777–800.

———. 2003 “Elections and Economics in Contemporary Latin America.” InPost-Stabilization Politics in Latin America, edited by Carol Wise and RiordanRoett. Brookings.

Roberts, Kenneth M., and Erik Wibbels. 1999. “Party System and ElectoralVolatility in Latin America: A Test of Economic, Institutional, and StructuralExplanations.” American Political Science Review 99(3): 575–90.

Stokes, Susan. 2001a. Mandates and Democracy. Neoliberalism by Surprise inLatin America. Cambridge University Press.

———. 2001b. Public Support for Market Reforms in New Democracies. Cam-bridge University Press.

Thaler, Richard H., and others. 1997. “The Effect of Myopia and Loss Aversionon Risk Taking: An Experimental Test.” Quarterly Journal of Economics112(2): 647–61.

Tversky, Amos, and Daniel Kahneman. 1991. “Loss Aversion in Riskless Choice:A Reference-Dependent Model.” Quarterly Journal of Economics 106(4):1039–61.


Weyland, Kurt. 2002. The Politics of Market Reform in Fragile Democracies.Argentina, Brazil, Peru and Venezuela. Princeton University Press.

Williamson, John. 1990a. Latin American Adjustment: How Much Has Happened.Washington: Institute of International Economics.

———. 1990b. “What Washington Means by Policy Reform.” In Latin AmericanAdjustment: How Much Has Happened? edited by J. Williamson. Washington:Institute for International Economics.

Wolfers, Justin. 2002. “Are Voters Rational? Evidence from Gubernatorial Elec-tions.” Harvard University, Department of Economics, mimeo (April).


Rules of Origin in Preferential Trading Arrangements: Is All Well with

the Spaghetti Bowl in the Americas?

Preferential trading arrangements (PTAs) have proliferated spec-tacularly over the past decade around the world.1 The number of PTAs in force soared from fifty in 1990 to some 230 by the end of

2004, and it is expected to rise to 300 in the course of 2005. Governingmore than a third of global trade, PTAs have sparked intense policy inter-est at the multilateral level. They are among the top priorities of the ongo-ing Doha Round of trade negotiations of the World Trade Organization(WTO).2

The Western Hemisphere has been a major source for the expansion ofthe world’s PTAs. The region’s countries have signed some forty freetrade agreements with each other or with extrahemispheric parties since1990. Mexico and Chile have been particularly prolific integrators: Mexicohas signed twelve PTAs and Chile seven.3 While the bulk of their agree-ments are with partners in the Americas, both countries have also inte-grated with European and Asian economies. For its part, the United Stateshas concluded four free trade agreements in the Americas and six with

63

A N T O N I E S T E V A D E O R D A LK A T I S U O M I N E N

Estevadeordal and Suominen are with the Inter-American Development Bank’s Integrationand Regional Programs Department, Integration, Trade, and Hemispheric Issues Division.

We are grateful to Andrés Rodríguez, Pablo Sanguinetti, Alberto Trejos, and AndrésVelasco for outstanding comments.

1. PTAs include free trade agreements, customs unions, common markets, and singlemarkets.

2. The Doha Declaration states, “We also agree to negotiations aimed at clarifying andimproving disciplines and procedures under the existing WTO provisions applying toregional trade agreements. The negotiations shall take into account the developmentalaspects of regional trade agreements.”

3. The figures refer to formal free trade agreements and exclude the economic comple-mentation agreements.

nonhemispheric partners, and it is proceeding toward another six. In total,the countries of the Americas are negotiating or opening negotiations formore than two dozen new PTAs. Prominent ongoing initiatives includethe Free Trade Area of the Americas (FTAA) talks encompassing thirty-four countries and the negations between the Southern Common Market(Mercosur) and the European Union aimed at connecting the world’s twolargest customs unions.

The hemispheric PTA spree has forged a veritable spaghetti bowl ofmultiple and often overlapping agreements (figure 1). The various rulesincluded in each PTA—such as standards, safeguards, government pro-curement, and investment—entangle the bowl further. While PTAs cangenerate important economic benefits, the PTA spaghetti bowl carries tworisks. First, the manifold trade rules of PTAs can introduce policy frictionsthat increase the costs of trading. Each new rule in each PTA represents anew policy for firms to consider in their export, outsourcing, and invest-ment decisions. Each also has legal, administrative, and economic impli-cations for the PTA partner countries. Not all PTA rules necessarily work toexpand trade from its pre-PTA levels. Second, differences in rules acrossPTAs can translate into transaction costs to countries dealing on two or more


F I G U R E 1 . The Americas and Trans-Pacific PTA Spaghetti Bowl

Source:–Devlin and Estevadeordal (2004).

Canada

USA

Mexico

Chile

Uruguay

Paraguay Brazil Argentina

Mercosur

Bolivia

Colombia

Venezuela

Peru

Ecuador

CostaRica

Nicaragua

El Salvador

Honduras Guatemala

CACM

Trinidad & Tobago Dominica

Guyana Antigua & Barbuda

BarbadosGrenadaSuriname Jamaica St. Vincent & Grenadines

Belize St. Kitts & Nevis

CARICOM

Panama

Dominican Republic

Andean Community

Bahamas

St. Lucia Haiti

Brunei

Cambodia

Thailand

Laos

Malaysia Philippines

Myanmar

Singapore

Indonesia Vietnam

Japan

New Zealand

Australia

ASEAN

Korea

PR China Hong Kong

Taiwan

Russia

Papua New Guinea

APEC

Intra-Asia-Pacific in forceIntra-Asia-Pacific signed

FTAA

Intra-Americas in force

Trans-Pacific signed

APEC

FTAAEU, EFTA

PTA fronts simultaneously. This is a particular consideration in the Amer-icas, where each country belongs to an average of four PTAs.4

Rules of origin are a key market access rule (or discipline, in the jargonof trade negotiators) in PTAs. Rules of origin are the crucial gatekeepersof commerce: a product shipped from an exporting PTA member mustmeet the corresponding rule of origin to receive preferential treatmentfrom the importing member. Rules of origin epitomize the hemisphere’spolicy problem: a growing number of the region’s PTAs carry complexand restrictive rules of origin, and the many rules-of-origin regimes dif-fer from each other. Consequently, the rules-of-origin spaghetti couldhold back the trade-creating potential of the hard-earned PTAs.

This paper presents an in-depth diagnosis of rules-of-origin regimes inthe Americas and offers policy recommendations for the region tocounter the potential negative effects of rules of origin. We hope to maketwo contributions: to deepen understanding of the types and effects of rules of origin used in Western Hemisphere PTAs, and to add rigor tothe policy debate on the implications of PTAs to the multilateral tradingsystem.5

The paper is organized in four parts. The first part surveys the state andlatest trends in the rules-of-origin regimes in the Americas. The next sec-tion summarizes the recent research on the political economy reasonsbehind the choice of rules-of-origin instrument in PTAs. The third sectiondoes the same for the economic effects of rules of origin, and discusses theimplications of the research findings to the hemisphere’s PTA spaghettibowl. The fourth part contains our policy recommendations, and a finalsection concludes.

The Current Status of Rules-of-Origin Regimes in the Americas

Rules of origin can be divided into nonpreferential and preferential rules oforigin. Individual countries use nonpreferential rules of origin to distinguishforeign from domestic products when applying other trade policy instruments,

Antoni Estevadeordal and Kati Suominen 65

4. The calculation includes continental Latin America, Canada, and the United States,but not the countries of the Caribbean. The figures exclude partial scope agreements andeconomic complementation agreements.

5. See Estevadeordal and Suominen (2005a), Suominen (2004), and WTO (2002b) forfurther discussions on rules-of-origin regimes around the world.

such as antidumping and countervailing duties, safeguard measures, ori-gin marking requirements, discriminatory quantitative restrictions or tar-iff quotas, and rules on government procurement. The WTO is in the finalstages of the decade-long process of harmonizing nonpreferential rules oforigin at the multilateral level.6 Preferential rules of origin, the focus ofthis paper, are employed in PTAs and in the context of a generalized systemof preferences. They define the processes to be performed and inputs to beincorporated in a product in the territory of an exporting PTA member in order for the product to qualify for preferential access to an import-ing PTA member. The justification for preferential rules of origin is tocurb trade deflection—to prevent products originating from non-PTAmembers from being transshipped through a low-tariff PTA partner to ahigh-tariff one under the PTA-provided preferential treatment. Rules oforigin, in short, are tools for keeping non-PTA parties from free riding onthe PTA preferences. They are an inherent feature of free trade agree-ments in which the members’ external tariffs diverge or in which themembers wish to retain their individual tariff policies vis-à-vis the rest ofthe world. Rules of origin are also used in aspiring customs unions to gov-ern sectors for which the members have yet to establish a common exter-nal tariff.7

Rules of origin have become increasingly important over the past decade.This is due both to the globalization of production—the growth of inter-national trade in goods manufactured in multiple countries—and to thefact that today’s PTAs quickly reduce the preferential tariff, the more tra-ditional tool regulating preferential market access, to zero across mostproduct categories. Indeed, rules of origin are currently a key arbitrator ofthe effectiveness of multilateral trade rules requiring PTAs to cover “sub-stantially all trade” between the partner countries and not to raise barriersvis-à-vis third parties.

Preferential rules of origin have thus far eluded multilateral regulation.As a result, a wide repertoire of rules-of-origin types and combinationshas developed around the world. This section surveys the rules of originemployed in the Western Hemisphere’s PTAs.


6. See Estevadeordal and Suominen (2005a) and Suominen (2004) for details on theharmonization process.

7. Rules of origin are thus employed in the vast majority of PTAs. The Asia-PacificEconomic Cooperation (APEC) forum is a prominent exception in that it operates on theconcept of open regionalism, with the preferential tariffs essentially being extended also tononmembers.

Product-Specific Rules of Origin: Toward Product-Specific Tailoring

Preferential rules of origin were a simple affair in most of the world until afew years ago.8 In the Americas, trade agreements formed before the 1990sgenerally put in place one, often vaguely defined, rule of origin applicableto all products. Over time, the lack of precision in origin requirementsbecame much criticized for allowing—and indeed requiring—subjectivecase-by-case origin determinations.9 The growth in international trade andthe globalization of production expanded the constituency that paid atten-tion to rules of origin, while paradoxically complicating correct judgmentson origin.

Perhaps the single most important event that raised the profile of rules oforigin in the Americas was the U.S. Customs finding in 1991 that the domes-tic content in Honda Civics imported from Canada fell below 50 percent—the threshold required for claiming origin under the U.S.-Canada free tradeagreement of 1989 and the preceding U.S.-Canada Auto Pact of 1965. Thefinding fuelled the U.S. automotive industry’s concerns about the intensi-fying Japanese competition and the moves by Japanese companies to useCanada as a production base. For its part, Canada, which was concernedabout the loss of foreign investment, claimed that U.S. origin determina-tions were arbitrary at best. The Honda case had important repercussionsfor the 1994 North American Free Trade Agreement (NAFTA) nego-tiations. Pressured by vehicle lobbies, NAFTA negotiators put in place ahighly precise regional value content of 62.5 percent—a level that Japanesefirms were not expected to meet. The rules of origin lobbying spreadthrough other economic sectors. Fearing that Asian and European firmswould establish simple touch-up assembly operations in Mexico in orderto gain duty-free access to the North American markets, U.S. industriescalled for tailor-made rules of origin that would be stringent enough tokeep extraregional parties out, yet lenient enough to allow U.S. multi-nationals to retain their extraregional outsourcing linkages. Mexico, inturn, generally pushed for rules of origin that would not deter foreigninvestment. This bargaining resulted in the 150-page NAFTA rules-of-origin protocol, which carries individualized rules of origin for some5,000 different products.


8. The exception to the global pattern of general and vaguely defined rules of origin wasthe European Community, which as early as the 1970s had preferential rules-of-originregimes with different rules of origin governing the various product categories.

9. Reyna (1995, p. 7).

Product-by-product rules-of-origin tailoring became the norm in PTAtalks across the hemisphere and around the world. The drive toward pre-cise rules of origin was reflected by the 1999 multilateral Kyoto Conven-tion, which established five main criteria for determining origin.10 The firstis the wholly obtained or produced criterion, which asks whether the com-modities and related products have been entirely grown, harvested, orextracted from the soil in the territory of the exporting PTA member ormanufactured in that member from any of these products. This rule is met through not using any second-country components or materials. Theremaining four criteria are more complex and are packaged together underthe substantial transformation criterion. Of the four individual criteria, thefirst involves a change in tariff classification in the territory of a PTA mem-ber between a product imported from an extra-PTA party and the productthat is being exported within the PTA. The change may be required to occurat the level of chapter (two digits under the Harmonized CommodityDescription and Coding System), heading (four digits), subheading (sixdigits), or item (eight to ten digits). The second criterion is an exceptionattached to a change in tariff classification rule, which generally prohibitsthe use of nonoriginating materials from a specific subheading, heading,or chapter. The third defines the value content and prescribes either a min-imum percentage of the product’s value that must originate in the territoryof the exporting PTA member (domestic or regional value content) or amaximum percentage of the product’s value that can originate outside thePTA member’s territory (import content). Finally, technical requirementsprescribe or prohibit the use of certain inputs or the realization of certainprocesses in the production. Technical requirements are particularly promi-nent in rules of origin governing apparel products.

Rules-of-origin regimes use these four substantial transformation rulesboth alone and in combination with each other. The staple of regimes is thechange in tariff classification. This rule is used at different levels: more thanhalf of NAFTA rules of origin are based on a change in chapter, while manyother regimes use mainly a change in heading. Table 1 displays the per-centage shares of various combinations of rules-of-origin components inselected PTAs in the Americas and elsewhere. The table reveals the high


10. The Revised Kyoto Convention is an international instrument adopted by the WorldCustoms Organization (WCO) to standardize and harmonize customs policies and proce-dures around the world. The WCO adopted the original Convention in 1974. The revisedversion was adopted in June 1999.


T A B L E 1 . Distribution of Rules-of-Origin Combinations, Selected PTAs in the Americas

U.S.- Mercosur- Andean Chile- U.S.- E.U.- E.U.-Requirementa NAFTA Chile G3 Chile Community Korea Jordan Mexico Chile

NC 0.54 0.51 4.05 0.51 0.39 0.39NC + ECTC 2.04 2.39NC + TECH 0.02 1.39 1.39NC + ECTC + TECHNC + VC 100.00 0.78 83.94 10.91 11.90NC + ECTC + VC 1.57 1.57NC + VC + TECH 0.20 0.20NC + Wholly obtained

chapter 16.06 7.62 7.62NC + Wholly obtained

heading 0.70 0.70

Subtotal 0.54 0.53 4.05 0.00 100.00 1.29 100.00 24.82 26.16

CICI + ECTC 0.02CI + TECHCI + ECTC + TECHCI + VCCI + ECTC + VC 0.02CI + VC + TECH

Subtotal 0.04 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

CS 1.29 16.56 1.54 1.68 0.20 0.20CS + ECTC 2.52 5.57 0.73 0.47CS + TECH 0.04 0.14 0.10 1.90 1.78CS + ECTC + TECH 0.40 0.04 0.04CS + VC 0.42 4.60 2.11 0.27 0.27CS + ECTC + VC 0.10 0.04 0.16CS + VC + TECH 0.04CS + ECTC + VC + TECH 0.83

Subtotal 4.35 22.77 7.88 0.00 0.00 4.42 0.00 2.37 2.25

CH 17.09 23.70 16.45 46.00 46.87 32.99 32.86CH + ECTC 19.18 11.19 13.45 9.12 5.13 4.56CH + TECH 0.02 0.34 0.97 20.04 0.14CH + ECTC + TECH 0.14 0.44 0.26 6.66 6.66CH + VC 3.54 3.25 2.01 9.99 2.95 12.68 12.78CH + ECTC + VC 0.58 0.48 0.49 0.86 0.37CH + VC + TECH 0.10 8.06 23.97CH + ECTC + VC + TECH 4.82 0.02 0.02

Subtotal 40.65 39.40 46.02 100.00 0.00 59.57 0.00 58.34 57.25

(continued)


CC 30.95 23.18 21.09 22.49 2.16 2.16CC + ECTC 17.71 5.83 5.90 4.71 1.02 1.02CC + TECH 0.02 0.06 5.43 0.08 0.04 0.04CC + ECTC + TECH 5.76 8.08 6.65 5.67 11.25 11.02CC + VC 0.06 0.14 1.80CC + ECTC + VCCC + VC + TECH 2.67CC + ECTC + VC + TECH 0.20

Subtotal 54.44 37.21 42.08 0.00 0.00 34.75 0.00 14.47 14.24

Total 100 100 100 100 100 100 100 100 100

Source: Adapted from Estevadeordal and Suominen (2005a); Suominen (2004).a. The notation on requirements is as follows: NC: No change in tariff classification required; CI: Change in tariff item; CS: Change in

tariff subheading; CH: Change in tariff heading; CC: Change in tariff chapter; ECTC: Exception to change in tariff classification; VC: Valuecontent; and TECH: Technical requirement. Calculations are made at six-digit level of the Harmonized System.

T A B L E 1 . Distribution of Rules-of-Origin Combinations, Selected PTAs in the Americas (continued )

U.S.- Mercosur- Andean Chile- U.S.- E.U.- E.U.-Requirementa NAFTA Chile G3 Chile Community Korea Jordan Mexico Chile

degree of diversity in rules-of-origin regimes in the Americas. Neverthe-less, four main hemispheric rules-of-origin families can be identified.11 Oneextreme is populated by the older trade agreements such as the Latin Amer-ican Integration Agreement (LAIA), which uses one general rule applicableto all products (either a change of heading level or a 50 percent regionalvalue content). The LAIA model has been the point of reference for theAndean Community and Caribbean Community rules-of-origin regimes.

At the other extreme lie the so-called new generation PTAs such asNAFTA. The NAFTA model served as the reference point for numerousrecent bilateral agreements, including the U.S.–Central America free tradeagreement (CAFTA) and the U.S.-Chile, Chile-Canada, Mexico-Bolivia,Mexico-Chile, Mexico–Costa Rica, Mexico-Nicaragua, Mexico–NorthernTriangle (El Salvador, Guatemala, and Honduras), and the Group of Three(or G3, encompassing Mexico, Colombia, and Venezuela) free trade agree-ments. The model, particularly the versions employed in the U.S.-Chilefree trade agreement and CAFTA, is also widely viewed as the likeliestblueprint for the FTAA rules of origin. The NAFTA-based rules-of-originregimes are complex: depending on the product, the rules of origin mayrequire a change of chapter, heading, subheading, or item, and the change

11. Garay and Cornejo (2002).


of tariff classification is often combined with an exception, regional valuecontent (generally ranging from 35 to 60 percent), or technical require-ments. Like many other rules-of-origin regimes in the world, the NAFTA-model regimes contain an alternative list of product-specific rules of originfor selected products, which enables an exporter to choose between twotypes of rules of origin. The list is relatively extensive in NAFTA, cover-ing nearly 40 percent of the products, and its rules of origin are as complexas those on the main list.

Mercosur rules of origin and the rules of origin in the Mercosur-Boliviaand Mercosur-Chile free trade agreements are based on the change-of-heading criterion and different combinations of regional value content andtechnical requirements. They fall between the LAIA and NAFTA extremes intheir degree of complexity. The Central American Common Market’s rules-of-origin regime can be placed between those of Mercosur and NAFTA.12

U.S. bilateral free trade agreements with some extrahemispheric partners—such as Jordan and Israel—diverge markedly from the NAFTAmodel, incorporating value content rules of origin alone. The rules of originof the U.S.-Singapore and U.S.-Australia free trade agreements, however,resemble NAFTA in their complexity. The recently forged Chile–SouthKorea and Mexico-Japan free trade agreements also feature sectoral selec-tivity à la NAFTA. The future Canada-Singapore, Mexico-Singapore, andMexico-Korea free trade agreements, among others, will likely compoundthe spread of the NAFTA model in Asia and the Pacific. Meanwhile, theEuropean Union’s free trade agreements with Mexico and Chile carry theEuropean Union’s standard, harmonized pan-European rules of origin.13

How Restrictive Are Rules of Origin?

Making meaningful cross-product comparisons across the many types ofrules of origin requires a parsimonious tool. Estevadeordal’s restrictive-ness index provides such a tool.14 The index’s observation rule is based on

12. The Central American Common Market chiefly uses a change in tariff classificationonly. The regime is more precise and diverse than Mercosur, however, because it requiresthe change to take place at the chapter, heading, or subheading level, depending on the prod-uct in question.

13. See Estevadeordal and Suominen (2003).14. Estevadeordal (2000). The index was subsequently made more generalizable in

Estevadeordal and Suominen (2005a) and Suominen (2004). Carrerè and de Melo (2004)compare Estevadeordal’s index with an ordering emerging from cost estimates of differenttypes of rules of origin; they find the index to be consistent with the cost ranking.


the length of the jump over the Harmonized System’s tariff lines requiredby rules of origin: a change of chapter is more restrictive than a change ofheading, a change of heading more restrictive than a change of subhead-ing, and so on. Value content and technical requirements add to the rule’srestrictiveness.

Figure 2 reports the restrictiveness values of rules of origin in some of themain PTAs. Since it is based on coding at the six-digit level, it also revealsthe degree of interproduct dispersion of restrictiveness values, which servesas a measure of the selectivity of regimes. The final bar represents the like-

Nonpreferential

AFTAE.U.-Chile

E.U.-Mexico

Pan-Europe

US-Israel

US-Jordan

Andean Comm.

Mercosur-Bolivia

Mercosur-Chile

Chile-Korea

Mexico-Bolivia

Mexico-Costa Rica

G3CAFTAU.S.-Chile

NAFTA

8

7

6

5

4

3

2

1

0

Source:–Adapted from Estevadeordal and Suominen (2005a); Suominen (2004). a.–The box plots represent interquartile ranges (IQR), with the box extending from the twenty-fifth percentile to the seventy-fifth

percentile. The line in the middle of the box represents the median fiftieth percentile of the data. The whiskers emerging from the boxes extend to the lower and upper adjacent values. The upper adjacent value is defined as the largest data point less than or equal to x(75) + 1.5 IQR. The lower adjacent value is defined as the smallest data point greater than or equal to x(25) + 1.5 IQR. Observed points more extreme than the adjacent values are individually plotted (outliers and extreme values are marked using—–and—–symbols, respectively).

F I G U R E 2 . Restrictiveness of Rules of Origin in Selected PTAsa

liest outcome of the harmonization process of nonpreferential rules of origin.Two issues stand out. The first is the presence of rules-of-origin families.Regimes drawing on the NAFTA model are highly similar in terms of over-all restrictiveness and selectivity, as are regimes drawing on the Mercosurand European Union models. Second, the NAFTA-type rules-of-originregimes are by far the most restrictive and selective in the hemisphere and,indeed, the world. This finding is particularly important in light of the spreadof the NAFTA-model rules of origin across the hemisphere.

Research shows that food, textiles, and apparel products tend to havethe highest restrictiveness values across regimes.15 This provides precur-sory evidence that rules of origin may be arbitrated by the same politicaleconomy variables that drive tariffs, particularly in the industrialized coun-tries. Nonpreferential rules of origin feature some selectivity, which sug-gests the operation of political economy dynamics also at the multilaterallevel—and the endogeneity of the nonpreferential rules of origin to theexisting preferential rules-of-origin regimes.

Comparing Regimewide Rules of Origin

Rules-of-origin regimes also vary in their use of general, regimewide rulesof origin—that is, rules of origin that apply similarly to all products in aregime. Some of the most commonly used regimewide rules of origin includethe following:

—De minimis levels, which allow a specified maximum percentage ofnonoriginating materials to be used without affecting origin.

—Cumulation provisions, which enable producers of one PTA memberto use materials from other members without losing the preferential statusof the final product. The three types of cumulation are bilateral cumulation,which operates between two PTA partners (that is, firms operating in onepartner country can use products that originate in the other and still qualifyfor preferential treatment when exporting the product), diagonal cumula-tion, which allows countries tied by the same set of preferential origin rulesto use products that originate in any part of the common rules-of-originzone as if they originated in the exporting country, and full cumulation,which extends diagonal cumulation to allow the use of goods processed inany part of the common rules-of-origin zone even if these do not qualify asoriginating products.


15. See, for example, Estevadeordal (2000); Estevadeordal and Suominen (2005a);Suominen (2004); and Sanguinetti and Bianchi (2005).

—Prohibition of duty drawback, which precludes the refunding of tar-iffs on nonoriginating inputs that are subsequently included in a final prod-uct that is exported to a PTA partner. Drawback in the context of a PTA isviewed as providing a cost advantage to producers who gear their finalgoods to export over producers who sell their final goods in the domesticmarket. However, ending drawback increases the cost of nonoriginatingcomponents to producers who have thus far benefited from it.16

—Certification method, which defines the instance authorized to certifyan origin claim. The main methods are self-certification by exporters; certi-fication by the exporting country’s government or a certifying agency; and atwo-step combination of the private self-certification and the public govern-mental certification. High bureaucratic hurdles for obtaining a certificate oforigin lower the incentives for exporters to seek preferential treatment.

Whereas de minimis and cumulation clauses insert leniency in the appli-cation of product-specific rules of origin, drawback prohibition and com-plex certification methods may have the opposite effect, namely, increasingthe difficulty of complying with the rules-of-origin regime.17

Table 2 compares the regimewide rules of origin in the various rules-of-origin regimes. Bilateral cumulation is applied in virtually all regimes,but use of other regimewide components varies considerably. Again, the dif-ferent rules-of-origin families stand out. The NAFTA-model regimes set deminimis levels at 7–10 percent, preclude diagonal and full cumulation, donot permit drawback (often after a certain transition period), and are basedon self-certification.18 There are exceptions; for example, CAFTA, the latestof the NAFTA-model regimes, allows cumulation within Central America,


16. Many PTAs in the Americas include duty drawback provisions in the market accesschapter rather than in the rules-of-origin protocol. However, the implications of endingdrawback are very similar to the implications of stringent rules of origin, namely, increas-ing production costs for exporters. Cadot, de Melo, and Olarreaga (2001) show that dutydrawback may have a protectionist bias due to reducing producers’ interest in lobbying againstprotection of intermediate products.

17. Nonmembers of a cumulation area may view the cumulation system as introducinganother layer of discrimination in that it provides incentives for member countries to out-source from within the cumulation zone at the expense of extrazone suppliers.

18. Two qualifications are in order. First, the de minimis principle has numerous excep-tions in most regimes. For example, in NAFTA, it does not extend to dairy products, edibleproducts of animal origin, citrus fruit and juice, instant coffee, cocoa products, and somemachinery and mechanical appliances. Many regimes also calculate de minimis levels intextile products as the percentage of weight rather than the value of the final product. Second,although NAFTA prohibits drawback, it has launched a refund system, whereby the pro-ducer will be refunded the lesser of the amount of duties paid on imported goods and on theexports of the good to another NAFTA member.

TA

BL

E 2

.Re

gim

ewid

e Ru

les o

f Orig

in in

Sel

ecte

d Pr

efer

enti

al T

rade

Arr

ange

men

ts

De m

inim

is Dr

awba

ck

Regi

on a

nd p

refe

rent

ial t

rade

agr

eem

ent

(per

cent

)Ro

ll-up

Bila

tera

lDi

agon

alal

low

ed?

Amer

icas

NAFT

A7

Yes (

exce

pt au

tom

otiv

e)Ye

sNo

No af

ter 7

year

sU.

S.-C

hile

10Ye

sYe

sNo

No af

ter 1

2 ye

ars

CAFT

A10

Yes

Yes

Yes i

n ch

ap. 6

2 w

ith M

exico

No

t men

tione

dan

d Ca

nada

G37

Yes

Yes

NoNo

t men

tione

dM

exico

–Cos

ta R

ica7

Yes

Yes

NoNo

afte

r 7 ye

ars

Mex

ico-B

oliv

ia7

Yes

Yes

NoNo

afte

r 8 ye

ars

Chile

-Kor

ea8

Yes

Yes

NoNo

t men

tione

dCa

nada

-Chi

le9

Yes

Yes

NoNo

t men

tione

dM

erco

sur-C

hile

Not m

entio

ned

Yes

Yes

NoNo

afte

r 5 ye

ars

Mer

cosu

r-Bol

ivia

Not m

entio

ned

Yes

Yes

NoNo

afte

r 5 ye

ars

Cent

ral A

mer

ican

Com

mon

Mar

ket (

CACM

)10

Not m

entio

ned

Yes

NoYe

sU.

S.-J

orda

nNo

t men

tione

dNo

t men

tione

dYe

sNo

Not m

entio

ned

U.S.

-Isra

elNo

t men

tione

dYe

sYe

sNo

Yes

U.S.

-Sin

gapo

re10

Yes

Yes

No (O

P an

d IS

I allo

wed

)aNo

t men

tione

dE.

U.-M

exico

10Ye

sYe

sNo

No af

ter 2

year

sE.

U.-C

hile

10Ye

sYe

sNo

No af

ter 4

year

sOt

her r

egio

nsPa

n-Eu

rope

an10

Yes

Yes

Yes (

full

in EE

A)b

NoAS

EAN

Free

Trad

e Agr

eem

ent (

AFTA

),No

t men

tione

dNo

t men

tione

dYe

sNo

Yes

Aust

ralia

–New

Zeal

and

Clos

er Ec

onom

ic Re

latio

ns Tr

ade

Agre

emen

t (AN

ZCER

TA)

2Ye

sYe

sYe

s (fu

ll)Ye

sSo

uthe

rn A

frica

n De

velo

pmen

t Com

mun

ity (S

ADC)

10Ye

sYe

sNo

Not m

entio

ned

Com

mon

Mar

ket f

or Ea

ster

n an

d So

uthe

rn A

frica

(COM

ESA)

2cYe

sYe

sNo

Not a

fter 1

0 ye

ars

Sour

ce:

Adap

ted

from

Este

vade

orda

l and

Suom

inen

(200

5a);

Suom

inen

(200

4).

a. B

oth

orig

inat

ing

prod

ucts

(OP)

and

the i

nteg

rate

d so

urcin

g in

itiat

ive (I

SI) o

pera

te in

the U

.S.-S

inga

pore

Free

Trad

e Agr

eem

ent.

ISI a

pplie

s to

nons

ensit

ive, g

loba

lized

sect

ors,

such

as in

form

atio

n te

chno

logi

es. U

nder

the

sche

me,

certa

in in

form

atio

n te

chno

logy

com

pone

nts a

nd m

edica

l dev

ices a

re n

ot su

bjec

t to

rule

s of o

rigin

whe

n sh

ippe

d fro

m ei

ther

of t

he p

artie

s to

the

FTA.

The

sche

me

is de

signe

d to

refle

ct th

e ec

onom

icre

aliti

es o

f glo

bally

dist

ribut

ed p

rodu

ctio

n lin

kage

s and

to en

cour

age U

.S. m

ultin

atio

nals

to ta

ke ad

vant

age o

f ASE

AN co

untri

es’ r

espe

ctiv

e com

para

tive a

dvan

tage

s.b.

Euro

pean

Econ

omic

Area

(EEA

).c.

The s

ectio

n on

trad

e rem

edie

s men

tions

that

one

of t

he cr

iteria

for i

mpo

sing

a cou

nter

vaili

ng d

uty w

ithin

the b

lock

is th

at th

e tar

gete

d su

bsid

y is n

ot le

ss th

an th

e 2 p

erce

nt d

e min

imis.

Cum

ulat

ion

19. See chapter 62 of CAFTA. (The full text of CAFTA is available at www.sice.oas.org/Trade/CAFTA/CAFTADR_e/CAFTADRin_e.asp.)

20. The Mercosur rules-of-origin regime is similar, but allows for drawback. Drawbackis, however, prohibited for Argentine and Brazilian imports of intermediate automotiveproducts when the final product is exported within Mercosur.

21. The absorption by the European Union customs union of the ten new member coun-tries implied that thirty-four of the pan-European free trade agreements vanished overnight.Prior to the accession, the diagonal cumulation incorporated sixteen partners and coveredno fewer than fifty free trade agreements.

22. Estevadeordal and Suominen (2005a); Suominen (2004).23. For a more thorough treatment, see Estevadeordal and Suominen (2005a).


Mexico, and Canada of materials that Central America may use for pro-ducing U.S.-bound goods.19 The clause covers only a limited quota, how-ever, and it enters into force only after Canada and Mexico agree on it. TheCentral American Common Market’s regimewide rules of origin resemblethose of the NAFTA model, but they do not prohibit drawback. Merco-sur’s free trade agreements do not have de minimis levels or cumulationprovisions, they phase drawback out in five years, and they are based onpublic certification.20 The European Union’s rules-of-origin regimes standout for employing diagonal cumulation extensively across Europe.21

As in the case of the restrictiveness of product-specific rules of origin,the facilitation provided by regimewide rules of origin to the applicationof the rules-of-origin regime can be systematically assessed through anindex. The facilitation index developed by Estevadeordal and Suominenincorporates de minimis levels, diagonal cumulation, full cumulation, anddrawback (all of which can be expected to cut producers’ production costsby amplifying their pool of low-cost inputs), as well as self-certification(which can keep producers’ administrative costs lower than the othermethods).22 Figure 3 shows the behavior of the index. Regimes styledafter NAFTA and the European Union feature the highest levels of facil-itation, while the Mercosur- and LAIA-based rules of origin score rel-atively low. The result suggests some correlation between the restric-tiveness and facilitation indexes: regimes with the highest restrictivenessof product-specific rules of origin tend to also have the highest facilita-tion values.

Many rules-of-origin regimes have devised further, more idiosyncraticad hoc mechanisms to help the members adjust to the rigors of rules oforigin.23 Some such mechanisms include phase-in periods for stringentvalue content rules of origin; a number of different options for calculat-ing value content rules of origin; and tariff preference levels, which allow

goods that would not otherwise satisfy the rules-of-origin protocol toqualify for preferential treatment up to a yearly quota. While most regimesthat employ these mechanisms make them available to all members, someregimes provide them to one or some of the PTA partners only (for instanceto accommodate country-specific endowments, production structures, anddevelopment levels).

Trends in Rules of Origin in the Hemisphere

The main finding of the above analysis is that rules-of-origin regimesbased on the NAFTA model are among the most restrictive and selectivein the world. The analysis also reveals two key temporal trends in theWestern Hemisphere. First, the so-called new generation regimes of the1990s score the highest for restrictiveness, selectivity, and facilitation val-ues. Second, the restrictiveness of NAFTA-style agreements has fallensomewhat over time. Some of the most marked declines are in the mineral,


0

1

2

3

4

NAFTA U.S.-Chile

CAFTA G3 Mexico-CostaRica

Mexico-Bolivia

Chile-Korea

Merco-sur-Chile

Merco-sur-

Bolivia

AndeanComm.

U.S.-Jordan

U.S.-Israel

Pan-Europe

E.U.-Mexico

E.U.-Chile

AFTA

Source:—Adapted from Estevadeordal and Suominen (2005a); Suominen (2004).

F I G U R E 3 . Facilitation of Regimewide Rules of Origin in Selected PTAs

leather, plastic, apparel, and footwear sectors. In very general terms, thismeans, for instance, that some producers based in Costa Rica have greaterleeway to procure inputs or perform operations outside the PTA zoneunder the recently signed CAFTA than they do under the 1995 CostaRica–Mexico free trade agreement. What is more, NAFTA itself is liber-alizing some of its rules of origin.24 The Working Group in charge of therules-of-origin review process is designing new rules of origin on the basisof consultations with consumers and producers and a review of the rules-of-origin protocols that each NAFTA member country has negotiated intheir post-NAFTA free trade agreements, such as the United States withSingapore or Mexico with the European Union. If this latter process resultsin interregime rules-of-origin borrowing, it could enhance convergencebetween the NAFTA rules of origin and the rules of origin of other regimesaround the world.

The Political Economy of Rules of Origin

This section examines why rules of origin are chosen as policy instrumentsin preferential trade. After all, integrating governments could simply excludethe potentially trade-deflecting sectors from the PTA’s coverage, or put inplace a common external tariff covering all products. We also considerwhy restrictive and selective rules-of-origin regimes have gained groundover the past few years.

The Choice of Rules of Origin as a Policy Instrument

Studies on political economy widely concur that using rules of origin as akey policy instrument serves to pay off import-competing lobbies jeopar-


24. The initial set of revised NAFTA rules of origin took effect on 1 January 2003; theyinvolve alcoholic beverages, petroleum/topped crude, esters of glycerol, pearl jewelry,headphones with microphones, chassis fitted with engines, and photocopiers. See “Regula-tions Amending the NAFTA Rule of Origin Regulations,” Canada Gazette, 1 January 2003(available at canadagazette.gc.ca/partII/2003/20030115/html/sor24-e.html). In July 2004,the trade ministers of the NAFTA countries instructed the trilateral Working Group onRules of Origin to extend the liberalization drive to chemicals, pharmaceuticals, plasticsand rubber, motor vehicles and their parts, footwear, copper, and all items with a zero most-favored-nation tariff for all of the NAFTA members. See “A Decade of Achievement,”NAFTA Free Trade Commission, 16 July 2004 (available at www.freetradealliance.org/pdf/2004%20Advocacy/JointStatement.pdf).

dized by PTA formation.25 Rules of origin can be employed to favor intra-PTA industry linkages over linkages between the PTA and the rest of theworld and thus to indirectly protect PTA-based input producers vis-à-vistheir rivals outside the PTA.26 If rules of origin provide captive marketsdownstream, they may even be superior for the import-competing inter-mediate producer lobbies than exclusions of their products from the PTA.27

Furthermore, stringent rules of origin can also extend protection to uncom-petitive intra-PTA final-good producers. This happens when their extra-PTA competitors are too hard-pressed to switch to the componentsprescribed by the rules of origin. Even if an extra-PTA firm were to moveoperations to the PTA market, the edge of producers with existing intra-PTA supply links would continue until the new entrant’s regional sourc-ing met the rules of origin.28

Rules of origin, in short, enable governments to balance the competingclaims of export lobbies, which seek a liberalizing PTA in which all prod-ucts are subjected to tariff phase-outs, and import-competing lobbies,which are intent on halting all liberalization. Rules of origin compensateand can even benefit import-competing lobbies, while export interestsaccept stringent product-specific rules of origin as a preferable and politi-cally attainable alternative to a PTA rife with exclusions.29 Indeed, regimeswith the most stringent rules of origin also tend to feature the highest facil-itation values, which may be a sign of counter-lobbying by exporters threat-ened by the restrictive rules of origin.

Empirical work supports the hypotheses about the protectionist impulsesbehind rules of origin. Both Estevadeordal and Suominen find that restric-tive rules of origin tend to be put in place in sectors that are also marked byhigh most-favored-nation tariffs and long preferential tariff liberalization


25. Rules of origin are a particularly useful trade policy instrument for two reasons.First, like tariffs, rules of origin are a highly targetable instrument because they are oftennegotiated at the product level. Second, unlike the tariff, rules of origin can be defined intechnical and diverse terms, so they can be tailored differently for each individual good, whiletheir presumed protection can be hidden since rules of origin are not as immediately quan-tifiable as a tariff.

26. Krueger (1993); Krishna and Krueger (1995).27. Suominen (2003).28. Graham and Wilkie (1998). Given that rules of origin hold the potential for increas-

ing local sourcing, governments can also use them to encourage investment in sectors thatprovide high value-added or jobs (Jensen-Moran, 1996; Hirsch, 2002).

29. Suominen (2003).

phase-out schedules.30 Sanguinetti and Bianchi encounter similar evidencein Mercosur’s rules-of-origin regime.31

Explaining Rules of Origin Trends

Because they use rules of origin as a trade policy instrument, governmentsexpend considerable time and resources on the tedious, technical, andoften highly contentious crafting of the rules-of-origin protocols. But whyhave rules of origin become more restrictive over time? One possibility isthat the liberalization of most-favored-nation treatment and the growth ofglobal trade have strengthened export lobbies, while antagonizing import-competing interests. Governments find themselves under growing pres-sures from export interests to produce deeper trade liberalization, so theyhave had to develop targetable and effective tools, such as product-specificrules of origin, to compensate the potential losers from liberalization. Inthe case of NAFTA, for example, neither the deep preferences nor the sus-tained political support for the agreement would have been possible with-out a stringent rules-of-origin regime. Earlier integration schemes, such asLAIA, were less liberalizing than NAFTA; they managed the potentiallosers’ concerns in the tariff schedules, which obviated the need to createa sturdy set of new compensation tools within the PTA. It is thus no acci-dent that the ambitious liberalization of today’s PTAs is accompanied byrestrictive rules of origin.

Another, complementary explanation is that the growing propensityto fragment global production presents a threat to import-competingintermediate-good providers, who, in turn, see stringent rules of origin asan opportunity to discourage final-good producers from outsourcing orshifting production abroad.32 This notion implies that PTA formationcould be driven by protectionist interests.

As noted above, however, the restrictiveness of rules of origin appearsto have declined somewhat in the microcosm of NAFTA-model regimesover the past decade. This trend has three potential explanations beyondthe potential strengthening of export lobbies in the Americas since themid-1990s. First, NAFTA partners have had time to learn about the impli-cations of the different types of rules of origin. NAFTA is one of the firstregimes in the world to establish rules of origin tailored individually for


30. Estevadeordal (2000); Suominen (2003, 2004).31. Sanguinetti and Bianchi (2005).32. Suominen (2003).

each product; both governments and business lobbies thus lacked information on the effects of rules of origin when the NAFTA rules werefirst negotiated. NAFTA-based exporters and producers are widely per-ceived as having grown to find the rules-of-origin regime excessivelyrestrictive.33

Second, the newer regimes may be endogenous to the prior ones. Coun-tries integrating with the United States after Mexico did so—namely,Chile and the Central American countries—may have sought terms thatare more favorable than those attained by Mexico in order to rapidly bringthemselves on a par with the existing U.S. partners in the U.S. preferentialmarket.34

The third explanation negates the other two: the reduced restrictivenessmay have little to do with temporal dynamics, but rather may simply becaused by other variables, such as bilateral trade volumes and the types ofgoods produced by the different partners. One hypothesis is that newerregimes may have achieved the same level of protection provided byNAFTA through using less stringent rules of origin. Detailed time-seriesdata on the utilization rates of tariff preferences in the different NAFTA-model PTAs would help illuminate whether this is the case.35

The Effects of Rules of Origin

We now turn to the potential economic effects of rules of origin. Recentresearch indicates that rules of origin can increase firms’ administrativeand production costs, and both costs can introduce protectionist biases thathamper the free flow of trade and investment. The differences across rules-of-origin regimes may generate transaction costs, but these have yet to receiveempirical scrutiny. We consider the three costs in turn.


33. In theory at least, stringent NAFTA rules of origin may have caused competitiveextraregional producers to move their production facilities to the NAFTA region. Inresponse, affected intraregional producers who initially favored restrictive rules of originmay have grown disposed to loosening the rules-of-origin regime.

34. Perhaps less plausibly, the fact that all recent free trade agreements have been nego-tiated in the shadow of the FTAA process may have provided the NAFTA-model adherentsincentives to define a rules-of-origin model that is more acceptable to all countries of thehemisphere than the FTAA rules-of-origin regime. This assumes that the adherents to theNAFTA model favor the adoption of the FTAA.

35. The pattern would not be universal, however, given that the European Union hasimplemented the identical rules-of-origin regime across all its partners.

Administrative and Production Costs

The administrative costs of rules of origin stem from the proceduresrequired for ascertaining compliance with the regime. They are essentiallybookkeeping costs for the exporter—the paperwork and costs associatedwith certifying origin—and the costs incurred by the partner country’s cus-toms in verifying origin. The administrative costs can be considerable even in regimes using self-certification.36 Cadot, Estevadeordal, and Suwa-Eisenmann disentangle NAFTA’s administrative costs into those associatedwith rules of origin and those that are not; they find the former to approximatetwo percent of the value of Mexican exports to the U.S. market.37

The production costs of rules of origin arise from the various technicalcriteria imposed by the rules-of-origin regime. If rules of origin encour-age final-good producers to use intra-PTA sources even when cheapersupplies are available in the rest of the world, then the rules of origin raiseproduction costs and thus likely dampen the PTA’s trade-creating poten-tial. Rules of origin can also create trade diversion in intermediates ifthey give an unnatural boost to intra-PTA purchases of intermediategoods. However, if the costs of complying with the rules-of-origin regimeexceed the benefits of the PTA-conferred preferences, then final-goodproducers might cease to use the preferential channel, obtaining interme-diates from the rest of the world and exporting final goods under the most-favored-nation regime instead. Status quo would ensue, with the PTAhaving no effect on trade. Meanwhile, the various facilitating regimewiderules of origin should have the opposite effect, helping the PTA channelto flourish.


36. Many regimes call for self-certification, including NAFTA, CAFTA, and the U.S.-Chile, Mexico–Costa Rica, Canada-Chile, Central American Common Market(CACM), CACM-Chile, Chile-Korea, U.S.-Singapore, and U.S.-Jordan agreements. TheMexico-Bolivia agreement implements self-certification after an initial four-year period oftwo-step private and public certification. The pan-European, European Union–Mexico,and European Union–Chile agreements are mostly based on two-step private and publiccertification, with limited self-certification. The G3 agreement, LAIA, the Common Mar-ket for Eastern and Southern Africa (COMESA), and the Southern African DevelopmentCommunity (SADC) specify two-step private and public certification, whereas Merco-sur, Mercosur-Chile, Mercosur-Bolivia, Andean Community, Caribbean Community(CARICOM), and most rules-of-origin regimes in Asia and the Pacific rely on public certification or delegate certification to a private entity. See Estevadeordal and Suominen(2005a); Suominen (2004).

37. Cadot, Estevadeordal, and Suwa-Eisenmann (2005).

Estevadeordal and Suominen provide the most comprehensive analysis todate on the implications of rules of origin for trade, based on a 155-countrygravity model spanning twenty-one years.38 They reach four main conclu-sions. First, restrictive and selective product-specific rules of origin—that is,rules of origin that can be expected to increase production and administra-tive costs—undermine bilateral trade flows. This indicates that stringent rulesof origin do undermine PTAs’ trade-creating potential. Second, de minimislevels, diagonal and full cumulation, drawback, and self-certification—whichcan be expected to reduce a rules-of-origin regime’s production and adminis-trative costs—foster bilateral trade. This suggests that lenient regimewiderules of origin may counteract the negative effects of stringent product-specific rules of origin. Third, restrictive rules of origin in final goodsencourage bilateral trade in intermediate goods. As such, restrictive rules oforigin may result in trade diversion to the PTA area. Fourth, the trade effectsof rules of origin change over time: the negative effects of stringent rulesof origin gradually decrease, while the positive effects of permissiveregimewide rules of origin increase. This suggests that exporters learn tocomply with product-specific rules of origin and to take advantage ofregimewide rules of origin.

Other, single-regime studies on the trade effects of rules of originreach similar results, as do the closely related studies on usage rates ofPTA preferences.39 Estevadeordal and Miller document missed prefer-ences (or utilization rates below 100 percent) between the United Statesand Canada, which they attribute to the tightening of the rules of originunder NAFTA in 1994.40 Cadot, Estevadeordal, and Suwa-Eisenmann


38. Estevadeordal and Suominen (2005b); Suominen (2004).39. Cadot, Estevadeordal, and Suwa-Eisenmann (2005), focusing on NAFTA, show

that stringent rules of origin have undermined Mexico’s aggregate exports to the UnitedStates. The United States played a key role in establishing NAFTA’s Uniform Regulationsand rules-of-origin enforcement mechanisms. In January 1995, the United States found ahigh compliance rate among Mexican and Canadian exporters and producers on rules of ori-gin, at 90 percent and 80 percent, respectively (Reyna, 1995, pp. 37–38). Appiah (1999) alsoexamines NAFTA, but using a three-country, multisector computable general equilibrium(CGE) model; he finds that rules of origin distort trade flows, diverting resources from theirmost efficient uses and undercutting global welfare. James (2004) posits that NAFTA pref-erences and restrictive rules of origin have undercut Asian textile and apparel exports to theUnited States. Flatters and Kirk (2005) find that restrictive South African DevelopmentCommunity (SADC) rules of origin work against efficiency gains that the members couldreach through outsourcing outside the PTA area.

40. Estevadeordal and Miller (2002).

link the mere 64 percent utilization rate of NAFTA preferences to strin-gent rules of origin.41

In addition to their short-run trade effects, stringent rules of origin maycause investment diversion in the long run. This occurs when extra-PTAfinal-good producers move production to the PTA area with the sole pur-pose of meeting the rules of origin, even if the PTA is not the most effi-cient location for production. Rules of origin can also divert investmentwithin the PTA. Final-good producers may want to get around rules of ori-gin by moving production to the territory of the PTA partner that has thelargest domestic demand or the lowest external tariff on third-countryinputs (or both)—such as the United States in NAFTA.42 From a theoreti-cal perspective, requirements for a high regional value content can para-doxically encourage investment to the PTA country that has the highestproduction costs (that is, is the most inefficient producer), because goodsmade in member countries with low production costs may be hard-pressedto meet the rules of origin. Rodriguez theorizes that stringent rules of origincan lead to distortions in production structures within the PTA, while Este-vadeordal, López-Córdova, and Suominen encounter preliminary empiricalevidence that flexible rules of origin are conducive to foreign direct invest-ment (FDI) inflows.43

Transaction Costs

Analysts have yet to understand whether differences among rules-of-origin regimes generate transaction costs and impart economic effects.44


41. Cadot, Estevadeordal, and Suwa-Eisenmann (2005). Krueger (1993) reports thatunder NAFTA’s predecessor (the U.S.-Canada free trade agreement), Canadian producersopted to pay the tariff rather than go through the administrative hurdles to meet the rules oforigin. Brenton (2003) and Inama (2004) show that rules of origin shape developing coun-tries’ odds of qualifying for treatment under the generalized system of preference.

42. For example, a Mexican and a U.S. firm selling on the U.S. market and purchasingtheir inputs from outside the NAFTA region would be treated unequally under NAFTA: theMexican firm would be disadvantaged vis-à-vis the U.S. firm because it fails to meet therules of origin required to export to the U.S. market (Graham and Wilkie, 1998, p. 110).

43. Rodriguez (2001); Estevadeordal, López-Córdova, and Suominen (2004).44. Garay and Cornejo (2002) provide the only rigorous examination of the diversity in

rules of origin across regimes. They evaluate the correlations between types of rules of ori-gin in NAFTA, CACM, Mercosur’s free trade agreements, and Mexico’s free trade agree-ments. The study finds that only 10 percent of the product-specific rules of origin are exactlyidentical or highly similar between the regimes, although up to 75 percent of the rules of ori-gin in most chapters within both the Mercosur and Mexican regimes are highly similar.

Any adverse effects would clearly be heaviest for countries that are partyto several relatively different rules-of-origin models, such as Chile andCosta Rica. These so-called spoke countries require customs that are well-equipped to verify and implement the different rules-of-origin regimes, andthey may eventually have to tailor their production structures differentlyfor each PTA market.45 This generates transaction costs that would be nilin a world with one rules-of-origin model. The costs will be highest forsmall producers in spokes with a narrow domestic sourcing base. In con-trast, producers and customs alike in rules-of-origin hubs—such as theEuropean Union, Mercosur, Mexico, and the United States—escape mostof these costs. If the transaction costs of operating on many PTA frontsbecome excessive, then producers in spoke countries may be compelled tospecialize for one preferential channel over the others.46 At the globallevel, the market specialization induced by rules of origin could give riseto policy-driven, trade-diverting PTA hubs.

Other factors, however, could mitigate the costs associated with cross-regime differences. First, a small producer generally produces only a fewitems and would thus need to apply only a couple of different rules oforigin when exporting to the various preferential markets. Multinationalcompanies selling a variety of goods in different markets may face greatercomplexity, but they are also better equipped to economize any transac-tion costs given their superior human, technical, and financial capacities.Second, even when rules of origin differ across PTA markets, a singleproduction process may qualify for preferential treatment in each market.


45. Consider a Chilean producer of typewriters (heading 8469): the firm will have tocomply with rules of origin that stipulate a ceiling of 50 percent import content to enter theEuropean Union; a change of subheading (except from subheading 8469.12) to enter theUnited States; a change of heading to enter Korea (except from heading 84.13 or, alterna-tively, a change from heading 84.13, provided the regional value content is not less than 45 percent using the build-down method or not less than 30 percent using the build-upmethod); and a 60 percent regional value content (that is, a ceiling of 40 percent import con-tent) to enter Mercosur. Meanwhile, a European Union producer in the same heading canuse the same rules of origin—50 percent import content—to enter Mexico, Chile, SouthAfrica, and the whole pan-European system. This example also illustrates the comparativecomplexities faced by customs: if each rules-of-origin regime stipulates rules of origin for5,000 products, the Chilean customs would basically have to verify 20,000 different rulesof origin, whereas customs in the European Union countries would only need to verify5,000 rules of origin.

46. Inter-PTA divergences also allow countries wishing to join these preferentialarrangements to engage in PTA shopping, choosing to join the agreements that best accom-modate their existing domestic standards and interests, rather than joining PTAs that are lib-eralizing, neutral vis-à-vis third parties, and welfare-enhancing.

Much depends on the idiosyncrasies of the product and production processin question. Finally, regimewide facilitation mechanisms can go a longway toward reducing the effects of the cross-regime incompatibilities inproduct-specific rules of origin.

Avenues for Future Research

Our understanding of the effects of rules of origin is far from complete. Thecosts of differences across rules-of-origin regimes await analysis. Threeavenues for future empirical research would be particularly fruitful. Thefirst involves the long-term effects of rules of origin, particularly in lightof the interplay of intermediate and final goods markets. While restrictiverules of origin may initially dampen intra-PTA trade in final goods byincreasing the cost of intermediate goods, the subsequent decline in thedemand of intermediates should lower their price and thus revive both thedemand for them and the intra-PTA trade in final goods.47

The second avenue for research is the economic impact of the certifica-tion and verification costs of rules of origin, together with the potentialtrade-off between the different certification methods, on the one hand, andverification costs, on the other. For example, do regimes using the self-certification method increase the costs of verifying origin, such that thelow transaction costs of certifying origin translate into high transactioncosts of verifying origin?

Finally, the welfare effects of rules of origin remain uncharted.48 Captur-ing welfare effects will undoubtedly prove challenging, given that rules-of-origin regimes carry frictions—including restrictiveness, selectivity, andvarious regimewide components—that work in different directions.

Policy Recommendations on Rules-of-Origin Systems for the Americas

This study has analyzed the structure and evolution of rules-of-originregimes in the Americas and reviewed the latest research results on theeffects of rules of origin. The main findings are three-fold: stringent rulesof origin can be used as a tool to pay off protectionist sectors in a PTA andthus to foster the political prospects of PTAs; the NAFTA rules-of-origin


47. See Ju and Krishna (1998) and Duttagupta and Panagariya (2001).48. Appiah (1999) finds that rules of origin undermine welfare in the case of NAFTA,

although his operationalization of rules of origin in a CGE framework is rather crude.

model gaining force in the Americas carries relatively restrictive rules oforigin; and restrictive rules of origin can undercut the liberalizing poten-tial of PTAs. Taken together, these findings raise concerns about the ulti-mate economic effects of the Americas’ expanding PTA bowl. They alsoraise legal concerns: stringent rules of origin may breach Article XXIV ofthe General Agreements on Tariffs and Trade (GATT), which in paragraph8(b) defines a free trade area as “a group of two or more customs territoriesin which the duties and other restrictive regulations of commerce . . . areeliminated on substantially all the trade between the constituent territoriesin products originating in such territories.” Indeed, the WTO has recentlyrecognized rules of origin as constituting part of “other regulations ofcommerce.”49 Since rules of origin have implications for extra-PTA par-ties’ access to the PTA market, they also risk violating paragraph 5 of Arti-cle XXIV, which prohibits PTAs that raise barriers toward the rest of theworld from the pre-PTA levels.50

The evolution of the hemisphere’s rules-of-origin regimes also pro-vides reasons for optimism, however, and the region’s countries have anumber of policy options for reducing the potential negative effects ofrules of origin. The rest of this section addresses these two issues.

Encouraging Patterns in Rules of Origin in the Americas

The countries of the Americas have five reasons to be optimistic about theevolution of the regional rules-of-origin regimes. Each also augurs wellfor the design and implementation of the FTAA rules of origin. First, themost recent rules-of-origin regimes based on the NAFTA model—namely,the U.S.-Chile free trade agreement and CAFTA—incorporate simpler,more practical, and less restrictive product-specific rules of origin thanNAFTA. This evinces a trend toward market-friendly rules of origin in thehemisphere. The NAFTA review process will provide a further boost tothe NAFTA system’s liberalization of its rules of origin.

Second, the various regimes designed after NAFTA are fairly similarvis-à-vis each other, in both the types of rules of origin specified and theirlevel of restrictiveness. This can help reduce any potential transaction


49. See, for instance, WTO (2002a). Ambiguities remain as to the meaning of “sub-stantially all the trade.”

50. The WTO Negotiation Group on Rules is advocating a case-by-case analysis of thepotentially restrictive effects of preferential rules of origin on extra-PTA parties (WTO,2002a).

costs for NAFTA-model adherents that export under preferential terms totwo or more NAFTA-model PTAs. NAFTA’s review of its rules of originmay engender further interregime compatibilities, thereby paving the wayfor diagonal cumulation linking the NAFTA-model free trade agreements.

Third, the NAFTA-style regimes apply relatively lenient facilitation terms.This helps alleviate the compliance costs of the product-specific rules of ori-gin. Even more encouraging is the movement toward somewhat higher deminimis levels and the willingness to experiment with diagonal cumulation,as evidenced in CAFTA. Cumulation is crucial even in the presence of iden-tical product-specific rules of origin across PTAs. Augier, Gasiorek, andLai-Tong find that bilateral trade is up to 52 percent lower than expectedbetween two spoke countries that have identical rules-of-origin protocolswith the same hub, but that are not linked by diagonal cumulation.51

Fourth, the NAFTA model has now been adopted in numerous free tradeagreements. The current adherents will thus find it fairly easy to negotiate,adopt, and implement future free trade agreements. Should the FTAA cometo carry NAFTA-type rules of origin, the costs of adjusting to its rules-of-origin regime would be low for a good part of the hemisphere.

Finally, negotiators on rules of origin throughout the Americas, andparticularly in free trade agreements based on the NAFTA model, haveproved their willingness to revise existing rules-of-origin regimes to makethem more flexible. NAFTA’s review of its rules of origin is the clearestexample, demonstrating commitment to keeping North America’s rules oforigin apace with changes in technology and the globalization of produc-tion, and potentially marking a growing role of export interests in settingtrade policy.

More generally, the precision of the NAFTA-model rules of origin issuperior to the vaguely defined and subjective rules of origin of the past.Precision provides clarity and certainty to traders and customs alike. Becausethe NAFTA regime is based on the change in tariff classification, it pro-vides a fairer, more transparent, and more easily verifiable rules-of-originmodel than regimes based on value content, which paradoxically can behard to meet in countries with low production costs and are difficult toimplement in the face of fluctuations in exchange rates and changes in pro-duction costs. Precise rules of origin do not need to be restrictive rules oforigin; the NAFTA review process may well yield rules of origin that areboth precise and flexible.


51. Augier, Gasiorek, and Lai-Tong (2004).

Tackling the Negative Effects of Rules of Origin: Flexible Rules of Origin plus Hemispheric Cumulation

The positive trends in the Americas notwithstanding, potential sources offriction remain: stringent and selective rules of origin still govern many sectors, and the various regimes differ markedly, even across the subset ofregimes based on the NAFTA model. How can entrepreneurs obtain inputsfrom the cheapest sources, firms exploit cross-border economies of scale, andmultinational companies make sweeping investment decisions based on eco-nomic efficiency? How can producers in spoke countries qualify for all thepreferential markets simultaneously without undue transactions costs? Whatare the best ways to counter the rise of trade- and investment-diverting hubs?

The simplest way to preempt the negative effects of rules of originwould be to bring most-favored-nation tariffs to zero globally, althoughthis is not likely to become politically palatable in the near future. A furtheroption would be to move from free trade agreements to customs unionswith low common external tariffs, thereby eliminating rules of origin alto-gether, or, alternatively, to harmonize preferential rules of origin at themultilateral level, which would ensure compatible requirements acrossspoke producers’ export markets. However, the founding of customsunions with an across-the-board common external tariff has proved diffi-cult outside the European Union; rules of origin will thus remain an issueas long as a common external tariff does not cover all product categories.52

Meanwhile, the prospect of global harmonization of preferential rules oforigin is still relatively distant.

Two shorter-term policy options are more realistic. First, the existingregional rules-of-origin spaghetti bowl can be revised. PTA membersshould strive to design and revise their rules-of-origin regimes to establishtransparent, simple, precise, nonrestrictive product-specific rules of origin,such as a change in heading or subheading, and they should put in placelenient regimewide rules of origin, in particular a high de minimis level.Such rules of origin alone would reduce the frictions within and betweenPTAs. The hemisphere’s PTAs should be interconnected through diagonalcumulation—a task that would be relatively uncomplicated to implement inthe presence of readily harmonized origin regimes and would pave the way


52. NAFTA contains a small sectoral customs union, with a common external tariffgoverning certain automatic data-processing goods and their parts. The tariff ranges fromzero to 3.9 percent. See NAFTA Annex 308. (The full text of NAFTA is available atwww.sice.oas.org/trade/nafta/naftatce.asp.)

to a regionwide trade and production base. The countries of the Americasshould also improve training for exporters and customs about the technicalrequirements and implementation of rules of origin. These measures wouldhelp shorten the learning lags associated with rules of origin, reduce theadministrative hurdles facing both exporters and customs, provide smallcountries access to larger pools of intermediate goods, and allow spokeeconomies to trade on several different fronts by applying the same rules oforigin. This, in turn, would ensure that the hemisphere continues to enjoythe benefits of open regionalism. The movement from the complex andrestrictive NAFTA rules-of-origin regime to the simpler and less restrictiveU.S.-Chile free trade agreement and CAFTA is an encouraging step, and itshould be furthered in future regimes.

Second, the FTAA would automatically sort out the rules-of-originspaghetti bowl and put in place a hemispherewide cumulation zone—nosmall feat given that the countries of the Americas contain a sizable sub-sample of the world’s PTAs.53 An optimal FTAA rules-of-origin outcomewould establish simple, nonrestrictive product-specific rules of origin and,again, lenient regimewide rules of origin. The overall framework could bebuttressed with ad hoc innovative measures designed to accommodate thepartners’ idiosyncratic production patterns and capabilities. Thus con-strued, the FTAA would also prove that the hemisphere’s existing PTAsrepresent genuine building blocks for regionwide trade liberalization.54

To be sure, all hopes should not be pinned on the FTAA. Much workremains to be done to reconcile the various partner countries’ rules-of-origin preferences, and the FTAA project per se has been troubled overthe past several months. Nonetheless, the FTAA might prove to be theonly way to integrate the NAFTA- and Mercosur-model rules-of-origin


53. Countries (and regions such as Mercosur) have thus far submitted rules-of-originproposals for chapters 1–40. Each product tends to feature five to ten different proposals.

54. The hemisphere’s trade ministers proposed in November 2003 a two-tiered FTAA,with the first tier of keen integrators adopting deep commitments and wide tariff liberaliza-tion across the tariff universe and the second tier opting for shallower commitments and anarrower list of liberalized products. Two cumulation zones would likely result—one withall member countries and a narrower range of goods, and another with the wider liberaliz-ers in the additional set of goods. Cumulation in both tiers could be complemented withsome ad hoc tools, such as phasing in the rules-of-origin regime, particularly for the smallercountries. Should this structure result, countries in the two tiers would be able to cumulatein the products they have liberalized with partners that have liberalized the same goods.Wider liberalizers would thus cumulate among each other in a broad range of goods, whileall countries would cumulate in the narrower range to which the second-tier countries haveacceded. See Blanco, Zabludovsky, and Gómez Lora (2004).

regimes.55 It could also facilitate the prospects of multilateral harmoniza-tion of preferential rules of origin.

Whether accomplished through interlinking PTAs or through the FTAA,a hemispherewide cumulation zone appears to be the most promisingoption—particularly when its rules of origin are flexible enough to preventtrade diversion. A sustained fluid operation of a hemispherewide cumula-tion zone will require solid verification tools. Poor verification is a majorproblem in most of Latin America, and it has been accentuated by thegrowing inflows of goods, particularly from Asia. This situation couldprovoke a backlash against regional trade liberalization. The strong verifi-cation regime that CAFTA introduces in the textile and apparel sectorcould serve as a starting point, along with technical assistance to countrieswith the most feeble verification systems. Information technology shouldbe fully harnessed to facilitate verification.56

The countries of the Americas cannot afford to pursue new policiesonly within the hemisphere, but should push their WTO partners on twofronts. First, they should call for launching the harmonization of the world’spreferential rules-of-origin regimes. This option is increasingly timelygiven the proliferation of free trade agreements with different rules of ori-gin around the world and, in particular, the establishment of free tradeagreements between the hemisphere’s countries and extraregional part-ners. Harmonizing multilateral rules of origin is hardly a novel idea, butrather is a long-standing international commitment: the Uruguay RoundAgreement on Rules of Origin stipulates that once the signatories con-clude the harmonization of nonpreferential rules of origin, they will moveto harmonize preferential rules of origin, using the relatively flexible andsimple harmonized nonpreferential rules of origin as a blueprint.

The second multilateral policy that the countries of the Americas shouldpursue is the lowering of tariffs and nontariff barriers. The higher the PTApartners’ most-favored-nation barriers, the wider the preferential marginsand the greater the willingness of firms in the partner countries to comply


55. Even if the different hemispheric rules-of-origin regimes were left to coexist withthe FTAA rules of origin (as occurred with the Central American Common Market andCAFTA rules of origin), exporters would be better off for two reasons: first, firms couldchoose between two alternative rules of origin when trading with their pre-FTAA PTA part-ners, and second, the FTAA rules of origin could be less restrictive, in practice, than theprior PTA rules of origin—even if they are more restrictive on paper—because the FTAAcumulation zone is vastly expanding the pool of inputs available to any member country.We thank Jeremy Harris for pointing this out.

56. See Cornejo (2004).

even with costly and distortionary rules of origin. The expansion of thePTA spaghetti bowl must be accompanied by open regionalism, in whichmost-favored-nation liberalization proceeds hand-in-hand with preferentialopening.57

Conclusion

This paper has analyzed the various rules-of-origin regimes in the Ameri-cas, reviewed the latest research findings on the effects of rules of origin,and provided policy recommendations for the region’s countries to reducethe adverse economic impact of rules of origin. We have found that theNAFTA rules-of-origin model, which is expanding in the hemisphere, car-ries restrictive and complex rules of origin, and such rules of origin cancounteract PTA-inspired trade liberalization. These findings raise con-cerns about the hemisphere’s increasingly complex rules-of-origin bowl.The worrisome features can be tamed, however, through regional cooper-ation, in particular the adoption of simple and transparent product-specificrules of origin, the incorporation of mechanisms to promote regimewideflexibility, and the implementation of cross-PTA diagonal cumulation.Given the globalization of regional integration—that is, the movement ofregional partners to negotiate interregional agreements—the countries of theAmericas should also live up to the Uruguay Round commitment of harmo-nizing preferential rules of origin at the global level.

Preferential rules of origin matter only as long as there are multilateralbarriers to trade. If there is a silver bullet for reducing the negative effectsof rules of origin, it is the multilateral liberalization of tariffs and nontariffbarriers. If the Doha Round negotiators succeed in producing deep cuts inmost-favored-nation tariffs and nontariff barriers, and if the proliferation ofPTAs engenders a dynamic of competitive liberalization worldwide, theimportance of preferential rules of origin as gatekeepers of commerce willprogressively dissolve.


57. See Bergsten (1997) and Wonnacott (1996). Wonnacott suggests that free tradeagreements should be replaced by customs unions or a hybrid arrangements of customsunions and free trade agreements, lest the benefits of preferential opening be lost.

Comments

Pablo Sanguinetti: This is a very interesting paper that deals with animportant and often neglected aspect of preferential trading agreements(PTAs), namely, the determination of rules of origin. Rules of origin arethe regulations that determine under what circumstances a good is consid-ered to be produced in the region and thus able to enjoy the preferentialtariff treatment. The definition of these regimes, which is mainly the con-cern of lawyers and policy practitioners, could have important economicimpacts on trade and investment flows. Rules of origin have thereforebecome an alternative trade policy instrument targeted by governmentsand especially by the private sector in the integrating countries.

The paper does four things. First, it offers a very complete and detailedsurvey of the various rules-of-origin regimes that have been put in place inthe context of the huge increase in PTA initiatives for the world economyand the Americas in particular over the last fifteen years. Second, the paperdraws on the political economy literature to examine why the use of rulesof origin has become such an important policy for government and privatesector lobbies and why the level of restriction implied by rules of originhas increased over time. Third, given what the (positive) theory predictsregarding why rules of origin are established, the paper summarizes theevidence about the effects of these regulations on trade and investmentflows. Finally, the paper ends with policy recommendations. I concentratemy comments on the first two of these issues: the features of the variousrules-of-origin regimes in the Americas and the political economy aspectsof these rules.

On the Extension of the Restrictive NAFTA Model in the Americas

The paper concludes that the NAFTA model of rules of origin, which hasbeen widely applied in the Americas since the second half of the 1990s,is much more restrictive and selective than the rules included in previous

93

agreements like the old LAIA system and also those applied in Mercosurand in the free trade agreements between Mercosur and Chile and Bolivia.Is this bad news for free trade in the region? To a certain extent, it is notsurprising that the NAFTA model has been extended to various free tradeagreements in the Americas. Many of these new free trade agreementswere signed by NAFTA member countries (including all the bilateral freetrade agreement signed by the United States), which presumably wouldestablish similar rules in their new agreements in the interest of internalconsistency and for the same political reasons that originated the NAFTArules-of-origin system. It is also not surprising that the NAFTA-type rules-of-origin regime is more restrictive than those established in previous pref-erential trade initiatives. As the authors mention, initiatives such as LAIAwere much less ambitious than NAFTA, and many sectors and goods wereexempted from free trade. Import-competing producers did not have to askfor an alternative mechanism to receive some sort of import relief becausethey were already excluded from the agreements. The free trade agree-ments signed since the beginning of the 1990s, however, are more inaccordance with Article 24 of GATT in that they cover a significant partof trade and go much deeper in terms of eliminating trade barriers (evencompared to unilateral or multilateral liberalization schemes). Govern-ments and import-competing sectors naturally try to target additionalmeasures like rules of origin to ease the cost of adjustment for sensitivesectors. This reasoning implies that this development is not necessarilybad news for free trade in the Americas, since the extension of rules of ori-gin is precisely a reaction to further trade integration. On the other hand,the impact of these added restrictions may partially undo the gains fromliberalization resulting from decreasing tariffs.

Mercosur differs from NAFTA in that it is an incomplete customsunion, which has certain advantages. Since the main normative argumentfor adopting rules of origin is to avoid trade deflection (that is, importsentering the member country with the lowest tariff and being reshipped tothe other partners with no additional tariffs), rules of origin are not rele-vant for items that have already converged to the common external tariffof the trade union. One would therefore expect a more lenient regime inMercosur than in NAFTA. By the same argument, Mercosur rules of ori-gin should also be less restrictive than those included in the free tradeagreements signed by Mercosur with other countries, like Chile and Bolivia.Table 3 presents an index that measures the degree of restrictiveness of theMercosur, Mercosur-Bolivia, and Mercosur-Chile regimes. The index


ranges from one to four, with one being the most lenient regime and fourthe most restrictive.1 The overall level of restriction implied by rules-of-origin rules is 1.7 for intra-Mercosur trade, 2.2 for Mercosur-Chile trade,and 2.3 for Mercosur-Bolivia. The table also shows that sectors like tex-tiles, chemicals and basic metal products (steel) are among those mostaffected by these regulations. Estevadeordal and Suominen find similarresults for NAFTA.

Despite the fact that Mercosur is an incomplete customs union (so thatrules of origin should only matter for items that are exempted from the com-mon external tariff, as mentioned above), in practice, the rules-of-originregime is applied to all items independently of whether they are included inthe common external tariff. This evidence confirms that these rules are usednot only for the normative prescription of avoiding trade deflection, but alsoas a policy tool that could potentially offer some type of import protection.

On the Political Economy of Rules of Origin

Given that rules of origin can function as a protectionist device within thecontext of free trade agreements, how does a political economy approachchange the normative prescription about the emergence of free trade agree-ments and the role of these regulations? What are their determinants, and


T A B L E 3 . Mercosur: Rules-of-Origin Index, by Manufacturing Sectora

Sector Mercosur Mercosur-Bolivia Mercosur-Chile

Food, beverages, and tobacco 1.3 1.7 1.4Textiles, apparel, and leather 1.8 2.9 2.9Wood products 1.0 1.7 1.4Paper and printing 1.2 1.4 1.3Chemicals 2.5 2.7 2.6Nonmetallic products 1.1 1.2 1.1Basic metal products 1.7 2.6 2.6Metal products, machinery, and equipment 1.6 2.0 1.9Other manufacturing products 1.0 1.3 1.2

Total 1.7 2.3 2.2

Source: Sanguinetti and Bianchi (2005).a. The index ranges from one to four, with one being the most lenient regime and four the most restrictive.

1. This index is developed in Sanguinetti and Bianchi (2005) and closely follows themethodology presented in Estevadeordal (2000).

how do they relate to other key trade policy variables like tariff preferences?The paper addresses some of these concerns, but I wish to offer some com-ments to complement the authors’ discussion.2

Grossman and Helpman provide a political economy model of the emer-gence of free trade agreements.3 According to their approach, the decisionof whether to form a free trade agreement is subject to political pressuresfrom the potential losers and winners of trade creation and trade diversion.Grossman and Helpman use the term enhanced protection to describe tradediversion and reduced protection for trade creation (relative to the tariff-ridden situation prevalent before the free trade agreement). This approachsuggests that exporters that stand to gain the most from trade diversion inthe partner country will be most in favor of establishing the trade agree-ment, while import-competing sectors that will suffer from trade creationoriginating in imports from the other members will most vividly opposethe free trade agreement. Thus producers will support a free trade agree-ment when the probability of generating trade diversion is maximizedand trade creation is minimized. This is the case when, from a normativepoint of view, a free trade agreement is not fully justifiable. In practice,the final result will depend on how efficient these different groups are ininfluencing government policy through lobby activity and how the gov-ernment objective function weights consumer welfare vis-à-vis that of pro-ducer groups.

The original Grossman and Helpman model does not address the issueof intermediate inputs, so it cannot be easily applied to study the endoge-nous determination of rules of origin. This extension is provided by Cadot,Estevadeordal, and Suwa-Eisenmann, who present a simple partial equi-librium model in which two countries (North and South) engage in a freetrade agreement and both tariff preferences and rules of origin are jointlydetermined.4 They focus on a case in which intermediate-good interests inthe North wish to use the free trade agreement to create a captive marketfor their product. These interests lobby their government (though politicalcontribution, as in Grossman and Helpman) to establish strong rules of ori-gin to obligate Southern final-good producers to source in the North inorder to qualify for preferential access. This clearly reduces the effectiveprotection that the Southern producers receive for entering into the final-good market in the North. The authors assume that the South is always on


2. My comments are based on Sanguinetti and Bianchi (2005).3. Grossman and Helpman (1995).4. Cadot, Estevadeordal, and Suwa-Eisenmann (2003).

its participation constraint (that is, effective protection is zero).5 In thiscontext, deeper tariff preferences for the final goods can sustain stricterrules of origin. This, in turn, favors the Northern producers because it raises both the demand for their product and, more important, theintermediate-good price. The model thus delivers the interesting predic-tion that this price is not tariff ridden, but depends on demand and supply(as if the market for this product were closed). This is not surprising; rulesof origin function as a type of quantitative restriction. This frameworkleads to the testable implication that the restrictiveness of rules of originand tariff preferences are positively associated. This positive associationis documented in Estevadeordal for NAFTA and in Sanguinetti andBianchi for Mercosur.6

Summary Remarks

As I indicated at the beginning, this paper by Estevadeordal and Suominenis a very interesting piece of work that carefully analyzes the political, eco-nomic, and policy implications of rules-of-origin regimes in the Americas.I hope this survey-type of work encourages further research on the topic.

Alberto Trejos: I quite like this paper, which thoroughly addresses thetopic of rules of origin in current and future free trade agreements in theAmericas. Motivations for this kind of work include concerns that the grow-ing complexity of the administration of rules-of-origin regimes will becompounded as very disparate rules are implemented across differentagreements; the problem that many free trade agreements may use strin-gent rules of origin as an alternative (and less visible) mechanism formaintaining high rates of protection; and the possibility that such disparaterules of origin will turn free trade agreements into a stumbling block,rather than a building block, in the process of world trade liberalization.Understanding this topic is necessary if governments are to design thecorrect policies, including better free trade agreements, in the future. The


5. In this case, exports of the final good will not increase significantly as a consequenceof the free trade agreement initiative. Thus the lobby for stronger tariff preferences by theintermediate-good industry in the North will not face strong opposition from the final-goodindustry in the same country. There will be very low trade creation in final goods and astrong trade diversion in intermediates.

6. Estevardeordal (2000); Sanguinetti and Bianchi (2005).

majority of world trade (especially within the Americas) happens today inthe context of free trade agreements or other preferential arrangements inwhich rules of origin are applied. Previous work by the same authors illus-trates that in the Western Hemisphere the prevalent rules-of-originregimes are indeed more restrictive and heterogeneous than in the rest ofthe world.

When rules of origin are binding, they can have some of the sameeffects as tariffs and other barriers to trade. They discourage trade, requirelearning, reduce the rate of utilization of free trade agreements, and re-direct investment and trade. Furthermore, the costs of compliance can bevery high, reaching 2 percent of the total value of trade in some cases.While not as effective as tariffs when used as trade barriers (especially incomprehensive free trade agreements in which tariff phase-out takes placeacross almost all goods), they provide protectionist measures that the gen-eral public does not always see and that policymakers have a hard timequantifying.

The authors measure and assess rules-of-origin regimes according tothe stringency of the rules, the cost of implementing them, their nature,and their heterogeneity within and across agreements. They find a veryhigh diversity of rules of origin in the existing free trade agreements andpreference regimes in the Americas, both across agreements and acrossgoods within a given agreement. The rules of origin can also be very strin-gent, especially in older free trade agreements.

At the same time, the authors demonstrate that there are some sourcesof optimism on this topic. First, newer agreements are less restrictive. Sec-ond, as economies become more open and the results of enhancing tradeare appreciated, it becomes easier for governments to negotiate agreementsthat boldly go beyond their predecessors. Third, countries that are nownegotiating new free trade agreements show signs of significant learningfrom a decade or so of implementing their older agreements. Fourth, toremain competitive in an environment where others are doing the same,negotiators of new agreements are producing further liberalization than inprevious agreements. Finally, the most recent free trade agreements havebeen negotiated in the context of an imminent Free Trade Area of theAmericas (FTAA), which would much reduce the effectiveness of rules oforigin as trade barriers. (This factor will probably be less meaningful infree trade agreements negotiated after the modest results of the Miamiministerial of 2003, which much delayed the expected time of completionof a comprehensive FTAA.)


I would add to these causes for optimism the fact that recent agreementsinclude a variety of new flexibilities to make rules of origin less stringent.De minimis clauses, phase-ins, tariff preference levels, and, most impor-tant, accumulation of origin are the most important of such flexibilities.The authors similarly mention the possibility of building on the progressat the WTO on multilateral harmonization of rules of origin in a most-favored-nation basis; I am not optimistic about achieving relevant progressthere at this time.

While criticisms of the restrictiveness of rules-of-origin regimes arelargely valid, the political economy of trade negotiations is such that restric-tive rules of origin are often the only way to maintain a particular productin the tariff phase-out commitments of a free trade agreement. Not only dorules of origin give the local producer of the good more protection (inwhich case the rules of origin undo some of the progress attained in thephase-out), but restrictive rules of origin create other winners (the regionalproducers of the key inputs to that good), often tilting the balance. Tradediversion toward the parties involved in a free trade agreement is alwayspolitically more feasible than trade diversion away from them, and this isused in negotiations to generate political backup for further liberalization.A flexible rule of origin (which is always preferable, of course) mayreduce the feasibility of achieving a quick tariff reduction in the first placeby shifting the sourcing of materials to third countries. Under that light,one may see restrictive rules of origin as a necessary, and transitory, evilin some cases.

The authors neglect to look carefully at the growing web of subregionalagreements in the hemisphere. Mercosur, the Andean Pact, CARICOM, andthe Central American Common Market involve plans of economic integra-tion that go much further than current free trade agreements. These effortswill probably converge to a situation in which nations that belong to thesame subregional group, in their efforts to construct customs unions, willhomogenize their existing bilateral agreements with third parties, commit-ting to the same rules of origin and allowing for origin accumulation amongthe subregional partners. This will probably take a long time to come tofruition, but when it does it will significantly simplify the “spaghetti bowl”problem and reduce the distortionary impact of rules of origins.

The authors should also address the question of how rules-of-originregimes differ across free trade agreements in another way: while rules-of-origin procedures may be very heterogeneous across different goodswithin a given free trade agreement, specific goods might be treated similarly


across different free trade agreements. My impression is that this is thecase for some of the problematic goods, so the effects of current rules-of-origin regimes on FTAA and on future integration are less daunting than afirst read of the paper may suggest.

In general, accumulation of origin that is not limited to subregionalpartners is a significant source of optimism that the hemisphere’s rules oforigin will become less onerous, both as trade barriers and as administra-tive costs. For example, four distinct (but quite similar) agreements exist-ing today bind together, in all directions, a group of four nations (namely,Canada, Chile, Mexico, and the United States). Costa Rica will join thisgroup with the enactment of CAFTA, as will the other Central AmericanCommon Market partners once their agreement with Canada is in place. Itshould be feasible and desirable for nations in this list to allow, in theirbilateral agreements, origin accumulation with other nations in the list, asthe direct market access to those other parties has already been granted.That bottom-up mechanism may result in a better way to construct hemi-spheric integration and solve the problems of the complexity and strin-gency of rules of origin.

In conclusion, this is a very good and important paper. It is not easyto figure out how to address this question systematically, and the techni-cal work required for that purpose is certainly daunting. The authorsclearly do a good job there. They ask the right questions and raise many keypoints. Perhaps some topics (origin accumulation, in particular) deservemore attention than was given to them, but the effort clearly achievesprogress.

References

Appiah, Alex Jameson. 1999. “Applied General Equilibrium Model of NorthAmerican Integration with Rules of Origin.” Ph.D. Dissertation, Simon FraserUniversity (Canada).

Augier, Patricia, Michael Gasiorek, and Charles Lai-Tong. 2004. “The Impact ofRules of Origin on Trade Flows.” Paper prepared for the conference Rules ofOrigin in Regional Trade Agreements: Conceptual and Empirical Ap-proaches. Inter-American Development Bank (Integration and Regional Pro-grams Department), Centre for Economic Policy Research, and DELTA/INRA,Washington, 20–21 February.

Bergsten, C. Fred. 1997. “Open Regionalism.” In Whither APEC: The Progress toDate and Agenda for the Future, edited by C. Fred Bergsten. Washington:Institute of International Economics.


Blanco M., Herminio, Jaime Zabludovsky K., and Sergio Gómez Lora. 2004.“Una llave para la integración hemisférica.” Documento de divulgaciónIECI-03. Buenos Aires: Instituto para la Integración de América Latina y elCaribe (INTAL) and Inter-American Development Bank, Integration andTrade Division.

Brenton, Paul. 2003. “Integrating the Least Developed Countries into the WorldTrading System: The Current Impact of EU Preferences under Everything butArms.” Policy research working paper 3018. Washington: World Bank.

Cadot, Oliver, Jaime de Melo, and Marcelo Olarreaga. 2001. “Can Duty Draw-backs Have a Protectionist Bias? Evidence from Mercosur.” Working paper2523. Washington: World Bank.

Cadot, Olivier, Antoni Estevadeordal, and Akiko Suwa-Eisenmann. 2003. “Rulesof Origin as Export Subsidies.” Washington: Inter-American DevelopmentBank. Mimeographed.

———. 2005 (forthcoming). “An Assessment of Rules of Origin: The Case ofNAFTA.” In The Origin of Goods: A Conceptual and Empirical Assessment ofRules of Origin in PTAs, edited by Olivier Cadot and others. Oxford Univer-sity Press and Centre for Economic Policy Research.

Carrerè, Céline, and Jaime de Melo. 2004. “Are Different Rules of Origin EquallyCostly? Estimates from NAFTA.” Discussion paper 4437. London: Centre forEconomic Policy Research.

Cornejo, Rafael. 2004. “Recientes innovaciones en los regímenes de origen y suincidencia en el proceso de verificación: el caso del CAFTA.” Mimeographed.

Devlin, Robert, and Antoni Estevadeordal. 2004. “Trade and Cooperation: ARegional Public Goods Approach.” In Regional Public Goods: From Theory toPractice, edited by Antoni Estevadeordal, Brian Frantz, and Tam R. Nguyen.Washington: Inter-American Development Bank and Asian Development Bank.

Duttagupta, Rupa, and Arvind Panagariya. 2001. “Free Trade Areas and Rules ofOrigin: Economics and Politics.” Working paper 03/229. Washington: Interna-tional Monetary Fund.

Estevadeordal, Antoni. 2000. “Negotiating Preferential Market Access: The Caseof the North American Free Trade Agreement.” Journal of World Trade 34(1):141–66.

Estevadeordal, Antoni, José Ernesto López-Córdova, and Kati Suominen. 2004.“Impact of NAFTA on the Location of Foreign Direct Investment in Mexico.”Washington: Inter-American Development Bank. Mimeographed.

Estevadeordal, Antoni, and Eric Miller. 2002. “Rules of Origin and the Pattern ofTrade between the U.S. and Canada.” Washington: Inter-American Develop-ment Bank, Integration, Trade and Hemispheric Issues Division. Mimeographed.

Estevadeordal, Antoni, and Kati Suominen. 2003. “Rules of Origin in FTAs inEurope and in the Americas: Issues and Implications for the EU-MERCOSURInter-Regional Association Agreement.” In Market Access for Goods and Ser-vices in the EU-Mercosur Negotiations, edited by Alfredo G. A. Valladão andRoberto Bouzas. Paris: Chaire Mercosur de Sciences Po. Also published as


“Regras de origem em áreas de livre comércio: implicações para negociações,”Revista Brasileira de Comercio Exterior (July/September 2004).

———. 2005a (forthcoming). “Rules of Origin: A World Map.” In The Origin ofGoods: A Conceptual and Empirical Assessment of Rules of Origin in PTAs,edited by Olivier Cadot and others. Oxford University Press and Centre forEconomic Policy Research.

———. 2005b. “What Are the Effects of Rules of Origin on Trade?” Washington:Inter-American Development Bank. Mimeographed.

Flatters, Frank, and Robert Kirk. 2005 (forthcoming). “Rules of Origin as Tools ofDevelopment? Some Lessons from SADC.” In The Origin of Goods: A Concep-tual and Empirical Assessment of Rules of Origin in PTAs, edited by Olivier Cadotand others. Oxford University Press and Centre for Economic Policy Research.

Garay, Luis, and Rafael Cornejo. 2002. “Metodología para el análisis de reg-imenes de origen: aplicación en el caso de las Américas.” INTAL-ITD-STAworking paper 8. Washington: Inter-American Development Bank.

Graham, Edward M., and Christopher Wilkie. 1998. “Regional Economic Agree-ments and Multinational Firms: The Investment Provisions of the NAFTA.” InGlobal Competitive Strategies in the New World Economy, edited by HafizMirza. Cheltenham: Edward Elgar.

Grossman, Gene M., and Elhanan Helpman. 1995. “The Politics of Free TradeAgreements.” American Economic Review 85(4): 667–90.

Hirsch, Moshe. 2002. “International Trade Law, Political Economy and Rules ofOrigin: A Plea for a Reform of the WTO Regime on Rules of Origin.” Journalof World Trade 36(2): 171–89.

Inama, Stefano. 2004. “Rules of Origin in GSP and ACP.” United Nations Con-ference on Trade and Development. Mimeographed.

James, William. 2004. “Rules of Origin, Tariff Discrimination and Trade Diver-sion: A Case Study of Asian Textiles and Apparel Exports.” Paper prepared forthe conference Rules of Origin in Regional Trade Agreements: Conceptualand Empirical Approaches. Inter-American Development Bank (Integrationand Regional Programs Department), Centre for Economic Policy Research,and DELTA/INRA, Washington, 20–21 February.

Jensen-Moran, Jeri. 1996. “Trade Battles as Investment Wars: The Coming Rulesof Origin Debate.” Washington Quarterly 19(1): 239–53.

Ju, Jiandong, and Kala Krishna. 1998. “Firm Behavior and Market Access in a FreeTrade Area with Rules of Origin.” Working paper 6857. Cambridge, Mass.:National Bureau of Economic Research.

Krishna, Kala, and Anne O. Kruger. 1995. “Implementing Free Trade Areas:Rules of Origin and Hidden Protection.” In New Directions in Trade Theory,edited by Alan Deardorff, James Levinsohn, and Robert Stern. University ofMichigan Press.

Krueger, Anne O. 1993. “Free Trade Agreements as Protectionist Devices: Rulesof Origin.” Working paper 4352. Cambridge, Mass.: National Bureau of Eco-nomic Research.


Reyna, Jimmie V. 1995. Passport to North American Trade: Rules of Origin andCustoms Procedures under NAFTA. Colorado Springs: Shepard’s/McGraw-Hill.

Rodriguez, Peter. 2001. “Rules of Origin with Multistage Production.” WorldEconomy 24(1): 210–20.

Sanguinetti, Pablo, and Eduardo Bianchi. 2005 (forthcoming). “ImplementingPreferential Trade Agreements in the Southern Cone Region of Latin America:Rules of Origin.” In The Origin of Goods: A Conceptual and Empirical Assess-ment of Rules of Origin in PTAs, edited by Olivier Cadot and others. OxfordUniversity Press and Centre for Economic Policy Research.

Suominen, Kati. 2003. “Selective Liberalization in Response to Globalization: Rulesof Origin as Determinants of Market Access Provisions in PTAs.” Integration andTrade 19(7): 153–85.

———. 2004. “Rules of Origin in Global Commerce.” Ph.D. Dissertation, Uni-versity of California at San Diego.

Wonnacott, Paul. 1996. “Beyond NAFTA—The Design of a Free Trade Agree-ment of the Americas.” In The Economics of Preferential Trading Agreements,edited by J. Bhagwati and A. Panagariya, pp. 79–107. Washington: AEI Press.

WTO (World Trade Organization). 2002a. “Coverage, Liberalization Process andTransitional Provisions in Regional Trade Agreements: Background Survey bythe Secretariat.” Working document WT/REG/W/46. Geneva: World TradeOrganization, Committee on Regional Trade Agreements.

———. 2002b. “Rules-of-Origin Regimes in Regional Trade Agreements: Back-ground Survey by the Secretariat.” Working document WT/REG/W/45. Geneva:World Trade Organization, Committee on Regional Trade Agreements.


Reforming Bank Capital Requirements:Implications of Basel II for

Latin American Countries

The appropriate regulation of banks is a hotly contested topic in bothindustrialized and developing countries. This year the Basel Commit-tee on Banking Supervision put forward a controversial proposal to

overhaul the 1988 Basel Accord that has long guided the regulation of bankcapital across over a hundred countries.1 The thirteen member countries ofthe Basel Committee on Banking Supervision are due to apply Basel II, asthe new accord is called, by 2007. If and how other countries should applyBasel II—and therefore whether the new Accord will be successful as astandard—remains an open question. If Basel II is applied across the globe,then its details will be extremely relevant; if not, it will be important to under-stand why many emerging countries decided to retain Basel I despite itswell-known drawbacks.2

105

G I O V A N N I M A J N O N IA N D R E W P O W E L L

Majnoni is with the World Bank; Powell is with the Universidad Torcuato Di Tella. Thispaper is part of a project on “Credit Information, Credit Risk Measurement, and SolvencyRatios in Emerging Countries” supported by a grant from the World Bank Research Com-mittee. We thank Margaret Miller and Nataliya Mylenko for their collaboration on this pro-ject. Andrew Powell gratefully acknowledges the World Bank, which he visited betweenAugust and December 2003, for financial support.

We are grateful to Matías Gutierrez from the Central Bank of Argentina, José LuisNegrín, Javier Márquez, and Alberto Romero from the Bank of Mexico, and RicardoSchectman from the Central Bank of Brazil for invaluable help in the quantitative analysis.We thank Jerry Caprio, Mark Carey, Patricia Correa, Charles Goodhart, Michael Gordy,Patrick Honohan, Patricia Jackson, Rafael Repullo, Roberto Steiner, Kostas Tsatsaronis,and Andrés Velasco for very helpful comments.

1. See the Financial Stability Forum website on financial standards (www.fsforum.org).2. Below we discuss the possibility that too many countries will adopt Basel II, in

which case its role as a standard in creating peer group pressure would have been too great!

It is widely accepted that bank capital should be regulated, but how to doso remains open to debate.3 The simple approach of Basel I divides assetsinto very broad risk categories and establishes an 8 percent minimum cap-ital requirement for risky assets.4 However, as bank risk management hasbecome more sophisticated and as the possibilities for transforming assetrisk have grown, the potential distortions created by these simple rules andthe opportunities for arbitraging across them have multiplied.5 By contrast,Basel II goes well beyond simply recasting quantitative requirements.Making capital requirements more risk sensitive and reducing regulatoryarbitrage are main objectives of the new accord.6 Basel II proposes twobasic approaches: the standardized approach, which uses external creditrating agencies together with a table that maps those ratings directly intocapital requirements; and the internal ratings-based (IRB) approach, inwhich the banks themselves estimate their customers’ default probability—without relying on external rating agencies—and then use a particular for-mula specified in Basel II to determine capital requirements as a function ofthe default probability and other parameters.7

This paper focuses on one specific but critical issue and on a set of moregeneral questions. We analyze whether the IRB approach as calibrated isappropriate for the Latin American context. We believe that this is the firstpaper to estimate credit risk across a set of emerging economies using a sim-ple and homogeneous methodology. We find significant differences betweenour estimates from the region and those from the Group of Ten (G10) coun-


3. Supporters see bank capital regulation as a response to the moral hazard of aninevitable public safety net for banks (see, for example, Mishkin, 2001; Goodhart andothers, 1999) or as part of a scheme to emulate the incentives for owners and managers infirms where debt holders are more sophisticated than bank depositors (see Dewatripont andTirole, 1994, for their representation hypothesis). Members of the free-banking school dis-agree with bank capital regulation (for excellent reviews, see Freixas and Rochet, 1999,pp. 260–65; Berger, Herring, and Szego, 1995).

4. Lower risk categories include mortgages, contingent facilities, short-term loans toother banks, and lending to members of the Organization for Economic Cooperation andDevelopment (OECD). See the original Basel Committee on Banking Supervision docu-ments and the literally hundreds of comments on Basel II at www.bis.org.

5. The standard criticism is that banks have incentives to sell or securitize assets forwhich capital requirements do not bind and buy assets when requirements would bind. Inthis way, banks would transform the risk on their balance sheets to ensure that capitalrequirements were always binding.

6. The new Accord has three pillars: quantitative requirements, supervisory review,and market discipline. While we focus on the first pillar, we briefly discuss the other pillarsin the next section.

7. See Basel Committee on Banking Supervision (2003).

tries. These differences have strong implications for the application of theIRB approach.

We also discuss Basel II implementation for Latin America and moregenerally across all emerging countries. Typically there are few externalrating agencies in these countries, so the standardized approach would havelittle effect in linking regulatory capital to risk.8 But the IRB approach maynot be calibrated appropriately for emerging markets, and its implementa-tion and supervision may stretch limited supervisory resources. Given thedata on compliance with the Basel Core Principles for Effective BankingSupervision across Latin America, it may not be advisable for many coun-tries to implement the IRB approach for a long time.

Given this situation, we suggest an innovative simplification of the IRBapproach that builds on current policies regarding provisioning in someemerging countries and that may be used as a transition arrangementtoward the IRB approach. We call this the centralized ratings-based (CRB)approach. Under this approach, banks would rate their clients, but the reg-ulator would determine the rating scale and the way in which the banks’ratings map into default probabilities. The use of a centralized scale wouldfacilitate comparison across banks and greatly ease the monitoring ofbanks’ ratings.9 Those requirements would also be easier to monitor, sincethe regulator would determine how banks’ ratings would feed into capitalrequirements.

Countries must choose whether to stay on Basel I or, if not, which BaselII alternative to apply (here we include our proposed CRB approach). Todate there is little guidance on this important decision. We therefore developa Basel II decision tree to assist countries deciding whether to adopt BaselII and, if so, how. Our broad advice is that many countries should stay onBasel I or only adopt Basel II for a subset of banks at least for several yearsbeyond 2007. Regulators should not move to complex rules too quickly sim-

Giovanni Majnoni and Andrew Powell 107

8. This may also be the case for smaller and regional banks in G10 countries. Thesebanks are unlikely to be systemic, however, whereas systemic banks in emerging countrieswill typically have mostly nonrated assets. See Ferri, Liu, and Majnoni (2001) for a discus-sion on the global pattern of ratings.

9. Bank ratings could be compared directly in the case of banks lending to the samecorporate client. Bank ratings for similar types of loans (to companies in the same economicsector, business line, or region) could also be compared and outliers investigated. Some G10regulators informally acknowledge that even where the IRB approach is likely to be employed,supervisors will compare banks’ internal ratings of important corporate clients (as they dotoday) and for that purpose will no doubt map ratings into a centralized scale.

ply because of peer group pressure or pressure from the large internationalbanks. We also argue that countries in the region should seriously considerthe CRB approach, and we suggest ways in which this may be made com-patible with Basel II for the purpose of assessing standards.10

The paper proceeds as follows. In the next section we provide a highlysynthetic account of the new Accord. The paper then introduces our method-ology for testing the calibration of the proposed requirements for the IRBapproach and also presents and discusses these results. We go on to con-sider broader questions regarding Basel II implementation. The closingsection concludes with a discussion of policy.

Basel II: A Synthetic Account

At first sight, the Basel II documentation is daunting. While the Accorditself is less than 300 pages, fully understanding those 300 pages requiresstudying several hundred pages of supporting documents. The Accordspecifies a set of new alternative approaches for minimum capital require-ments (Pillar 1: Quantitative Requirements), states how those requirementsshould be supervised (Pillar 2: Supervisory Review), and finally defines whatbanks should reveal to the market regarding the risk of their assets and how(including whether) they satisfy regulatory requirements (Pillar 3: MarketDiscipline). The idea is that the three pillars are complementary and mutu-ally reinforcing.11

The Basel Core Principles for Effective Banking Supervision alreadyencapsulate Basel I and much of the second pillar of Basel II.12 Moreover,for the simpler Basel II approaches, the third pillar (on market discipline)


10. Countries may be concerned about how banking regulations and supervision will beassessed by the International Monetary Fund and the World Bank, in the context of theFinancial Sector Assessment Programs (FSAPs).

11. The definition of capital has not changed from Basel I, but the new Accord includesimportant changes in the level of consolidation that banking supervisors should apply(scope of application) and for the first time introduces rules on lending to affiliated compa-nies (related lending).

12. The Core Principles refer to supervision. They cover what banks should report to thesupervisor, but not what banks should disclose to the market (Basel II, pillar 3). Strictlyspeaking, following Basel I is neither necessary nor sufficient for a country to be compliantwith the sixth Core Principle (on capital adequacy). In practice, however, Basel I is nor-mally considered a necessary condition, and the Financial Stability Forum deems it one ofthe critical financial standards that countries should implement (see www.fsforum.org).

is reduced to the bank’s obligation to publish its capital requirement andits actual level of capital. Thus what is really new in Basel II are the vari-ous Pillar 1 alternatives regarding actual capital requirements. We there-fore focus on the first pillar in this paper.13

Pillar 1 contains three main approaches: the simplified standardizedapproach, the standardized approach, and the internal ratings-based (IRB)approach, which breaks down into two options (the foundation IRB andthe advanced IRB approaches). Pillar 1 also covers alternatives for the mea-surement of basic credit risk, credit risk mitigation techniques, securitiza-tion risk, and operational risk.14 With regard to the first three alternatives,Basel II attempts to improve on the treatment in Basel I; in the case of thefourth alternative, this is the first time that an actual quantitative require-ment for operational risk has been included in the Basel recommendations.The alternatives are illustrated in table 1. In practice, a relatively simpleapproach for underlying credit risk assessment would normally be com-bined with simple approaches for the other topics. The simplified standard-ized approach, for example, explicitly combines the simplest approachesfor credit risk evaluation, credit risk mitigation, securitization risk, andoperational risk. At the other extreme, the advanced internal ratings-basedapproach would normally be accompanied by advanced approaches else-where, in particular the advanced measurement approach for operational risk.

One potentially important issue for emerging economies is that the newcapital requirements are calibrated so that, on average, the capital require-ment for a standard G10 bank would remain around 8 percent under thestandardized approach, and capital requirements for an average G10 bankwould fall under the IRB approach. This implies that under the standard-ized approach, the increase in the new requirement for operational riskwould be roughly offset by the reduction in requirements for credit risk,given the ratings distribution in a typical G10 country. In the case of adeveloping country with low ratings penetration, the proportion of unratedclaims on bank balance sheets is likely to be much higher than for G10


13. This does not imply that the second pillar is unimportant. Compliance with the CorePrinciples is weak in developing countries, and the second pillar’s tighter definitions onaspects of the supervisory process highlight the importance of making progress in these areas.

14. Credit risk mitigation techniques mainly refer to contracts that use securities asguarantees, such as repurchase agreements (repos) and credit derivatives; they do not referto real guarantees such as mortgages, for which there are rules under basic credit risk eval-uation. Securitization risk covers both investment in a securitized instrument and theretained risk of originating a securitization of assets on a bank portfolio.

banks, and the distribution of rated claims will probably also be different.15

This implies that the operational risk requirement may approximate a sim-ple add-on, increasing overall capital requirements. While this may not beundesirable, it may be an impediment to the implementation of Basel II insome countries.


15. Local supervisors may also employ “local ratings” (that is, ratings conductedaccording to a national or local scale rather than an international scale). However, the threemajor rating agencies (Fitch, Moody’s, and Standard and Poor’s) all warn customers thatlocal ratings are not necessarily comparable across countries. This raises an important issuefor the use of the standardized approach as a financial standard.

T A B L E 1 . Different Options Proposed in the First Pillar of Basel II

Basic credit risk Approach measurement technique Credit risk mitigation Securitization risks Operational risk

Simplifiedstandardized

Standardized

Internalratings-based

Foundation

Advanced

Export credit agencies(www.oecd.org,trade directorate, ECApage)

Export credit agencies orcredit rating agencies(such as S&P,Moody’s, Fitch)

Banks’ internal ratingsfor defaultprobability and BaselII formula sets capitalrequirement (lossgiven default 45% forsenior and 75% forsubord).

Banks set internal rating(default probability),Loss given defaultexposure at defaultand maturity. Capitalrequirement stillgiven by Baselformula.

Simple: risk weight ofcollateral substitutesthat of claim

Simple: same as simplifiedstandardized approach.

Comprehensive: exposureamount reducedsubject to claim andcollateral haircuts.

Comprehensive, loss givendefault adjusted givenreduction in exposureand capitalrequirement given byBasel formula.

Own model determinesLoss given default andexposure at default;capital requirementgiven by formula.

SSA banks can onlyinvest (cannot offerenhancements orliquidity facilities).Risk weight = 100percent.

Standardized: usesexport credit agencyratings (onlyinvesting banks canuse below BB+)

IRB approach: Investingbanks may use bankratings according to astandard scale.Originators may usesupervisory formula.

As for Foundation IRBapproach.

Basic indicator:Capital = 15% gross

income

Basic indicator, orstandardizedapproach wherebank capital =weighted sum ofgross incomeacross activities.

More sophisticatedbanks will beexpected tograduate to theadvancedmeasurementapproach wherecapital require-ment is given byown risk measure-ment system.

As for Foundation IRBapproach.

This paper focuses on Pillar 1—namely, the measurement of underlyingcredit risk. Given the fundamental problem of asymmetric informationbetween the regulator and the regulated institution, banks will generallyhave superior measures of clients’ risks than will the regulator. However,if the underlying motive for regulation is moral hazard, then it is clearlyproblematic to allow the bank itself to use its own estimates of client riskor its own assessment of a portfolio of such risks.16 The solution to thisconundrum in the new Accord is to use either external credit rating assess-ments or bank ratings subject to supervision of the rating methodology anda specific formula that maps those ratings into capital requirements.

Since rating penetration is typically low in developing countries, thestandardized approach will buy very little in terms of linking regulatorycapital to risk. This argument implies that many regulators may well beinterested in considering the IRB approach or our proposed CRB approach.Countries may also adopt a mixed approach whereby some banks remainon Basel I or adopt the standardized approach while more sophisticated orlarger banks adopt the IRB approach. The United States may providesomething of a model in which very few banks will be forced to adopt theIRB approach, some others may be permitted to adopt the IRB model, andthe vast majority of banks will remain on Basel I. When we translate theU.S. model to an emerging market, we find that if the regulator’s aim is forregulatory capital in the financial system as a whole to reflect risk moreclosely, then the proportion of assets that will be covered by the IRBapproach will be relatively large (that is, the banks that adopt the IRB modelare likely to be few in number but large in size). These arguments suggestthat the calibration of the IRB approach is an important issue for emergingeconomies considering whether and how to adopt Basel II. This is the topicof the next section.

On Basel II Calibration: Methodology

The new capital Accord’s internal ratings-based approach suggests a for-mula for calculating a bank’s capital requirement as a function of threebasic variables: default probability, exposure at default, and loss given


16. Assessing credit portfolio risk implies assumptions not only on individual defaultprobabilities, but also on the multivariate distribution of those default probabilities. For sim-pler assumptions on distributions, this implies assumptions on the mean and variance-covariance matrix of default probabilities.

default.17 The expected loss for a single claim is the multiplication of thethree variables if expressed in appropriate units. However, the provisionsa bank holds are typically identified with expected losses, and capital isidentified with a value-at-risk (VaR) concept. A regulator might then aska bank to hold provisions and capital to cover a specified percentage of thedistribution of losses to ensure the continued solvency of the bank exceptin highly extraordinary circumstances.

This is illustrated in figure 1, where the segment OB represents the totalvalue at risk associated with a specific probability of occurrence, that is,the level of losses that corresponds to a certain percentile of the distribu-tion function of expected losses (99 percent in the figure). A regulatormight then ask a bank to hold general provisions to cover the expected lossrepresented by the segment OA, where A is the mean of the loss distribu-tion, and to hold capital to cover the unexpected loss represented by thesegment AB (that is, the difference between the total value at risk definedwith respect to a particular percentile of the distribution and the mean orexpected loss). The calibration of the Basel II IRB formula employs a valueat risk of 99.9 percent with a horizon of one year. Hence a bank is onlyexpected to use up its capital in one year with a probability of 0.1 percent,or once every 1,000 years.18

The inputs for the Basel II IRB formula are the parameters for a singleloan or claim, and the output is the capital requirement for that instru-ment. Subject to an underlying assumption regarding the correlation ofasset risks, this single-instrument approach should approximate the resultof a portfolio credit risk model. An econometric methodology is typicallyemployed to estimate individual instrument default probabilities, and theseestimates are then fed into a model with other parameter estimates to obtainthe loss distribution curve for the portfolio.19 Simplifying assumptions areemployed in both estimating the parameters and developing the model.

Commonly used models include Moody’s KMV option-based model,the McKinsey macroeconomic simulation model, the CreditMetrics model


17. See Altman and Saunders (1997) for a useful discussion.18. This assumes that draws from the distribution are independent over time.19. This may be for a particular bank or a specific business line of a bank. As we directly

mimic the portfolio of a bank, we do not discuss further the important issue of aggregation.Suffice to say that the Basel II IRB formula is by business line (sovereign, commercial, andretail) and is calibrated with particular assumptions regarding asset correlations in each sec-tor. The results are simply added, implying an assumed perfect positive correlation betweenbusiness lines.

from J. P. Morgan’s RiskMetrics division, and the Credit Suisse FirstBoston’s CreditRisk+ model. The latter of these models is arguably thesimplest to implement, but even here implementation relies on key addi-tional assumptions such as the number of risk factors, the estimation offactor volatilities and loadings, and the correlation of default probabili-ties.20 Moreover, the estimation of each model relies on a set of quite spe-cific data requirements and assumptions that make cross-country and evencross-institution comparisons problematic. These model-based methodolo-gies of estimating the credit loss distributions of a loan portfolio are thussubject to both estimation risk of the parameters of a single instrument andmodel risk (in that the assumptions of the portfolio model may be incor-rect). Furthermore, the Basel II IRB formula introduces approximationrisk. Capital requirements are calculated on each single instrument modeland then they are simply added across all instruments. The aggregated sin-gle instrument formula yields the appropriate loss distribution for the port-folio only for a particular correlation of risks between instruments. Ifactual correlations differ substantially from this assumption, then thisapproximation to the risk of a portfolio may cease to be valid.


20. See Balzarotti, Falkenheim, and Powell (2002) and Balzarotti, Castro, and Powell(2004) on the implementation of CreditRisk+ in Argentina. See Márquez and others (2003)for the case of Mexico and Foglia (2003) for a discussion and model-based estimates ofcredit risk using Italian credit registry data.

O A Loss B

Prob

abili

ty

99th percentile

Expected losses

CapitalGeneralprovisions

Unexpected losses

Figure 1. Loan Loss Probability Distribution Stemming from Credit Risk

In this paper we take a quite different approach: we adopt a bootstrap-ping technique that essentially enables us to mimic the shape of the lossdistribution function of any specified loan portfolio.21 This approach mini-mizes the impact of estimation errors and maximizes the degree of compa-rability across countries. Even without an underlying model, bootstrappingtechniques can be used to simulate the frequency distribution of creditlosses. The resampling approach is very flexible and lends itself to manyalternative simulation exercises aimed, for example, at measuring the expo-sure to credit losses of portfolios characterized by different loan sizes,maturities, ratings, geographic locations, or economic sectors.

The empirical exercises performed in this paper are for Argentina, Brazil,and Mexico. In each of the three countries, the central bank maintains apublic credit registry that contains information on a very large number ofloans in the financial system.22 Each financial system requires a clearamount of capital plus provisions every year to confront total credit losses.However, that observation tells us little about the required capital and pro-visions for an average bank in that year. Conditional on the overall macro-economic conditions, the losses suffered by an average bank depend onthe sensitivity of the bank’s loan portfolio performance to the prevailingeconomic conditions and the idiosyncratic risk of the portfolio. The BaselII IRB approach assumes that the correlation structure of a bank portfoliois known and summarizes credit loss correlations as sensitivities to a sin-gle factor, but credit risk correlations are not known with certainty and asingle factor model can at best be thought of as an approximation to a morecomplex reality.

The technique we employ generates conditional loss distribution func-tions based on overall economic performance, the correlation of credit losses,and any residual idiosyncratic risks in a large number of sample portfolios.We then use these distributions to measure the expected and unexpectedlosses. In other words, conditional on the overall performance of the finan-cial system over the period of analysis, our results provide a measure ofthe level of expected and unexpected losses of a bank of average size with


21. Here we are following Carey (2002). Also see Carey (1998) for further analysis ofcredit risk in G10 portfolios.

22. See Miller (2003) for details on public credit registries around the world; see Powelland others (2004) for an empirical analysis of the value of public credit registries inArgentina, Brazil, and Mexico and a discussion of their use for predicting credit losses.

a loan portfolio randomly drawn from the universe of loans within thefinancial system.23

We limit our empirical analysis to only one specific year owing to changesin definitions, the scope of coverage, and data quality across the credit reg-isters from the three countries. Our findings should therefore be regardedas illustrative of a methodology that needs to be repeated over severalyears to achieve its full empirical relevance. This point is shown in greaterdetail in figure 2, which shows a sequence of conditional distributions esti-mated at different points in time (namely, t1, t2, and t3, which represent asequence of good-bad-good years over a hypothetical economic cycle) andthe unconditional distribution resulting from pooling the data from all theconditional distributions. Our estimates reflect the events of the chosenyear and thus cannot be taken as representative of the unconditional dis-tribution. However, we chose a year close to the cyclical trough for eachcountry (the period t2 in the characterization of figure 2), so our estimatesmay properly reflect those observations that carry a greater weight in shap-ing the right tail of the unconditional distribution of credit losses. Forinstance, the Argentine data are for 2001. In that year, a recession led to afall of GDP equal to 4.4 percent in real terms and a deepening crisis. Bankdeposits were frozen in December 2001, and there was considerable eco-nomic and political uncertainty that resulted in the removal of the presi-dent amid riots. Over the same period, Mexico experienced a stagnation ofeconomic activity with zero GDP growth and a reduction of the ratio ofbank credit to GDP to 11.9 percent, the lowest value of the last decade. InBrazil, a slowdown in economic activity brought GDP growth down toonly 1.3 percent and led to a contraction of bank credit in real terms. Tosummarize, while the results naturally reflect a period in time, the snapshotcaptures economic stress in all three countries.


23. Our one-point-in-time distributions might be thought of as distributions across idio-syncratic risk or, alternatively, of distributions of correlations of asset risk within our sam-ple portfolios. For a one-factor model, the systemic risk of a portfolio might be approximatedby average correlations as portfolio size increases. In practice, however, asset correlationsmay differ substantially across bank portfolios if asset correlations depend on many factors,including sector, loan, and borrower characteristics, and if portfolios are lumpy in terms oftheir exposures across these factors. Indeed, one common explanation of why one bank mayfail during a recession whereas another does not is based on differences in exposure to sys-temic factors rather than pure idiosyncratic risk. The Basel II IRB formula assumes thatthere is a single systemic factor, that bank portfolios have zero idiosyncratic risk, and thatasset correlations are identical for companies of the same size (and decrease with companysize), but correlations are always assumed to be known and stable.

The first step of the procedure consists of extracting from the publiccredit registry a large pool of performing loans to the nonfinancial corpo-rate sector at a particular date. This pool reflects the overall risks of lend-ing to the corporate sector in each particular country. Second, we definedefault as the event of payments that are over ninety days past due.24 Third,we classify loans into two categories according to whether they maintaintheir initial status or default over the following twelve months. Fourth,from this pool of loans, we randomly sample a predefined number of loans(in our case 500), intended to mimic the loan portfolio of a medium-sizedbank.25 Given a predefined recovery ratio, we compute the value of thelosses of the sampled portfolio, expressing this as a fraction of the facevalue.26 Fifth, we replicate the last step a large number of times (20,000 in


24. This follows the typical definition of a nonperforming loan according to interna-tional best practices and to one of the criteria set out by the Basel Committee on BankingSupervision.

25. Modifying the sampling procedures would enable the selection of predefined riskprofiles and the analysis of different risks embedded in bank portfolios (see Carey, 2002).

26. We assume a predefined recovery ratio of 50 percent of the face value of a defaultedloan. The Basel Committee on Banking Supervision employs a loss given default of 45 per-cent for the foundation IRB approach. Our conversations with the central banks indicatedthat this might be somewhat low for emerging markets, so we selected the figure of 50 percentfor the purposes of our calibration exercises. The bootstrapping methodology would be con-siderably more precise if data on loss given default were available at the individual loan level.

Figure 2. Conditional and Unconditional Credit Loss Distributions

Conditional Unconditional

Time

EL1

EL2

EL3

Losses(%)

UL1

UL2

UL3

Frequency

EL

O O t1 t2 t3

UL

this case) to generate a frequency distribution of credit losses. This fre-quency distribution simulates the actual distribution of credit losses facedby banks in that financial system at that time. Finally, we use the frequencydistribution of credit losses to calculate a number of statistics in a reason-ably homogeneous way across countries. In particular, we calculate theexpected and unexpected losses up to different statistical tolerance valuesacross the different portfolios. We compare these estimates in each coun-try against current regulations (inspired by Basel I), the actual provisionsand capital of banks, and a simulated capital requirement using estimateddefault probabilities and following the Basel II IRB formula.

Having described the methodology, we reiterate the caveat that theresults reflect a snapshot of a particular country in a particular year. Theresults cannot and should not be interpreted as average values represent-ing credit risk exposure over different time horizons or over the full busi-ness cycle. Moreover, the results are dependent on the universe of loanscollected by the public credit registry. For Argentina and Mexico, we areconfident that this universe of commercial loans is representative of thefinancial system as a whole, but in the case of Brazil, we could only accessthe universe of larger corporate loans, for reasons explained below.27 Ourresults should thus be interpreted as tentative and conditional on the timeand loan universes obtained. Nevertheless, they are highly suggestive, tend toback up other evidence, and could provide the catalyst for similar studiesin other countries or in the same countries over longer time periods.

Calibrating Basel II for Emerging Countries: The Results

This section first discusses the data that we use for the analysis and thendetails the results of the bootstrapping sampling methodology describedabove. Finally, we draw the main implications of our results for emergingeconomies.

Description of the Data

Figures 3 to 5 illustrate, for the three countries, the frequency distributionof three variables involved in the experiment. Panel A reports the frequencydistribution of the size of individual loans, which are extracted from the pub-


27. The Basel II formula was calibrated on commercial loans, so we feel that this choiceis appropriate.


Figure 3. Argentina: Relevant Frequency Distributions from the Resampling Exercise

Source: Central Bank of the Argentine Republic.

A. Frequency distribution of bank loan size

0

10

20

30

40

50

60

21171

321471

621771

9211,071

1,2211,371

1,5211,671

1,8211,971

2,1212,271

2,421

Exposure size (thousands of U.S. dollars)

Freq

uenc

y (pe

rcen

t)Fr

eque

ncy (

perc

ent)

Freq

uenc

y (pe

rcen

t)

B. Frequency distribution of 20,000 sampled portfolios

0.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

73116

159202

245289

332375

418462

505548

591634

678721

764807

851894

937

Portfolio size (millions of U.S. dollars)

C. Frequency distribution of credit losses

0.00.51.01.52.02.53.03.54.04.5

0.000.02

0.030.05

0.060.08

0.090.10

0.120.13

0.150.16

0.180.19

0.200.22

0.230.25

0.260.28

0.29

Portfolio losses (percent)

Figure 4. Brazil: Relevant Frequency Distributions from the Resampling Exercise


Source: Central Bank of Brazil.

A. Frequency distribution of bank loan size

0

5

10

15

20

25

30

37 148

259 370

481 592

703 814

925 1,036

1,147

1,258

1,369

1,480

1,591

1,702

1,813


Freq

uenc

y (pe

rcen

t)

B. Frequency distribution of 20,000 sampled portfolios

0. 0

0. 5

1. 0

1. 5

2. 0

2. 5

3. 0

275 334

392 451

509 568

626 685

743 802

860 919

977

1,036

1,094

1,153 1,211

1,270

Portfolio size (milions of U.S. dollars)

Freq

uenc

y (pe

rcen

t)

C. Frequency distribution of credit losses

0. 0

0. 5

1. 0

1. 5

2. 0

2. 5

3. 0

0.781.70

2.623.53

4.455.37

6.297.21

8.139.05

9.9610.88

11.8012.72

13.6414.56

15.4816.39


Freq

uenc

y (pe

rcen

t)


Source: Bank of Mexico.

A. Frequency distribution of bank loan size: Large banks

05

1015202530354045

16 114

212 310

408 506

604 702

800 898

996

1,094

1,192

1,290

1,388

1,486

1,584

Freq

uenc

y (pe

rcen

t)


B. Frequency distribution of 20,000 sampled portfolios: Large banks

05

10152025303540

26 95

164 233

302 371

439 508

577 646

715 784

853 922

991

1,060

1,129

Freq

uenc

y (pe

rcen

t)Fr

eque

ncy (

perc

ent)

Portfolio size (millions of U.S. dollars)

05

10152025303540

C. Frequency distribution of credit losses: Large banks

0.4 3.0 5.5 8.0 10.613.1

15.618.2

20.723.3

25.828.3

30.933.4

35.938.5

41.0


Figure 5. Mexico: Relevant Frequency Distributions from the Resampling Exercise

lic credit registries and which represent the universe from which our sam-ples of 20,000 loans are drawn; panel B shows the frequency distributionof the dollar value of the 20,000 randomly selected portfolios; and panelC shows the distribution of credit losses of the 20,000 randomly selectedportfolios as a fraction of the face value of the respective portfolios.

A visual inspection of the charts shows clear differences among thethree samples. Mexico has the lowest concentration in terms of loan size,and about 80 per cent of all the loans extracted from the credit register aresmaller than U.S.$100,000. The same figure for Argentina is about 60 per-cent. We adopted a different sampling procedure for Brazil because of thehuge size of the credit market and following advice from the Central Bank.Specifically, we included only those companies whose gross exposureswith the financial system were above U.S.$300,000, and hence the onlysmaller loans included are those to companies that had other larger loansoutstanding (about 40 percent of loans were less than U.S.$100,000). Wewere advised that this sampling methodology would capture the majorcredit risks in the Brazilian financial system. The charts for Argentina andMexico thus show considerably more skewed distributions for both thevalue and credit losses of the portfolios than the comparable distributioncomputed for Brazil. As a supplement to the visual information providedby the charts, table 2 summarizes a set of descriptive statistics of the dis-tribution of the 20,000 randomly sampled portfolios.


T A B L E 2 . Simulated Loan Portfolios: Descriptive Statisticsa

Millions of U.S. dollars

No. StandardCountry and period observations Mean Median Mode deviation Minimum Maximum

Argentina (Dec 2000 to Dec 2001) 70,017 242 215 182 104 72 943Brazil (Oct 2001 to Oct 2002) 41,784 551 538 510 110 275 1,306Mexico (Dec 2000 to Dec 2001) 188,165 85 62 33 89 16 1,477

Source: Authors’ calculations.a. Data refer to the twelve-month period indicated for each country. The number of observations are the number of bank loans to

nonfinancial entities above a minimum amount, which were extracted from the national credit registers of each country at the begin-ning of the twelve-month period. The criteria underlying the selection of loans from the credit registers differ slightly across countries.For Argentina and Mexico, a loan refers to the overall position of a single borrower with the banking system as a whole. For Argentina,the positions selected are those equal to or larger than U.S.$21,000. For Mexico, we include both loans equal to or larger thanU.S.$20,000 (which are reported on a compulsory basis) and smaller loans that are reported on a voluntary basis. For Brazil, the mini-mum size is U.S.$300,000, but positions with different banks that concur to define the total exposure are treated as distinct individualloans. The descriptive statistics (mean, median, mode, standard deviation, minimum, and maximum) refer to the distribution of thevalue of the 20,000 portfolio of 500 loans each, randomly sampled from the pool of loans described above. Exact sources and definitionsof each variable can be found in the main text.

The Main Results

Table 3 presents the main results of the bootstrapping exercise for eachcountry, including the value of expected losses and the value of unex-pected losses associated with different percentile levels of the right tail ofthe simulated distribution of credit losses. Here we refer to the total losses(equal to the sum of the expected and unexpected components) as the valueat risk. For the three countries considered, expected losses proved to be inan interval between 1 percent and 5 percent. If we assume provisions to coverthis amount, the amount of capital necessary to provide protection for unex-pected losses up to 99 percent of the distribution is about 15 percent forboth Argentina and Mexico. The capital that would be required to cover99.9 percent of the distribution is significantly higher, at 21 percent forArgentina and 31 percent for Mexico. In the case of Brazil, the results sug-gest that capital of just over 6 percent is required for the 99 percent confi-dence limit and 10.5 percent for the 99.9 percent confidence value.28


28. These lower values may reflect the fact that the loan size for Brazil is significantlygreater than that for Argentina and Mexico.

Table 3. Capital and Provisions: Unexpected and Expected Losses Estimations Based onSimulation Resultsa

Percent

Default ExpectedUnexpected loss

Country and period probability loss 95% 99% 99.9% Basel (Jan 2004)

Argentina (Dec 2000 to Dec 2001) 9.60 4.80 7.30 14.80 21.80 14.93Brazil (Oct 2001 to Oct 2002) 8.32 4.16 3.51 6.07 10.46 14.15Mexico (Dec 2000 to Dec 2001) 2.70 1.35 4.44 16.58 31.64 9.68United States (1989–91)b 3.00 1.50 1.62 2.55 3.91 10.07United States (1929)b 6.24 3.12 2.54 3.80 5.36 12.70

Source: Authors’ calculations and Carey (2002).a. Expected loss is given by the mean value of the simulated distribution of credit losses. Simulations are based on the random

extraction without replacement of 500 loans from the pool of loans registered in the credit register of each country, to simulate a stan-dard bank loan portfolio. The extraction is repeated 20,000 times (this time with replacement) to obtain 20,000 portfolios. The distri-bution of credit losses for each portfolio provides the 20,000 observations used to simulate the distribution of credit losses. Unexpectedlosses at different levels of probability represent the value of credit losses (as a percentage of the face value of the portfolio) corre-sponding to the percentile on the right tail of the distribution minus the expected loss given by the mean value of the distribution. Baselunexpected losses indicates the value of unexpected losses computed according to the algorithm proposed by the Basel Committee onBanking Supervision (2004b). The value of default probabilities used in the algorithm is given by twice the value of the expected loss,assuming the same 50 percent loss given default used in the simulation exercise.

b. Results for the United States are from Carey (2002), who simulates credit losses based on a loan portfolio that mimics the riskexposure of a medium-sized bank in the United States and the default probability values observed at times of systemic distress, such asthe moderate recession of 1989–91 and the severe 1929 recession.

The results for Brazil reflect the fact that the category of commercialloans is restricted to borrowers with consolidated borrowing of more thanU.S.$300,000. To test the effect of this different definition of the loan cat-egory, we report a summary of the results for the Argentine bootstrappingwith the same restriction as Brazil (see table 4). In this case, the capital thatwould be required for Argentine banks is reduced to 5.6 percent (for 99 per-cent unexpected losses) or 8 percent (for 99.9 percent unexpected losses),which is actually somewhat lower than the estimated requirements forBrazil. These results suggest that Brazil’s risks are roughly in line withthose of Argentina when we take into account the different definition ofthe loan universe. They also call into question the adjustment made forlending to small and medium-sized enterprises in Basel II, which reducescapital requirements for this sector. In other words, the reduction in requiredcapital, which reflects the additional diversification of risks, appears to bemore than outweighed by increased default probabilities in our sample ofemerging economies.

We compared our results with the level of capital that would be gener-ated using the estimated probability of default and the formula proposedby the Basel Committee for the foundation IRB approach.29 The Basel for-


29. We used the formula for assessing the capital requirement for the corporate portfo-lio as described in the Basel Committee’s third consultative paper (Basel Committee onBanking Supervision, 2003) and revised in January 2004 (Basel Committee on BankingSupervision, 2004b).

Table 4. Capital and Provisions in Argentina: Unexpected and Expected Losses EstimationsBased on Loan Universes with Different Minimum Exposure Thresholds

Percent

Minimum size of individual Default Expected Unexpected loss

borrower’s loan exposure probability loss 95% 99% 99.9%

U.S.$20,000 9.6 4.8 7.3 14.8 21.8U.S.$300,000 8.2 4.1 3.6 5.6 8.0

Source: Authors’ calculations.a. Expected loss is given by the mean value of the simulated distribution of credit losses. Simulations are based on the random

extraction without replacement of 500 loans from the pool of loans registered in the credit register of each country, to simulate a stan-dard bank loan portfolio. The extraction is repeated 20,000 times (this time with replacement) to obtain 20,000 portfolios. The distri-bution of credit losses for each portfolio provides the 20,000 observations used to simulate the distribution of credit losses. Unexpectedlosses at different levels of probability represent the value of credit losses (as a percentage of the face value of the portfolio) corre-sponding to the percentile on the right tail of the distribution minus the expected loss given by the mean value of the distribution. Baselunexpected losses indicates the value of unexpected losses computed according to the algorithm proposed by the Basel Committee onBanking Supervision (2004b). The value of default probabilities used in the algorithm is given by twice the value of the expected loss,assuming the same 50 percent loss given default used in the simulation exercise.

mula applied to simulated default probabilities generates capital require-ments of 14.9 percent, 14.1 percent, and 9.7 percent for Argentina, Brazil,and Mexico, respectively (see the last column of table 3). These are con-siderably lower than our simulations at the 99.9 percent level of confidencefor Argentina and Mexico and higher than the 11.5 percent requirementthat we computed for Brazil.

We also compared our results to those of a similar exercise conductedby Carey, which was intended to mimic the risk exposure of a representa-tive U.S. bank in the period 1989–91 and also in a period of high stress forthe financial system (namely, 1929).30 These results are included in table 3.The expected loss for a U.S. bank in 1929 was about 3.1 percent, whichfalls between our estimates of 1.4 percent for Mexico and 4.2 percent forArgentina and Brazil. However, the estimates of unexpected loss for theUnited States in 1929 are significantly below our estimates for the threeemerging countries at each statistical confidence level. Finally, our resultscan also be compared to the capital requirements generated by the Baselformula, using a one-year default probability appropriate to the Standardand Poor’s sovereign rating in domestic currency (see table 5).31 The resultsfor expected and unexpected losses are also given. Table 6 provides detailson the domestic currency sovereign ratings across the whole region, aswell as the simulated Basel II IRB capital requirements.


30. See Carey (2002).31. The default probabilities are estimated by Standard & Poor’s based on their histor-

ical data of defaults by rating category, including corporate claims.

Table 5. Capital and Provisions: Unexpected and Expected Losses Estimations Based onStandard and Poor’s Observed Average Default Frequency

Percent

Basel unexpectedCountry and period Rating Default probability Expected loss loss (Jan 2004)

Argentina (Dec 2000 to Dec 2001) CCC 27.87 13.94 19.57Brazil (Oct 2001 to Oct 2002) BB+ 1.38 0.69 8.07Mexico (Dec 2000 to Dec 2001) BBB+ 0.37 0.19 4.75

Source: Authors’ calculations, based on data from Standard and Poor’s.a. Rating represents the lowest value of domestic currency sovereign rating expressed by Standard and Poor’s over the time period

considered. This is only partially true in the case of Argentina, where we have conventionally selected a rating of CCC, although formalrating was suspended in November on the eve of the government default. The default probability is computed on the basis of thehistorical average one-year default frequency on Standard and Poor’s–rated corporate bonds; loss given default is assumed equal to50 percent. Basel unexpected loss refers to the value of unexpected loss computed according to the algorithm proposed by the BaselCommittee on Banking Supervision (2004b) and using the default probability value corresponding to the value of expected losses overloss given default.

Three conclusions can be drawn from the results. The first is that theBasel formula, applied to Argentina and Mexico, generates levels of pro-tection inferior to the advertised 99.9 percent.32 In the case of Brazil, ourresults indicate the opposite, but they are affected by the different universeof loans used. As the IRB approach is currently calibrated, the degree ofprotection would be in the range of 95–99 percent of the credit loss distri-bution for Argentina and Mexico. Another way to state this result is thatachieving the advocated 99.9 percent level of protection would requiresubstantially higher capital requirements than those advocated in Basel II.33

Moreover, the fact that different levels of capital are required to achievethe level of protection theoretically granted by the Basel IRB curve callsinto question the curve calibration. In other words, given the types of default


32. This conclusion is conditional on events in the year chosen for the analysis.33. The Basel II IRB formula in fact calculates assets at risk, and capital requirements

are defined as 8 percent of those assets at risk. Assets at risk are then 12.5 times requiredcapital. Our results may be interpreted as saying that assets at risk must be larger (main-taining the 8 percent capital requirement) or that the 8 percent should be increased toachieve 99.9 percent protection. This discussion also assumes that provisions cover expectedlosses.

Table 6. Capital and Provisions in Latin American Countries Based on Standard andPoor’s Ratings

Percent

Standard and Expected Basel unexpected Sum of expected andPoor’s classification Country loss loss (Jan 2004) unexpected losses

AAA - 0.00 0.34 0.34AA Chile 0.00 0.61 0.62A Barbados, Mexico, and Trinidad

and Tobago 0.02 1.56 1.59BBB Colombia 0.17 4.75 4.91BB Belize, Brazil, Costa Rica, El Salvador,

Guatemala, Panama, and Peru 0.62 8.07 8.69B Bolivia, Jamaica, Suriname, Uruguay,

and Venezuela 2.71 12.54 15.25CCC Ecuador, Paraguay, and Dominican

Republic 12.54 19.57 32.11

Source: Authors’ calculations.a. Country classification refers to Standard and Poor’s domestic currency sovereign rating as of December 2003. The expected loss is

the product of the default probability and the loss given default. The default probability is computed on the basis of the historical aver-age one-year default frequency on Standard and Poor’s–rated corporate bonds; the loss given default is assumed equal to 50 percent.Basel unexpected loss refers to the value of unexpected losses computed according to the algorithm proposed by the Basel Committeeon Banking Supervision (2004b) and using the default probability value corresponding to the value of expected losses over loss givendefault.

probabilities in emerging countries, it would not necessarily be appropriateto apply the IRB curve, as written, to individual instruments to ensure agood approximation to portfolio risk.

The second conclusion is that emerging countries face a difficult choiceif they wish to apply the IRB approach. They may either implement theIRB curve as it is written and hence very likely opt for a lower degree ofprotection than that envisaged by the Basel Committee on Banking Super-vision, or they must attempt to recalibrate the curve to obtain a degree ofprotection closer to the 99.9 percent as suggested by the Basel Committee.This should not come as a surprise, since the same issue is present withBasel I. Many countries adopted the Basel I methodology, but they applieda higher minimum than the recommended 8 percent.34

A third conclusion of our results is that for emerging countries, thefoundation IRB level of capital requirement is (notwithstanding its benev-olent risk calibration) likely to give higher capital requirements than theexisting 8 percent minimum of the Accord. In the case of Argentina, thefoundation IRB approach gives a requirement of about 15 percent, but infact it is close to actual capital requirements in Argentina.35 In Mexico andBrazil, the foundation IRB approach would, according to our simulations,yield requirements of around 10 percent and 14 percent—higher thanBasel’s 8 percent and higher than current levels in both Mexico (8 percent)and Brazil (11 percent).36

In the discussion above we compared the foundation IRB approachwith the actual default experience to measure the default probabilities. Weobtain much lower figures, however, when we use the Standard and Poor’shistorical mappings of ratings to default probabilities and the relevant sov-ereign rating. Rating agencies typically interpret such a rating as the floorto nongovernmental ratings, and therefore the associated default probabil-


34. A country could simply state that capital requirements are, for example, 10 percentof Basel II IRB-calculated assets at risk, but there seems little point in adopting a relativelysophisticated formula and then applying an ad hoc adjustment without considering whatthe effect of that adjustment would be on the level of protection within, say, a value-at-riskframework.

35. Argentina had a baseline 11.5 percent capital requirement, but various add-onsimplied that the overall requirement was close to 15 percent of assets at risk as calculatedunder Basel I.

36. Moreover, this does not take into account the additional operational risk capitalrequirement (set equal to 15 percent of gross income for the basic indicator approach; seefigure 1). On the other hand, we have not computed whether the enhanced rules on creditrisk mitigation techniques or securitization risk would significantly change this conclusion.

ity, expected loss, and unexpected loss can be considered as a floor for thecorporate sector’s default probability, expected loss, and unexpected loss.37

Table 4 shows that, notwithstanding the lower level of capital require-ments derived from Standard and Poor’s ratings and default probabilities,increased capital charges are likely to emerge for most Latin Americanand Caribbean countries—where the sum of expected and unexpectedlosses already exceeds the value of the all-encompassing 8 percent capitalrequirement of the current Accord.

In short, our results suggest that while the foundation IRB approachimplies a rather generalized increase in capital requirements, it may notafford the 99.9 percent protection advocated by the Basel Committee onBanking Supervision given the default probabilities encountered inemerging countries. This result also calls into question the calibration ofthe published curve for use in these environments. If we assume that pro-visions cover expected losses, then achieving 99 percent protectionwould require capital levels significantly higher than the Basel I recom-mendation of 8 percent and around a 15 percent overall requirement.Achieving 99.9 percent protection would require even higher levels ofcapital. While these levels were close to Argentina’s overall capitalrequirement in 2001, they represent a steep increase in capital require-ments for many countries.

Policy Implications for Latin America and Emerging Countries

The previous section focused on the appropriate calibration of the Basel IIIRB approach. The results are highly relevant for the wider discussion ofappropriate Basel II application to Latin America and emerging coun-tries.38 That is the focus of this section. The IRB approach reflects recentdevelopments in the internal risk management of larger G10 banks. Manylarge banks have developed their own rating methodologies and have testedhow their own ratings map into default probabilities and value at risk—both on an individual claim and on a portfolio basis—using their own


37. Having said that, each of the three major rating agencies now allows private institu-tions to break through the sovereign floor, although each is subject to slightly different rules.

38. Kupiec (2001) also discusses Basel II calibration for developing countries. How-ever, he considers the original Basel II curve and examines specific assumptions on partic-ular types of loans in a model-based approach.

credit risk portfolio models.39 Supervisors across G10 countries havelargely been playing catch-up in their understanding of these models.40

The starting point is very different, however, in a typical country in LatinAmerica.

Is Latin America Ready for the IRB Approach?

Banks in emerging countries are generally less advanced in terms of devel-oping and using internal rating methodologies, mappings those ratingsinto default probabilities, and establishing portfolio models of credit risk.In many emerging countries, the supervisory agency’s main motivation formoving towards the Basel II IRB approach may be to improve banks’ owninternal risk management, rather than to catch up with what banks are alreadydoing.41 Moreover, supervisors tend to have significantly less resources inemerging countries, and they lack supervisory human capital, informationsystems, and both legal and real power.42

The statistics on compliance with the Basel Core Principles for Effec-tive Banking Supervision convey a picture of inadequate banking supervi-sion across many emerging countries worldwide and across Latin Americain particular. The average emerging country is compliant with just sevenof the thirty Basel Core Principles.43 Figure 6 illustrates emerging econ-omies’ compliance with a set of critical Basel Core Principles. Emergingcountries fare poorly, to say the least, and Latin America performs worsethan the average of this group. This suggests that it may be many yearsbefore supervisors in these countries would be advised to adopt the IRBapproach, given the heavy burden on scarce supervisory resources implied.


39. In fact, the concern of many larger G10 banks is that Basel II does not give themsufficient freedom to use their own portfolio models of credit risk and that they must use theIRB formula to approximate the risk of a loan portfolio. See the comments by several largebanks on the proposals at www.bis.org.

40. The Basel Committee on Banking Supervision has decided to maintain the formularather than allow banks to use internal models for multiple reasons, including the issues ofparameter and model risk and perhaps the fundamental moral hazard reasons discussed inthe introduction.

41. The more sophisticated emerging markets will present exceptions to this, and localbanks that are branches or subsidiaries of large G10 banks are likely to have benefited fromthe risk management methodologies implemented across the globe.

42. See Pagano (2001) for a set of papers on issues related to the legal system and creditrisk in Latin America.

43. There are actually twenty Core Principles; here we count the subprinciples of prin-ciple 1 as principles in their own right to obtain thirty.

010

2030

4050

6070

8090

100

Inde

pend

ence

& re

sour

ces (

1.2)

Capi

tal a

dequ

acy (

6)

Conn

ecte

d le

ndin

g (1

0)

Coun

try ris

k (11

)

Mar

ket r

isks (

12)

Othe

r risk

s (13

)

Cons

. sup

ervis

ion

(20)

Rem

edia

l mea

sure

s (22

)

Glob

ally

Cons

. Sup

ervis

ion

(23)

Com

plia

nt

Larg

ely c

ompl

iant

M

ater

ially

non

com

plia

nt

Nonc

ompl

iant

No

t app

licab

le

Perce

nt

Figu

re 6

.Em

ergi

ng C

ount

ries’

Com

plia

nce

wit

h Sp

ecifi

c Bas

el C

ore

Prin

cipl

es fo

r Eff

ecti

ve B

anki

ng S

uper

visi

on

As discussed earlier, however, the standardized approach may yield lit-tle in relating regulatory capital to risk because of the low penetration ofrating agencies in emerging countries. Therefore an intermediate approachis warranted to serve as a transition measure to the IRB model. We referto this as the centralized ratings-based approach.

The Centralized Ratings-Based Approach

Our proposed centralized ratings-based (CRB) approach is similar in spiritto the IRB framework in that banks would place their clients into a set ofrating buckets based on their estimated default probabilities, and then eachrating bucket would translate into an average default probability that is thenmapped to a capital requirement using a formula along the lines of theBasel II IRB model. We simplify the methodology significantly, however,because the rating methodology and the mapping to capital requirementsare determined by the regulator. The Basel II IRB approach stipulates thatthere would be a minimum of seven rating buckets. Under the CRB approach,the regulator would define the default probabilities (mean, minimum, andmaximum) that would correspond to each of a minimum of seven buckets.This scale might conform to one used by a leading rating agency, but banksupervisors may wish to define the scale to reflect the risk characteristicsof their own country and any objectives they wish to achieve.44 Bankswould then simply slot their clients into the buckets suggested by the reg-ulator based on their estimation of each borrower’s probability of default.

This approach suffers from one disadvantage—namely, that each bankwould be forced to use the same rating scale (though not necessarily thesame rating for each client, since banks’ opinions might differ). This meansthat a bank specializing in one type of business or region of a country wouldhave to use the same rating scale as a bank in another line of business orregion. Put another way, because the buckets would essentially be definedby the minimum and maximum default probabilities, the default probabil-ity range of each bucket may not be ideal for every bank. Some banks mayhave a large number of clients in one or two buckets of a CRB approach,whereas if they used an internal scale, they could break those buckets down


44. For a supervisory-based application using cluster analysis, see Foglia, Iannotti, andMarullo Reedtz (2001). Rating agency scales are typically through the cycle, whereas inter-nal bank rating scales tend to specify a twelve-month or other horizon. Supervisors maywish to adopt a through-the-cycle scale to reduce concerns of procyclicality.

into finer ones with a smaller range of default probabilities, thereby achiev-ing a more precise measure of required bank capital. The supervisor wouldmost likely define the buckets to be appropriate for the largest, systemicinstitutions in the banking sector.45 Consequently, these problems wouldbe limited to relatively small institutions. One solution would be to adoptthe U.S. model, which calls for such institutions to remain on Basel I oradopt the standardized approach of Basel II.

Countries may be concerned that the CRB approach would not be seenas compatible with Basel II. Some emerging countries, especially in LatinAmerica, already use a type of CRB approach for calculating provisions,and the level of provisions tends to be high in the region.46 In our simula-tions we defined default as more than ninety days past due and a loss givendefault of 50 percent, whereas many countries in Latin America ask for100 percent provisions for noncollateralized loans in this category.

This discussion underlines the need for a highly coordinated system forloan loss reserves and capital requirements. It is the sum of provisions andcapital that should be compared against the value at risk (the sum of expectedand unexpected losses)—and not necessarily provisions against expectedloss and capital against the unexpected component.47 If for some reason(legal or otherwise) there are impediments to increasing capital to coverunexpected losses relative to the desired level of protection, then provi-sions might be increased over and above the level of expected loss.

The methodology should gauge the overall value at risk of loans ratherthan their expected or unexpected loss components. This calls into ques-tion the common system combining a general loan loss reserve, a specificloan loss classification and provision depending on past performance (say,according to the traditional five-category classification), and a specified levelof capital.


45. In a more complex proposal, the regulator could allow the use of more than one cen-tralized scale to reflect different banking specializations. The Hong Kong Monetary Author-ity planned the introduction of a loan classification regulation similar to that described inthis paper (with loan grades characterized by an upper and lower default probability for eachgrade; see Hong Kong Monetary Authority (2002)).

46. Colombia, for example, is developing a system labeled SARC (Sistema de Admin-istración de Riesgos de Crédito) to quantify loan loss reserves based on individual banks’assessments of expected losses (internal models). Argentina and Brazil also have databasesthat include a rating scale determined by the regulator, which is used to monitor provisioning.

47. We abstract here from a discussion of the potential dynamic aspects of banks’ pro-visioning policies or counter cyclical regulations regarding loan loss reserves; see Cavalloand Majnoni (2002) and Laeven and Majnoni (2003) for discussions.

How would capital requirements be defined under a CRB approach?One possibility would be to rely on observed default probabilities, use theBasel II IRB curve to calculate unexpected losses, and set provisions andcapital to cover expected and unexpected losses, respectively.48 In this case,provisions could be derived as a residual from the following expression,based on the estimated value at risk (VaRe):

General Provisions = VaRe − Capital

If a regulator could not alter provisioning rules for some reason, a secondapproach could set capital equal to value at risk minus the allowable pro-visions along the lines suggested by the Basel Committee in the revisionsto the third consultative paper.49 This approach is intended to ensure a morerational integration of bank capital requirements and loan loss reserves.In this case, capital would be computed as a residual:

Capital = VaRe − General Provisions

A third approach, which is appropriate for regulators who have the free-dom to alter provisions but who prefer compatibility with Basel II and asimple rule for capital, would be to adopt the Basel II standardized approachbut then establish forward-looking provisions determined by the value atrisk minus capital requirement specified by the Basel II standardizedapproach. This method thus uses the CRB approach to enhance forward-looking provisioning rules in a fashion that is totally consistent with BaselII.50 Moreover, banks’ internal rating methodologies should develop overtime, allowing such banks to move over to the full IRB approach when theyare ready.

The CRB approach has many advantages as a transition tool. First, it low-ers the monitoring costs for bank supervisors. Supervisors would have toverify the quality of banks’ methodologies for slotting clients into the rele-


48. Whether a regulator is free to do this would depend on the particular constraints,legal or otherwise. However, if regulators do not have the freedom to determine provision-ing or capital rules, then they would not comply fully with the first Basel Core Principle onindependence and autonomy.

49. Basel Committee on Banking Supervision (2004a, 2004b).50. In practice, regulators in Latin America tend to have more freedom to determine

provisions than capital. It thus is not surprising that provisions are typically higher than astandard expected loss calculation.

vant buckets; to do so, they could very easily establish a homogeneous scalefor comparing different banks’ ratings of particular corporate clients, cor-porate clients of similar characteristics, corporate clients in particular eco-nomic sectors or regions, and so forth. The homogeneous scale would alsoallow the supervisor to easily verify loan classifications based on the defaultprobabilities. Second, the CRB approach provides a consistent treatment ofcapital and loan loss reserves, which is a vital component of Pillar 2 andBasel Core Principle 8. Third, a risk-based capacity is developed within thebanking system, independently of each country’s decision to officially remainin Basel I or to adopt the different options of Basel II. Fourth, the homo-geneity of bank classification schemes implies that bank data could easily beaggregated at a country level, thereby generating an important (and thus farlargely missing) source of data for prudential monitoring at the macroeco-nomic level. These data would be useful for analyzing changes in the assetstructure of the financial system, developing tools to consider aggregatefinancial sector risk, and predicting where problems might occur and theirpotential depth. Finally, if countries (in a region or more widely) could coor-dinate the number and definitions of their rating buckets, then this wouldenhance aggregation and comparability across countries. Under the standard-ized approach, local regulators will likely use incomparable local ratings,and comparability will undoubtedly be lost under the IRB approach, in whichindividual banks will use their own rating methodologies. From this per-spective, the CRB appears particularly suited for ensuring the disseminationof a new risk-based regulatory standard.

Developing a Basel II Decision Tree

Many emerging countries face a difficult decision of whether to stay withBasel I or move to Basel II.51 If they choose the latter, they must considerwhich of the many alternatives to adopt, or whether to implement a mixedapproach leaving some banks on a simpler approach and allowing or forc-ing a selection of banks to incorporate a more advanced alternative. Fivecountry characteristics that may aid to guide these important choices: thedegree of compliance with the Basel Core Principles and hence with BaselII Pillar 2; the penetration of rating agencies and the operation of the rat-ings market in general; the current level of bank capital and the feasibilityof increases in bank capital ratios in the short term; the size of, or the


51. This section draws on Powell (2002, 2004).

desire to develop, domestic capital markets; and the availability of infor-mation and the degree of sophistication of both banks and the supervisorin assessing and monitoring loan loss provisioning.

As discussed above, evidence from Financial Sector Assessment Pro-grams completed by the International Monetary Fund and the World Bankillustrates that many countries are far from being fully compliant with theBasel Core Principles for Effective Banking Supervision and, on average,emerging countries lag behind their G10 counterparts.52 Of particular con-cern is the lack of (i) effective consolidated supervision, (ii) supervisoryindependence, resources, and authority, and (iii) effective prompt correc-tive action. If supervisors lack resources and the basics of effective banksupervision, then correcting this deficiency should be the first priority, andthe implementation of complex rules on capital requirements may well becounter productive. Basel II also introduces a significant change in thelevel of consolidation required for banking supervision—from the bankitself to its holding company. Since many countries do not comply with moremodest versions of consolidated supervision, these countries remain farfrom the spirit of the Basel II proposals.

However, full compliance with the Basel Core Principles for EffectiveBanking Supervision is too strict a precondition for moving to Basel II.After all, many G10 countries are not compliant with all the Core Princi-ples. A country should be compliant with the Core Principles to the degreerequired to implement the appropriate alternative chosen within the BaselII framework. For example, if a supervisor does not have the resources(including data, information, technical competence, staffing, and manage-ment) to consider whether the calibration of the Basel II IRB approach isappropriate to that country, or to monitor effectively how banks wouldapply the IRB methodology, then a simpler alternative should be adopted.

Many emerging countries will probably opt for the simpler Basel IIapproaches, including the simplified standardized approach and the stan-dardized approach. An important difference between the two is that thelatter allows for the use of credit ratings from private agencies, whereasthe former only uses the ratings of official export credit guarantee agenciesfor sovereign risk assessment. The problem for many emerging countries,however, is that markets for credit ratings are shallow, so the standardizedapproach would not improve much on either the simplified standardizedapproach or Basel I in terms of aligning capital requirements with risk.


52. See the joint IMF–World Bank project (IMF and World Bank, 2002)

Adopting the standardized approach may create incentives for such rat-ings’ markets to develop, but this brings its own dangers in terms of com-panies buying a good rating and provoking a “race to the bottom” in ratingsquality. The second key characteristic, then, is the state of the ratings mar-ket. The standardized approach makes little sense for a country with noratings market to speak of. Such a country should stick with Basel I oradopt the simplified standardized approach. Alternatively, if the country’scompliance with the Core Principles is reasonably high, the authoritiescould consider the CRB approach as a potential precursor to the Basel IIIRB approach. For a country with an active ratings market, the standard-ized approach makes more sense.

If a country adopts either the simplified standardized approach or thestandardized approach, Basel II will likely increase bank capital require-ments.53 The source of the extra capital charge is operational risk. Anincrease in capital requirements may not be a bad thing, but an emergingcountry deciding whether to adopt Basel II should carefully consider thecurrent level of bank capital and the feasibility of increasing required bankcapital. This is the third characteristic listed above.

Basel II also includes enhancements for the credit risk implications ofsecuritization risk and for credit risk mitigation techniques. A countrywith a fairly inactive ratings market may still benefit from the use of rat-ings in these areas. For example, if a country has an active market for secu-ritized claims (a market that is currently growing in importance in somecountries), then those claims will most likely be rated and the Basel IIstandardized approach regarding securitization risk might be gainfullyadopted. This may not seem to be a critical feature, but if a country wishesto develop capital markets, then it needs to ensure that banks have the rightincentives to securitize claims. Basel II does a better job here than Basel I.A similar argument can be made for credit risk mitigation techniques.Basel II makes useful improvements in this area, so it may be appropriateif markets using securities as collateral are important or if a country wishesto develop them. The fourth characteristic listed is thus the importance oflocal capital markets and the desire to develop them.

The final characteristic suggested above is the sophistication of the super-visor and banks in terms of provisioning rules, monitoring, and control. Basel


53. This may also depend on the long list of items under national discretion. Two suchissues are the risk weights for mortgages and for retail exposures. Capital requirements aremore likely to rise if the more generous treatments are not applied. We do not go further intothe specific items left to national discretion.

II replaces a set of ad hoc rules regarding capital requirements with a morerobust estimate of credit risk reflecting value at risk. Value at risk may bedecomposed into expected and unexpected losses subject to a statisticaltolerance value. As discussed, current theory holds that provisions shouldreflect expected loss, whereas capital should reflect unexpected loss. Foran economist, the appropriate level of provisioning and capital for creditrisk then both come from the same probability distribution; they simplyreflect different statistics of that same distribution.

Under this more general approach, a supervisor that has advanced in termsof more forward-looking provisioning rules has also advanced in terms ofconsidering finer risk-based capital rules. In several countries in the region,supervisors have set up centralized databases to monitor the large debtorsof the financial system and ensure that each lender knows the total debtoutstanding of large borrowers. In some cases these databases have beenexpanded to cover most loans of the financial system and are used to mon-itor and control provisioning requirements. Miller presents a review of thedesign and uses of these databases.54 While in most countries such require-ments are not forward looking but reflect arrears, the move to a forward-looking system for provisioning and capital is certainly made more feasibleif such a database is in place. For example, some countries have now incor-porated a bank rating into these databases that includes not only backward-looking variables, but also cash-flow-type analyses.

In sum, the key characteristic is the sophistication of the supervisor andbanks in terms of information on provisioning and loan losses. A super-visor that has regularly tracked loan losses across banks and has developedmonitoring tools such as transition probability matrixes and simple creditscoring techniques to monitor provisioning rules is in a much better posi-tion to implement the Basel II IRB approach or our simpler centralizedratings-based approach than a supervisor that has no experience in theseareas. Still, the IRB or CRB approach will probably only be appropriatefor large, relatively sophisticated banks. A country with a highly concen-trated banking sector in which a few large, sophisticated banks control alarge percentage of the sector will encounter added benefits in moving tothe CRB or IRB formula, at least for those banks.

The decision tree in figure 7 illustrates how the above five characteris-tics may affect the Basel decision and provides a simple navigational aidfor countries regarding the Basel standards. Countries that do not comply


54. Miller (2003).

FIG

UR

E 7

.A

Base

l II D

ecis

ion

Tree

Base

l I

Very

wea

k com

plia

nce w

ithCo

re P

rincip

les

Base

l I

Not f

easib

le/d

esira

ble t

oin

crea

se ca

pita

lre

quire

men

ts

Base

l II A

Pilla

r I: S

impl

ified

stan

dard

ized

appr

oach

Base

l II B

Pilla

r I: S

tand

ardi

zed

appr

oach

Com

plia

nce w

ith Co

rePr

incip

les i

s ade

quat

e for

mon

itorin

g ex

tern

alra

tings

Mor

e liq

uid

capi

tal

mar

kets

or r

easo

nabl

era

ting

pene

tratio

n

Cent

raliz

ed ra

tings

-ba

sed

appr

oach

astra

nsiti

on to

inte

rnal

ra

tings

-bas

ed ap

proa

ch

Rela

tivel

y hig

h co

mpl

ianc

ew

ith Co

re P

rincip

les,

but

not h

igh

enou

gh fo

rin

tern

al ra

tings

-bas

edap

proa

ch

Inte

rnal

ratin

gs-

base

d ap

proa

ch

Mor

e sop

hist

icate

dpr

ovisi

onin

g ru

les

Very

goo

d co

mpl

ianc

ew

ith Co

re P

rincip

les

Ratin

g pe

netra

tion

may

be l

ow

Feas

ible

/des

irabl

e to

incr

ease

capi

tal

requ

irem

ents

Min

imum

com

plia

nce w

ith Co

rePr

incip

les t

o co

mpl

y with

Pill

ars I

Ian

d III

and

mon

itor o

pera

tiona

l risk

The B

asel

II D

ecisi

on

with many of the basic Basel Core Principles are probably advised to staywith Basel I. However, a country that wishes to increase bank capital require-ments should consider the Basel II simplified standardized approach if theextra burden of supervising operational risk is feasible. Countries thathave only a shallow market for ratings will get limited benefits from thestandardized approach and should be advised that this will also lead to anincrease in capital requirements. They should stick with the simplified stan-dardized approach if supervisory resources are limited. Countries that havedeeper capital markets or a strong desire to develop them should reconsiderthe standardized approach for its enhancements to securitization risk andcredit risk mitigation techniques. Finally, countries that have made advancesin terms of forward-looking provisioning rules and that have the informa-tion and systems to control banks’ provisioning practices are better placedto consider the CRB or even the IRB approach.55

Conclusions

In this paper we have discussed the implementation of the Basel II Accordin emerging countries, with an emphasis on Latin America. The discussionsuggests three broad concerns with the new accord. First, given the lowpenetration of rating agencies, the Basel II standardized approach (whichuses external ratings to gauge credit risk) will do little to link regulatorycapital to risk in Latin America. For countries adopting the standardizedapproach, moving to Basel II will imply only a marginal correction of otherproblems in Basel I, and it will not address the fundamental problems ofBasel I that motivated the new accord.56 Second, the more advanced BaselII internal ratings-based approach may require recalibration given our esti-mates of credit risk. It also appears complex and will stretch scarce super-visory resources in many countries. Finally, the essence of a standard maybe lost if many countries adopt the standardized approach (using incom-parable local ratings) or the IRB approach (using many different privatebanks’ ratings and default-probability estimation methodologies).

We propose an intermediate approach between the standardized andIRB approaches, which we call the centralized ratings-based approach.


55. We perceive a loss of comparability across countries as a cost of the IRB approach.56. Implementing the standardized approach has advantages stemming from improve-

ments in the treatment of securitization risk and credit risk mitigation techniques.

Our approach might be used as a transition measure to the IRB methodol-ogy; it might be employed to more fully integrate capital and provisioningregulations; and it might allow increased coordination on a standardizedrisk-based reserving policy across countries in the region or beyond. Amain difference with the IRB approach is that although banks would ratetheir clients (and estimate default probabilities), the regulator would definethe rating scale and the way in which the rating buckets would map todefault probabilities. This approach could be used to set forward-lookingprovisioning requirements only. A country could then adopt the Basel IIstandardized approach, set provisions using the CRB methodology to coverthe value at risk minus the standardized approach’s capital, and therebyensure that banks’ total reserves (provisions plus capital) covered the entirevalue at risk up to the desired level of protection.

We employed a homogeneous bootstrapping methodology to analyzecredit risk in three emerging markets in Latin America. The bootstrappingmethodology implies that our estimates are free from the usual problemsof parameter estimation error and model error that plague standard attemptsto measure portfolio credit risk. At the same time, our results should betaken as indicative only, and we hope that future research will attain fur-ther precision by extending the empirical methodology over time. The resultsindicate that to achieve a 99 percent level of protection (in other words,such that capital covers the unexpected loss to 99 percent of the distribu-tion), capital requirements would need to be significantly higher than the8 percent level recommended in Basel I and closer to 15 percent. Even higherlevels would be required to achieve 99.9 percent protection, as intended inBasel II. We also find that the Basel II foundation IRB approach, whileresulting in increases in capital requirements above Basel I, would resultin levels closer to the 90–95 percent protection rather than the 99.9 percentlevel stated as used in its calibration. We believe that further research isrequired in this area to consider if and how the Basel II IRB methodologymight be recalibrated for countries that have default probabilities signifi-cantly higher than G10 countries.

We also discuss Basel II implementation in Latin America more gener-ally and develop a simple Basel II decision tree. Countries should considerfive characteristics when deciding whether to stay on Basel I or implementBasel II—and if the latter, how. An important characteristic is how coun-tries comply with the Basel Core Principles for Effective Banking Super-vision and, hence, with the second pillar of Basel II. If compliance with theCore Principles is weak, then countries should consider staying on Basel I.


If operational risk can be monitored and capital requirements increased, thenthe simplified standardized approach may be in order. If the ratings mar-ket is reasonably deep, if the country has a strong desire to deepen capitalmarkets, and if supervising external ratings is feasible, then the standardizedapproach may be appropriate. As the degree of supervisory sophisticationimproves, especially in regard to the regulation and supervision of forward-looking provisions, then the CRB and, eventually, the IRB approachesmay be considered.

Basel II may imply the end of a standard rather than the establishmentof a new one, given the multiple Basel II alternatives, the reliance on incom-parable local ratings, and the use of individual bank-generated ratings anddefault probability estimations. Put simply, two countries with 14 percentassets at risk under Basel II may actually be quite different. The CRBapproach is an attempt to develop a more homogeneous system that is com-patible with Basel II, suitable to the context of emerging country super-visors, and consistent with the notion of maintaining a standard.57


57. We have not discussed a number of issues pertinent to Latin America that receiveless attention in the new Accord: namely, bank lending to its own government, lending inforeign currency (domestic dollarization), and related lending. The Basel Accord addressesall three cases, but the treatment should be tightened in each case.

Comments

Patricia Correa: Endless pages have been written on the potential flawsthat could prevent Basel II from achieving the ultimate goal of increasingfinancial stability worldwide by improving bank risk management andmaking capital requirements more sensitive to risk. Criticisms, which aresometimes contradictory, can be grouped into five categories. First, the newaccord offers alternative approaches for measuring capital requirements(two versions of the standardized approach and two of the internal ratings-based, or IRB, approach), but it does not create proper incentives to use themost risk-sensitive approach, thereby opening the door for arbitrage. Sec-ond, in countries with little capital market and financial development, reli-able external ratings are not available for most of the assets in the banks’credit portfolio; in such cases, the standardized approach will do little tolink better capital to risk and would be, at best, a poor substitute to Basel I.Third, Basel II relies heavily on methodologies that lead to capital require-ments that tend to accentuate the cycle, which can eventually increase,rather than diminish, financial instability. Whether this flaw can be attrib-uted to the methodologies themselves or to the length of the databasesemployed to make the respective estimates is a controversial issue, but thistopic is certainly an important one in the discussion. Fourth, the new frame-work will most probably increase the amount of minimum regulatory cap-ital regardless of the measurement methodology chosen, and many banks,particularly in emerging economies, are simply not ready to meet the addi-tional capital requirement. Finally, regulatory and supervisory bodies inmost emerging economies are not prepared to meet the challenges posedunder the second pillar of the accord, owing to lack of infrastructure, inad-equate human capital, and so forth.

Majnoni and Powell’s paper represents an important contribution to theunderstanding of the potential impact of Basel II on emerging economiesand sheds light on the validity or relevance of the aforementioned criti-cisms. The authors’ contribution is particularly valuable in two aspects: it

141

is the first attempt, to my knowledge, to estimate capital requirements inLatin American economies using the same methodology, enabling com-parison across countries; and the paper goes beyond simply criticizing thenew accord and, within the spirit of Basel II, constructively proposes a newapproach to capital requirements (namely, the centralized ratings-basedapproach). However, issues related to the estimation of recovery ratios,which represent a key component of capital requirement estimation, areleft out of both the quantitative and qualitative analyses in the paper. Thismakes the analysis of the challenges ahead incomplete, and it leads tosome erroneous conclusions about the relative advantages and disadvan-tages of Basel II over Basel I. The authors’ recommendations are also some-what unclear and self-contradictory regarding how far and how fast thesecountries should move toward implementing better risk management sys-tems and more risk-sensitive capital requirements. Below I present morespecific comments on each of the paper’s three main components: a quan-titative study estimating capital requirements that accomplish the objec-tives of Basel II in three emerging economies; a qualitative analysis of thepractical difficulties of implementing Basel II, with an assessment of acountry’s readiness to adopt Basel II; and policy recommendations.

With regard to the empirical analysis, the authors use a bootstrapmethodology—which they argue is free of the usual problems that plaguetraditional econometric techniques (namely, parameter estimation and modelerrors and risks)—to estimate default probabilities for the loan portfolio ofArgentina, Brazil, and Mexico. They describe their specific assumptionsabout the provisioning system and their allegedly representative sample ofcredit data.1 They then calculate the capital requirements that would benecessary to cover credit risk under two circumstances: first, using thefoundation IRB approach and the default probability risk weight mappingcurve proposed by Basel II; and second, using the bootstrapping method-ology to calculate the capital requirements needed to cover the value atrisk of those portfolios at a 99.9 percent confidence rate.

When comparing the two results, the authors conclude that, except forBrazil, the Basel II formula generates levels of protection inferior to theadvertised 99.9 percent. The authors therefore call into question the cali-


1. Their assumptions include a recovery ratio of 50 percent, a level of provisions thatcovers expected losses; and a definition of default as the event of more than ninety dayspast due payments. The sample periods are 2000–01 for Argentina and Mexico and 2001–02for Brazil.

bration of the Basel II IRB curve, stating that it is not clear that it wouldbe appropriate for emerging economies to apply the IRB curve to individ-ual instruments to ensure a good approximation of risk. They suggest thatas with Basel I, countries should or may choose to recalibrate the curve ormake the requirement more stringent (for example, establishing a 10 per-cent capital adequacy requirement instead of 8 percent). They also con-clude that the foundation IRB approach is, notwithstanding its benevolentrisk calibration, likely to set higher capital requirements than the existing 8percent under Basel I.2

While the methodology employed by the authors has its advantages, thedata limitations and the short period chosen for the estimations (issues thatare acknowledged in the paper) make it premature to categorically con-clude that the Basel II models need to be recalibrated. Only after manyyears of experience and collection of quality data would it be possible toquantify those default rates and value-at-risk levels appropriately and,incidentally, reduce the procyclicality inherent to all approaches based onempirical estimations (even those using bootstrapping). Several relatedquestions come to mind. Has the bootstrapping methodology been con-trasted with that of Basel II using G10 data? If so, for what periods? Howsensitive are the comparisons of bootstrapping and Basel II to the periodchosen? Are Brazil’s small credit loans necessarily more risky than largeones, as assumed by the authors?

In any event, these limitations should not serve as an excuse for regu-lators and banks to not move forward in refining internal risk measurements.Neither should the fact that Basel II implies more capital requirements. Toadvance Basel II’s general goal, the problem of raising additional capitalcould be solved by phasing in the meeting of the new requirements, ratherthan halting progress in the implementation of IRB systems.

I turn now to the paper’s qualitative analysis. In assessing the difficul-ties that emerging economies may have in applying the new capital accord,Majnoni and Powell outline the minimum conditions that countries haveto meet before they begin implementation, and they summarize documen-tation that proves that many emerging economies do not yet meet theserequirements. I disagree with the authors’ approach to tackling these prob-lems, which is basically to strengthen institutions before starting to imple-ment Basel II. It is true that the ideal conditions are not present in manycountries (I would dare say any country), and many institutional weaknesses


2. Specifically, 15 percent for Argentina and 10–14 percent in Mexico and Brazil.

prevail in the supervisory agencies and the banking industry. My experi-ence as a banking supervisor in Colombia, however, made me a firmbeliever that the only way to create such conditions, particularly regardingthe second pillar, is precisely by moving forward and setting clear goalsand deadlines. In this ever-evolving field, which is more art than science,learning by doing is the only way to succeed, as has long been the processin the developed world. A sure way to delay preparedness is to postponethe definition of policy goals such as the development of good risk assess-ment within banks and matching regulatory capital. Again, gradually phas-ing in the objective is preferable to not starting the run.

Finally, the paper’s policy recommendations aim to facilitate the tran-sition toward the Basel II IRB regime in emerging economies. The authorspropose a centralized ratings-based (CRB) approach to cover credit risks,which is compatible with the IRB model and which has the following char-acteristics: banks estimate default probabilities according to their owninternal models, as in Basel II; the regulator defines the rating scale to beused and the mapping of each rating bucket to a range of default probabil-ities (in Basel II this is done independently by each bank, not uniformly bythe regulator); and loan loss provisions are defined as the expected lossgiven default for each category of loans, and regulatory capital is definedas the total value at risk minus the expected loss (or, in the case of legal orother problems with changing the capital regime during the transitionperiod, provisions could be defined as the difference between the desiredtotal level of protection and the current capital requirement, and provisionscould thus be over the expected loss).

This transition regime has a number of advantages. It coordinates thesystem for loan loss provisioning and capital requirements, and it wouldfacilitate comparisons across banks and the handling and interpretation ofdata on credit risk. This proposal is in many ways similar to the systembeing implemented in Colombia. It is extremely appealing and shouldreceive more attention and backing by international regulators. While itcertainly simplifies matters for both regulators and banks, it is perfectlycompatible with the spirit and ultimate goal of Basel II.

Philip Brock: In 1988 the Basel Committee on Banking Supervision for-mulated the first Basel accord for bank capital requirements (Basel I). Itspurpose was to raise overall levels of capital adequacy in the thirteenmember countries while simultaneously homogenizing standards. Basel Iwas a great success, with over a hundred countries adopting the framework.


Concerns arose, however, over the side effects of the accord. Among otherissues, the capital standards of Basel I are relatively insensitive to the risk-iness of bank portfolios, and the accord creates incentives to engage in reg-ulatory arbitrage, whereby banks increase their risk within the parametersof Basel I without raising levels of capital.1

These concerns with the somewhat blunt nature of the Basel I capitalstandards led to the forging of a second accord (Basel II) in June 2004,which seeks to make bank capital more responsive to credit risk. Basel IIoffers four approaches to calculating bank capital. The first two, the stan-dardized and simplified standardized approaches, map the ratings of creditrating agencies into capital requirements. The second two rely on banks’own internal ratings-based (IRB) models to generate levels of capital ade-quacy. Although this menu of approaches to capital requirements addressesthe concern that Basel I is not sensitive enough to bank risk, Basel II hasits own flaws. The two standardized approaches have highlighted con-cerns about the ability of risk-rating agencies to provide meaningful assess-ments of bank risk.2 The two IRB approaches rely heavily on value-at-risk(VaR) models that only provide point estimates of the loss distribution,leaving substantial room for so-called spike-the-firm events involvinghigh losses with low probability.3 Regulatory arbitrage could also occuracross banks adopting different approaches (for example, the standardizedversus IRB approaches).4

Majnoni and Powell’s paper centers on the adaptation of Basel II toLatin American financial systems. Their first concern is the lack of pene-tration of credit-rating agencies in Latin America, which makes the imple-mentation of Basel II’s standardized approach difficult. The second is theaccuracy of the VaR approach for calculating capital adequacy levels. Acenterpiece of the paper is the use of a bootstrapping methodology to cal-culate levels of capital adequacy that cover losses in 99 and 99.9 percentof potential outcomes in any given year. The authors apply this bootstrap-ping methodology to loans from Argentina, Brazil, and Mexico. When theycompare the bootstrapping methodology with a VaR model calibrated usingBasel II values, they find that that the VaR model underestimates the amountof capital that banks should be holding in each of the three countries. The


1. Dewatripont and Tirole (1994); Saidenberg and Schuermann (2003).2. Danielsson and others (2001); Goodhart (2004).3. Danielsson and others (2001).4. Repullo and Suárez (2004).

authors find, among other factors, that the Basel II risk weights for smalland medium-sized business loans are too low for Latin America, in comparison with the empirically derived results from the bootstrappingexercises.

The authors’ concerns with the application of the standardized and IRBapproaches of Basel II leads them to propose a hybrid approach, whichthey call the centralized ratings-based (CRB) approach. This approachrelies on bank supervisors to develop risk ratings for banks based on theinformation that banks provide to the supervising authorities. In contrastwith Basel II, the CRB imposes uniform (rather than bank-specific) riskweights across categories of loans for all banks, but the risk weights aredetermined with the active interaction of the banks and the bank super-visor. The CRB approach is similar to approaches currently in use in sev-eral Latin American countries.

A primary purpose of the Basel capital accords is to promote the stabilityof financial systems. As with Basel I and II, there are some worries associ-ated with the CRB approach. Like Basel II, the CRB approach may lead toprocyclical capital requirements, since lower measured credit risk will leadto lower capital-asset ratios during extended periods of good banking per-formance. This is less apt to be the case with Basel I, in which capitalrequirements respond less to changes in risk. Regulatory capture is anotherconcern with the CRB approach. In particular, the regulator may come underpressure at times to be lenient in the classification of bank loans.

Any capital adequacy framework that a country adopts may destabilize,as well as stabilize, the financial system. Basel I, Basel II, and the CRBapproach all strengthen bank supervision, but they may result in unwantedrisk taking. Much risk faced by banks is macroeconomic, and this type ofrisk is underemphasized in Basel II.5 Other financial sector policies canpartially address this macroeconomic risk. For example, evidence indi-cates that policies geared toward reducing dollarization in Latin Americawould stabilize financial systems.6 Policy measures to cushion the impactof sudden stops of foreign capital would also increase the stability of thebanking systems.7 Ultimately, the success of Basel II or the CRB approachin Latin America will depend on the accompanying policy measures takento stabilize the economies against macroeconomic shocks.


5. Blaschke and others (2001); Carling and others (2002); Sorge (2004).6. Herrera and Valdes (2004); Levy-Yeyati (2004).7. Calvo, Izquierdo, and Mejía (2004).

References

Altman, Edward I., and Anthony Saunders. 1997. “Credit Risk Measurement:Developments over the Last Twenty Years.” Journal of Banking and Finance21(11–12): 1721–42.

Balzarotti, Veronica, Christian Castro, and Andrew Powell. 2004. “ReformingCapital Requirements in Emerging Countries: Calibrating Basel II Using His-torical Argentine Credit Bureau Data and CreditRisk+.” Business Schoolworking paper. Universidad Torcuato Di Tella.

Balzarotti, Veronica, Michael Falkenheim, and Andrew Powell. 2002. “On theUse of Portfolio Risk Models and Capital Requirements in Emerging Markets:The Case of Argentina.” World Bank Economic Review 16(2): 197–212.

Basel Committee on Banking Supervision. 2003. “Overview of the New BaselCapital Accord.” Consultative document 3 (April). Basel: Bank for Interna-tional Settlements.

———. 2004a. “An Explanatory Note on the Basel II IRB Risk Weight Func-tions.” Technical paper (October). Basel: Bank for International Settlements.

———. 2004b. “Modifications to the Capital Treatment for Expected and Unex-pected Credit Losses in the New Basel Accord.” Technical paper (January).Basel: Bank for International Settlements.

Berger, Allen N., Richard R. Herring, and Giorgio P. Szego. 1995. “The Role ofCapital in Financial Institutions.” Journal of Banking and Finance 19(3–4):393–430.

Blaschke, Winfrid J., and others. 2001. “Stress Testing of Financial Systems: AnOverview of Issues, Methodologies, and FSAP Experiences.” Working paper01/88. Washington: International Monetary Fund.

Calvo, Guillermo, Alejandro Izquierdo, and Luis-Fernando Mejía. 2004. “On theEmpirics of Sudden Stops: The Relevance of Balance-Sheet Effects.” Workingpaper 10520. Cambridge, Mass.: National Bureau of Economic Research.

Carey, Mark S. 1998. “Credit Risk in Private Debt Portfolios.” Journal of Finance53(4): 1363–87.

———. 2002. “A Guide to Choosing Absolute Bank Capital Requirements.”Journal of Banking and Finance 26(5): 929–51.

Carling, Kenneth, and others. 2002. “Capital Charges Under Basel II: CorporateCredit Risk Modelling and the Macroeconomy.” Working paper series 142.Stockholm: Sveriges Riksbank.

Cavallo, Michelle, and Giovanni Majnoni. 2002. “Do Banks Provision for BadLoans in Good Times? Empirical Evidence and Policy Implications.” In Rat-ings, Rating Agencies and the Global Financial System, edited by RichardLevich, Giovanni Majnoni, and Carmen Reinhart. New York: Kluwer Aca-demic Publishers.

Danielsson, Jon, and others. 2001. “An Academic Response to Basel II.” Specialpaper 130. London School of Economics, Financial Markets Group.


Dewatripont, Mathias, and Jean Tirole. 1994. The Prudential Regulation of Banks.MIT Press.

Ferri, Giovanni, Li-Gang Liu, and Giovanni Majnoni. 2001. “The Role of RatingAgency Assessments in Less Developed Countries: Impact of the ProposedBasel Guidelines.” Journal of Banking and Finance 25(l): 115–48.

Foglia, Antonella. 2003. “Using Credit Register and Company Account Data forMeasuring Credit Risk: A Supervisory Approach.” Rome: Bank of Italy.Mimeographed.

Foglia, Antonella, Simona Iannotti, and Paolo Marullo Reedtz. 2001. “The Defini-tion of the Grading Scales in Banks’ Internal Rating Systems.” Rome: Bank ofItaly. Mimeographed.

Freixas, Xavier, and Jean-Charles Rochet. 1999. Microeconomics of Banking,4th ed. MIT Press.

Goodhart, Charles. 2004. “Some New Directions for Financial Stability?” PerJacobsson lecture. London School of Economics, Financial Markets Group.

Goodhart, Charles, and others. 1999. Financial Regulation. London: Routledge.Herrera, Luis Oscar, and Rodrigo Valdes. 2004. “Dedollarization, Indexation and

Nominalization: The Chilean Experience.” Working paper 261. Santiago: Cen-tral Bank of Chile.

Hong Kong Monetary Authority. 2002. “Revision of Loan Classification System.”Mimeographed.

IMF (International Monetary Fund) and World Bank. 2002. “Implementation ofthe Basel Core Principles for Effective Banking Supervision: Experiences,Influences, and Perspectives.” Washington: International Monetary Fund. Avail-able at www.imf.org/external/np/mae/bcore/2002/092302.htm.

Kupiec, Paul. 2001. “Is the New Basel Accord Incentive Compatible?” Washing-ton: International Monetary Fund. Mimeographed.

Laeven, Luc, and Giovanni Majnoni. 2003. “Loan Loss Provisioning and Eco-nomic Slowdowns: Too Much, Too Late?” Journal of Financial Intermedia-tion 12(2): 178–97.

Levy Yeyati, Eduardo. 2004. “Financial Dollarization: Evaluating the Conse-quences.” Universidad Torcuato Di Tella. Mimeographed.

Márquez, Javier, and others. 2003. “Credit Information, Credit Risk Measure-ment, and the Regulation of Bank Capital and Provisions in Mexico.” MexicoCity: Bank of Mexico. Mimeographed.

Miller, Margaret. 2003. Credit Reporting Systems and the International Economy.MIT Press.

Mishkin, Frederic S., ed. 2001. Prudential Supervision: What Works and WhatDoesn’t. University of Chicago Press.

Pagano, Marco, ed. 2001. Defusing Default: Incentives and Institutions. JohnsHopkins University Press.

Powell, Andrew. 2002. “A Capital Accord for Emerging Economies?” Policyresearch working paper 2808. Washington: World Bank.


———. 2004. “Basel II and Developing Countries: Sailing through the Sea ofStandards.” Universidad Torcuato Di Tella. Mimeographed.

Powell, Andrew. and others. 2004. “Improving Credit Information, Bank Regula-tion, and Supervision: On the Role and Design of Public Credit Registries.”Policy research working paper 3013. Washington: World Bank.

Repullo, Rafael, and Javier Suárez. 2004. “Loan Pricing under Basel CapitalRequirements.” Madrid: Center for Monetary and Financial Studies.

Saidenberg, Marc R., and Til Schuermann. 2003. “The New Basel Capital Accordand Questions for Research.” Working paper 03-14. University of Pennsyl-vania, Wharton School, Financial Institutions Center.

Sorge, Marco. 2004. “Stress-Testing Financial Systems: An Overview of CurrentMethodologies.” Working paper 165. Basel: Bank for International Settlements.


The Economics of TransMilenio,a Mass Transit System for Bogotá

The provision of urban public mass transportation in Latin America hasexhibited marked swings during the last century. Various cities haveexperimented with public and private provision, as well as different

types of market regulation ranging from almost total liberalization to inter-vention in fares, route allocations, bus size, service quality, and exclusiveroad lanes. Despite the importance of this sector for the quality of sociallife and urban economic development, the literature contains little theo-retical research on the market failures characteristic of this industry. Thereare also few studies on the consequences of alternative provision and reg-ulation arrangements. This paper contributes to this second area of research.We describe the experience of Bogotá, Colombia, where a new hybrid sys-tem of urban public transport was put in place at the beginning of thiscentury—hence its name, TransMilenio. The new system was designed afterdecades of learning about the failures of both publicly and privately ownedsystems. It currently supplies more than 20 percent of daily trips. By 2015,the complete TransMilenio system is expected to transport 80 percent ofthe city’s population at an average speed of 25 kilometers per hour with aservice quality similar to an underground metro system.

Several market failures affect the provision of urban mass transit: un-clear definition of property rights on the curbside and on the road; collusion,

151

J U A N C A R L O S E C H E V E R R YA N A M A R Í A I B Á Ñ E Z

A N D R É S M O Y AL U I S C A R L O S H I L L Ó N

Echeverry, Ibáñez, and Moya are with the Universidad de los Andes, Department of Eco-nomics, in Bogotá, Colombia. Hillón is with the Colombian Ministry of the Environment.

We express our thanks to Marcela Meléndez and Edgar Sandoval, who worked on a firstdraft of this paper. We are particularly grateful to Andrés Gómez-Lobo for on-going cooper-ation with successive drafts, as well as to Mauricio Cárdenas, Kenneth McConnell, NormanOffstein, and Andrés Velasco. Excellent research assistance was provided by Ángela Fonsecaand Mónica Hernández.

which results in fares set above competitive equilibrium levels; misalign-ment of the incentives of bus drivers and owners (a typical principal-agentproblem); and congestion and pollution.1 In many developing countries,these market failures are exacerbated by weak regulation and enforce-ment. In Bogotá over the last three decades, city transportation was in thehands of private entrepreneurs, and local authorities were in charge of reg-ulating the system and maintaining the road infrastructure. This arrange-ment suffered to some degree from all of these market failures. At the endof the 1990s, the system had an excess supply of usually empty and slowbuses, low-quality service, and widespread inefficiency. Average traveltime to work was one hour and ten minutes; obsolete vehicles were used;average speed was only ten kilometers per hour during peak hours; 70 per-cent of air pollution in the central corridors was generated by traffic; andaccidents were frequent.

TransMilenio, a hybrid public-private scheme, was designed to over-come these market failures and improve urban transport quality. The firstmarket failure, the unclear definition of property rights on the curbsideand on the road, was solved via exclusive lanes and the construction ofrestricted-access elevated stations. Passengers pay for access to a systemof buses and stations, as in an underground metro system, and not for accessto the vehicles, as was previously the case. TransMilenio uses the left laneof the streets, and there is no staircase for accessing the buses, which facil-itates transfers in different directions and improves the speed of passen-gers’ movements.

The second failure TransMilenio addressed involved fares. These weretraditionally set above competitive equilibrium levels, largely as a resultof capture of the regulator. The theoretical literature identifies reasons formarket power by the bus owners that can also support overpricing. Over-investment in busses was profitable, leading to excess supply, low use, over-crowding of streets, low speeds, and pollution. In contrast, TransMileniofares are set at the level at which they finance the long-term cost of provi-sion, defined through a route-tendering process in which potential pro-viders compete for the exclusive use of the roads based on the lowest costbidding. Fares evolve based on the change in input prices and the numberof passengers transported.

TransMilenio also faced a third market failure, which resulted from theprivate solution to the agency problem between affiliating firms and bus


1. Estache and Gómez-Lobo (2005).

driver-owners. Under the traditional scheme, firms extracted their revenuesfrom affiliated buses, while the driver-owners’ revenues depended on thenumber of passengers. This arrangement caused an excess supply of busesand infighting among them for each extra passenger, with undesirable effectson safety, pollution, and congestion. TransMilenio introduced a new arrange-ment in which bus owners are private firms that charge per kilometertraveled, not per passenger, and drivers are employees whose salary is deter-mined by a labor contract that is unaffected by the number of passengerstraveling. The public sector participates in the network configuration andsystem regulation and supervision, and the private sector operates the buses.In addition, a different private company is in charge of fare collection.

After its launch in January 2001, TransMilenio significantly improvedtraveling conditions for its users and reduced traffic on its corridors. Thesystem reduced travel times for TransMilenio users by 32 percent, partic-ulate matter pollution fell by 9 percent in some areas of the city, and acci-dent rates dropped by 90 percent in the TransMilenio corridors.

But there have been problems, as well. Perhaps the most significantshortcoming of TransMilenio is related to the transition between the tradi-tional transportation system and the new one on currently unserved corri-dors. To secure support for TransMilenio, the local government providedpolitically powerful traditional transportation firms with more routes in theremaining corridors, where they relocated most of the vehicles displaced byTransMilenio. Congestion, pollution, and travel times have worsened forriders of the traditional system.

We carry out a cost-benefit analysis to assess the impact of the firstphase of TransMilenio on travel times, the environment, and traffic acci-dents in the whole city. Despite the sizable benefit that TransMilenio hasbestowed on its users, the net benefits for the whole city of implementingthe first phase appear negative. This is the result of the spillovers causedby the slow scrapping rates and the maintenance of weak regulation of thetraditional system.2

This paper provides policy lessons for cities in developing countries plan-ning to undertake similar reforms in this industry. The cost-benefit analysisidentifies policy mistakes during the implementation of TransMilenio that

Juan Carlos Echeverry, Ana María Ibáñez, Andrés Moya, and Luis Carlos Hillón 153

2. Contracts for TransMilenio private operators required them to buy traditional systembuses and scrap them, to prevent those buses from being relocated to unserved TransMileniocorridors. The authorities did not effectively enforce this process, however, so traditionalsystem buses that were displaced by TransMilenio’s operation were, in fact, relocated toother corridors.

led to negative spillovers and reduced the benefits of TransMilenio. Rec-ognizing these risks up front is useful for avoiding similar mistakes incities that are planning to reform their mass transit system. The paper isorganized as follows. The next section reviews the market failures pre-vailing in the provision of public transport, describes Bogotá’s mass tran-sit system before TransMilenio, illustrates the main characteristics of thenew system, and discusses some political economy issues of its imple-mentation. Both positive and negative changes in the quality of life result-ing from the new mass transit system, as well as its particular adoption anda cost-benefit analysis, are included in the subsequent section. A final sec-tion presents our concluding remarks.

Mass Transit Market Failures in Bogotá and the TransMilenio Reform

Many cities in developing countries have experienced a pendulum in theoperation of their public transit system. Bogotá, for instance, evolved fromprivately owned trolley buses at the end of the nineteenth century to a fullypublic bus system in the first half of the twentieth. In the second half of thelast century, developing countries witnessed a slow transition back to pri-vate ownership owing to the lack of flexibility and productive inefficiencythat characterized state ownership and operation. The mixed success of thelast decades regarding private provision, liberalization, and market compe-tition, especially in cities in developing countries, led to a reconsiderationof this model and the adoption of so-called hybrid models that combinepublic and private features. The main reason for this paradigm shift is theevidence of market failures that hinder a fully private provision from reach-ing socially optimal outcomes.

Market failures in the provision of urban mass transit include uncleardefinition of property rights on the curbside and on the road; the fact thatfares are set above competitive equilibrium levels; a principal-agent prob-lem stemming from a possible misalignment of the incentives of bus driversand owners; and externalities of street congestion and air and noise pollu-tion. In addition to these market failures, developing countries frequentlysuffer policy failures in the form of weak regulation and enforcement.

Property rights usually are not clearly defined on the road or on thecurbside.3 The absence of properly defined “rights to waiting passengers”


3. Klein, Moore, and Reja (1997).

implies that the fare received from a pedestrian at the curb does not belongexclusively to the firm authorized to operate on that route. In the absenceof regulation and control, other means of transportation can interlope andoffer pedestrians a ride, which they may accept because waiting time iscostly.4 The implication of this lack of property rights differs in developedand developing countries. Demand for mass transit in developed countriescan be low (that is, markets are thin), in which case it needs to be inducedby a regular, high-quality service. Once this demand is created, incentivesarise for illegal interlopers to start operating. Hence investments by legaloperators may not be recuperated. This induces underinvestment. In con-trast, demand in developing countries tends to be high (that is, markets arethick).5 Excessive bus entry is the norm, resulting in strong competition forpassengers on the curb. This strong competition spurs distortions ininvestment, such as deficient service quality and the use of small vehicles,which are more maneuverable but produce more pollution, congestion,larger investment per seat, and safety problems.6

In developing countries, fares are often set above competitive equilib-rium levels. This promotes excessive entry of buses. Because buses are notperfect substitutes, price competition is not an effective mechanism forregulating the optimal quantity of buses in the market. To minimize wait-ing time, riders prefer to use the first bus that arrives even though a cheaperbus may come along in a few minutes. Time, not fares, might be the mostimportant decision variable for the rider, so the bus can exercise its mar-ket power by raising fares. As a result, prices are set above efficient levelsand returns on investment are high, creating incentives for entry and anexcessive number of buses on the road.7 This feature can also result from,or be exacerbated by, the capture of the regulator. The excess supply ofbuses, paired with deficient service, leads to congestion, pollution, and traf-fic accidents.

Another source of market failures results from the private solution to theprincipal-agent problem. This solution, which tries to cope with the mis-alignment between the interests of the bus driver and the bus owner, intro-duces further inefficiencies in the provision of mass transit. The profits ofbus owners depend on the number of passengers carried per bus. To alignthe interests of bus owners and drivers, the owners typically pay the drivers


4. Evidence shows that passengers consider waiting time more costly than travel time.5. Most of the population in these countries lacks transportation alternatives.6. Estache and Gómez-Lobo (2005).7. Evans (1987).

based on the number of passengers carried. This contract between the ownerand the driver, however, introduces further complications for the provisionof mass transit. On the one hand, bus drivers compete for passengers tomaximize their payments, causing negative externalities in terms of safetyproblems and congestion. On the other, this compensation scheme pro-motes the use of smaller buses, despite the congestion and pollution prob-lems, because they are better equipped to compete for passengers.

The problems described above are deepened and new inefficienciesarise when drivers are responsible for fare collection. High monitoringcosts promote the sale of vehicles to the drivers and cause the atomizationof the bus industry in terms of bus ownership. Although this atomizationis effective for aligning the interests of the owner and drivers, it leads toadditional costs. The large number of owners makes it difficult to exploiteconomies of scale on coordination and economies of density. Deficientregulation, weak enforcement, and capture of institutions also worsen mar-ket failures.

The Mass Transit System in Bogotá before TransMilenio

Essentially all of the market failures mentioned above characterized thepublic transportation system of Bogotá in the second half of the twentiethcentury. This industry went through three periods in terms of ownershipand regulation: (i) simultaneous provision and ownership by the state andthe private sector (through the mid-1980s); (ii) private firms specialized inintermediating routes between the regulator and the bus owners and drivers(mid-1980s to the early 1990s); and (iii) private provision and the emer-gence of different types of buses, with fares dependent on the quality ofservice, such as carrying seated passengers only, providing clear bus stops,and opening new routes (early 1990s to the present).

In an initial phase, mass transit was exclusively provided by a publiccompany, resulting in economic inefficiencies, an excess of drivers pervehicle, and inflexibility in supplying sufficient transport services and newroutes for a growing city.8 Increasingly, the regulatory agencies found thatprivate firms could supply new routes and either complement or replace theexisting ones at cost-efficient levels. The resulting private scheme hadthree types of actors: the local government, bus owners, and firms acting asintermediaries between them. Each bus owner operated a specific routeafter purchasing one of the slots approved by the authority and then paying


8. Urrutia (1981).

a monthly affiliation fee to the intermediary.9 Originally, the intermediaryfirms owned the buses, and the drivers were in charge of fare collection.The drivers’ income depended on the number of passengers carried, whichserved to align the interests of drivers and bus owners. However, this solu-tion led to strong competition among drivers of different companies. More-over, most routes sought to cover the central part of the city, where thehighest number of passengers is concentrated. Overlapping routes exacer-bated traffic congestion, street competition, and traffic accidents.

In the early 1980s, fare setting depended on recurrent negotiations withthe regulator and local authorities and on a subsidy system implementedto avoid transferring high fuel costs to passenger fares. The high return onthe bus investment promoted oversupply, as fares and subsidies surpassedmarginal costs. The gradual elimination of subsidies in the 1980s led totwo developments. First, higher fares were granted to unsubsidized vehi-cles, which were obliged to comply with higher quality standards thantheir subsidized counterparts. The average fare increased from U.S.$0.10in 1975 to U.S.$0.40 in 2004, in constant 1999 dollars (see figure 1). SinceBogotá did not experiment with fare liberalization, these fare incrementswere mostly the result of the strong bargaining power that private trans-port firms attained in the 1980s.10 Second, the absence of price competi-tion made the market a poor regulating mechanism for the optimal quantityof buses. Returns to investment were high, which created incentives forthe entry of an excessive number of buses. Figure 2 illustrates the evolu-tion of the public bus fleet in Bogotá. The increasing trend accelerated inthe first half of the 1990s, peaking at approximately 22,000 vehicles. Thiswas accompanied by a sharp decline in average occupancy after 1985. Thedaily number of passengers per vehicle stabilized briefly in the early 1990sand then fell dramatically from 538 in 1992 to 294 in 2003. Frequent farehikes offset the subsequent income losses and served as an incentive forthe entry of more vehicles, resulting in further reductions in the number ofpassengers per vehicle.

The industry further separated the affiliating firms from the bus ownersand drivers. The atomization of the ownership structure was caused by thehigh costs of monitoring the drivers’ revenue collection. The solution of


9. In 1995 the cost per slot was between U.S.$2,300 and U.S.$4,600, the affiliation feeranged from U.S.$100 to U.S.$600, and the monthly fee varied between U.S.$5 and U.S.$35(Lleras, 2003).

10. Private transporters started to participate in politics either directly via their own can-didates in the city council or indirectly via funding the campaigns of traditional politicians.

this agency problem led the affiliating firms to specialize in obtainingroutes from the authorities and affiliating as many buses as possible. Sincethe aging of the existing fleet increased operating costs, the firms sold oldvehicles, usually to the drivers themselves, and owned only new buses.11 Atthis point, the number of vehicles on the streets was completely determinedby the affiliating firms, because the public company had been liquidated.The firms obtained as many routes as possible and earned their incomebasically from affiliation rather than from the number of passengers trans-ported. The firms also established a cartel to consistently push for bus farehikes and the authorization to extend the service life of buses.12 This struc-


11. The increase of driver-owners as a solution to the limited monitoring capacity of theaffiliating firms is also documented for the case of Santiago de Chile (Estache and Gómez-Lobo, 2005).

12. Bus fares were historically set through negotiation processes with the local author-ities. Only in 1997 was an effort put forth to tie them to a transport costs basket (Decree3109 of 1997). Estache and Gómez-Lobo (2005) propose a model for explaining the over-provision of services in a private equilibrium when fares are too high. Their key argumentis the fact that buses are not perfect substitutes, but differentiated products. Hence the busstopping in front of a passenger enjoys a “market power” over its competitors due to the cost

300

350

400

450

500

550

600

650

1970 1975 1980 1985 1990 1995 2000 2005Year

Average occupancy

10

15

20

25

30

35

40

45Fare (1999 U.S.$ cents)

Passengers per vehicleFare

Source: Authors’ calculations, based on data from the National Administrative Department of Statistics (DANE).

F I G U R E 1 . Bus Fare and Average Occupancy, 1973–2003

ture intensified the struggle for economic rents, increased the number ofbuses on the streets, and contributed to the aging of the vehicle fleet. Mar-ket and policy failures thus accentuated the problem of oversupply.

We estimated the economic rents for Bogotá’s traditional mass transitsystem by simulating the contracts between affiliating firms and owner-drivers (see appendix A for a description of the methodology). This indus-try operated at efficiency levels (nearly zero profits) in only two periods:the end of the 1970s and the beginning of the 1990s (see figure 3). Outsideof these two periods, economic rents were positive, substantial, and highly


of waiting, which is highly valued by eventual passengers; this permits the charging of ahigher fare. Consequently, rising tariffs create excessive returns to investment and promotethe entry of new buses beyond the socially efficient level (see pp. 12–15). The capture of (orpolitical influence on) the regulator is yet another reason for high fares. As a result of exces-sive entry, buses in Bogotá and Santiago, Chile, have witnessed a sharp decline in occu-pancy, as illustrated below.

1,000

2,000

3,000

4,000

5,000

6,000

7,000

8,000

9,000

10,000

1970 1975 1980 1985 1990 1995 2000 2005Year

Busetas

5,000

10,000

15,000

20,000

25,000Vehicles

BUSETAEJECUTIVASUPEREJECUTIVATOTAL BUSETASVEHICLES

Source: Authors’ calculations, based on data from DANE. a. While busetas carry an average of 25 passengers and operate with low-quality standards, busetas ejecutivas and superejecutivas, introduced in the early 1990s, offer an average capacity of 50 seated passengers and better service and, consequently, require higher fares than traditional busetas.

F I G U R E 2 . Public Transit Vehicle Fleet in Bogotá, 1970–2003a

variable. These rents rose in the 1980s, when fares grew from U.S.$.10 toU.S.$.30, which was associated with the introduction of smaller capacityvehicles. Rents accrued to both the fleet administrating firms and the busowner-drivers.

As figure 3 illustrates, however, the composition of rents changed in the1990s, favoring the affiliating firms vis-à-vis bus owner-drivers, and totaldaily rents never returned to the peak observed at the end of the 1980s. Anew development was observed at the beginning of the 1990s, when theaffiliating firms lobbied for further fare increases based on different levelsof service quality. The so-called executive buses carried only seated pas-sengers, used relatively new vehicles, and offered faster service at higherfares than traditional buses. This behavior reveals either the capture of theregulator or the authorities’ ignorance regarding the true costs of bus trans-portation, or both.

In sum, the market failure stemming from the absence of price competi-tion in urban bus transport was exacerbated by a policy failure that granted


2.00

4.00

6.00

8.00

10.00

12.00

14.00

16.00

18.00

20.00

19731974

19751976

19771978

19791980

19811982

19831984

19851986

19871988

19891990

19911992

19931994

19951996

19971998

19992000

20012002

2003

Year

Bus drivers (1999 U.S.$)

4.00

8.00

12.00

16.00

20.00

24.00

28.00

32.00

36.00

40.00Fleet administrator (1999 U.S.$)

Driver-owners

Fleet administrator

Source: Authors’ calculations, based on DANE (2003).

F I G U R E 3 . Daily Simulated Economic Rent per Vehicle, 1973–2003

free affiliating privileges and periodic fare negotiation with a cartel, lead-ing to systematic economic rents. In addition, this incentive scheme forcedbus drivers to compete for passengers along the central corridors. Driverstherefore did not respect assigned bus stops or delineated areas for bustransit in the pursuit of passengers.13

At the end of the 1990s, the excess bus supply, coupled with a substan-tial increase of private vehicles, exceeded the traffic capacity of Bogotá.From 1991 to 1995, the total number of cars registered in Bogotá increasedby 75 percent, and 40 percent of the country’s vehicles were circulating inthe city.14 By 1998, private cars occupied 64 percent of the road space andmobilized only 19 percent of the population.15 As a result, gridlock wascommonplace, accidents abounded, and travel times were unbelievablyhigh. Estimates show that in 1995 the average number of daily trips perhousehold was 11.9, the average number of daily trips per person was 1.7,and the total number of daily trips was about 10 million.16 The averagespeed for public transportation during peak hours was ten kilometers perhour, which could drop to five kilometers per hour in the center of thecity.17 Inequality of travel times between public transport users and carowners was sizeable. Mean travel time was 66.8 minutes for public trans-port users, whereas car owners faced a mean travel time of 42.6 minutes.18

These problems finally led to a rethinking of the entire bus transport sys-tem in Bogotá, which culminated in the design and implementation of thefirst phase of the TransMilenio system in 2001.

The Features of TransMilenio

The authorities designed and adopted TransMilenio to tackle the conse-quences of the market and policy failures described above. The key ele-ments of the new system are concession contracts for service providers(namely, TransMilenio transporters and feeder buses); vertical separationof the transportation service and fare collection; bus remuneration basedon kilometers traveled rather than passengers transported; fare setting based


13. The road infighting among drivers for every extra passenger is locally known as thepenny war (la guerra del centavo).

14. Lozano (2003).15. Chaparro (2002).16. In 2000, Hong Kong, which has a population similar to Bogotá (7,394,170 inhabi-

tants), reported 12 million total daily trips.17. Chaparro (2002).18. Lleras (2003).

on long-term investment recovery following a tendering process (that is,competition for the road); exclusive road lanes (the left lane of each road,to facilitate turning at intersections by private automobiles); and exclusivecurbside service in metro-like stations, which constitute an organized sys-tem of express and slow routes and facilitate transfers.

As Lleras states, TransMilenio is “a flexible, rubber-tired rapid transitmode that combines stations, vehicles, services, and driving lanes . . . intoan integrated system.”19 The first phase of TransMilenio was designed for35,000 passengers per hour per direction; it covers 42.4 kilometers ofexclusive bus lanes along three of the main transit corridors. In these cor-ridors, central lanes are dedicated exclusively to TransMilenio operations,and passengers at the stations are their exclusive customers. The stationsare in the middle of the road like a metro system, which facilitates bustransfers. The bus fleet consists of 470 buses for the first phase of the system,with better mechanical conditions and environmental performance thanthose of the traditional system. The bus stops include fifty-seven stations,located every 700 meters, equipped with pay booths, registering machines,surveillance cameras, and infrastructure such as bridges, pedestrian cross-ings, and traffic lights designed to ease the entrance of passengers into thesystem. At the end of the corridors, three principal access stations serve asa meeting point for feeder buses and buses from the traditional system thatwork in neighboring municipalities. Feeder bus passengers have an inte-grated tariff, so riders do not have to pay twice for using the feeder systemand TransMilenio. Feeder buses, which share corridors with the traditionalsystem, have bus stops every 300 meters in the lower socioeconomic areasof the city and are synchronized by the operators with a satellite system tominimize travel and wait times for passengers.20

TransMilenio solved the agency problem present between owners anddrivers by establishing a prepayment scheme, in which users buy tickets inbooths located in the stations, as in underground systems around the world.Therefore, bus drivers do not have to deal with collecting payment fareslike they do in the traditional system. Previously, drivers’ incomes dependedon the number of passengers traveling, whereas TransMilenio drivers aresalaried employees with no direct relation to fare collection.

The market failure deriving from an opaque property right definition forroads and curbsides, which is present in different degrees in thick and thin


19. Lleras (2003).20. Each TransMilenio bus has a capacity of transporting 160 passengers, while feeder

buses can transport seventy passengers.

mass transit markets, was solved via exclusive bus transit lanes and exclu-sive stations.

Finally, the recurrent problem of setting fares above market equilibriumlevels was solved with concession contracts awarded through public compe-tition. The concessionaires’ income is no longer associated with the numberof passenger-trips serviced. Transport providers, both local and foreign, arecalled to associate and participate under a new set of rules. The new transportfirms own a number of buses that meet certain specifications and whose oper-ation is subject to the leadership of a central authority, TransMilenio S.A.These buses have a specified service life stipulated at the outset of the con-tracts. The concessions expire when the vehicle fleet reaches an averagemileage of 850,000 kilometers, with no individual bus reaching more than1,000,000 kilometers. If the average mileage threshold is reached in less thanten years, then the concession extends to the tenth year. If the averagemileage threshold is not reached by the tenth year, the concession lasts untilthis happens or until the fifteenth year, whichever comes first.21 The new con-tracts also establish a payment per kilometer traveled. The number of kilo-meters traveled depends on the manner in which the central authoritydispatches service and, ultimately, on demand. The reward per kilometertraveled is the central variable by which the firms compete for these contracts.

TransMilenio S.A. carries out thorough calculations of the costs in-volved in the provision of the transportation service, as well as demandforecasts, which enables the authority to set the range of acceptable feesper kilometer traveled. This range guarantees a fair return to the partici-pating investors and is used as a parameter in the selection process of theconcessionaires.22 TransMilenio thus represents a hybrid private-publicmodel that replaces competition on the road by competition for the road.23

These measures turned the entire incentive scheme of the previous sys-tem upside down. They eliminated the affiliating firm and the license busi-ness. They also removed any gains from cutting maintenance costs, sincethe service is to be provided over a specified period under safety and effi-ciency standards: cutting down on costs today implies incurring greatercosts tomorrow to comply with the norms. These measures gave rise to a


21. There is a restriction by which the concessionaire is forbidden to add new vehiclesto the fleet to deliberately bring its average mileage down when a certain portion of the con-cession period has expired. This contract resembles those analyzed by Engel, Fischer, andGaletovic (2001).

22. The range is given by a real return on investment between 14 percent and 16 percent.23. Estache and Gómez-Lobo (2005).

true transportation firm able to earn a return on its capital comparable towhat it would receive in any other business of comparable risk.

Feeder buses are also organized through concession contracts awardedthrough competition. These contracts are slightly different from the mainservice contracts: they are awarded for a ten-year period subject to the con-dition that no individual vehicle should exceed 950,000 kilometers duringthe time of the concession, and income to the feeder buses is defined as acombination of revenues paid per kilometer traveled and per passengerserved. The reward system is designed to prevent opportunistic behaviorfrom the transport firms. Feeder buses bring passengers from outskirt loca-tions into the TransMilenio system, but most of the time they do not tran-sit on exclusive public transport lanes, making it difficult to control thekilometers they travel. The regulating authority, TransMilenio S.A., hascalculated the maximum fixed operation costs that a firm will incur perbus, and it agrees to remunerate the firms based on the number of passen-gers carried and kilometers traveled. Under this rule, traveling withoutpassengers becomes unprofitable for the feeder bus.

Taking the responsibility of collecting fares away from the bus driversand centralizing the management of fare revenues under an independentfiduciary entity further contribute to the improved operation of the new sys-tem.24 Booths at the TransMilenio stations collect the system’s revenuesand are managed through a fiduciary contract. This not only relieves thesystem of information asymmetries and aids in the collection of taxes, butalso has a considerable impact on road safety.25

The last important element of the new scheme is the fare-setting proce-dure. Under TransMilenio, travel fares cease to be negotiable. The writtencontracts subject the concessionaires to operate under the fares set by thepublic regulating authority, TransMilenio S.A. The fare-setting procedureand fare adjustment over time are part of the contracts.26 TransMileniotravel fares are set to cover the long-run average costs of operation, includ-


24. In the previous system, passengers paid the drivers on entering the buses, which wasa distraction to the drivers and contributed to congestion and accidents.

25. At the moment, the regulator lacks a direct monitoring instrument for the fare col-lection contract.

26. The procedure for calculating the technical fare for TransMilenio comprises foursteps. First, the basis of the calculation is the pesos per kilometer offered by the main trans-port and feeder transport concessionaires, and the pesos per ticket offered by the revenuecollection concessionaire. The calculation also contemplates the cost of the fiduciary rev-enue management and TransMilenio’s management (both of which enter the formula as per-centage shares). Second, the weight of each input in total operation costs is calculated using

ing the administration costs of the regulating authority and costs of thefiduciary contract through which fare revenues are managed.

Authorities decided during the design stage that fares would not be setto recover the infrastructure investments, since that would render the sys-tem unfeasible. TransMilenio fares set to recover the system’s operatingcosts were already significantly higher than the current bus fares. Imposingon TransMilenio the obligation of recovering the infrastructure investmentsor its maintenance costs would have resulted in negative profitability,thereby eliminating the participation of the private sector. In addition, theintervention on TransMilenio corridors includes a full renovation of twolanes for private cars and the curbside; benefits from this investment accruenot only to TransMilenio providers and users, but also to other populationgroups. Thus investment in the infrastructure required to operate the systemwas said to be social investment.

After the opening of the first TransMilenio lines, traditional buses con-tinue to mobilize nearly 80 percent of the population. The TransMilenioconcession contracts obligated the operators to purchase and scrap busesthat previously operated in TransMilenio corridors. Only 1,410 of thesebuses were effectively scrapped, however, and the remaining 4,670 buseswere relocated into unserved TransMilenio corridors. Although thedemand for the old system dropped, causing revenues per vehicle to fall, aloose regulation of the system permitted the excessive entry and overpro-vision of services in these corridors. Moreover, the fares for the traditionalsystem were still allowed to rise after TransMilenio came into operation,as illustrated in figure 4. The introduction of TransMilenio thus did notspur a substantial reduction in the transport fleet of buses from the tradi-tional system.

Some Political Economy Considerations of TransMilenio’s Adoption

Before the adoption of TransMilenio, the local authorities’ biggest concernwas the opposition the system could face from traditional transporters.These companies had derived substantial rents over the decades, and they


the numbers contained in the contracts, and costs are adjusted assuming a monthly averageof 6,400 km per vehicle. Third, to conform to available coin values, the passengers’ fare isrounded to the nearest multiple of fifty. Finally, the difference per ticket goes into a fundthat compensates for cost increases not reflected in the fare. Costs are revised monthly, andthe fare is revised (up or down) to the next closest multiple of fifty when the technical farechanges by more than twenty-five pesos.

might feel threatened by the TransMilenio intervention. However, the firmswere also aware that problems related to public transportation, such as con-gestion, pollution, and accidents, had reached worrying levels. A growingconsensus thus emerged in the late 1990s in favor of a regime change, inthe form of either a subway or a transformation of the bus system.

Bogotá witnessed several episodes of protest against the new system.27

The strikes and protests lost momentum quite rapidly, however, anddemonstrations on other issues consistently outnumbered TransMilenio-related incidents. Between 1999 and 2000, five events were organized againstTransMilenio, and TransMilenio-related protests represented only one-fifthof total political demonstrations in the sector even at their peak. Activityagainst this initiative had already ceased by 2001.

An important factor facilitating the transition was the national govern-ment’s decision to support a new transportation system for Bogotá. Thecentral government’s willingness to partly finance the associated infra-structure investments became a critical argument to convince transportersthat this was the right time for reform.28


27. Estache and Gómez-Lobo (2005) document similar reactions in the case of Santiago,Chile.

28. In the 1990s, Colombian presidents supported either a subway or a bus solution.Elected Bogotá mayors followed suit because any solution would require substantial financialresources. Consequently, local authorities’ choices regarding the new public transport systemwere partly determined from the outside. The city authorities revealed the TransMilenio blueprint in 1999 after a new national government rejected the subway alternative for Bogotá.

0

200

400

600

800

1000

1200

Dec-95Apr-9

6

Aug-96

Dec-96Apr-9

7

Aug-97

Dec-97Apr-9

8

Aug-98

Dec-98Apr-9

9

Aug-99

Dec-99Apr-0

0

Aug-00

Dec-00Apr-0

1

Aug-01

Dec-01Apr-0

2

Aug-02

Dec-02Apr-0

3

Month

Fare

Buses

TransMilenio

Aug-03

Dec-03

Source: TransMilenio S. A.

F I G U R E 4 . Nominal Fares: Buses versus TransMilenio, 1995–2003

Finally, the municipal government decided that any solution would bepursued in association with the traditional transportation firms of Bogotá,and they used the allocation of new routes in the non-TransMilenio corri-dors to reward affiliating firms for their willingness to participate in Trans-Milenio. The traditional transporters relocated their buses to new routesgranted by TransMilenio authorities themselves, and not by the traditionalregulator. The municipal representatives thus chose not to abandon the oldbusinesses in the proportion in which TransMilenio replaced them, butsimply to relocate them. Unfortunately, this produced damaging effects onthe other corridors in terms of speed, congestion, pollution, and riding time,as illustrated below.

Cost-Benefit Analysis of the First Phase of TransMilenio

The ultimate goal of TransMilenio is to improve the quality of life inBogotá. Although only 25 percent of the system has been put in place sofar, its impact is indicative of what can be achieved. Travel times for Trans-Milenio passengers have dropped, and traffic congestion, air pollution, noiselevels, and frequency of traffic accidents have decreased significantly inTransMilenio corridors. However, some unexpected negative spillovershave emerged. Traffic congestion and pollution have heightened alongcorridors not served by TransMilenio owing to the slow scrapping rate ofbuses in the traditional public transit system. This section presents a cost-benefit analysis of the first wave of TransMilenio contracts.

The Impact of TransMilenio on the Quality of Life in Bogotá

Soon after TransMilenio came into operation, many public transporta-tion users switched to this new public transit mode. The year before theintroduction of TransMilenio, 69 percent of individuals relied on publictransportation. After TransMilenio became available in 2001, 6 percentof individuals switched from traditional public services to the new alter-native. This share of the population using TransMilenio has expanded con-tinuously, and the new system served 13 percent of the population in2003.29 TransMilenio has not increased the overall attractiveness of publictransit, however, since the share of the population using cars has remained


29. The first phase of TransMilenio, which represents 25 percent of the total expectednetwork, covers 13 percent of the public transportation demand in Bogotá.

constant since its adoption. Thus the demand for public transport as wholedid not expand, but rather the appearance of a new transportation mode ledto a distribution of this demand between TransMilenio and the traditionalsystem.

Demand for public transportation stems mainly from the lower-incomehouseholds, yet TransMilenio users are concentrated in the fourth andfifth income quintiles, as seen in figure 5. In contrast, demand for tradi-tional buses is primarily from the first, second, and third quintiles. Thehighest income quintiles thus use TransMilenio more than the lower in-


A. Transport Mode by Income Quintile

0

25

50

75

Income quintile

Percent ofpopulation

Bus

Private vehicle

B. TransMilenio Usage by Income Quintile

0

3

5

8

10

1 2 3 4 5

1 2 3 4 5

Percent of population

Income quintile

Source: Authors’ calculations, based on DANE (2003).

F I G U R E 5 . Use of Public Transportation, by Income Quintile

come quintiles.30 The top two quintiles are also the main users of privatevehicles.

TransMilenio’s impact on travel times and average speed varies widelyacross users of TransMilenio, private vehicles, and the traditional publictransport system. Two years after TransMilenio began operating, the aver-age trip time in Bogotá decreased to 35 minutes, from 44 minutes in 2001.The average speed of cars increased to 20 miles per hour, from 16.8 milesper hour in 2000.31 Table 1 shows that the percentage of individuals whospent more than an hour traveling from home to work fell from 23 percentto 17 percent, while the frequency of trips under one hour rose during 2003.

The benefits of travel time reductions accrued mainly to TransMileniousers. The average speed of other forms of public transportation actuallydropped, which caused travel times to increase by 10 percent.32 Slow scrap-ping of buses from the traditional system may be causing the uneven dis-tribution of benefits. An important share of these buses were relocated tocorridors not served by TransMilenio, worsening congestion. Attributingworsening congestion solely to the relocation of buses is not accurate,however; slow adoption of other measures to control traffic congestionand the increase in the number of taxis, as a result of simultaneous inter-ventions, are partially responsible, as well.

Lleras studies TransMilenio’s impact on public transportation users,mainly through travel times.33 In 2002 a revealed preference survey was


30. Of all TransMilenio users, 62 percent (325,925 users) correspond to the fourth andfifth income quintile, while 38 percent (203,830 users) correspond to the first three incomequintiles (DANE, 2003).

31. See www.transitobogota.gov.co.32. To date, the impact on travel times for riders using unserved corridors has not been

officially measured. The union of small providers of public transportation argues that traveltimes in unserved corridors increased by 10 percent as a consequence of bus relocation(“Pico y Placa seguirá en discussion,” El Tiempo, 11 March 2001).

33. Lleras (2003).

T A B L E 1 . Time Spent Traveling from Home to WorkPercent of individuals

Time spent 2001 2002 2003

Less than one hour 74 68 74Between one and two hours 23 29 17More than two hours 4 2 5

Source: Authors’ calculations, based on Napoleon Franco (www.napoleon.com.co)

applied to 2,095 public transport riders who could choose between Trans-Milenio and the traditional system for their transportation. Respondentswere interviewed in the street near areas where routes from the two sys-tems run. The survey elicited information about the transportation modeselected, the expected attributes of the trip, and various socioeconomiccharacteristics. The data were used to estimate random utility models,which provide the coefficients to calculate the value of time for users ofboth systems in the different stages of the process (for example, walk-inand walk-out times). The study shows that declines in travel times werenot uniform across TransMilenio users. TransMilenio passengers startingthe trip in the vicinity of the main corridors travel 12 minutes less per tripthan passengers of the traditional system. In contrast, passengers requiringone or more transfers did not experience drops in travel times. In fact, totaltravel time is two minutes shorter in the traditional system because of thewaiting time required for the TransMilenio feeder routes.34

Waiting-in and waiting-out time increased for all TransMilenio users,however, because these times are extremely low for the traditional bussystem.35 TransMilenio users have to buy tickets, wait in line, exit the sta-tion, and walk to the final destination, whereas users of the traditional sys-tem enter the bus at any location, pay the bus driver directly, and stop thebus at the point nearest to their destination, since there are no official busstops in Bogotá. Improvements in TransMilenio travel times arise, there-fore, from in-vehicle travel time.

Estimations of the value of time for TransMilenio passengers, vis-à-visusers of the traditional system, show unambiguous improvements in trav-eling conditions for the former. Overall, people are willing to pay less forsavings in travel time than people using the traditional system, indicatingthat TransMilenio is a less “painful” experience. For example, the valueof waiting time for the traditional system is U.S.$3.08 per hour, while inTransMilenio it is U.S.$1.14 per hour.36

These figures indicate that the first phase of TransMilenio enhancedtravel conditions for a certain population group, although improvements


34. Lleras (2003).35. Waiting-in time refers to the time spent waiting for the bus; waiting-out time is cal-

culated from the time the passengers step off the bus until they arrive at their destination.Lleras (2003) estimates that waiting-in and waiting-out times increased 2.95 and 5.16 min-utes, respectively, for TransMilenio users vis-à-vis users of the traditional public transportsystem.

36. Lleras (2003).

should not be entirely attributed to the new mass transit system.37 Never-theless, benefits from the new system are not distributed widely across theentire population; gains were perceived largely by TransMilenio users, inparticular downtown dwellers.

The quality of services also improved, and TransMilenio corridorsexperienced a decline in the external costs caused by public transport—namely, accidents and pollution. Traffic accidents decreased dramaticallyin TransMilenio corridors. Figure 6 depicts the incidence of traffic colli-sions, pedestrian accidents, injuries, and fatalities in TransMilenio corridorsbefore and after the system came into operation. After two years, trafficcollisions and pedestrian accidents decreased by 94 percent, injuries topassengers by 76 percent, and fatalities by 94 percent. This is, by all accounts,an impressive performance.

Finally, air pollution, which is a major concern in Bogotá, exhibited adeclining trend in TransMilenio corridors. Half of the districts in the citycurrently exceed the particulate matter (PM-10) and ozone pollution lim-its. Studies indicate that automobiles are the most significant emission


37. Bogotá’s authorities implemented several programs to transform traffic conditions,including mobilization restrictions during traffic peaks, investment in road infrastructure,and increased traffic fines.

0

200

400

600

800

1,000

1,200

1,400

1,600

Traffic collisions Pedestrian accidents Injuries Fatalities

Number of incidents

Before TransMilenio20022003

Source: TransMilenio S. A.

F I G U R E 6 . Traffic Accidents in TransMilenio Corridors before and after the Intervention

source in Bogotá, contributing 70 percent of air pollution.38 Reductions intraffic congestion increased the speed of other vehicles and thus curbedemissions in TransMilenio corridors. TransMilenio also appears to haveimproved air quality by transporting more passengers in less time with bet-ter vehicles.39 In 2000–02 the citywide average of PM-10 emissions grewby 23 percent, but it fell by 8 and 11 percent in the two TransMilenio cor-ridors shown in figure 7. Although the environmental authorities of Bogotáadopted a group of measures to control air pollution over the last decade,evidence indicates TransMilenio contributed substantially in this respect.40


38. Cavallazi (1996).39. This superior performance of TransMilenio vehicles contributes to the control of

mobile source emissions. TransMilenio buses transport 1,596 passengers daily, which is fivetimes the number of passengers for traditional public buses, and the average speed is 17 to 44 percent faster. The average age of public buses is fifteen years. TransMilenio buses havecatalytic converters, emissions are below the limits required by the Euro II norm, some vehi-cles use natural gas, and noise levels are less than ninety decibels (Ibáñez and Uribe, 2003).

40. Measures include the use of catalytic converters for new cars, mandatory inspectionand maintenance, and mobilization restrictions.

0

20

40

60

80

Daily average -Monitoringstation 1

Daily average - Monitoringstation 2

Daily average - Bogotá

Micrograms per cubic meter

199920002001

Source: Authors’ calculations based on data from the Bogotá Air Quality Monitoring Network (Red de Monitoreo de la Calidad del Aire de Bogotá, or RMCAB).

a. The RMCAB maintains thirteen monitoring stations, located throughout Bogotá, that measure hourly emissions for a group of pollutants.

F I G U R E 7 . Average Daily Readings of Particulate Matter, Citywide and in MonitoringStations near TransMilenioa

Méndez analyzes the impact of several pollution programs imple-mented in Bogotá.41 Her study evaluates the evolution of particulate mat-ter and ozone in monitoring stations located near TransMilenio corridorsand far from industrial areas from 1997 to 2002. Méndez defines categor-ical variables to denote the implementation of three policy interventions(namely, the introduction of TransMilenio, mobilization restrictions,and mandatory inspection and maintenance) and then estimates timeseries regressions to predict the trend of particulate matter and ozonelevels, controlling for these policy interventions. Her results reveal thatTransMilenio is the most effective program for curbing pollution, with amuch stronger impact than programs specifically designed to controlemissions. Mandatory inspection and maintenance reduced ozone levelsby 13.6 percent, while restrictions on car mobilization lowered ozonelevels by 21 percent. The first phase of TransMilenio, however, pro-duced a 28.8 percent decline in ozone levels. Similar estimations for par-ticulate matter indicate that TransMilenio abated PM-10 pollution levelsby 9.2 percent.

Air pollution levels rose in other areas of the city, because the slowscrapping rate of buses from the traditional system triggered the reloca-tion of buses to non-TransMilenio corridors. To evaluate the negativepollution spillovers of TransMilenio, we used a difference-in-differencesapproach to compare readings from a monitoring station located near aTransMilenio corridor to readings at a baseline monitoring station. Thebaseline monitoring station should have two characteristics: it shouldhave similar particulate matter readings to the TransMilenio monitoringstation before entry into operation of the system; and it should be locatednear a non-TransMilenio corridor without negative spillovers from busrelocation. Other interventions in the city can also affect the results,namely, the entrance of new taxis and the construction of new Trans-Milenio corridors. The evolution of particulate matter in both monitoringstations is illustrated in figure 8.

Readings of PM-10 soared in the monitoring stations with negative spill-overs after TransMilenio fully entered into operation in June 2001.42 Itremained constant in the monitoring station without the negative spill-overs. The difference-in-differences calculations indicate that bus reloca-


41. Méndez (2004).42. TransMilenio started partial operations in January 2001, and the full system with the

three trunk corridors was in place in June 2001.

tion caused particulate matter emissions to increase by 10.5 percent, morethan offsetting the reductions in PM-10 readings in TransMilenio corri-dors. Although the change cannot be attributed solely to TransMilenio,this was the main intervention in Bogotá’s traffic system in the period.

Air pollution can cause serious damages to health, and particulate mat-ter and ozone levels appear to increase the incidence of acute respiratoryillnesses in Bogotá. Lozano estimates a concentration response functionthat links daily respiratory hospital admissions in 1998; the study finds astrong link between incidence of respiratory health admissions and partic-ulate matter levels.43 In fact, a 25 percent increase in particulate mattercauses a 21.8 percent rise in respiratory health admissions, while decreas-ing particulate matter emissions by 25 percent results in a 17 percent re-duction in respiratory health admissions.

The impact of the first phase of TransMilenio reveals positive resultsfor the people attended by the system, but ambiguous results for the rest ofthe city. Travel times for TransMilenio users fell, safety improved, con-gestion dropped, and air quality increased, yet these benefits are restricted


43. Lozano (2003).

0

10

20

30

40

50

60

70

Mar

-98

May

-98

Jul-9

8Se

p-98

Nov-

98

Monthly average PM-10

WithWith

Jan-

99M

ar-9

9M

ay-9

9Ju

l-99

Sep-

99No

v-99

Jan-

00M

ar-0

0M

ay-0

0Ju

l-00

Sep-

00No

v-00

Jan-

01M

ar-0

1M

ay-0

1Ju

l-01

Sep-

01No

v-01

Jan-

02M

ar-0

2M

ay-0

2Ju

l-02

Month

Without negative spilloversWith negative spillovers

Source: Authors’ calculations, based on data from RMCAB.

F I G U R E 8 . Particulate Matter Readings from Monitoring Stations with and withoutNegative TransMilenio Spillovers, 1998–2002

to TransMilenio corridors. Meanwhile, the relocation of buses increasedcongestion in non-TransMilenio corridors, causing a deterioration in trav-eling conditions for passengers of at least some corridors of the traditionalsystem and a rise in pollution levels. Once the full system comes into oper-ation, the negative spillovers should decrease significantly or disappear.

Cost-Benefit Analysis

Understanding the welfare gains and losses from the introduction of Trans-Milenio is crucial for evaluating whether investments in this new schemeproduced net benefits for the population and identifying future adjust-ments to the system. Welfare changes can be approximated by measuringvariations in the consumer and producer surplus caused by the adoption ofTransMilenio. We used the above information and data on investments,costs, and revenues of the system’s operation to conduct a cost-benefitanalysis. Only the first phase of TransMilenio (25 percent of the system)was evaluated. Ideally, we should estimate the consumer and producersurplus before and after TransMilenio and then calculate the changes inboth surpluses; however, data restrictions limited our analysis to calculat-ing changes in both surpluses as a result of the adoption of the system. Adetailed description of the assumptions and methodology used to performthe analysis is included in appendix B. Table 2 presents a summary ofresults.44

Who wins and who loses with TransMilenio? The previous sectionsuggests how benefits and costs are distributed among different groups ofthe population. We now undertake a detailed cost-benefit analysis to clar-ify those findings. On the production side, benefits arise for the privateTransMilenio operators but are lost for the owners of the scrapped buses.The producer surplus of private TransMilenio operators represents wel-fare gains. In contrast, the former producer surplus that the owners of thescrapped buses no longer collect represents welfare losses. The shadedareas depicted in figure 9 denote those welfare gains and losses.

On the demand side, the evaluation of TransMilenio must account forchanges in travel times faced by TransMilenio users and users of the tra-ditional system. An appropriate welfare indicator for these changes is thewillingness to pay for improvements in travel time. The value of time isdefined as the marginal rate of substitution between travel time and a mon-


44. Detailed calculations of the costs and benefits are available on request.

etary cost. Figure 10 presents welfare changes resulting from the adoptionof TransMilenio. When TransMilenio started operations, TransMileniousers experienced two welfare enhancing effects: travel time decreased,and the ride was more pleasant than on the traditional system.45 The move-ment from point A to point B in the indifference curve map for Trans-Milenio users shows the combined effect on utility; X2

0 − X 21 measures

willingness to pay for these improvements in traveling conditions. Ridersof the traditional system faced costs from increases in travel time, while theirtraveling conditions were practically unchanged. As a result, travel timesrose, but the marginal rate of substitution between savings in travel timeand the numeraire good remained constant. By moving from the status quo(point C) to the new condition (point D), riders of the traditional systemexperienced welfare losses, represented by X 2

2 − X 23.

The adoption of TransMilenio also produced both positive and negativeimpacts on pollution readings, leading to gains and losses for the popula-


45. The improved quality of the ride reduces willingness to pay for improvements intravel time.

Number of trips

Fare

TATM

FATM

STM(F)

Number of trips

Fare

TBTM

FBTM

STS(F)

B. Losses for owners of scrapped busesA. Gains for private TransMilenio operators

a. The shaded area denotes welfare gains (panel A) and losses (panel B). The subscript BTM indicates the fare and number of passengers in TransMilenio corridors before TransMilenio, and the subscript ATM indicates the fare and number of passengers after TransMilenio. Supply curves for private TransMilenio operators and for owners of scrapped buses (STM and STS, respectively) are assumed to depend on fares (F).

F I G U R E 9 . Welfare Gains for Private TransMilenio Operators and Welfare Losses for Scrapped Busesa


A. TransMilenio users B. Traditional system passengers

U 0

U 1

X1

X2

12X

02X

Gain

A B

C

X1

X2

D

U 0

U 1

22X

32X

Losses

a. The horizontal axis represents savings in travel time (X1), and the vertical axis represents a numeraire good measured in monetary units (X2).

F I G U R E 1 0 . Welfare Effects from Impacts on Traveling Conditions and Travel Timesa

tion. Residents located near TransMilenio corridors experienced welfaregains from drops in pollution, while bus relocation heightened congestionand deteriorated air quality in some of the unserved TransMilenio corri-dors. To establish the welfare impact of these opposite effects, we use themarginal damage function (which corresponds to drops in utility as a resultof pollution, measured in monetary terms) to calculate welfare changesfrom the adoption of TransMilenio. Figure 11 shows how to value thesewelfare impacts for both groups of the population.

Finally, society as a whole benefits from the decrease in fatalities inTransMilenio corridors stemming from enhanced traffic and security con-ditions. We use the value of a statistical life, which shows aggregated indi-viduals’ willingness to pay for risk reductions, to measure reductions inthe risk of death in TransMilenio corridors.

This series of figures offers an insight into who wins and who loses afterthe adoption of TransMilenio. We now perform a cost-benefit analysis toestablish whether the benefits offset the costs or whether the costs fromnegative spillovers offset the benefits provided to TransMilenio users. Thecost-benefit analysis incorporates the revenues and costs from Trans-Milenio operations, as well as initial investment. Operational informationcomprises revenues from fare collections, operational costs for concession-

aires, fare collection expenses, administrative expenditures of TransMilenioS.A., and management costs for the fiduciary. Infrastructure investmentsand maintenance costs of corridors are also incorporated.

We also calculated the forgone revenues of firms operating previously inTransMilenio corridors. Before TransMilenio came into operation, nearly6,080 buses operated in those corridors; of those, approximately 1,410 werescrapped, and the remaining 4,670 were relocated to unserved TransMileniocorridors. Forgone revenues were estimated for the scrapped buses.

Our estimation of welfare losses and gains from variations in traveltimes for users of both TransMilenio and the traditional system considersall stages of traveling from one destination to another: waiting time, in-vehicle travel time, and walking-in and walking-out time. Lleras calculatesthe value of time for each stage.46 Public transportation demand encom-passes two groups: users that require one or more transfers and users thattravel directly to their destination. The estimation includes the values cal-culated for both segments.

We also consider the benefits and costs associated with changes in par-ticulate matter levels. Méndez provides figures from emissions reductions


PM-10

Marginal damage

EBTM EATM PM-10

Marginal damage

EATM EBTM

A. Gains to residents near TransMilenio corridors

B. Losses to residents near unserved corridors with negative spillovers

F I G U R E 1 1 . Welfare Gains for Residents near TransMilenio Corridors and Welfare Lossesfor Residents near Unserved Corridors

46. Lleras (2003).

in TransMilenio corridors.47 Emissions increases from negative spilloverswere calculated using a difference-in-differences approach. We calculatedthe impact of particulate matter pollution on the incidence of acute respi-ratory illnesses based on a concentration-response function estimated byLozano for Bogotá.48 We then valued the changes in the incidence ofacute respiratory illness as a consequence of TransMilenio using willing-ness to pay for reductions in respiratory health admissions, also reportedby Lozano.49

Finally, we calculated the gains from reductions in traffic fatalities inTransMilenio corridors. We estimated the risk of death before and afterTransMilenio based on the number of passengers and the number of deaths.Bowland and Beghin report the value of a statistical life (VSL) for Chile;we adjusted these values for Colombia using the GNP per capita for bothcountries.50

The analysis does not incorporate two main welfare benefits accruingfrom TransMilenio as a result of data restrictions, namely, gains fromreduced traffic collisions and injuries (reported in figure 6) and gains fromthe reduction in petty thefts and general insecurity in TransMilenio corri-dors. Data simply are not available for calculating the welfare benefitsfrom these variables.

Table 2 presents the net present value of the cost-benefit analysis fordifferent discount rates.51 The calculation of the net present value occursin two stages: we first calculate the net benefits for TransMilenio corri-dors, without accounting for negative spillovers on unserved corridors andthen incorporate the effects on unserved TransMilenio corridors. Resultsreveal that the benefits in TransMilenio corridors significantly offset costs;net benefits are U.S.$3.7 million for a 9 percent discount rate. Once nega-tive spillovers are incorporated in the analysis, net benefits are negative, at


47. Méndez (2004).48. Lozano (2003).49. Lozano (2003).50. Bowland and Beghin (1998).51. We also calculated net benefits, considering infrastructure investments (a net loss of

U.S.$64 million for a discount rate of 9 percent). However, there are several argumentsagainst including infrastructure investments in the cost-benefit analysis. First, the bulk ofinfrastructure investment is street pavement, most of which would have been done anywayand is the responsibility of city authorities. Second, a full renovation of two lanes for pri-vate cars and the curbside was required for TransMilenio’s operation, such that benefitsfrom this investment not only accrue to TransMilenio providers and users, but also to otherpopulation groups. Third, this infrastructure lasts for decades, allowing city dwellers to reapthe benefits over a very long period.

U.S.$8 million. The increased travel times for users of the traditional sys-tem, resulting from bus relocation, drive these results. Pollution in non-TransMilenio corridors also reduces benefits, but the contributions are notsignificant and are partially offset by drops in pollution in TransMileniocorridors. Benefits stemming from less in-vehicle travel time for Trans-Milenio users, operational revenues, and reductions in mortality, althoughsignificant, are not sufficient to cover these costs.

Because congestion costs are highly nonlinear, the congestion reliefbrought about by TransMilenio has been more than offset by increases incongestion in others areas. Currently, TransMilenio covers 13 percent ofthe public transportation demand, whereas the public transportation sys-tem covers the remaining 87 percent. TransMilenio has indeed improvedthe traveling conditions of its users, but it has worsened the conditions ofa high proportion of the population—namely, users of the traditional pub-lic system.

Negative spillovers have resulted from three features of the currentmass transit system of Bogotá. First, only 25 percent of the TransMilenio


T A B L E 2 . Cost-Benefit Analysis of TransMilenio’s First PhaseMillions of 2002 dollars

Discount rate

Item 7 percent 9 percent 12 percent

Forgone revenues, traditional system −3.628 −2.720 −2.114Forgone operational costs, traditional system 2.702 2.025 1.574Revenues, TransMilenio 6.858 5.046 3.870Operational costs, TransMilenio −6.005 −4.410 −3.378Costs from increased waiting time for TransMilenio, no transfer −0.608 −0.447 −0.343Costs from increased waiting time, transfer −1.686 −1.240 −0.950Costs from increased walking in and out time, no transfer −2.766 −2.034 −1.559Costs from increased walking in and out time, transfer −3.026 −2.225 −1.705Benefits from reductions in travel time, no transfer 6.217 4.572 3.504Benefits from reductions in travel time, transfer 5.852 4.303 3.298Benefits from a 9.1 percent reduction in PM-10 in TransMilenio corridors 0.283 0.212 0.165Benefits from reductions in mortality 0.832 0.625 0.486

Net present value, TransMilenio 5.03 3.71 2.85

Travel time costs for traditional system users, no transfer −5.014 −3.762 −2.925Travel time costs for traditional system users, transfer −10.239 −7.682 −5.973Costs from a 10.5 percent increase in PM-10 in traditional corridors −0.357 −0.268 −0.208

Net present value, traditional system −15.61 −11.71 −9.11

Total net present value −10.59 −8.00 −6.26

Source: Authors’ calculations.

network has been constructed so far. Because the TransMilenio network isonly partially operating, the benefits arising from the economics of coor-dination and density are not being fully exploited. Second, the scrappingrates required in the first phase of TransMilenio contracts are insufficient.Buses that were previously operating in TransMilenio corridors and thatshould have been scrapped were relocated to unserved corridors, whichheightened congestion. Finally, high fares and imperfect controls on theentry of new buses into the traditional public system promoted the entry ofnew buses and slowed the exit of old ones. In fact, the second wave of con-tracts defines more stringent scrapping rates than did the first phase. Faresare still excessively high, creating an incentive for the entry of new buses.52

In addition, controls on the entry of new buses are difficult to enforce,allowing so-called pirate buses to operate citywide.

There are two important caveats to these results. First, our analysis doesnot incorporate some important benefits, like fewer traffic accidents andbetter security in TransMilenio corridors, because of data restrictions. Thesebenefits would certainly change the results. Second, given that TransMileniowill eventually serve the vast majority of the road network, negative spill-overs from bus relocation should disappear once the network is complete.Nevertheless, the analysis sheds some light on drawbacks of the Trans-Milenio system that should be adjusted for the next phases. The design ofthe system should take into account negative spillovers to the traditionalpublic transportation system in order to minimize them. Moreover, Trans-Milenio and the traditional system should not operate separately, but shouldbe integrated into a single mass transit system. Finally, the prevailing in-centives (such as high fares) that spur the entry of new buses, despite thefact that many of them travel empty, should be eliminated.53

Concluding Remarks: The Road Ahead

TransMilenio has become a cornerstone in the history of urban publictransportation in Colombia. Today at least five Colombian cities with overhalf a million people each are applying for central government resourcesto replicate the Bogotá system. Other cities in Latin America are planningto follow suit. The design of TransMilenio mimicked the interventions in


52. Castro and others (2001).53. Castro and others (2001).

mass transit systems of Curitiba, Brazil, and Quito, Ecuador, but it intro-duced some novelties that have become its trademark. The new transitsystem is a hybrid model that combines public planning of the networkstructure, route tendering conditions, regulation, and supervision, with pri-vate operation of the separate functions of revenue collection and transportservice. Among the most salient features are the financial fiduciary man-agement; the division of services among separate private providers thatsimultaneously administer the firm TransMilenio S.A.; the flexible con-tracts for bus operation; the separation of concessions for feeder buses andregular buses; the payment per kilometer instead of per passenger; the def-inition of property rights for the road and the curbside; and the use of thefaster left lanes.

This paper is the first to provide a full economic account of the origins,design, political economy, and costs and benefits of TransMilenio. Thesystem had a sizeable impact on users by improving travel conditions sig-nificantly. Congestion, pollution, and traffic accidents fell sharply in Trans-Milenio corridors. But the type of transition adopted for corridors notcovered by TransMilenio caused unforeseen negative spillovers, as a con-sequence of slow scrapping rates and bus and route relocation. Althoughthe cost-benefit analysis for the first phase of the corridors covered byTransMilenio is positive, the citywide net effect is negative mainly as aresult of increases in travel time for passengers using the traditional trans-port system. To minimize the negative spillovers during the full imple-mentation of TransMilenio, integration of the traditional and new systemsshould continue, and strict regulation of the traditional public transporta-tion system should be crafted.

There remain potential vulnerabilities. As Estache and Gómez-Loboindicate, the institutional capacity of the national and municipal planningauthorities for defining the network configuration, quality requirements,and service levels are crucial.54 One of the advantages of the private systemis its flexibility to modify routes, cover new developing areas of the city,and bring bus routes closer to consumers. These challenges now need tobe answered satisfactorily by good planning and regulation. The tenderingsystem requires special care to ensure competition and avoid collusionamong potential bidding concessionaires, as has been argued in the case ofPereira, the second Colombian city to implement a TransMilenio-type sys-tem. The application of contracts also needs to be monitored. This is par-



ticularly true for the revenue collection contract, whose performance iscentral to the system’s profitability.

Recent events demonstrate other deficiencies in Bogotá. The number ofdaily passengers has increased from 800,000 to 900,000, and the numberof vehicles on the road remains the same. Passenger jamming has becomethe rule, and the problem is an especially severe problem during peakhours. Long lines to purchase access cards are coupled with crowds wait-ing at the curb to enter the buses. People find it difficult to board the buses.Waiting time, a critical variable in any public transportation system and oneof TransMilenio’s big achievements early on, has started to rise. Securityhas deteriorated, and theft is common. As a result, people have begun tocomplain about the TransMilenio monopoly.55

Additional vulnerabilities are related to financing and political support.TransMilenio depends on fiscal resources from the city and the nation.Both political will and financial difficulties could prove obstacles in thefuture. The last two municipal administrations strengthened the system’sinstitutional framework. The new national and municipal administrationsthat entered office in August 2002 and 2004 had earlier criticized aspectsof TransMilenio, but since taking office these new leaders have embraced itand promised to improve it.

Finally, the fact that a democratically elected mayor appoints the headof the public company, TransMilenio S.A., poses a potential problem. Oneor two individuals can decide the future of the entire system. The systemneeds to strengthen an impersonal and institutionalized regulatory arrange-ment. One such decision might involve the scrapping process of the firstlines of TransMilenio.

The implementation of TransMilenio holds several policy lessons forcities in developing countries planning to reform their public transporta-tion system. The design of TransMilenio successfully reduced many mar-ket failures plaguing the provision of public transportation in developingcountries. This hybrid system, in which the public sector participates in theregulation of the system and network configuration while the private sec-tor operates the buses, avoids the problems inherent in both public monop-oly or unregulated private provision. Breaking the link between passenger-trips serviced and profits eliminates the incentives to atomize the busindustry, removes the perverse signals to use smaller buses, and enhances


55. Transmi-Lleno (or Transmi-full) became slang for TransMilenio. On 10 March2004, passengers protested by sitting on the road after an accident delayed a long queue ofTransMilenio buses for hours.

traffic conditions. The allocation of exclusive property rights over a routeabolishes the excess supply of buses, and the competitive tendering of theright to operate on a route pushes fares towards efficient levels.

But the gradual implementation of the new system along with the paralleloperation of a poorly regulated traditional system brought unexpected nega-tive results. Since congestion costs are highly nonlinear, the welfare lossesfrom heightened congestion in unserved TransMilenio corridors more thanoffset the benefits from TransMilenio, even though those benefits are sizable.This suggests the adoption of a new public transportation system must runparallel to an appropriate regulation of all other public transit providers.

Appendix A: Econometric Estimates for Rents

To calculate the economic rents for affiliating firms and bus drivers, weestimated the demand per vehicle and the income per passenger using timeseries regressions, using two approaches. The first approach estimates anordinary least squares (OLS) regression that corrects for autocorrelation;the second approach estimates an autoregressive moving average exoge-nous variables (ARMAX) model.

The observed demand per vehicle is estimated with a distributed lagmodel in which a lagged variable for the number of passengers is includedas an explanatory variable:

where qt is the daily number of passengers per vehicle, qt −1 is the laggeddaily number of passengers, and vt represents the number of buses. Weestimated several models. The model that best approximates the functionalform is presented table A1.

The ARMAX model provides the best estimation, and autocorrelationis eliminated. The signs of the coefficients are as expected. β0 is equiva-lent to autonomous demand, β1 incorporates the influence of prior periods,and β2 includes the influence of size of the total fleet, which is negative inall the estimates.

To estimate income per passenger, the production function curve for acongestion good is estimated using a polynomial regression model.

Y V v vt t t= + + + + + −α α α α ε δε0 1 22

33

1,

q q vt t t t t= + + + +− −β β β ε δε0 1 1 22

1


where Y represents the income per passenger, which is a variable thatapproximates the fare. To correct for autocorrelation, ARMAX and first-order autoregressive, or AR(1), models were estimated. The results arepresented in table A2.

Calculating rents involves calculating the gains and costs to each vehi-cle. Gains are determined by the number of vehicles in circulation, forwhich a production function of the public transportation sector is defined.Costs per vehicle include fuel expenses, parts replacement, maintenance,depreciation, insurance costs, taxes, and purchase value of the vehicle.1 Wecalculated the average cost per vehicle for each year. Rents for the affili-ating firms were estimated by adding up the benefits for the buses they own,revenues from routes allocations, and operational costs.

Contracts between the affiliating firm and bus owners are simulated bymaximizing benefits from affiliating firms subject to positive benefits for the


T A B L E A 1 . Econometric Estimates of Demanda

Coefficient OLS model ARMAX model

β0 605.2 123.34β1 0.8064β2 −0.15 × 10−6 −0.167 × 10−

6

δ 0.3363**Sum of the square of the errors 40648.9 6003.83E(�t)b 0.24* = 0

*Durbin Watson; ** statistically significant at the 2 percent level.Source: Authors’ calculations.a. All figures are statistically significant at the 1 percent level unless otherwise specified.b. E(�t) = 0 implies that autocorrelation does not persist.

T A B L E A 2 . Econometric Estimates of Income per Passenger

Coefficient OLS model AR(1) model ARMAX model

α0 52.8123 21.4539 33.5797α1 −0.1212 × 10−1 −0.5918 × 10−3 −0.7134 × 10−2

α2 0.1062 × 10−5 0.1053 × 10−6 −0.6694 × 10−6

α3 −0.2484 × 10−10 −0.2776 × 10−11 −0.1538 × 10−

10

δ 0.9674 0.8316Sum of the square of the errors 284.49 139.77 137.78E(�t)a 0.7466* ≠ 0 = 0

*Durbin Watson.Source: Authors’ calculations.a. E(�t) = 0 implies that autocorrelation does not persist.

1. The prevailing interest rate for each year was used.

bus owners. We thus obtained the optimal number of vehicles per affiliatingfirm, the affiliation fee, and rents for the affiliating firm and bus owners.

Appendix B: Methodology of the Cost-Benefit Analysis

We performed the cost-benefit analysis for the length of the first wave ofthe contract: 2000–15. All values are in millions of 2002 U.S. dollars. Pop-ulation growth rates for 2004 and 2005 are projections from the NationalAdministrative Department of Statistics (DANE); for 2006 onwards, weused the average population growth for 2001–05.

We calculated TransMilenio revenues by multiplying the number ofTransMilenio users by the technical fare. For 2001, 2002, and 2003, weused observed values for the number of passengers, as reported by Trans-Milenio S.A. For 2004 onwards, the revenues were calculated assumingthat the number of passengers increased with the population growth rate.

We estimated the revenues for the traditional bus system as follows.First, the number of passengers on the Caracas and Avenida 80 corridorsbefore the implementation of TransMilenio was obtained from DANE for1996 and adjusted for 2001 using population growth rates for Bogotá.Castro and others provide the revenues and costs per passenger for the dif-ferent types of buses that operated in these trunk corridors before Trans-Milenio.2 We estimated the number of buses that operated in these trunkcorridors (approximately 6,080 buses) based on the number of routes inthese corridors and the average number of buses per route; this informa-tion was provided by Express del Futuro, one of the TransMilenio con-cessionaires. With regard to scrapping, TransMilenio S.A. reports that forevery TransMilenio bus that entered the system, three traditional systembuses were scrapped. Nearly 470 TransMilenio buses came into opera-tion, so 1,410 traditional buses (23.2 percent of the total) were effectivelyscrapped; we assume that the remaining 4,670 buses (76.8 percent) were re-located to other corridors. We calculated the forgone revenues of the tra-ditional buses as the number of passengers multiplied by the revenue perpassenger for buses in the Caracas and Avenida 80 corridors. We estimatedthe gains from forgone operational costs of traditional buses previouslyoperating in TransMilenio corridors as the number of passengers multi-plied by the operational cost per passenger for buses in the Caracas and


2. Castro and others (2001).

Avenida 80 corridors. Operational revenues (costs) for traditional buses thatrelocated to unserved TransMilenio corridors were calculated as the rev-enues (costs) of traditional buses multiplied by the percentage of busesthat relocated to other corridors (76.8 percent).

We calculated five types of operational costs: (a) feeder line costs: thenumber of passengers in the feeder lines multiplied by the feeder rate perpassenger; (b) trunk corridor cost: the number of kilometers multiplied bythe bus fare per kilometer; (c) fare collection costs: a rate per passengermultiplied by the number of passengers; (d) operational costs for Trans-Milenio S. A: 3 percent over the technical fare multiplied by the number ofpassengers; and (e) fiduciary contract cost: 0.0387 percent of total rev-enues. For the first four operational costs, we used observed values reportedby TransMilenio S.A. for 2001, 2002, and 2003. For 2004 onwards, thenumber of passengers increases by the population growth rate.

Infrastructure costs for the first phase of TransMilenio are based on val-ues reported by Lleras.3 Costs include trunk corridors, bus stations, bus ter-minals, access to stations, bus depots, operations control center, and buses.

Maintenance costs are the total maintenance costs reported by the Insti-tuto de Desarrollo Urbano (IDU) and TransMilenio.

Time costs break down into several components. The value of time isfrom Lleras.4 All times were assumed to remain constant over the periodof analysis. (a) Waiting time: we calculated increases in waiting time asthe difference in waiting time of TransMilenio and traditional systemusers and then multiplied this figure by the value of waiting time for Trans-Milenio users and the number of TransMilenio users. (b) Walking-in andwalking-out time: we calculated increases in walking-in and walking-outtimes as the difference between TransMilenio and traditional system usersand then multiplied this figure by the value of walking time for Trans-Milenio users times the number of TransMilenio users. (c) In-vehicle tra-vel time: we calculated reductions in in-vehicle travel time as the differencebetween TransMilenio and traditional system users and then multipliedthis figure by the value of travel time for traditional system users and thenumber of traditional system users. (d) Travel time for users of the tradi-tional system: increases in travel time for users of the traditional systemamount to ten percent according to Lleras.5 We then multiplied this increase


3. Lleras (2003).4. Lleras (2003).5. Lleras (2003).

by the value of travel time for traditional system users and the number oftraditional system users that transit throughout corridors to which tra-ditional buses were relocated. (e) Negative congestion spillovers in tradi-tional system corridors: five city corridors received the bulk of relocatedbuses when TransMilenio began its operation (namely, Avenida Calle 68,Carrera 7a, Carrera 13, Carrera 15, and Avenida 19). Information regard-ing the number of daily passengers that transit these selected corridors wasprovided by the District’s Transit Secretariat. (f ) TransMilenio times: wecalculated all TransMilenio times for two segments of the population. Seg-ment one includes users who do not use feeder lines, as their trips origi-nate in the vicinity of the Trunk Corridors (55 percent of the population,according to Lleras).6 Segment two encompasses the remaining 45 percent,who start their trips in areas served by feeder lines and thus have to engagein one or more transfers. (g) Traditional system times: we calculated alltraditional system times for two segments of the population. Segment oneincludes those who only take one bus to their destination (55 percent ofthe population), while segment two (the remaining 45 percent) includesthose who take two or more buses to their destinations.

We estimated the effects on air pollution as follows. Reductions in PM-10 emissions associated with TransMilenio operations in trunk corridorswere based on Méndez.7 Emission reductions or increases in TransMileniocorridors are assumed to affect 20 percent of the population. Finally, weconducted a difference-in-differences analysis to determine the increases inPM-10 emissions associated with traffic spillovers to non-TransMileniocorridors.

To determine reductions in mortality, VSL estimates for Chile wereadjusted for Colombia using the per capita GNP of both countries.8 Wethen calculated the risk of death for public transportation users before andafter TransMilenio as the number of deaths divided by the number of pub-lic transit users.

To calculate TransMilenio’s net present value, we used social discountrates of 7.00 percent, 9.33 percent, and 12.00 percent.


6. Lleras (2003).7. Méndez (2004).8. VSL estimates for Chile are from Bowland and Beghin (1998).

Comments

Mauricio Cárdenas: This is a very interesting paper on a subject that ishighly relevant and relatively ignored by the profession. The success storyof TransMilenio has received ample press coverage in Colombia and else-where, but little academic attention. This public policy initiative was ableto change an entrenched status quo and deliver positive results in a shortperiod. Since other large cities are following the example of Bogotá, it isvaluable to understand the main features of the new scheme, including itslimitations and main challenges.

The authors discuss many dimensions of the problem of organizingtransportation in large cities. The paper provides useful analysis and dataon the optimal amount of intervention in urban transportation and the typeof intervention that works best in developing countries. The paper dis-cusses the cost effectiveness of different kinds of intervention and theirimpact on living standards, particularly the health impact of air quality.

The experience of Bogotá is characteristic of many large cities in thedeveloping world. Bogotá tried for decades to deal with market failuresassociated with transportation systems, experimenting with solutions suchas trolley buses and publicly owned and operated buses. Based on theinformation shown in the paper, these policies tended to exacerbate, ratherthan resolve, market failures. This is what the authors call the pervasive-ness of huge service inefficiencies, and it is reflected in variables such asthe excessive number of buses, the low average speed of circulation, thehigh frequency of accidents, and the low air quality. TransMilenio has notsolved all the problems, however, and many inefficiencies are still presenttoday. For example, the authorities of Bogotá have been unable to regulatethe flow of buses originating in conjoining municipalities that circulate in thecity. Also, TransMilenio has increased multileg trips and, hence, the totalfare costs for the average user.

The paper emphasizes the excessive number of private cars in the city.However, contrary to the conventional wisdom that guides policies in thesector, Bogotá has only 130 cars per 1,000 inhabitants, far fewer than

189

other cities of similar size and per capita income. For example, Curitiba,Brazil, which was one of the early and influential adopters of a bus rapidtransit system, has 300 cars per 1,000 inhabitants. Other internationalcomparisons suggest that rather than having too many cars, Bogotá hastoo few kilometers of paved roads. Since per capita income is rising, vehi-cle demand will undoubtedly continue to increase rapidly, more so aftertrade liberalization has lowered the cost of many durable goods, includ-ing automobiles. Consequently, while TransMilenio can solve some traf-fic problems and reduce commuters’ travel time, the contribution will beshort-lived unless important investments are made in additional infra-structure or policies are adopted to discourage the use of cars throughmarket mechanisms.

The institutional aspects of TransMilenio are also of interest. The cre-ation of an entity relatively insulated from political competition was criti-cal for attracting individuals with better technical qualifications, who arepaid competitive salaries. The paper sometimes favors the role of individ-uals, but the fact that the system has been maintained and improved withthe passage of time speaks well of institutional features. The adoption andexpansion of the system has been the responsibility of three successivemunicipal administrations that have shown important ideological differ-ences on other issues. This suggests that the institutions were designed ina way that prevented policy volatility.

The public policies behind the TransMilenio system have not only beenstable, in the sense of remaining in place beyond the tenure of politicalofficeholders, but they have also been flexible. The system is not based onrigid rules, so the authorities have been able to fine-tune important aspects,most notably in relation to the bidding criteria and the contractual design.Recent contracts have thus transferred more risks and fewer revenues tothe private sector concessionaires. The various agents involved in the pro-cess have exhibited a good amount of coherence and coordination, includ-ing the different agencies in charge of building the infrastructure and thecompany that operates the system. The policymaking process has gener-ated incentives for increasing efficiency in delivery. Most important,policies have generated broad (as opposed to concentrated) benefits, re-sembling true public goods.

This brings me to the political economy of the implementation, whichis one the most interesting and relevant aspects of the paper. I agree withmuch of what is said and with the overall spirit of the argument, but fouraspects merit greater emphasis. First, the institutional reform in the city of


Bogotá (implemented in 1993) was a consequence of the 1991 Constitu-tion. This enhanced the powers of the mayor and reduced the influence ofthe city council. Prior to 1993, the city council coadministered the city,especially in relation to public utilities and transportation policies. Cor-ruption was rampant. Privatizing public utilities and changing transporta-tion policies would have been very difficult without that reform. Second,privatization and fiscal reform allowed the city to improve its finances,providing the necessary resources for a large investment project such asTransMilenio. Third, given the electoral importance of the capital city, thepresidential candidates in the 1998 election supported the idea of commit-ting national funds to the public transportation system for Bogotá. Thequestion was whether to build a metro or a bus-based system like Trans-Milenio. The economic technocracy of the national government ruled outthe metro and opted for TransMilenio because it was the only fiscally fea-sible alternative. In fact, the national government actually chose the Trans-Milenio solution while the city administration was still considering themetro as an option. This decision carried political costs because it wasseen by the public as an inferior solution to the transportation problems ofthe city, but in truth it is a much better solution. TransMilenio’s area ofinfluence (500 meters on each side of the busway) covers 85 percent of theurban area, whereas the metro would have covered only 8 percent of thecity. In addition, the metro’s infrastructure would have cost twice as muchas TransMilenio’s and its operation twenty times more. Fourth, multilateralbanks did not have an important role in this process. Conversations on acredit facility for the construction of infrastructure for a bus rapid transitsystem started in 1986. These loans never materialized and had endlessobstacles. Cities in need of overhauling their transportation systems cannotcount exclusively on the advice of multilateral banks.

As the authors rightly point out, congestion, pollution, and traffic acci-dents fell significantly in TransMilenio corridors, explaining why thepartial cost-benefit analysis is positive. When they look at the general equi-librium effects, they find that the opposite is actually true, mainly becausecongestion increased in the unserved corridors, where more buses continueto compete for passengers. The authors use difference-in-differences cal-culations to show that emissions have increased in unserved corridors,more than offsetting the reductions in TransMilenio corridors. These cal-culations are sensitive to the number of data points (before and after Trans-Milenio) and the number of monitoring stations, among other factors. Thevalue of time, the value of life, and the conversion of emissions into monetary


amounts also make the results sensitive to specific assumptions. There-fore, the results should not be taken as unambiguous proof that Trans-Milenio has had a short-run negative impact. However, I fully share theauthors’ view in favor of speeding up the process of implementation ofTransMilenio and improving the regulatory framework that applies to tra-ditional public transportation in order to minimize the potential down-sides of the scheme. These are important lessons for other cities buildingon the experience of Bogotá.

Andrés Gómez-Lobo: The paper analyzes the origins, justification, andresults of TransMilenio, an urban transport plan introduced in Bogotá,Colombia, in 2000. Urban transport policy is not a prominent issue in thediscussions of economists who study Latin America, but it nonethelessmerits more attention from specialists. The macroeconomic costs of havinga large portion of the population wasting several productive hours a daytraveling to work or the extra costs for companies that need to distributeproducts throughout a city are rarely measured. Urban transport efficiencymay be just as important for economic development as the sophisticationand efficiency of other network industries such as telecommunications.This paper, insofar as it presents a novel experience in urban transport pol-icy, is relevant for countries throughout the region.

Gwilliam points out that in developing countries urban bus services areusually characterized by competition in the market.1 Competition amongurban transport providers has been suspect as a welfare maximizing pol-icy, however, since Chadwick’s famous article advocating competition forthe market rather than in the market for certain industries, including urbantransport.2 Interestingly, the market failures of competitive urban bus trans-port markets are still not well understood from a theoretical perspective.Congestion in the absence of road pricing is an obvious and well-understoodproblem, but what is more puzzling is why fares seem to increase ratherthan decrease when competition is introduced.3 The ensuing high tariffsgenerate excessive entry into the industry, capacity utilization of each busdiminishes, and the final result is economically inefficient and sociallyharmful (for example, reduced safety and increased pollution).

This paper briefly reviews the arguments behind the observed marketfailures in urban bus markets. This is not the best part of the paper since


1. Gwilliam (2001).2. Chadwick (1859).3. Estache and Gómez-Lobo (2005).

some of the arguments are presented very succinctly and are thus out offocus. For example, the authors mention the principal-agent relationshipbetween bus owners and drivers as a market failure. When owners are in-terested in the number of passengers carried, they will naturally try to aligndrivers’ incentives with their own by offering drivers a share of each pas-senger’s fare. There is no market failure here. The problem is that theowner does not take into account the externalities that these incentivesgenerate in the form of more aggressive driving and frequency distortions.

The main aim of the paper, however, is not to review the theoretical lit-erature, but rather to describe the transport reform in Bogotá and under-take a cost-benefit analysis of the first stages of this reform. There is agrowing consensus among policymakers that state-of-the-art regulatorypolicy in the transport sector, at least for middle-income countries, in-volves what Estache and Gómez-Lobo call the hybrid system.4 In thisscheme, the authorities design the network and impose quality standards,frequencies, and tariff integration, while the private sector operates the ser-vices under a concession system. Revenues are centralized, and operatorsare not paid according to the number of passengers carried but receiveeither a fixed fee or rate per kilometer traveled. This hybrid scheme avoidsthe most important failures associated with state-provided monopoly trans-port services, as well as the market failures of competition in the streets.This policy has been adopted in London and Curitiba, Brazil, and it willalso be implemented in Santiago, Chile, in 2005. Bogota’s TransMilenioexperience provides another interesting example of the application of thishybrid model, and its documentation and evaluation provides interestinglessons for policymakers worldwide.

What can one learn, then, from the experience of Bogotá? The currentpaper raises two sets of issues that are particularly interesting: first, thedesign of the concessions and the way risks are allocated among agents;and second, the evaluation of the benefits and costs of the reform, whichproduces two unexpected results. With regard to the design issue, a keypoint is the way contracts were designed for the bus operators in the maincorridors. Since the authorities determine dispatch frequencies, the numberof kilometers served is not under the operator’s control. Operators are paidaccording to the kilometers actually served, however, to avoid competitionon the road. It seems rational that operators should not face a demand riskthat is not under their control, whereas the state is better suited to adminis-



ter this risk. This was achieved in Bogotá through a novel variable-lengthconcession that is reminiscent of Engel, Fisher, and Galetovic’s least-present-value-of-revenue concessions.5 If at the end of the tenth year theconcessionaire has not served the number of kilometers stipulated in thecontract, the concession is automatically extended. This design has severalattractive properties, and a similar system has been adopted for the futurereform in Santiago, Chile.

At the same time, this scheme has the disadvantage that since operatorsare shielded from demand risk, they will make no extra effort to take-onextra passengers or change route designs when demand patterns change.This will probably be a greater problem in routes outside the main corri-dors and in the periphery of the city where new neighborhoods may gen-erate new transport demands. The TransMilenio planners wisely relaxedthe payment method for feeder services by combining a payment per kilo-meter traveled with a payment per passenger carried, thereby providingincentives to cater to passengers’ needs and demands in the feeder zones.

On the evaluation of the first stage of TransMilenio, the paper presentstwo surprising results that should be given serious attention by policy-makers in other countries embarking on similar reforms. First, the reformwas regressive. Those that benefited (that is, TransMilenio users) includea higher proportion of wealthy individuals than those that lost (namely,users of the traditional system). Given that infrastructure investments,which represent a large fraction of the costs of the reform, will be paid fornot by users but by taxpayers in general, the distributional impact of thefirst stage of the reform is worrying. Second, the overall cost-benefit analy-sis shows that the extra pollution and congestion generated on unservedcorridors more than offset the benefits on TransMilenio corridors.

The above two results suggest that it may be socially preferable to under-take a more comprehensive reform rather than implementing the system ina piecemeal fashion as in Bogotá. In Santiago, Chile, the reform of the bustransport system scheduled to start in August 2005 is all-encompassing,simultaneously changing the whole transport system of the metropolitanarea. A few years will have to pass before it is evident whether the addedcosts and complexities of a citywide reform, as in Santiago, more than off-set the benefits of avoiding some of the negative effects of the piecemealapproach used in Bogotá.


5. Engel, Fisher, and Galetovic (2001).

References

Bowland, Brad, and John Beghin. 1998. “Robust Estimates of Value of a Statisti-cal Life for Developing Economies: An Application to Pollution and Mortalityin Santiago.” Staff general research paper 4046. Iowa State University, Depart-ment of Economics.

Cavallazi, Marcelo. 1996. “Contaminación atmosférica en Bogotá.” Revista Cámarade Comercio 97. Bogotá.

Castro, Raúl, and others. 2001. “Cálculo de la tarifa óptima para el sistema detransporte masivo—TransMilenio.” Final report presented to TransMilenioS.A. Universidad de los Andes, Department of Economics.

Chadwick, Edwin. 1859. “Results of Different Principles of Legislation and Ad-ministration in Europe; of Competition for the Field, as Compared with Com-petition within the Field, of Service.” Journal of the Statistical Society ofLondon 22(3).

Chaparro, Irma. 2002. “Evaluación de impacto socioeconómico del transporteurbano en la ciudad de Bogotá: el caso del sistema de transporte masivo, Trans-Milenio.” Serie recursos naturales e infraestructura 48. Santiago: United NationsEconomic Commission for Latin America and the Caribbean.

DANE (Departamento Administrativo Nacional de Estadística). 2003. Encuestade Calidad de Vida. Bogotá.

Engel, Eduardo M. R., Ronald D. Fischer, and Alexander Galetovic. 2001. “Least-Present-Value-of-Revenue Auctions and Highway Franchising.” Journal ofPolitical Economy 109(5): 993–1020.

Estache, Antonio, and Andrés Gómez-Lobo. 2005. “Limits to Competition in UrbanBus Services in Developing Countries.” Transport Reviews 25(2):139–58.

Evans, Andrew. 1987. “A Theoretical Comparison of Competition with OtherEconomic Regimes for Bus Services.” Journal of Transport Economics andPolicy 21(1): 7–36.

Gwilliam, Kenneth M. 2001. “Competition in Urban Passenger Transport in theDeveloping World.” Journal of Transport Economics and Policy 35(1): 99–118.

Ibáñez, Ana María, and Eduardo Uribe. 2003. “Medio ambiente y desarrollo eco-nómico: priorización de la inversión ambiental con criterios económicos.” CEDEDocument 2003-33. Universidad de los Andes. Bogotá.

Klein, Daniel B., Adrian Moore, and Binyam Reja. 1997. Curb Rights: A Foun-dation for Free Enterprise in Urban Transit. Brookings.

Lleras, Germán. 2003. “Bus Rapid Transit: Impacts on Travel Behavior in Bogotá.”Masters thesis, Massachusetts Institute of Technology, Department of UrbanStudies and Planning.

Lozano, Nancy. 2003. “Air Pollution in Bogotá, Colombia: A Concentration-Response Approach.” Masters thesis, University of Maryland at College Park,Department of Agriculture and Resource Economics.


Méndez, Mildred. 2004. “Análisis de intervención: efectividad de las políticaspara reducción de la contaminación por fuentes móviles en Bogotá.” Mastersthesis, Universidad de los Andes, Department of Economics. Bogotá.

Urrutia, Miguel. 1981. Buses y busetas: una evaluación del transporte urbano enBogotá. Bogotá: Fedesarrollo.


What Happens to Wages after Displacement?

Economic shocks and policy reforms can induce large changes inestablishment-level employment. Since wage losses from displace-ment can be large and long-lasting, policymakers often express a

desire to support displaced workers. When resources are limited, policy-makers need to target support to the workers who need it most.

But the academic literature offers little guidance on how to do this.There is little agreement on how wages change after displacement. Theinfluential works of Jacobson, LaLonde, and Sullivan document large ad-verse effects of displacement on workers in the United States.1 Subsequentstudies also find that displacement has significant long-term adverseeffects.2 More recent international comparisons, however, find zero orpositive wage changes following displacement. Abbring and others findno change in wages in the United States, and Bender and others find pos-itive wage changes following displacement in France and Germany.3

197

D A V I D S . K A P L A NG A B R I E L M A R T Í N E Z G O N Z Á L E Z

R A Y M O N D R O B E R T S O N

Kaplan is with the Instituto Tecnológico Autónomo de México (ITAM); MartínezGonzález is with the Inter-American Conference on Social Security; and Robertson is withMacalester College.

This paper is part of a project with the Inter-American Development Bank called “Mar-ket Institutions, Labor Market Dynamics, Growth and Productivity: An Analysis of LatinAmerica and the Caribbean.” We gratefully acknowledge the assistance of Hector Macíasof the Mexican Social Security Institute (IMSS), financial support from the AsociaciónMexicana de Cultura, and research assistance from Diago Dieye and Allison Hicks. We alsothank Omar Arias, Emek Basker, David Drukker, Francisco H. G. Ferreira, Tricia Gladden,John Haltiwanger, Daniel Hamermesh, Adrianna Kugler, Naércio Menezes-Filho, CarmenPagés, Ken Troske, Andrés Velasco, and Sarah West for extremely helpful comments.

1. Jacobson, LaLonde, and Sullivan (1993a, 1993b).2. Most studies focus on the United States, including Caballero, Engle, and Haltiwanger

(1997), Davis and Haltiwanger (1999), Stevens (1995, 1997), Revenga, Riboud, and Tan(1994), and Marcal (2001). Others focus on other developing countries, such as Menezes-Filho (2004), Burda and Mertens (2001), Couch (2001), Fallick (1996), Kletzer (1998), andRuhm (1991a, 1991b).

3. Abbring and others (2002); Bender and others (2002).

The literature also offers conflicting explanations of why these esti-mates vary. Kuhn suggests that differences in inequality and institutionsin France, Germany, and the United States can explain the different es-timates for these countries.4 Alternatively, Howland and Peterson, Car-rington, Jacobson, LaLonde, and Sullivan, and Farber suggest that labormarket conditions can affect postdisplacement wages.5 Since a wide vari-ation in displacement costs makes targeting aid difficult, the efficiency gainsfrom identifying determinants of postdisplacement wages are potentiallysignificant.

This paper studies the Mexican labor market to contrast various expla-nations for differences in postdisplacement wage changes. We hope toidentify patterns that may help policymakers target aid to displaced work-ers. An environment with varying temporal and regional economic condi-tions and with economic conditions and institutions substantially differentfrom those in the countries previously studied is ideally suited to identifysuch patterns. If institutions vary little across regions, then the institutionalhypothesis would be an unlikely explanation of differences across regionsin postdisplacement wages.

Mexico meets these conditions. Differences between Mexico and othercountries, as well as differences within Mexico over time and space, canhelp us identify these patterns in postdisplacement wage changes. First,wage dispersion is higher in Mexico than in France, Germany, or the UnitedStates.6 If inequality drives differences in postdisplacement wages, thenMexican workers should have much more negative postdisplacement ex-periences than observed in these countries.

Second, institutions such as workers’ separation costs, the legislatedcosts of displacement (to the firm), and unions are very different in Mexicothan in other countries. Mexican workers are much less likely to leave firmsvoluntarily than workers in other countries, which suggests that they have


4. Kuhn (2002).5. Howland and Peterson (1988); Carrington (1993); Jacobson, LaLonde, and Sullivan

(1993b, chap. 6); Farber (2003). When examining local labor market conditions, Jacobson,LaLonde, and Sullivan (1993b) compare two Pennsylvania regions over the same timeperiod. Carrington (1993) and Howland and Peterson (1988) provide much wider geo-graphic coverage, but these studies are not directly comparable to Jacobson, LaLonde, andSullivan because they use cross-section data that are subject to recall error rather than track-ing the actual wages of workers over time.

6. The Deininger and Squire data set (available at www.worldbank.org/research/growth/dddeisqu.htm) shows that Mexico’s historically averaged Gini coefficient (52.92) ishigher than that of the United States (35.79), France (37.71), and Germany (32.91).

higher separation costs. Mexico’s mandatory severance pay is higher thanwhat is legislated in France, Germany, and the United States (the latter twohave no legislated severance pay). Unions also have much less wage-settingpower in Mexico than in the comparison countries, which can lead to neg-ative union-wage differentials.7 If unions explain the difference in resultsacross countries, then Mexican workers should have much more negativepostdisplacement experiences than is the case elsewhere.

Third, Mexico’s geographic regions exhibit little variation in unioniza-tion and inequality, but large differences in economic conditions.8 Mexi-can labor laws regarding severance payments, strikes, prohibitions againstnominal wage reductions, the legal recourses of workers in case of unfairtreatment, and guarantees of profit sharing are determined at the federallevel and therefore do not vary across regions. Moreover, no importantlabor market reforms occurred in the period we study.

In short, we use this variation in economic conditions (but not institu-tions) to compare local labor market conditions and postdisplacement wagesover time and space. Since inequality and institutions vary less withinMexico than across the countries previously studied in the literature,heterogeneity in postdisplacement wages within Mexico probably cannotbe explained by inequality and institutions. This points instead toward animportant role for labor market conditions. At the same time, our resultshelp explain the variation found in the literature. It thus seems very likelythat they can be applied beyond Mexico to target aid to displaced workerswhen and where it is most needed.

Our approach differs from previous studies in two key ways. First, weuse a very simple, but formal, theoretical framework that illustrates howdifferences in institutions, such as separation costs for both the firm andthe worker, play a key role in the postdisplacement experience. The modelalso shows how a displaced worker might earn higher wages after beingdisplaced and yet not have wanted to leave the original job in search ofhigher wages prior to displacement. More important, the model showshow unemployment rates (through time) and differences in economic activ-ity (through space) can lead to negative, zero, or positive postseparationwage changes.9 Second, we employ a near-census-sized administrative dataset that allows us to directly compare displacement experiences across time

David S. Kaplan, Gabriel Martínez González, and Raymond Robertson 199

7. Panagides and Patrinos (1994).8. On unionization, see Fairris and Levine (2004).9. The model also shows how the comparison group plays a key role, as Kuhn (2002)

suggests.

and regions. Since we want our results to be as directly comparable withprevious studies as possible, we use the methodological “gold standard”established by Jacobson, LaLonde, and Sullivan.10 We use the same stan-dards as in previous studies to identify displaced workers, match workersto firms, and track workers as they move between firms. We are thereforeable to avoid the so-called recall bias from displaced-worker surveys anddirectly compare our results with earlier studies.

In line with the model, we find that different external conditions cancause wages to go up, go down, or stay constant after displacement. Work-ers who are displaced during good times experience higher wages than non-displaced workers (including both nondisplaced workers who remainedemployed in firms that underwent large employment contractions and non-displaced workers who never worked in these firms), while workers dis-placed in bad times can experience very large losses.11 Furthermore, theeffects of separating in bad times linger: workers who separate when un-employment is high never seem to catch up to workers who separate whenunemployment is low. This effect is most pronounced in relatively lesseconomically dynamic geographic regions.

Our basic results are robust to the effects of age, attrition, tenure, andswitching sectors. Like previous studies, we find that displaced workerswith longer tenure experience larger losses than workers with shorter tenurein some periods but not others. Variation in economic conditions throughtime may therefore explain why Kreichel and Pfann argue that tenure doesnot account for observed wage differences, while other studies, such asCarrington, support the tenure explanation.12 We also find effects of chang-ing sectors that are similar to previous studies. Our main conclusion is thatchanging local labor market conditions produce a wide range of displace-ment effects and therefore might be the key to understanding when dis-placement hurts workers.


10. For example, Jacobson, LaLonde, and Sullivan (1993a) analyze the effects of dis-placement on workers using matched firm-worker data from the United States. Their resultssuggest that workers begin to experience falling wages before they are displaced and that earn-ings recovery may take more than five years. Hamermesh (1989) and Davis and Haltiwanger(1992) show that adjustment costs at the firm level are generally nonlinear and significantlyaffect employment decisions. Other studies examine earnings losses before displacement(de la Rica, 1995) and how changing labor market conditions affect displacement (Stevens,2001; Clark, Herzog, and Schlottmann, 1998; Helwig, 2001).

11. We find loss levels that are very similar to those documented by Jacobson, LaLonde,and Sullivan (1993a, 1993b).

12. Kreichel and Pfann (2003); Carrington (1993).

We present our analysis in five sections. We start by presenting our sim-ple theoretical framework. The subsequent section describes the source,collection, and limitations of our data, discusses the Mexican economicenvironment, defines the term displaced worker, and finally describes thevarious comparison groups. We then explain our empirical approach andpresent our results. A final section concludes.

Theory

This section illustrates how differences in economic conditions can resultin either an increase or a decrease in wages following displacement. Themodel modifies McLaughlin’s theory of quits and layoffs by incorporatinga separation cost that the worker bears in the case of a quit but that the firmpays to the worker in the case of a layoff.13 We present the model in itssimplest form to illustrate the concepts that guide our empirical work.

Workers receive a wage, w, and produce value to the firm, W. Weassume that the value to the firm is a function denoted W(X, G) in which Xrepresents worker-specific characteristics and G represents firm-specificcharacteristics, including the firm’s output price and productivity shocks.Workers have the ability to search on the job. Denote as E(r) the expectedvalue of an outside wage offer from a firm that values the worker at R. Weassume the outside offer is costlessly verifiable once it is made, and thatthe expected value of the offer is a function of external characteristics,including the number of firms that value the worker’s skills (followingStevens), the unemployment rate (which reduces the expected value), andthe economic activity in the region (which increases the expected value).14

We employ McLaughlin’s important distinction between layoffs andquits. McLaughlin defines a quit as the result of a firm-refused, worker-initiated attempt to increase wages and a layoff as the result of a firm-initiated, worker-refused attempt to lower wages. Firms (workers) havethe option to accept proposals for changes in the wage, and they will do soas long as the value of the wage is not larger (smaller) than the value of theoutput to the firm, or the reservation wage. We modify McLaughlin’s modelby adding a separation cost. The separation cost may stem from the loss ofshared surplus from firm-specific training, an institutional arrangement that


13. McLaughlin (1991).14. Stevens (1994).

encourages long-term employment, or other reasons.15 The relevant char-acteristic is that this cost is paid to the worker in the event of a layoff. Thisis particularly relevant for the Mexican case, because Mexico, like someEuropean and many Latin American countries, follows a more preventa-tive stance.16 Article 50 of the Mexican Federal Labor Law mandates thatworkers hired for an indefinite period who are laid off (without cause) areentitled to twenty days pay for each year of service.17 This mandate maysignificantly increase the cost of separation in Mexico.18

The effect of this type of legislation on employment flows is still underdebate.19 One particularly relevant study argues that this kind of legislationin Brazil creates the incentive for workers to negotiate with firms to makequits look like layoffs, in order to receive this payment.20 This behaviorcreates a procyclical turnover pattern, because workers may be particu-larly interested in getting their separation payment in good times to startnew businesses or invest in areas with higher returns. Kaplan, MartínezGonzález, and Robertson examine job creation and job destruction inMexico.21 They find that the pattern of job destruction—especially thecomponent that is due to contraction (layoffs)—is weakly counter cycli-cal. Over the 1986–2001 period, the component of job destruction stem-ming from firm contraction moved negatively with the net growth rate ofemployment. We are therefore confident that the kind of adverse incen-tives and false layoffs documented in Brazil do not affect our results.22


15. Hashimoto (1979, 1981).16. Kuhn (2002).17. This provision applies to contracts of indefinite length. The United States and

Germany have no legislated severance pay, although in the United States the industry stan-dard is one to two weeks per year of service and in Germany severance pay is generallyincluded in the social plan. In France, workers with more than two years of service receive0.1 months of salary per year of service (Kuhn, 2002).

18. McLaughlin’s (1991) analysis of data from the Panel Study of Income Dynamics(PSID) suggests that separations in the United States are evenly divided between quits andlayoffs. Mexicans, in contrast, are much less likely to leave voluntarily, which may imply alarger separation cost. See figure 3.

19. See Heckman and Pagés (2000) and Robertson and Dutkowsky (2002) for examplesof estimates of labor market adjustment costs in Latin America and a discussion of their linkto labor market legislation.

20. Gonzaga (2003).21. Kaplan, Martínez González, and Robertson (2004).22. We do not claim that this behavior does not occur in Mexico or that these concerns

are not relevant for Mexico. The aggregate statistics, however, seem to suggest that suchbehavior does not have a significant effect on our results.

The worker’s decision rule is most clearly illustrated graphically, asseen in figure 1. Given an initial wage of w, the worker will voluntarilyleave the firm as long as the expected wage, E(r), minus the separationcost, C, is greater than the current wage. In figure 1, all workers withwages in region IV will quit and seek employment elsewhere. Workerswith wages in regions I, II, and III will remain in the firm. The separationcost will lead some workers to remain in the firm even if they have higherexpected wages elsewhere (as shown in region III).

Now consider the effect of an adverse shock that lowers the worker’svalue to the firm from w to w′. Such an adverse shock could stem from anadverse productivity or price shock or from some other factor. In this case,the firm will offer the worker a wage reduction, which the worker caneither accept or reject. If the worker accepts the lower wage, the workerremains with the firm; if the worker refuses the wage cut, the worker is laidoff. According to Mexican law, the worker must receive a separation pay-ment, which is presumably intended to compensate the worker for the


E(r)

E(r) E(r)

E(r) – C

w

w’

A B

I II III IV

F I G U R E 1 . Worker’s Decision Rule

separation. The effect of this payment means that workers with expectedwages above the new wage, w′, will be laid off. Some of those workers willreceive lower wages than in their prior employment (workers withexpected wages in region II), but other workers will now take jobs withhigher wages (workers with expected wages in region III). Workers withwages in regions II and III will both have higher postdisplacement wagesthan workers in the same firm (region I).

The effect of displacement on wages is therefore ambiguous: wagesmay either rise or fall after displacement. Wages may go up because theseparation cost keeps workers from voluntarily moving to take advantageof higher potential wages in other firms. This result illustrates why aworker who involuntarily separates from a firm may have higher wagesafter separating, while lacking the incentive to take a higher-paying job(before incurring the separation cost) prior to being displaced.

For a given level of separation costs, the value of wages in other jobs(that is, the expected value of the outside offers) depends on several factors.If labor markets are not perfect and worker experience is valued outside asingle firm, then increasing the number of firms that value the worker’sexperience or that would compete for workers will drive up the outsidewage offers into region III.23 Alternatively, a higher unemployment ratereduces the expected value of wage offers into region II, which impliesthat displaced workers would tend to enter firms with lower wages. Themodel thus illustrates that the heterogeneity of results found in the litera-ture (negative, zero, and positive) is consistent with a simple theory andthat this heterogeneity can be linked to institutions and labor market con-ditions in ways that can be empirically compared.

When assessing a theoretical framework such as this one, it is useful toconsider alternative explanations that may generate similar results. Mosttheory in this literature focuses on explanations for lower postdisplacementwages. Suggestions include loss of firm-specific capital and seniority.These concepts can be easily incorporated into the model above, but theyoffer little guidance for the case of higher postdisplacement wages. Higherpostdisplacement wages are somewhat more difficult to reconcile if work-ers can move to higher paying jobs. This argument, however, assumes thatmoving is basically costless. As explained above, a positive moving costdirectly addresses this concern. Furthermore, the model described abovediffers from previous approaches that focus on either positive or negative


23. See Stevens (1994).

wage changes because it shows how differing labor market conditionscould generate either positive or negative postdisplacement wages.

The Data

Mexican labor laws require all private sector firms to report wage andemployment information on all employees to the Mexican Social SecurityInstitute (Instituto Mexicano del Seguro Social, or IMSS). In practice, how-ever, firms report information on roughly half the private sector employees.Firms may choose not to formally register in order to evade taxes and socialsecurity contributions. Academic studies of Mexico’s informal sector usethe act of reporting to the IMSS as a criterion for formal sector participation.The IMSS records thus represent a census of private firms in the formal sec-tor of the Mexican economy.24 Our data come directly from these records.

The IMSS data are collected at the firm level rather than at the estab-lishment level. Each formal sector firm in Mexico has a firm identifiercalled its registro patronal. The registro patronal is similar to the employeridentification number (EIN) that is commonly used as a firm identifier inU.S. data sets. Just as several subsidiary EINs in the United States might beowned by one parent firm, several registros patronales might be owned bythe same parent company in Mexican data. The registro patronal mayincorporate more than one establishment in a single firm (again like EIN inU.S. data), but in almost all cases, we identify individual establishments (orplants in the case of manufacturing industries). We use the registro patro-nal to link observations over time, to follow workers as they move amongfirms, and to track workers’ wages within their given firm over time.25

As an initial check of data quality, we compare our sample with officialIMSS employment statistics. The motivation behind this comparison isthat the IMSS reports formal employment statistics based on their data,which are used as an indicator of Mexican employment, but their methodfor calculating these statistics is not known to us. A favorable comparison


24. Public sector workers and members of the military have social security accountswith other agencies.

25. Firms could potentially change their registros patronales from time to time foradministrative reasons, and this would generate false births and deaths stemming fromchanges in the registro patronal for continuing firms. In practice, however, very few firms(fewer than four) closed entirely and opened again in the next quarter with the same employ-ees. These firms were dropped from the sample.

with official statistics would indicate that we have a reliable data set, andin fact the figures match up quite well in a comparison with data from offi-cial IMSS statistics.26

Our data represent all sectors of the Mexican economy.27 To verify cov-erage, we compared the 1993 average employment in manufacturing in ourdata (2,958,715.5) with the 1993 average total employment in the 1993 Mex-ican Industrial Census (3,246,039.0). Our data thus cover about 91.1 percent oftotal manufacturing employment. This leads us to conclude that the distinc-tion between formal and informal labor markets, which is so important indeveloping economies, is mainly an issue outside the manufacturing sector.

Since our data are effectively a census of formal sector employment, weare particularly concerned about the rate of attrition in our sample. Work-ers may leave our sample for three reasons: they may leave the labor force,become unemployed, or enter the informal sector. To get a sense of therate of attrition in our sample, we focus on workers who worked at leastone quarter in 1993. Of these workers, 78 percent worked at least once in1994, and 57 percent worked at least once in 2000. About 87 percent of theworkers who appear in our sample for at least one quarter in 1993 appearin our sample for at least one quarter between 1994 and 2000.

Analytically, treating workers who leave the labor force or become un-employed is straightforward. The complication in our data arises because apotentially significant number of workers who leave our sample may enterthe informal sector. The informal sector has traditionally been thought of asan employer of last resort, in which workers earn lower wages and experienceinferior working conditions. Maloney challenges this view for Latin Amer-ica generally and for Mexico in particular.28 He shows that workers whobecome self-employed in the informal sector often earn 25 percent higherwages, on average, than they did as salaried workers in the formal sector.Salaried workers in the informal sector, however, always earn less than theirformal sector counterparts. This result suggests that there is no clear pre-sumption of bias, or, more specifically, the direction of a bias from not beingable to account for informal sector employment is unclear. In the empiricalsection below, we address this potential bias by comparing results acrosssamples that include and exclude workers who drop out of our sample.


26. The official data are from www.imss.gob.mx/ventunica/memoria_2001/2/024000.htm.A table showing this comparison is available on request.

27. Our data cover all economic sectors and are classified using a four-digit industrycode that is similar, but not identical, to the U.S. 1987 SIC code.

28. Maloney (2004).

We analyze employee-level records for the period 1993 to 2000, mea-suring wages on 31 March, 30 June, 30 September, and 31 December ofeach year. This yields thirty-two quarters of data. While the period is deter-mined by data availability, it is a particularly interesting time in which tostudy displacement because it encompasses several reforms and macro-economic events, including the implementation of the North AmericanFree Trade Agreement (NAFTA) in 1994 and the December 1994 collapseof the peso, which induced a serious recession that lasted until 1996.

The Mexican Economic Environment

In addition to the firm-worker identifier, the data also include details suchas geographic region, sector, and the gender and birth date (month and year)of each worker. Regional heterogeneity in Mexico emerges as a result ofhistorical differences in production (centered on Mexico City), concentra-tion of foreign investment in the U.S.-Mexican border region, and the per-sistent poverty and lack of growth in southern Mexico. In particular,Mexican regions have had very different experiences with adjustment.29

We therefore focus on four Mexican regions: the border, the north, centralMexico, and the south.30

The simple model presented earlier suggests that differences in the con-centration of economic activity can affect postdisplacement wages. Mexicoexhibits significant regional heterogeneity. Manufacturing is predomi-nantly located in the border, north, and central regions. The highest rates ofemployment growth and investment have been concentrated in the borderregion, possibly as a result of maquiladora investment.31 The south, in con-trast, concentrates on tourism (most notably in the state of Quintana Roo,home of Cancún) and oil. While aggregate statistics suggest that manufac-turing is similar in the border, north, and center regions, the regions areactually quite different. Figure 2 (panel A) shows the evolution of percapita gross domestic product (GDP) for each region over time. The central


29. Conroy and West (2000).30. We define the four regions as encompassing the following Mexican states: the bor-

der region: Baja California, Coahuila de Zaragoza, Chihuahua, Nuevo León, Sonora, andTamaulipas; the north: Aguascalientes, Baja California Sur, Durango, Guanajuato, Hidalgo,Jalisco, Nayarit, Querétaro de Arteaga, San Luis Potosí, Sinaloa, Veracruz-llave, andZacatecas; the central region: Distrito Federal, México, Morelos, Puebla, and Tlaxcala; andthe south: Campeche, Colima, Chiapas, Guerrero, Michoacán de Ocampo, Oaxaca, QuintanaRoo, Tabasco, and Yucatán.

31. Feenstra and Hanson (1997); Robertson (2000).


A. Per capita GDPa

Thousands 1993 pesos

Percent

Year

Border NorthCentral South

1993 1995 1997 1999 2001

14

19

24

29

34

a. The four lines represent GDP per capita for the different regions. The GDP of each region is calculated as the sum of the real state GDP of all states in each region. The population of each region in each year was calculated using a linear growth trend from data between the 1990 and 2000 population censuses. The real state GDP is in thousands of 1993 pesos. The distribution of states into regions is identified in the text.

b. Regional unemployment rates are the simple averages of city-level unemployment rates in each region, as identified in the text. Tic marks are at third quarter of given year.

B. Unemployment ratesb

Time Period

Border North Central South

1992q3 1993q3 1994q3 1995q3 1996q3 1997q3 1998q3 1999q3 2000q3 2001q3 2002q3

1.5

3

4.5

6

7.5

F I G U R E 2 . Regional Differences in per Capita GDP and Unemployment Rates

and border regions are much more economically dynamic than the north.The central region has the largest amount of measured economic activity,although it has the fewest states. The border and the north regions have sim-ilar levels of total GDP, but GDP per capita is much higher in the borderregion, suggesting that economic activity is more concentrated in the borderregion than in the north. The border region also had a higher GDP growthrate than the north for most of the sample period. All regions experienced asharp recession in 1995; the years following 1995 are recovery years. Weexpect that, to be consistent with the model, postdisplacement wages wouldgenerally be lower in the north than in the central and border regions.

The model also suggests that unemployment rates can affect postdis-placement wages. Panel B of the figure presents regional unemploymentrates calculated as a simple average of the official urban unemployment forthe main cities in each state. The recession is especially evident here.Unemployment rates are highest in the third quarter of 1995, and they fallsteadily in the third quarters of subsequent years in our sample. Unemploy-ment rates track each other quite closely across regions, but the south tendsto have lower unemployment rates than the rest of the country. The northhas the highest peak unemployment rate. Prior to the collapse, the borderregion had higher unemployment rates than other regions, but rates in theborder region fell faster than the rest of the country during the recoveryperiod. We expect that, to be consistent with the model, postdisplacementwages would generally be lower for workers who separated during theheight of the recession.

In contrast, institutions (such as unions) and inequality differ very littleacross regions in Mexico. Fairris and Levine find unionization rates for1998 of 0.21 both in states that share a border with the United States andin states that do not.32 In general, they find little heterogeneity in union-ization rates across regions in Mexico. Inequality varies little across re-gions, as well. As a measure of income inequality, we calculated the Ginicoefficient of the natural logarithm of the real daily wage (the wage mea-sure used in the empirical work below) for each region in our data. In thefirst quarter of 1993, the Gini coefficients for the border, north, central,and south regions were 0.424, 0.422, 0.422, and 0.439, respectively.33


32. Fairris and Levine (2004).33. For the first quarter of 1991, the Gini coefficients were, in the same order, 0.406,

0.409, 0.419, and 0.405. The regional Gini coefficients generally track each other closelyover time, rising after Mexico’s entrance into GATT and then leveling off when NAFTAwent into effect. See Robertson (2004) for further discussion of Mexican wage inequality.

These rates are much more similar to each other than they are to the Ginicoefficients of France, Germany, and the United States.

Description of Comparison Groups and Definition of Displaced Workers

We divide our sample of approximately 80 million observations in severalways. Given the overwhelming size of the data set, we focus on workersdisplaced between the third and fourth quarters of 1995, 1996, and 1997.These periods match three possibly distinct conditions: high unemploy-ment (1995), sharply falling unemployment (1996), and relatively low andstabilizing unemployment (1997). These three displacement periods alsomaximize the time necessary to identify wage movements before and afterdisplacement.

The issues related to defining displaced workers are widely discussedin the literature. Administrative data, such as those used by Jacobson,LaLonde, and Sullivan, generally do not include direct information on thecause of separation. The cause of separation is important because workerswho leave voluntarily are more likely to have more positive economic pros-pects beyond their current firm. Including voluntary separations wouldtherefore bias the estimated effects of displacement upwards.

To get a sense of the magnitude of voluntary and involuntary displace-ment, we draw from the National Urban Employment Survey (EncuestaNacional de Empleo Urbano, or ENEU). This household survey is like theU.S. Current Population Survey in that it is used to calculate measures ofunemployment. The survey contains a question that can be used to deter-mine whether a worker separated voluntarily or involuntarily from the firm.The average responses over time are shown in figure 3. Two facts areimmediately apparent. First, workers in Mexico are more likely to leavetheir firm involuntarily than voluntarily, which might suggest a high vol-untary separation cost for workers. Second, the rate of involuntary (vol-untary) separation is highest (lowest) during the three years on which wefocus (1995, 1996, and 1997). These rates follow the business cycle (theyare consistent with figure 2, panel B) and suggest that our focus years arethe least susceptible to selection bias.34


34. Gonzaga (2003) suggests that workers may negotiate with firms to create theimpression that they were fired when they actually quit, in order to receive severance pay-ments mandated by Brazilian labor law. Our Mexican data do not seem to exhibit the samekinds of patterns (for example, the cyclicality of separations) that this phenomenon appar-ently causes in Brazil.

We follow previous studies in our attempt to minimize this bias byfocusing on workers who left firms with significant contractions. We cre-ated two samples to identify displaced workers. We first identified firmsthat contracted more than 60 percent between the third and fourth quartersof the reference year, from an initial employment of 50 or more workers.Our second sample uses a 30 percent contraction threshold. We labeledthese displacing firms.35 The logic behind this decision, which is well es-tablished in the literature, is that workers leaving these distressed firms aremost likely to be immune from the selection bias that arises from volun-tary separation.

We broke down each of the two samples into four subsamples of dis-placed workers to examine the possible effects of tenure, sample attrition,


35. Jacobson, LaLonde, and Sullivan (1993a) label all firms that contract more than 30 percent from an early sample average as contracting firms.

Time Period

Involuntary Separation Rate Voluntary Separation Rate

87.1 89.1 91.1 93.1 95.1 97.1 99.1 01.1

8

16

24

32

40

48

56

64

72

80

a. Separation rates are calculated based on the Mexican quarterly Encuesta Nacional de Empleo Urbano. The two rates do not add up to 100 because we excluded separation resulting from injury and other exogeneous factors.

F I G U R E 3 . Separation Rates by Motivea

age, and sector changes. To focus on the effects of tenure, we restricted oursample to workers who were in the displacing firm for the entire period upuntil the displacement event, left the firm at the time of displacement,found work in another firm either immediately or after spending some timeout of our sample, and then worked in one firm until the end of the sam-ple period. We refer to this group as C1. Workers who are out of the sam-ple may be either unemployed or working in the informal sector.36 Sincewe cannot distinguish between these two conditions, we created anothersubsample of displaced workers who found jobs immediately after the dis-placement event and then remained employed for the rest of the sample.We label this group C2. Our third group comprises workers who workedat the same (displacing) firm for less than two years prior to displacementand then worked at one firm following displacement (C3); these workersmay not be in the sample for the entire period. This short-tenure criterioncontrasts directly with the long-tenure criterion for workers in the firstgroup. The final group consists of workers who worked at one firm priorto displacement, are in the sample in all periods, but may have worked atseveral firms following displacement (C4).

We contrast the wage patterns of these workers against two compari-son groups. Employees in the first group (A) worked in every quarter atfirms that did not experience large contractions in any quarter during ourperiod of study. Given the large size of the data set, we selected theseworkers from a 1 percent sample of all workers in nondisplacing firms.For 1995, our sample of this group begins with 3.87 million observations,or about 121,000 per quarter. The second group (B) consists of workerswho worked at a displacing firm in every period of the sample (that is,workers in displacing firms who remain with the firm after the displace-ment event).37

About 18 percent of the observations represent multiple firms per workerin each quarter. This could be due to the fact that workers could hold sev-eral formal sector jobs, change jobs frequently within the quarter, or arenot coded correctly. The problem of multiple jobs becomes slightly moreserious when considering displacement because being displaced from yoursecond job may not have the same implications as being displaced from


36. See appendix A for a discussion of the age differences of those who are displacedand those who leave the sample.

37. Since we want to compare the wages of workers who remain in displacing plants,we omit plants that shut down completely.

your primary job. Since we do not have data on hours, it is difficult todetermine which is the primary job, which entries are errors, and whichentries represent changing jobs. We therefore drop all workers with mul-tiple observations per quarter from the sample.

We impose additional sample restrictions. We restrict the sample toworkers between fifteen (in the first year) and seventy-one years of age (inthe last year), those with positive earnings, and those who are not missingsectoral information. We also restrict the sample to those who are notmissing any geographic data.

Summary Statistics

Tables 1 and 2 contain summary statistics for the 1995 ABC1 and ABC2samples (that is, samples covering groups A, B, and either C1 or C2) bysector and region. The “1995 sample” refers to the sample for the 1995 dis-placement event and contains observations for every quarter between 1993and 2000. The summary statistics in tables 1 and 2 summarize data for1996 from these samples. Our measure of wages is the natural log of thereal daily wage.38 Several interesting results emerge. Workers are gener-ally youngest in the central and border regions, and wages are lowest in theborder. The border and the north have higher employment shares in man-ufacturing than in services.

Table 3 disaggregates the A, B, and C components of the sample andcompares the sample summary statistics before and after the 1995 dis-placement event by summarizing the data for 1994 and 1996. The tableshows that the wages of all workers fell between 1994 and 1996. Interest-ingly, the average wages of workers who remained in displacing firms fellby more than workers who were displaced.

One potential concern is that the ages of displaced workers and work-ers who left the sample affect our results. Appendix A formally comparesthe ages of workers in each subsample and those who leave our data. Thiscomparison reveals two main results. First, for workers who remained inthe sample, there is no statistical difference between workers who weredisplaced and workers who remained in displacing firms in the 1995 and1996 sample (workers who remained in displacing firms were, on average,


38. We converted nominal wages to real wages using the national-level consumer priceindex available at www.banxico.org, based on the index values corresponding to the monthof observation.


T A B L E 1 . Summary Statistics for the ABC1 Displacement Sample, 1995:3 to 1995:4a

Sector and summaryGeographic region

statistic Border North Central South Total

AgricultureWage 2.94 2.87 2.80 2.46 2.84Age 37.51 37.52 39.51 44.2 38.46Percent female 8.79 13.11 10.59 18.78 12.1No. observations 2,674 3,096 236 905 6,911

MiningWage 3.34 3.36 3.04 3.63 3.40Age 33.62 35.06 36.92 35.79 34.66Percent female 1.94 1.8 11.11 2.96 2.4No. observations 1,029 778 72 540 2,419

ManufacturingWage 3.25 3.52 3.39 3.47 3.42Age 30.4 33.18 33.63 32.69 32.74Percent female 33.04 29.89 29.61 17.67 29.36No. observations 22,455 45,223 36,741 9,909 114,328

Transport equipmentWage 3.50 3.61 3.86 . . . 3.65Age 29.4 32.72 33.64 . . . 31.65Percent female 28.03 14.91 10.2 . . . 18.77No. observations 2,472 1,254 2,107 0 5,834

ConstructionWage 3.21 3.23 3.20 3.46 3.26Age 35.81 33.89 33.82 35.05 34.79Percent female 3.05 2.78 11.62 9.01 5.33No. observations 15,042 12,888 5,997 8,033 41,960

UtilitiesWage 4.03 4.32 4.15 3.96 4.18Age 38 39.51 39.01 38.36 39.02Percent female 14.44 16 12.6 12.59 14.25No. observations 561 2,025 1,619 588 4,793

ServicesWage 3.20 3.35 3.49 3.32 3.37Age 35.38 36.09 34.81 36.04 35.49Percent female 35.75 42.42 39.76 42.19 40.28No. observations 19,492 35,706 42,777 17,636 115,611

TotalWage 3.23 3.42 3.45 3.38 3.38Age 33.58 34.57 34.33 35.17 34.36Percent female 24.91 29.66 32.43 27.06 29.14No. observations 63,725 100,970 89,549 37,612 291,856

. . . No observations in this category.a. Data are for 1996.


T A B L E 2 . Summary Statistics for the ABC2 Displacement Sample, 1995:3 to 1995:4a

Sector and summary Geographic region

statistic Border North Central South Total

AgricultureWage 2.89 2.71 2.79 2.39 2.66Age 40.21 41.33 44 45.72 42.53Percent female 14.49 13.31 7.32 18.33 14.63No. observations 552 1,052 164 720 2,488

MiningWage 3.70 3.49 3.28 3.92 3.67Age 37.59 36.75 38.71 34.81 36.58Percent female 3.41 1.01 14.29 1.47 2.29No. observations 352 396 28 272 1,048

ManufacturingWage 3.63 3.79 3.70 3.70 3.71Age 33.94 35.76 37.73 36.28 36.11Percent female 32.79 25.82 24.8 17.31 26.36No. observations 8,344 12,020 13,500 2,888 36,752

Transport equipmentWage 3.77 3.84 4.18 . . . 3.94Age 30.93 33.5 35.94 . . . 33.27Percent female 35.73 14.68 7.21 . . . 21.54No. observations 1,500 436 1,276 0 3,212

ConstructionWage 3.45 3.38 3.74 4.02 3.62Age 39.22 35.32 34.19 36.54 36.09Percent female 14.56 7.78 7.69 47.77 17.34No. observations 632 1,028 780 628 3,068

UtilitiesWage 4.16 4.04 4.27 4.01 4.16Age 39.79 39.83 40.37 40.4 40.13Percent female 17.27 20.97 6.95 14.12 13.32No. observations 440 744 1,208 340 2,732

ServicesWage 3.55 3.38 3.81 3.48 3.58Age 38.33 38 38.15 37.84 38.09Percent female 44.8 45.72 39.7 48.66 43.79No. observations 7,884 13,936 15,112 5,540 42,472

TotalWage 3.59 3.55 3.79 3.53 3.64Age 36.01 37.08 37.91 37.88 37.23Percent female 35.85 33.5 29.94 35.43 32.98No. observations 19,704 29,612 32,068 10,388 91,772

. . . No observations in this category.a. Data are for 1996.

1.32 years older than displaced workers in 1997). The point estimatessuggest that workers displaced in 1995 and 1996 were slightly older thanworkers who remained in displacing firms. Second, workers who left thesample were younger than those who remained in the sample. Other stud-ies suggest that workers in Mexico often leave the formal sector to becomeentrepreneurs in the informal sector.39 Since risk is often associated with


39. Maloney (1998, 2004); Maloney and Krebs (1999).

T A B L E 3 . Descriptive Statistics by Displacement Status before and after Displacement Eventa

Year before displacement (1994) Year after displacement (1996)

Statistic A B C1 C2 A B C1 C2

Wage 3.820 4.189 3.541 3.907 3.569 3.854 3.278 3.607(0.829) (0.806) (0.785) (0.774) (0.854) (0.848) (0.737) (0.804)

Age 35.962 33.519 33.042 34.301 37.962 35.519 33.134 36.301(9.902) (9.311) (10.641) (9.455) (9.902) (9.311) (10.474) (9.455)

Percent female 0.349 0.299 0.342 0.248 0.349 0.299 0.273 0.248(0.477) (0.458) (0.474) (0.432) (0.477) (0.458) (0.446) (0.432)

Agriculture 0.035 0.002 0.018 0.019 0.035 0.002 0.023 0.042(0.184) (0.047) (0.133) (0.135) (0.184) (0.047) (0.149) (0.202)

Mining 0.012 0.014 0.008 0.000 0.012 0.014 0.007 0.000(0.107) (0.117) (0.088) (0.000) (0.107) (0.117) (0.081) (0.000)

Manufacturing 0.330 0.640 0.319 0.207 0.330 0.640 0.382 0.192(0.470) (0.480) (0.466) (0.405) (0.470) (0.480) (0.486) (0.394)

Transport equipment 0.049 0.007 0.010 0.000 0.049 0.007 0.013 0.000(0.216) (0.081) (0.101) (0.000) (0.216) (0.081) (0.112) (0.000)

Construction 0.012 0.049 0.179 0.074 0.012 0.049 0.194 0.193(0.109) (0.216) (0.383) (0.261) (0.109) (0.216) (0.396) (0.395)

Utilities 0.038 0.014 0.008 0.001 0.038 0.014 0.010 0.002(0.192) (0.119) (0.087) (0.036) (0.192) (0.119) (0.100) (0.045)

Services 0.524 0.274 0.459 0.699 0.524 0.274 0.372 0.570(0.499) (0.446) (0.498) (0.459) (0.499) (0.446) (0.483) (0.495)

Border 0.263 0.103 0.181 0.145 0.263 0.103 0.218 0.145(0.441) (0.304) (0.385) (0.352) (0.441) (0.304) (0.413) (0.352)

North 0.271 0.468 0.374 0.284 0.271 0.468 0.355 0.288(0.445) (0.499) (0.484) (0.451) (0.445) (0.499) (0.478) (0.453)

Central 0.369 0.255 0.305 0.523 0.369 0.255 0.294 0.519(0.482) (0.436) (0.460) (0.500) (0.482) (0.436) (0.456) (0.500)

South 0.097 0.174 0.140 0.048 0.097 0.174 0.133 0.048(0.295) (0.379) (0.347) (0.213) (0.295) (0.379) (0.340) (0.213)

No. observations 62,260 23,480 251,313 6,032 62,260 23,480 206,116 6,032

a. Displacement occurred between 1995:3 and 1995:4. The groups are defined as follows: group A: workers who are not in displacingfirms and remain in the same firm; group B: workers who are in displacing firms but do not separate from displacing firms; group C1:workers who are in displacing firms, separate from those firms, and are not necessarily employed in every period in the sample; andgroup C2: workers who are in displacing firms, separate from those firms, and are employed in every period in the sample. Standarderrors are in parentheses.

youth, our results seem to be consistent with the idea that when displaced,young workers may find the informal sector attractive and therefore re-main out of our sample.

Table 3 also shows that sample ABC2 is balanced, but ABC1 is not,since workers in C1 are allowed to exit and return to the sample after thedisplacement event. When we consider only a balanced panel of workers,the average age of all workers is two years higher in the later period. Thetable also shows that displaced workers who found a job right away weremost likely to be male, while the difference in the percent of females in C1and C2 suggests that workers who were displaced and exited the samplewere more likely to be female.

The table also includes information about the sectoral distribution ofeach group in each period (the sum over the sectors in each columnequals one hundred percent). Since workers who did not change firmsdid not change sectors, the sectoral distribution of the first two groupsremains constant. One might expect that the sectoral distribution of thetwo groups of workers who were in displacing firms would be identical,but we restrict the sample to workers who were employed in every period.Therefore, differences in the sectoral distribution between the second twogroups reflect the differences in future employment patterns. In our sam-ple, no workers who were displaced from transportation equipment ormining remained in those sectors when they were displaced. They couldbe excluded from the sample as a result of extended search times or theycould have moved to other sectors, such as construction or agriculture.The percent of displaced workers in construction and agriculture morethan doubles following displacement (for workers who immediatelyfound employment).

Regional differences in displacement patterns are also evident in table 3.The majority of employment is in the central region, but so are most of thedisplaced workers who immediately found jobs. The north has the highestshare of workers who either were in displacing firms and did not leave orwere displaced and exited the sample at some point. This may be consis-tent with the shift in production from the central region to the north, asdescribed by Hanson.40 On the other hand, the overall regional pattern ofemployment in table 3 displays a large degree of stability, suggesting thatfew workers who were displaced in a particular region moved to otherregions in the very short run to begin other jobs.


40. Hanson (1998).

Empirical Approach

To maximize comparability with studies in developed countries, we employthe methodological gold standard established by Jacobson, LaLonde, andSullivan.41 We first define displacement indicators as Djit, which equalsone if the worker separated from a displacing firm (and zero otherwise) tocompare the wages of displaced workers with all other workers. After pre-senting these initial results below, we redefine the displacement indicatorto identify workers in each of three groups ( j = 1, 2, 3). The first variabletakes on the value of one for workers who were not in displacing firms, andzero otherwise (group A). The second takes on a value of one for workersin displacing firms who remained with the same firm, and zero otherwise(group B). The third variable takes on a value of one if the workers leftfirms that contracted more than 60 percent in the quarter in which they sep-arated (that is, they are in one of the C samples). We estimate each aggre-gated sample separately (ABC1, ABC2, and so forth). We begin with thefollowing specification.

The dependent variable is the natural log of the real wage, which is cal-culated by adjusting the nominal wages variable by the Mexican nationalconsumer price index using 1994 as the base year. The ai term capturesindividual-specific fixed effects that take on a value of one for each indi-vidual in the sample. The parameter γt represents time-specific effects.Each estimated equation includes a dummy variable for each quarter-year(for thirty-one of thirty-two periods, omitting the first quarter in the sam-ple). The vector xit represents other time-varying characteristics of work-ers, including age. We also include the indicator for the individual’sdisplacement group status, excluding the workers not in displacing firmsas a control group. We then interact the time effects with the displacementgroup indicators to compare wages in each group before and after the dis-placement event. We estimate this equation separately for each of the fourgeographic regions in our sample.42

By fully interacting displacement status with the time effects (dummyvariables for each quarter), we allow the time trend for displaced workers

( ) .1 w a D Dit i t jj ji ji itt jt itj= + + + + +∑ ∑∑γ β ϑ γ δ εxit


41. Jacobson, LaLonde, and Sullivan (1993a, 1993b).42. We estimate separate equations for each region because the sample sizes are so

large.

to differ from the time trend for nondisplaced workers. These differentialtime trends are identified off differences over time in wage changes betweendisplaced workers and nondisplaced workers. We would expect, for exam-ple, that wage changes over time would be fairly similar for displacedworkers and nondisplaced workers before the displacement event, but thatwage changes would begin to differ sharply after the displacement event.This is, in fact, what we normally observe.43

Results

We begin by estimating equation 1 by ordinary least squares (OLS) foreach region. All but thirteen of the 124 estimated marginal effects of thedisplacement x time variables (thirty-one coefficients for each of fourregions) for sample ABC1 are statistically significant at the 5 percentlevel. The R2 are all 0.90 or higher. A nearly identical pattern of signifi-cance emerges for sample ABC2.44 The standard errors are generally verysmall. Since we are particularly concerned about the pattern of the relativewages of displaced workers’ wages over time, however, a graphical pre-sentation may more effectively facilitate comparisons across years andsectors.45 Figures 4 and 5 graph the estimated coefficients for the 1995,1996, and 1997 displacement samples. The patterns of standard errors anddiagnostic statistics are similar for the other years.46

As in Jacobson, LaLonde, and Sullivan, wages in all periods and all re-gions fell prior to displacement; in contrast with Jacobson, LaLonde, andSullivan, no region exhibits a sharp drop in wages at the time of dis-placement.47 Figures 4 and 5 do show significant effects of displacement,but these effects vary by region and time of displacement. Figure 4, forexample, reveals that workers who were displaced in 1995, the trough ofthe recession, did worse than other workers. Workers displaced in lateryears, however, recovered. The heterogeneity across time is especiallyevident in the central and border regions. Workers in the relatively poor


43. We are required to use some normalization for both time trends. For both displacedworkers and nondisplaced workers, we set the coefficient equal to zero for the dummy vari-able corresponding to nine quarters before the (potential) displacement event.

44. The tables are available on request.45. Appendix B provides a more extensive evaluation of statistical significance and

standard errors.46. These results are also available on request.47. Jacobson, LaLonde, and Sullivan (1993a, 1993b).

north show few effects of displacement and fewer of recovery. While thewage trends in the border and central regions become positive at the timeof displacement, wage trends remain flat in the north and (to a lesserdegree) the south.

The second important message emerging from figures 4 and 5 is thatthe effects of being displaced in 1995 seem to be permanent, or at leastlong-lasting. That is, there is little evidence of recovery over the sampleperiod. This is especially true in the border and central regions. The re-sults of being displaced at times of peak unemployment are therefore sim-ilar to the findings of Jacobson, LaLonde, and Sullivan.48 Being displacedat different times generates different results that include positive post-displacement wages.


48. Jacobson, LaLonde, and Sullivan (1993a, 1993b).

Quarters from Displacement

1995 1996 1997

Border

-.3-.15

0

.15.3

.45.6

North

Central

-16 -12 -8 -4 0 4 8 12 16 20

-.3-.15

0

.15

.3

.45.6

South

-16 -12 -8 -4 0 4 8 12 16 20

a. The sample is defined as workers who were in the displacing firm for the entire period up until the displacement event, left the firm at the time of displacement, found work in another firm either immediately or after spending some time out of our sample, and then worked in one firm until the end of the sample period (sample ABC1 in the text). Estimated coefficients of the time x displacement status effects for displaced workers are from equation 1. Groups A and B were both omitted, so these coefficient estimates represent the difference between the wages of displaced workers and all other workers in the sample. The reference time period is nine quarters prior to the displacement event for each sample.

F I G U R E 4 . Effects of Displacement by Region: Sample ABC1a

The differences between figures 4 and 5 suggest that workers who maynot be in the sample in all periods may suffer more serious repercussionsfrom displacement than other workers. In general, workers who were em-ployed immediately did better than workers who were out of the samplefor any length of time. The difference between 1995 and 1996 becomesmore pronounced in the north and somewhat less pronounced in other re-gions when we focus on those in the sample in all periods. Even workerswho were employed immediately in the north continued to experiencefalling wages, whereas workers in the dynamic central and border regionsdid much better when employed immediately.

Figures 4 and 5 compare the wages of displaced workers with all otherworkers, which allows us to compare our results with other studies in the



1995 1996 1997

Border

-.3-.15

0

.15.3

.45.6

North

Central

-16 -12 -8 -4 0 4 8 12 16 20-.3

-.15

0

.15.3

.45.6

South

-16 -12 -8 -4 0 4 8 12 16 20

a. The sample is defined as workers who were in the displacing firm for the entire period up until the displacement event, left the firm at the time of displacement, found work in another firm immediately (that is, they spent no time out of our sample), and then worked in one firm until the end of the sample period (sample ABC2 in the text). Estimated coefficients of the time x displacement status effects for displaced workers are from equation 1. Groups A and B were both omitted, so these coefficient estimates represent the difference between the wages of displaced workers and all other workers in the sample. The reference time period is nine quarters prior to the displacement event for each sample. The omitted category is workers who were not in displacing firms and remained in the same firm for the entire sample.

F I G U R E 5 . Effects of Displacement by Region: Sample ABC2a

literature. We can also divide the comparison groups into nonseparatingworkers who are in distressed firms and those who are not in distressedfirms. Figure 6 contains the results from the comparison of displaced work-ers with workers who remain in contracting firms for the entire sample.49

The main result of this comparison is that workers who were displacedexperienced large wage gains relative to workers who remained in dis-tressed firms. In nearly every region and every time period, workers whoseparated from distressed firms experienced higher wages relative to


A. Sample BC1a


1995 1996 1997

Border

-.3-.15

0.15

.3.45

.6

North

Central

-16 -12 -8 -4 0 4 8 12 16 20-.3

-.150

.15.3

.45.6

South

-16 -12 -8 -4 0 4 8 12 16 20

a. The sample is defined as workers who were in the displacing firm for the entire period up until the displacement event, left the firm at the time of displacement, found work in another firm either immediately or after spending some time out of our sample, and then worked in one firm until the end of the sample period (sample ABC1 in the text). Estimated coefficients of the time x displacement status effects for displaced workers are from equation 1. Group B was omitted, so these coefficient estimates represent the difference between the wages of displaced workers who remained in the displacing firm. The reference time period is nine quarters prior to the displacement event for each sample.

F I G U R E 6 . Effects of Displacement by Region Relative to Workers Who Remained in the Firm

49. The results from comparisons with workers in nondisplacing firms only are verysimilar to those described above, so we do not discuss them in detail.

(continued )

workers who stayed behind. This result is consistent with the model pre-sented earlier (specifically, regions II and III of figure 1).

The second result that emerges from figure 6 is that many of the pat-terns described in the analysis of figures 4 and 5 remain: wages fall priorto displacement and the long-run effects of being displaced in 1995 areless positive than the effects of being displaced in recovery years. Dis-placement in the border and central regions is followed by higher wagesthan in the north and south. Workers displaced in 1995 in the north andsouth took much longer to recover than their counterparts in the borderand central regions. As the economy recovered, so did the prospects ofdisplaced workers.


B. Sample BC2b


1995 1996 1997

Border

-.3-.15

0.15

.3.45

.6

North

Central

-16 -12 -8 -4 0 4 8 12 16 20

-.3-.15

0.15

.3.45

.6

South

-16 -12 -8 -4 0 4 8 12 16 20

b. The sample is defined as workers who were in the displacing firm for the entire period up until the displacement event, left the firm at the time of displacement, found work in another firm immediately (that is, they spent no time out of our sample), and then worked in one firm until the end of the sample period (sample BC2 in the text). Estimated coefficients of the time x displacement status effects for displaced workers are from equation 1. Group B was omitted, so these coefficient estimates represent the difference between the wages of displaced workers who remained in the displacing firm. The reference time period is nine quarters prior to the displacement event for each sample.

F I G U R E 6 . Effects of Displacement by Region Relative to Workers Who Remained in the Firm (Continued )

One of the findings in the current literature is that tenure increases theadverse effects of displacement. To investigate the effects of tenure, wecreated a sample similar to the first except that we dropped all workerswith more than two years tenure in the displacing firm. We performed thesame empirical exercise using this sample and present the results in fig-ure 7. The results in panels A and B can be directly compared with theresults in figures 4 and 6 (panel A). Panel A of figure 7 suggests thatshort-tenure workers did better than workers with longer tenure. This isconsistent with worker training and other hypotheses in the literature.Short-tenure workers displaced in 1995, however, did worse than work-ers displaced in 1996 and 1997 in the border and central regions, since the


A. Sample ABC3a


1995 1996 1997

Border

-.3-.15

0.15

.3.45

.6North

Central

-16 -12 -8 -4 0 4 8 12 16 20

-.3-.15

0.15

.3.45

.6South

-16 -12 -8 -4 0 4 8 12 16 20

a. The sample is defined as workers who may not have been in the sample for the entire period, who worked at the same (displacing) firm for less than two years prior to displacement, and then worked at one firm following displacement (sample ABC3 in the text). This short tenure contrasts directly with the long tenure of workers in the first group. Estimated coefficients of the time x displacement status effects for displaced workers are from equation 1. Groups A and B were both omitted, so these coefficient estimates represent the difference between the wages of displaced workers and all other workers in the sample. The reference time period is nine quarters prior to the displacement event for each sample. The omitted category is workers who were not in displacing firms and remained in the same firm for the entire sample.

F I G U R E 7 . Effects of Displacement by Region: Short-Tenure Workers

(continued )

latter immediately earned higher wages and the former experienced a down-ward trend in wages. Time of displacement also induces more heterogene-ity in the northern region: short-tenure workers in the north who separatedin 1995 did much worse than long-tenure workers who displaced at thesame time.

Tenure does not seem to matter when comparing displaced and non-displaced workers from displacing firms, in the sense that the overall re-sults in figures 6 (panel A) and 7 (panel B) are very similar. Workers in allperiods and regions eventually did better than workers who stayed behind.Workers displaced during recovery periods did the best in all regions, andworkers displaced in the border and central regions did better than work-ers in the north and south regardless of tenure. Therefore, differences in


B. Sample BC3b


1995 1996 1997

Border

-.3-.15

0.15

.3.45

.6

North

Central

-16 -12 -8 -4 0 4 8 12 16 20

-.3-.15

0.15

.3.45

.6

South

-16 -12 -8 -4 0 4 8 12 16 20

b. The sample is defined as workers who may not have been in the sample for the entire period, who worked at the same (displacing) firm for less than two years prior to displacement, and then worked at one firm following displacement (sample BC3 in the text). This short tenure contrasts directly with the long tenure of workers in the first group. Estimated coefficients of the time x displacement status effects for displaced workers are from equation 1. Group B was omitted, so these coefficient estimates represent the difference between the wages of displaced workers who remained in the displacing firm. The reference time period is nine quarters prior to the displacement event for each sample.

F I G U R E 7 . Effects of Displacement by Region: Short-Tenure Workers (Continued )

the effects of displacement do not seem to be driven by tenure so much asby local labor market conditions.

Figure 8 presents our results for the sample in which workers are allowedto change firms several times following displacement but remain in thesample. We focus on this sample because workers who switch more oftenmay have either lower search costs, which would suggest that their wageswould be higher, or less potential to accumulate firm-specific capital, whichwould suggest that their wages would be lower. The results in panels Aand B of the figure suggest that, generally, workers who switch more oftenmay be no less susceptible to the effects of the time of displacement thanworkers who switch less often. In fact, the results are almost identical tofigures 5 and 6 (panel B), respectively, which implies that the results are


A. Sample ABC4a


1995 1996 1997

Border

-.3-.15

0

.15.3

.45.6

North

Central

-16 -12 -8 -4 0 4 8 12 16 20-.3

-.15

0

.15.3

.45.6

South

-16 -12 -8 -4 0 4 8 12 16 20

a. The sample is defined as workers who worked in one firm prior to displacement, were in the sample in all periods, and may have worked at several firms following displacement (sample ABC4 in the text). Estimated coefficients of the time x displacement status effects for displaced workers are from equation 1. Groups A and B were both omitted, so these coefficient estimates represent the difference between the wages of displaced workers and all other workers in the sample. The reference time period is nine quarters prior to the displacement event for each sample. The omitted category is workers who were not in displacing firms and remained in the same firm for the entire sample.

F I G U R E 8 . Effects of Displacement by Region: Postdisplacement Movers

(continued )

not driven by the restriction that workers stay in the same firm for theremainder of the sample. Overall, the regional and temporal heterogeneityseem to matter more than the worker’s switching cost.

One of our concerns about displacement is that workers may lose spe-cific human capital. This problem may increase with the degree of changea worker experiences from the original position. To consider some of thepotential effects of the loss of such capital, we differentiated the effectsof displacement for workers who remained in the same two-digit sectorand those who changed sectors. To isolate the comparison, we focused onsample ABC4, which is the sample in which workers are observed for allperiods but may change firms more than once following displacement. Wethen created an indicator variable based on whether the worker changedsectors at the time of displacement. The results are presented in figure 9.


B. Sample BC4b


1995 1996 1997

Border

-.3-.15

0.15

.3.45

.6

North

Central

-16 -12 -8 -4 0 4 8 12 16 20

-.3-.15

0.15

.3.45

.6

South

-16 -12 -8 -4 0 4 8 12 16 20

b. The sample is defined as workers who worked in one firm prior to displacement, were in the sample in all periods, and may have worked at several firms following displacement (sample BC4 in the text). Estimated coefficients of the time x displacement status effects for displaced workers are from equation 1. Group B was omitted, so these coefficient estimates represent the difference between the wages of displaced workers who remained in the displacing firm. The reference time period is nine quarters prior to the displacement event for each sample.

F I G U R E 8 . Effects of Displacement by Region: Postdisplacement Movers (Continued )


A. Workers who changed two-digit sectors


1995 1996 1997

-16 -12 -8 -4 0 4 8 12 16 20

-.3

-.15

0

.15

.3

.45

.6

B. Workers who did not change two-digit sectors


1995 1996 1997

-16 -12 -8 -4 0 4 8 12 16 20

-.3

-.15

0

.15

.3

.45

.6

F I G U R E 9 . Sectoral Effects of Displacement

Panel A compares workers who change sectors and all other workers, whilepanel B compares workers who remain in the same sector and all otherworkers. For these regressions, we pooled all of the regions and controlledfor region-specific effects using regional dummy variables.

As in previous studies, our results suggest that Mexican workers experi-ence some loss to capital from changing sectors or, perhaps more precisely,a gain from remaining in the same sector. Workers who changed sectors atthe time of displacement do not seem that much different from other work-ers for the majority of the sample. Displaced workers who remained in thesame sector, however, follow a pattern similar to those workers in previoussamples. Wages fell prior to displacement, and rose following displace-ment if the workers were displaced after 1995.

As an additional robustness check, we considered all of the samples andresults described above using a contraction of 30 percent, rather than 60 per-cent, as our criterion for identifying displacing firms. If selection biasseverely affects our sample, then the bias would be larger with firms underthe 30 percent contraction criterion, because workers leaving firms thatcontract by 30 percent would probably include a higher proportion of vol-untary separations. These results are nearly identical numerically and qual-itatively to the results presented above.50 We find no evidence of a risingproblem of selection bias when we expand the sample. This may be due tothe fact that we include individual-specific fixed effects in all of the regres-sions, and these effects may effectively be capturing unobserved charac-teristics that are correlated with ability and other features that could driveselection bias. This result, along with the fact that we follow establishedapproaches designed to minimize selection bias, leads us to believe thatselection bias does not significantly drive our results.

Conclusions

Given limited resources and a desire to support displaced workers, policy-makers could increase the efficiency of support programs if they targetedaid when and where it is needed most. Studies on the effects of displace-ment on wages have generated a wide range of results, but they do littletheoretically or empirically to formally explain the underlying sources ofthis heterogeneity. Previous studies suggest that differences in institutions,

50. The results are available on request.


inequality, or labor market conditions might explain the results, but no studythat we are aware of compares these possibilities using matched worker-firm data over time.

In this paper we examine the costs of displacement to workers using anadministrative data set that allows us to follow workers over thirty-twoquarters and four regions that vary significantly in labor market conditions.By following an established empirical methodology to estimate postdisplace-ment wages, we focus on the differences in institutions, inequality, andlabor market conditions in a single study in an attempt to understand thedifference in results.

Several findings emerge. First, our results exhibit the same heterogeneityfound in the current literature. We find a range of postdisplacement experi-ences from negative (such as those documented by Jacobson, LaLonde, andSullivan) to positive (such as those documented in Kuhn).51 Since inequal-ity and institutions (unions) are similar throughout Mexico but the empir-ical results vary through time and space, we therefore conclude that ouranalysis provides little support, if any, for the institutional explanation.This conclusion is further backed by international comparisons. If nationalinstitutions alone explained the differences in results between Germanyand the United States, we would expect to see little heterogeneity withinMexico rather than the very wide range of results we find.52 Furthermore,if France and Germany have positive postdisplacement because wages aremore compressed than in the United States, then we would expect the dis-placement effects in Mexico to be mainly negative because Mexico’sinequality is greater than that of the United States. Instead, we find muchheterogeneity in the results, with many instances of positive postdisplace-ment wages.

We conclude that differences in local labor market conditions (overboth space and time) are most consistent with our results. We do find large,negative, and lasting effects of displacement on wages for workers whoare displaced during times of high unemployment and in less economi-cally active regions.53 Postdisplacement wage changes are typically zeroor positive in good times and in the most economically active regions. This


51. Jacobson, LaLonde, and Sullivan (1993a, 1993b); Kuhn (2002).52. The difference in the results of Jacobson, LaLonde, and Sullivan (1993b) and

Abbring and others (2002), which both focus on the United States, also weighs against theinstitutional explanation.

53. Jacobson, LaLonde, and Sullivan (1993b) find similar results.

is similar to recently documented patterns in France, Germany, and theUnited States.

Our results are robust to changes in the definition of displaced worker.For example, we consider the implications of displacement for workerswho may not be employed immediately following displacement, workerswho are employed after possibly being outside the labor force, workers withdifferent levels of tenure, and workers who change sectors. Our results aregenerally consistent with other studies that focus on tenure and sectoralchanges. We also explore different reference groups and find strong andconsistent evidence that displaced workers earn significantly higher wagesthan their coworkers that were left behind, which seems consistent with avery simple theoretical model and Kuhn’s reminder that comparison groupmatters.54

We also find that of our four regions (the border, the north, centralMexico, and the south), the border region has the displacement wage pat-tern most like that of the United States. Other studies have shown thatlabor markets in this region are the most integrated with the United States,so this similarity may not be surprising. The main example of this is thedecline in wages prior to displacement documented by Jacobson, LaLonde,and Sullivan.55 This feature is not present in all regions or at all times ofdisplacement in Mexico, but it emerges most frequently in Mexico’s borderregion.

The main policy recommendation that emerges from our results is thattargeting aid to displaced workers during recessions and in less economi-cally active areas has potentially significant efficiency gains. These work-ers tend to suffer larger and more lasting adverse effects from displacementthan other workers, which suggests that targeted aid may be especially valu-able. This recommendation clearly assumes that behavior does not changewith policy. But clearly, potential changes in behavior must be taken intoaccount when considering changes to policy.

Appendix A: Age, Displacement, and Attrition

We formally compared the ages of displaced workers and workers wholeft our sample. We are concerned about any difference in age because it


54. Kuhn (2002).55. Jacobson, LaLonde, and Sullivan (1993a, 1993b).

could affect our results through a sort of selectivity bias. If workers wholeft the sample are systematically younger than the workers who stayed,for example, then the remaining workers’ wages may appear higher sim-ply because these were older workers with more experience.

The results illustrate several points. First, workers in displacing firmsare generally younger than workers in nondisplacing firms. The age of dis-placed workers is not statistically different from that of workers who re-mained in displacing firms for 1995 and 1996, although the point estimatessuggest that displaced workers were slightly older. In 1997, displacedworkers were younger than workers who remained in displacing firms, andthe difference (about 1.32 years) is statistically significant.

Second, we find that workers who left the sample were significantlyyounger than workers who remained in the sample. As discussed in thetext, this seems to be consistent with other papers that find relatively highrates of entrepreneurship in the informal sector. Our results may be biaseddownwards (upwards) if these workers earn higher (lower) wages, onaverage, than workers in the formal sector. These results are shown intable A1.


T A B L E A 1 . Age Comparisons across Samplesa

Age

Explanatory variable 1995 1996 1997

Age and attritionAlways in sample 2.308 5.814 6.174

(0.215)** (0.129)** (0.146)**Constant 32.503 30.937 30.313

(0.032)** (0.031)** (0.029)**No. observations 112,032 116,437 136,321R2 0.00 0.02 0.01

Age relative to displaced workersAll other workers 2.222 1.034 2.186

(0.207)** (0.140)** (0.152)**In displaced firms −0.108 −0.171 1.324

(0.230) (0.183) (0.193)**Constant 34.811 36.751 36.487

(0.191)** (0.116)** (0.130)**No. observations 23,711 26,537 24,579R2 0.01 0.00 0.01

*Statistically significant at the 5 percent level; ** statistically significant at the 1 percent level.a. Standard errors are in parentheses.

We used two main approaches to address this issue. First, we includedage and individual-specific fixed effects in our wage regressions to controlfor differences in age. Second, we explored the robustness of our results todifferent samples. The basic patterns emerge in all samples, regardless ofhow we control for experience or attrition. Changing samples affects theabsolute, but not the relative, magnitude of our results.

Appendix B: The Statistical Significance of the Differences

We undertook a simple analysis to determine whether the differences be-tween regions and periods are statistically significant. Given our large sam-ple sizes, the standard errors are generally quite small, suggesting that thedifferences between regions and periods are often statistically significant.For example, figure B1 graphs the 95 percent confidence intervals for theestimates of the three periods (1995, 1996, and 1997) for the border regiongraph in figure 4. The graph suggests that the differences across periodsare probably not statistically significant prior to displacement, but clear


time-18 -12 -8 -4 0 4 8 12 16 20

-.3

-.15

0

.15

1997

1996

1995

F I G U R E B 1 . Ninety-Five Percent Confidence Intervals for the Border Region in Figure 4


time-12 -8 -4 0 4 8 12 16 20

-.3

-.15

0

.15

Border

Central

North

F I G U R E B 2 . Ninety-Five Percent Confidence Intervals for Three Regions in the 1995Sample in Figure 4

differences emerge after displacement. With regard to differences acrossregions, figure B2 graphs the effects of displacement in 1995 for the bor-der, north, and central regions from figure 4 with the 95 percent confidenceintervals. The differences between the lines again suggest that the differ-ences between regions are statistically significant, especially in the post-displacement period.

Comments

Naércio Menezes-Filho: This interesting paper on the costs of job dis-placement in Mexico is commendable for several reasons. First, it dealswith an important issue, namely, the earnings trajectory of individuals whochange their jobs. While there is an established literature on this subjectfor developed countries, studies on developing countries are scant, despitethe fact that most of the recent reforms that provoked labor reallocationoccurred in the latter, where the problems of poverty and inequality aresevere. Second, the paper addresses this issue using very good data—essentially a census of private firms in the formal sector of the Mexicaneconomy—over a long period. Finally, the analysis is thorough, as theauthors submit their results to various robustness tests using different sub-samples. The main drawback of the paper lies in the interpretation of theresults, as I detail below.

The paper does not fully distinguish among the different explanationsfor the costs of displacement. It presents interesting graphical analyses ofthe wage changes for different periods and regions of displacement, but itoffers very little formal statistical analysis as to whether these differencesare statistically significant (apart from two figures in appendix B). Theauthors could have pooled the data and explicitly tested for differences inthe displacement effect across regions and periods or included indicatorsof labor conditions at the time of displacement, such as regional unem-ployment, directly in the regression. Moreover, institutions and inequalitymay differ across regions in Mexico. Their explanatory power should betested as well, if the aim is to provide a formal test of the different ex-planations for the displacement effect. It does not suffice to state thatinequality and institutions vary less within Mexico than across countries,so they are not likely to be the main explanations for the different costs ofdisplacement.

The authors could also have spent more time interpreting the results,since reconciling them with the theory is not straightforward. The model

235

predicts, for example, that postdisplacement wages should be lower in thenorth than in the central and border regions, but figures 4 and B2 showexactly the opposite, as wages fall by more in the border and central areas.The results change completely in figure 5, where the panel sample is used,(and again in figure 6, panel A), yet the authors do not present a fully con-vincing explanation for these changes.

The main problem with the data is that, as in other studies, the authorscannot distinguish between displacements and voluntary separations. Asfigure 3 shows, the rate of involuntary displacements varies over time,reaching its peak in 1995, when displacement seems to be most damagingto the worker. Does this rate vary across regions, as well? This selectionproblem could explain some of the variation in displacement effects doc-umented in the paper. To minimize this selection problem, the authors focuson workers from displacing firms, that is, firms that contracted more than60 percent between the third and fourth quarters of a given year (30 per-cent in another subsample). According to the authors, these workers areless likely to have voluntarily separated from their firm than workers infirms that did not implement such massive layoffs. But why were theseworkers displaced instead of the workers who continued working at thedisplacing firm? According to the model, the displaced workers were thosewith expected wages above the new proposed wage. If this is the case, whywere their wages falling prior to displacement, as the various figures show?

Another question that deserves a more careful explanation is why prod-uct and labor market conditions vary so much across regions and over time.The paper does not investigate the reasons for such differences in anydetail. Do good firms and workers, for example, endogenously locate in theborder regions so as to enjoy its good prospects? Is this choice driven byunobservables? In other words, an endogeneity problem may underlie thedifferences in the displacement effect across regions.

In sum, this paper represents an important step toward better under-standing the displacement problem in developing countries. The results asa whole are very interesting, but they deserved a more careful explanation,especially in view of the selection problems mentioned above.

Omar Arias: The paper discusses the impact of displacement (resultingfrom layoffs or voluntary separation) on future earnings performance usingMexican data. The topic is certainly of utmost relevance for Latin Amer-ica and the Caribbean in light of the limited reform of overly protective jobregulations and the need for well-designed support for displaced workers.


The paper provides an extensive and concise review of the literature in thisarea and presents novel results that highlight the heterogeneous potentialearnings impacts of displacement depending on labor market conditions.The paper will be useful for both researchers and policymakers to betterunderstand the role of factors mediating the impact of displacement onearnings and factor these into policy design.

The paper uses a unique panel data set for a large sample of Mexicanworkers registered in social security over a reasonably long period. Theauthors discuss the methodological difficulties of isolating the impact ofdisplacement on future earnings. Three sets of issues merit special atten-tion: the problems caused by omitted variables and self-selection (samplecomposition biases); attrition or incomplete employment spells (censoringbiases); and the existence of heterogeneous impacts. The first two refer tothe inability to appropriately control for worker and firm characteristicsthat may be correlated with both displacement probabilities and post-displacement earnings, to the restriction to workers with social securityregistration (that is, formal sector), and to the possibility that workers whodrop out of the sample may have different characteristics and earnings per-formance than those who stay. The paper proposes several ways to addressthese issues and discusses the implications for the robustness of the re-sults. The third point relates to the fact that average postdisplacement earn-ings may vary widely across workers depending on context-specific factorsand workers’ skills. The paper argues that the empirical results favor animportant role for varying labor market conditions over that of local insti-tutions and inequality. I focus my comments on some questions for futureresearch with regard to the methodological approach and the robustness ofthe empirical evidence to discern competing explanations of impacts.

Although not framed in this way, the paper deals with an impact evalua-tion problem, in which the treatment effect corresponds to the change indisplaced workers’ earnings. The counterfactual is given by the change inearnings that would have occurred had these workers not been displaced, andit is approximated by the change in earnings of comparable nondisplacedworkers. This raises issues familiar from the impact evaluation literature:identifying the parameter(s) of interest, whether the control (comparison)groups are good proxies of the counterfactual, and validity of the identify-ing assumptions. The recent evaluation literature highlights that alterna-tive treatment (impact) parameters could be of interest, although they arenot always identifiable. For example, one may want to measure the aver-age impact of displacement (the effect on any randomly selected worker),


the average effect on the treated (the impact on formal workers actuallydisplaced), or a local average effect on the treated (the impact for workersclose to displacement thresholds, such as those fired first in a recession).1

These parameters have different interpretations and, more important, leadto different implications regarding the impact of displacements. For exam-ple, the latter parameter tends to capture impacts on marginal workers(that is, those displaced at the margin during layoffs). These impacts maydepend on both observed skills (for example, human capital measures likeyears of education or tenure) and unobserved skills (such as individualability or labor market connections). The parameter thus fails to fully cap-ture the impact of large-scale layoffs such as those that would occur inmajor recessions or economic restructuring. The paper analyzes multipletreatment groups that seem to resemble local average impacts, and it is notentirely clear that the estimated effects readily generalize to the impacts ofdisplacements of any size or to well-defined groups of workers. Conse-quently, the results may have limited application for interventions targetedto massive numbers of workers.

The question of the validity of comparison groups is fundamental to theresults. Lacking other identifying restrictions, this boils down to whetherearnings trends prior to displacement were the same in the displaced(treated) and nondisplaced (control) samples. The paper does not explic-itly discuss this identifying condition. It does not seem to hold for all ofthe displaced samples considered since wages decline prior to displace-ment in some regions or periods. If these trends were not matched in thecorresponding comparison groups of nondisplaced workers, it would raisequestions of possible biases arising from dissimilar composition of the sam-ples (that is, differences in worker characteristics across groups) or meanreversion (in which earnings eventually move back towards their mean). Itwould be useful for future work to discuss these issues in detail.

The paper maintains that the results support a greater role for labor mar-ket conditions in mediating the impacts of displacements vis-à-vis otherfactors such as labor institutions and inequality. While well argued andsuggestive, this claim deserves further exploration in future studies. First,the reported similarity of inequality levels within Mexico does not con-form to results from other studies that find significant differences in in-equality levels across Mexican regions.2 The reported Gini coefficients are


1. For a clear exposition of this, see Moffitt (1999).2. See, for example, Andalón-López and López-Calva (2002).

obtained from the sample of formal sector workers under study (who arelikely to have equally dispersed earnings across regions), while the rele-vant statistics should cover the entire local labor markets. Second, varyingregional capacities to enforce labor legislation may lead to de facto re-gional differences in relevant regulations. Finally, it is ultimately difficultto separate labor market conditions (outcomes), such as unemployment,from the characteristics of labor institutions. For example, differences inthe enforcement of regulations that prescribe high severance payments ornonwage benefits correlate with differences in the rates of unemploymentor informal employment.

Thus the reported variation in displacement effects across regions andtime does not support definitively disregarding the potential role ofinequality and institutions in mediating the impacts of displacement.Future empirical research should delve further into the questions raised bythe new results of the paper and their implications for informing the designof policies to better balance protection against job loss and more flexiblelabor regulations in the region.


References

Abbring, Jaap H., and others. 2002. “Displaced Workers in the United States andthe Netherlands.” In Losing Work, Moving on: International Perspectives onWorker Displacement, edited by Peter J. Kuhn, pp. 105–94. Kalamazoo: W. E.Upjohn Institute for Employment Research.

Andalón-López, Mabel, and Luis López-Calva. 2002. “The Many Mexicos:Income Inequality and Polarization in Urban Mexico during the 1990s.” Paperprepared for the Cornell-LSE-WIDER Conference on Spatial Inequality. Cor-nell University, London School of Economics, and the World Institute forDevelopment Economics Research, London, 28–30 June.

Bender, Stefan, and others. 2002. “Worker Displacement in France and Ger-many.” In Losing Work, Moving on: International Perspectives on Worker Dis-placement, edited by Peter J. Kuhn, pp. 375–470. Kalamazoo: W. E. UpjohnInstitute for Employment Research.

Burda, Michael C., and Antje Mertens. 2001. “Estimating Wage Losses of Dis-placed Workers in Germany.” Labour Economics 8(1): 15–41.

Caballero, Ricardo J., Eduardo M. R. A. Engle, and John C. Haltiwanger. 1997.“Aggregate Employment Dynamics: Building from Microeconomic Evidence.”American Economic Review 87(1): 115–37.

Carrington, William J. 1993. “Wage Losses for Displaced Workers: Is It Reallythe Firm that Matters?” Journal of Human Resources 28(3): 435–62.

Clark, Don P., Henry W. Herzog Jr., and Alan M. Schlottmann. 1998. “Import Com-petition, Employment Risk, and the Job-Search Outcomes of Trade-DisplacedManufacturing Workers.” Industrial Relations 37(2): 182–206.

Conroy, Michael E., and Sarah E. West. 2000. “The Impact of NAFTA and theWTO on Chiapas and Southern Mexico.” In Poverty or Development: GlobalRestructuring and Regional Transformations in the U.S. South and the Mexi-can South, edited by Richard Tardanico and Mark Rosenberg, pp. 41–55. NewYork: Routledge.

Couch, Kenneth A. 2001. “Earnings Losses and Unemployment of DisplacedWorkers in Germany.” Industrial and Labor Relations Review 54(3): 559–72.

Davis, Steven J., and John C. Haltiwanger. 1992. “Gross Job Creation, Gross JobDestruction, and Employment Reallocation.” Quarterly Journal of Economics107(3): 819–63.

———. 1999. “Gross Job Flows.” In Handbook of Labor Economics, vol. 3B,edited by Orley Ashenfelter and David Card. Amsterdam: Elsevier.

De la Rica, Sara. 1995. “Evidence of Preseparation Earnings Losses in the Dis-placed Worker Survey.” Journal of Human Resources 30(3): 610–21.

Fairris, David, and Edward Levine. 2004. “Declining Union Density in Mexico,1984–2000.” Monthly Labor Review 127(9): 10–17.

Fallick, Bruce C. 1996. “A Review of the Recent Empirical Literature on Dis-placed Workers.” Industrial and Labor Relations Review 50(1): 5–16.


Farber, Henry S. 2003. “Job Loss in the United States, 1981–2001.” Workingpaper 471. Princeton University, Industrial Relations Section.

Feenstra, Robert C., and Gordon H. Hanson. 1997. “Foreign Direct Investment andRelative Wages: Evidence from Mexico’s Maquiladoras.” Journal of Inter-national Economics 42(3): 371–93.

Gonzaga, Gustavo. 2003. “Labor Turnover and Labor Legislation in Brazil.”Economía 4(1): 165–222.

Hamermesh, Daniel S. 1989. “Labor Demand and the Structure of AdjustmentCosts.” American Economic Review 79(4): 674–89.

Hanson, Gordon H. 1998. “North American Economic Integration and IndustryLocation.” Oxford Review of Economic Policy 14(2): 30–44.

Hashimoto, Masanori. 1979. “Bonus Payments, On-the-Job Training, and Life-time Employment in Japan.” Journal of Political Economy 87(5, part 1):1086–104.

———. 1981. “Firm-Specific Human Capital as a Shared Investment.” AmericanEconomic Review 71(3): 475–82.

Heckman, James, and Carmen Pagés. 2000. “The Cost of Job Security Regulation:Evidence from Latin American Labor Markets.” Working paper 7773. Cam-bridge, Mass.: National Bureau of Economic Research.

Helwig, Ryan T. 2001. “Worker Displacement in a Strong Labor Market.” MonthlyLabor Review 124(6): 13–28.

Howland, Marie, and George E. Peterson. 1988. “Labor Market Conditions andthe Reemployment of Displaced Workers.” Industrial and Labor RelationsReview 42(1): 109–22.

Jacobson, Louis S., Robert J. LaLonde, and Daniel G. Sullivan. 1993a. “EarningsLosses of Displaced Workers.” American Economic Review 83(4): 685–709.

———. 1993b. The Costs of Worker Dislocation. Kalamazoo: W. E. Upjohn Insti-tute for Employment Research.

Kaplan, David, Gabriel Martínez González, and Raymond Robertson. 2004.“Worker and Job Flows in Mexico.” Macalester College.

Kletzer, Lori G. 1998. “Job Displacement.” Journal of Economic Perspectives12(1): 115–36.

Kreichel, Ben, and Gerard A. Pfann. 2003. “Heterogeneity among DisplacedWorkers.” Paper prepared for the conference on Analyses of Labor Realloca-tion. W. E. Upjohn Institute and the University of Michigan Business School,William Davidson Institute, Augusta, Mich., 21–23 August.

Kuhn, Peter J. 2002. “Summary and Synthesis.” In Losing Work, Moving on:International Perspectives on Worker Displacement, edited by Peter J. Kuhn,pp. 1–104. Kalamazoo: W. E. Upjohn Institute for Employment Research.

Maloney, William F. 1998. “Are Labor Markets in Developing Countries Dualis-tic?” Policy research working paper 1941. Washington: World Bank.

———. 2004. “Informality Revisited.” World Development 32(7): 1159–78.


Maloney, William F., and Tom Krebs. 1999. “Quitting and Labor Turnover: Micro-economic Evidence and Macroeconomic Consequences.” Policy research work-ing paper 2068. Washington: World Bank.

Marcal, Leah E. 2001. “Does Trade Adjustment Assistance Help Trade-DisplacedWorkers?” Contemporary Economic Policy 19(1): 59–72.

McLaughlin, Kenneth J. 1991. “A Theory of Quits and Layoffs with EfficientTurnover.” Journal of Political Economy 99(1): 1–29.

Menezes-Filho, Naércio. 2004. “Costs of Displacement in Brazil.” University ofSão Paulo.

Moffitt, Robert A. 1999. “ Models of Treatment Effects When Responses AreHeterogeneous.” Proceedings of the National Academy of Sciences 96(12):6575–76.

Panagides, Alexis, and Harry A. Patrinos. 1994. “Union-Nonunion Wage Differ-entials in the Developing World: A Case Study of Mexico.” Policy researchworking paper 1269. Washington: World Bank.

Revenga, Ana, Michelle Riboud, and Hong Tan. 1994. “The Impact of Mexico’sRetraining Program on Employment and Wages.” World Bank Economic Re-view 8(2): 247–77.

Robertson, Raymond. 2000. “Wage Shocks and North American Labor MarketIntegration.” American Economic Review 90(4): 742–64.

———. 2004. “Relative Prices and Relative Wages: Evidence from Mexico.”Journal of International Economics 64(2): 387–409.

Robertson, Raymond, and Donald H. Dutkowsky. 2002. “Labor Adjustment Costsin a Destination Country: The Case of Mexico.” Journal of Development Eco-nomics 67(1): 29–54.

Ruhm, Christopher J. 1991a. “Are Workers Permanently Scarred by Job Displace-ments?” American Economic Review 81(1): 319–24.

———. 1991b. “Displacement Induced Joblessness.” Review of Economics andStatistics 73(3): 517–22.

Stevens, Ann Huff. 1995. “Long-Term Effects of Job Displacement: Evidencefrom the Panel Study of Income Dynamics.” Working paper 5343. Cambridge,Mass.: National Bureau of Economic Research.

———. 1997. “Persistent Effects of Job Displacement: The Importance of Multi-ple Job Losses.” Journal of Labor Economics 15(1, part 1): 165–88.

———. 2001. “Changes in Earnings Instability and Job Loss.” Industrial andLabor Relations Review 55(1): 60–78.

Stevens, Margaret. 1994. “A Theoretical Model of On-the-Job Training withImperfect Competition.” Oxford Economic Papers 46(4): 537–62.


Economia: Spring 2005 (Economia: Journal of the Latin American & Caribbean Economic Association)

Documents

Transcript of Economia: Spring 2005 (Economia: Journal of the Latin American & Caribbean Economic Association)