Three Essays in InternationalIntegration

Alexander Fraser McQuoid

Submitted in partial fulfillment of the

requirements for the degree of

Doctor of Philosophy

in the Graduate School of Arts and Sciences

COLUMBIA UNIVERSITY

2012

c� 2012Alexander Fraser McQuoid

All rights reserved

Abstract

Three Essays in International Integration

Alexander Fraser McQuoid

In this dissertation, I consider multiple dimensions of international integration. In chap-

ter one, I consider the impact of immigration on public finance. In chapter two, I study

capacity constrained firms and the transmission of foreign shocks to the domestic market

through these firms. In chapter three, I focus on the importing behavior of firms and how

macro and micro patterns of trade and production diverge.

In the first chapter, I investigate the role diversity plays in the provision of public goods.

The conventional wisdom holds that diversity is a significant hindrance to collective action

and the provision of public goods. Empirical support for this view comes primarily from the

observation that measures of diversity are negatively correlated with provisions of public

goods in the cross-section. The generally held conjecture is that this negative relationship is

true within countries over time as well. I address this belief directly by exploiting a natural

migration experiment and a unique IV strategy to causally identify the impact of diversity

on public goods expenditures and revenues. With the political collapse of the Soviet Union

in the fall of 1989, mass migration to Israel increased the population there by roughly seven

percent over two years. This led to substantial changes in diversity in local communities,

with some becoming more homogeneous and others becoming more diverse. I confirm the

usual negative relationship in the cross-section by using data on local government budgets

at disaggregated levels. However, I find limited evidence that increased diversity leads to

lower expenditures on local public goods when I instrument for changes in diversity using

historic settlement patterns. Local revenue generating mechanisms do respond to changes

in diversity, but are offset by national government transfers.

Chapter two challenges a central assumption of standard trade models: constant marginal

cost technology. We present evidence consistent with the view that increasing marginal cost

is present in the data, and further identify financial and physical capacity constraints as

the main sources of increasing marginal costs. To understand and quantify the importance

of increasing marginal costs faced by financially and physically constrained exporters, we

develop a novel structural estimation framework that incorporates these micro frictions.

Our structural estimates suggest that the presence of such capacity constrained firms can

(1) reduce aggregate output responses to external demand shocks by 30% and (2) result in

welfare loss by around 23%.

Chapter three contributes to the understanding of a long-running puzzle in international

trade. For more than 40 years, economists have analyzed the phenomenon that trade is

excessively volatile relative to GDP, with a recent revival of interest following the “Great

Trade Collapse”. This well-documented phenomenon of excess sensitivity of trade has been

observed in numerous countries and across multiple time periods. A variety of explanations

have been considered, but none have satisfactorily solved the puzzle. The point of departure

for the present study is to match theory and empirics explicitly by using plant level data

on imported intermediate inputs and production to evaluate the theory. Bringing both

macro and micro data from Indonesia to bare on the question, I find the import elasticity

puzzle is more accurately characterized as an aggregation puzzle. While aggregate national

accounts data exhibit the typical excess sensitivity of trade, I find no such excess sensitivity

of imports at the plant level. I estimate the income elasticity of imports to be one, precisely

as standard theory predicts. Explanations for this aggregation puzzle are considered.

Contents

1 Does Diversity Divide? Public Goods Provision and Soviet Emigration to

Israel 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Related Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.3 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1.4 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

1.5 Identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

1.6 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

1.7 Social Cleavage and Role of Local Government in Israel . . . . . . . . . . . 23

1.7.1 Social Cleavage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

1.7.2 Local government . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

1.8 Empirical Design and Implementation . . . . . . . . . . . . . . . . . . . . . . 27

1.8.1 Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

1.9 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

1.9.1 Preliminary Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

1.9.2 Pre and Post-Shock Approach . . . . . . . . . . . . . . . . . . . . . . 36

1.9.3 Revenues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

1.10 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

1.11 Figures and Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

i

2 Capacity Constrained Exporters: Micro Evidence and Macro Implica-

tions 64

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

2.2 Illustrative Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

2.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

2.4 Reduced Form Evidence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

2.5 Structural Form Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

2.5.1 Structural Framework . . . . . . . . . . . . . . . . . . . . . . . . . . 84

2.5.2 Structural Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . 86

2.5.3 Counterfactuals I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

2.5.4 Counterfactuals II . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

2.7 Figures and Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

3 The Import Elasticity Puzzle: An Aggregation Puzzle? 106

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

3.2 What’s so puzzling? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

3.3 Related Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

3.4 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

3.5 Empirical Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

3.6 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

3.7 Macro Puzzle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

3.8 Micro Evidence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120

3.9 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

3.10 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

3.11 Figures and Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

ii

References 142

A Appendix: Capacity Constrained Exporters: Micro Evidence and Macro

Implications 143

A.1 Underlying Model for Welfare Loss Evaluation . . . . . . . . . . . . . . . . . 143

iii

List of Figures

1.1 Israeli Immigration by Month, 1970-2010 . . . . . . . . . . . . . . . . . . . . 47

1.2 Israeli Population Growth, 1970-2010 . . . . . . . . . . . . . . . . . . . . . . 47

1.3 Voter Participation, Municipal and Knesset Elections, 1949-2003 . . . . . . 48

1.4 Total per capita spending in High and Low Migration Intensity Localities . . 48

1.5 Soviet Settlement by Initial Immigrant Share . . . . . . . . . . . . . . . . . 49

1.6 Soviet Settlement by Initial Population . . . . . . . . . . . . . . . . . . . . . 49

1.7 Soviet Settlement by Initial Religious Fragmentation . . . . . . . . . . . . . 50

1.8 Soviet Settlement by Initial Ethnic Fragmentation . . . . . . . . . . . . . . . 50

2.1 Constant Marginal Cost and Production . . . . . . . . . . . . . . . . . . . . 97

2.2 Increasing Marginal Cost and Production . . . . . . . . . . . . . . . . . . . . 97

2.3 Infinite Marginal Cost and (Sub) Optimal Production . . . . . . . . . . . . . 98

2.4 Cross Correlation of Constraint Measures . . . . . . . . . . . . . . . . . . . . 98

2.5 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

2.6 Summary Statistics for Constrained Firms . . . . . . . . . . . . . . . . . . . 99

2.7 Domestic and Export Sales Tradeoffs . . . . . . . . . . . . . . . . . . . . . . 99

2.8 Capacity Constraints and Domestic-Export Sales Trade Offs . . . . . . . . . 100

2.9 Robustness Check with Productivity as TFP . . . . . . . . . . . . . . . . . . 100

2.10 Robustness Check with Productivity as Levinsohn and Petrin Methodology . 101

2.11 Robustness Check with Alternative Physical Capacity Constraint Measure . 102

2.12 Robustness Check with Alternative Financial Capacity Constraints Measure 103

iv

2.13 Robustness Check with Inventory Adjustments . . . . . . . . . . . . . . . . . 104

2.14 One Percent Positive External Demand Shock . . . . . . . . . . . . . . . . . 105

2.15 One Percent Negative External Demand Shock . . . . . . . . . . . . . . . . . 105

3.1 Real GDP and Trade Growth, 1959-2010 (percent changes year to year) . . 127

3.2 Real GDP and Trade Growth, 1993Q1-2003Q4 (percent changes quarter to

quarter) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

v

List of Tables

1.1 Summary Statistics, Demographic . . . . . . . . . . . . . . . . . . . . . . . 51

1.2 Summary Statistics, Expenditures . . . . . . . . . . . . . . . . . . . . . . . 51

1.3 Ln (Total Spending Per Capita), Religious Fragmentation . . . . . . . . . . 52

1.4 Ln (Education Spending Per Capita), Religious Fragmentation . . . . . . . 53

1.5 Ln (Welfare Spending Per Capita), Religious Fragmentation . . . . . . . . . 54

1.6 Ln (Total Spending Per Capita), Ethnic Fragmentation . . . . . . . . . . . 55

1.7 Ln (Education Spending Per Capita), Ethnic Fragmentation . . . . . . . . . 56

1.8 Ln (Welfare Spending Per Capita), Ethnic Fragmentation . . . . . . . . . . 57

1.9 Expenditures per capita (Religious Fragmentation) . . . . . . . . . . . . . . 58

1.10 Expenditures Per Capita (Ethnic Fragmentation) . . . . . . . . . . . . . . . 59

1.11 Sources of Revenue Per Capita, Religious Fragmentation . . . . . . . . . . . 60

1.12 Sources of Revenue Per Capita, Ethnic Fragmentation . . . . . . . . . . . . 61

1.13 Own Revenue Sources Per Capita, Religious Fragmentation . . . . . . . . . 62

1.14 Own Revenue Sources Per Capita, Ethnic Fragmentation . . . . . . . . . . . 63

2.1 Implied Parameter Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

3.1 Summary Statistics, All Importing Firms . . . . . . . . . . . . . . . . . . . 128

3.2 Income Elasticity, Annual Frequency, 1959-2010 . . . . . . . . . . . . . . . . 128

3.3 Income Elasticity, Quarterly Frequency, 1993-2002 . . . . . . . . . . . . . . . 128

3.4 Import Demand Elasticities, Plant Level, 1990-1999 (Annual) . . . . . . . . 129

vi

Acknowledgements

A dissertation can never be completed in isolation, and my experiences are no different.

The resulting document was influenced, in small and large, by many people. While I can’t

thank them all here, I would like to single out a few for special consideration.

Without the support and insight of Don Davis, my dissertation would never have existed.

I would learn more about life, research, and myself in an hour of conversation with him than

I ever dreamed possible. A mentor in the truest and most encompassing sense conceivable.

I would also like to thank the Columbia faculty who took the time to listen, consider,

and critique my work. Columbia’s International Economics Seminar is the hot coals over

which all who want to claim the mantle of trade economist should walk. I’m proud to be

part of such a distinguished lineage.

Thanks to my fellow Trade students for teaching me so much about the world. A special

thanks to my co-author and co-conspirator JaeBin Ahn for showing me how soju brings out

the best ideas. Paul Landefeld and Guru Sethupathy, I propose we continue our arguments

about offshoring over 18 down in Miami.

Lastly, my Columbia classmates were what made this whole process bearable. If not for

David Grad, I might have finished the dissertation sooner, but at a personal cost I dare not

calculate. If you ever need a coffee break, just let me know. Joao Salles introduced me to

Peking Duck, and my world has never been the same. Thank you both for participating

in the Summer of Lunch. Ryan Chahrour kept me from becoming obese with our gym

commitment and ethical pickup basketball games. While I take most of the blame for this

document, Cyntia Azevedo, Evan Borkum, Fang He, Anne J., Kyle Jurado, and Arunima

vii

Sinha must share somewhat in the infamy.

As always, my mother suffered far more than I during the process. While I appreciated

the regular reminders that "there is still only one Ph.D. in the family", I’m sorry I put you

through it. Now scoot over, there is finally another (fake) doctor in the family.

viii

Mom, as you requested, I finished the damn thing before you died

ix

Chapter 1

Does Diversity Divide? Public Goods Provision

and Soviet Emigration to Israel 1

1.1 Introduction

"The differences in attitude towards redistributive taxes are not just betweencountries but also within them, and economists have several explanations as towhy. When it comes to differences between countries, social cohesion plays amajor role. Broadly speaking, countries that are more ethnically orracially homogeneous are more comfortable with the state seekingto mitigate inequality by transferring some resources from richer topoorer people through the fiscal system. This may explain why Swedescomplain less about high taxes than the inhabitants of a country of immigrantssuch as America. But it also suggests that even societies with a tradition ofhigh taxes (such as those in Scandinavia) might find that their citizens wouldbecome less willing to finance generous welfare programmes were immigrantsto make up a greater share of their populations.” - The Economist, EconomicFocus, August 13th 2011 (emphasis added)

The role diversity plays in the provision of public goods has been an important issue

for many of the social sciences. The provision of public goods, and collective action more

generally, have been challenging problems to understand, given that individual incentives

are often misaligned with social incentives. While theory has tended to focus on forces that1I would especially like to thank Adi Brender for his help in acquiring the Local Authorities Financial

and Physical Data.

1

2

either exacerbate or alleviate the collective action problem, the empirical observation has

been that while some collective action is often observed, the provision of public goods seems

to depend greatly on the specific characteristics of a given community.

The difficulty in generalizing the important forces driving the provision of public goods

has led to researchers stressing different elements of diversity in trying to understand when

collective action is likely to fail. Studies have looked at a variety of cleavages in society,

including age, religion, income, ethnicity, and race. The general findings support the view

that increased diversity is correlated with lower provisions of public goods, though the

mechanisms behind these failures are less clear.

The conventional wisdom, as summed up by the opening quotation from a recent The

Economist article (August 13th, 2011), notes that there is robust cross-country evidence

that suggests individuals are more comfortable with fiscal redistribution policy when a

country is more socially cohesive, and racial or ethnic diversity has been identified as a key

source of social fracture. While cross-country evidence is relatively robust, there is a logical

gap in extending the results to fiscal policy within a country. The quotation above offers

only that this evidence is “suggestive” of within country behavior. The quotation hints

at a larger issue - namely, in a world with increasing diversity and mass migration, will

robust fiscal redistribution mechanisms survive? If there is a true causal force leading from

increased diversity to diminished provision of public goods, then a significant reorganization

of societies will be necessary in an increasingly diverse world.

The main challenge associated with this observation is that it is empirically difficult

to identify a causal mechanism leading from diversity to the provision of public goods.

While there are abundant theories making this claim, it is empirically challenging to iden-

tify causality without some sort of exogenous shock to diversity in a community. To be truly

confident about the assertion that diversity leads to a lower provision of public goods, rather

than some alternative factor driving both the provision of public goods and observed diver-

sity, requires a suitable experimental approach to study the impact of exogenous changes in

3

diversity on the provision of public goods.

This paper attempts to identify this causal channel by studying an episode of mass

migration that led to an exogenous increase in diversity in Israel after the collapse of the

Soviet Union in 1989. The wave of migration that took place from the Soviet Union to

Israel following the collapse was intense in both size and swiftness. Nearly 400,000 immi-

grants entered Israel in 1990 and 1991, compared to only a couple of thousand in the years

preceding. This initial shock represented about seven percent of the population in Israel at

the time. Over the entire decade, nearly a million Soviet Jews would eventually move to

Israel. The social and political ramifications of this immigration experience are still being

felt today.

This migration episode has a number of features that make it an ideal natural experiment

for studying the impact of diversity on public goods provision. Besides the swiftness and

size already mentioned, the migration phase was unexpected. The collapse of the Soviet

Union occurred suddenly, and for many Soviet Jews the opportunity to leave Russia was

unplanned. Given the political and social uncertainty following the collapse, emigration

was a new and necessary option for many Soviet Jews. From the point of view of empirical

identification, the collapse of the Soviet Union was uncorrelated with local conditions in

Israel, and represents an exogenous migration shock to Israel.

Furthermore, this was the second large Soviet migration shock to hit Israel in the last

half of the 20th century. Following international outcry in response to the social repres-

sion brought about by the Dymshits–Kuznetsov Hijacking Affair in 1970, the Soviet Union

temporarily relaxed emigration rules. This led to significant emigration of Soviet Jews to

Israel from 1972 to 1975, representing an increase of about four percent of the Israeli pop-

ulation. This emigration wave and settlement pattern created networks of Soviet Jews in

Israel, which were especially strong because the migrants spoke little Hebrew. To deal with

concerns about settlement patterns in 1990 being driven by unobservable characteristics, I

construct instruments for changes in diversity based on actual settlement patterns in the

4

1970’s. Identification then depends upon the large immigration shock and instrumented

settlement patterns.

I focus on two measures of social fragmentation, religious and ethnic, which most closely

capture the social divisions within Israeli culture. First, I construct a measure of religious

fragmentation. Israel is primarily a Jewish nation, but there is significant variation in

religious populations at Local Authority level. There is little doubt that religious identity

represents a key source of division within this society. If anything, the worry may be

that religious fragmentation is too extreme in Israel, and hence not generally applicable.

While religious contestation over scarce resources is a common feature of many migration

episodes (e.g. Muslims in Europe), questions about general applicability may remain since

the religious identity of the migrants was the same as the dominant culture.

To address this question further, I focus on intra-Jewish ethnic divisions. Focusing

on localities that are predominantly Jewish, I construct a measure of ethnic fragmentation

based on geographic branching. The major ethnic division in Israel is between the Ashkenazi

and the Mizrahim. The Ashkenazi have geographic roots in Western Europe, and make up

the majority culture in Israel. The Mizrahim are composed of a variety of different Jewish

traditions, and are often associated with Jews from Muslim or Middle Eastern cultures.

The history of Judaism, with its exodus and return, has been shaped by the traditions and

cultures that were developed while in exile in countries around the world. This geographic

branching manifests itself in different linguistic, culinary, and liturgical practices. What

is important to note is that the society itself has identified ethnicity as a salient social

difference. All cultures struggle with their own sense of in-group and out-group, and the

drawing of social boundaries. In this sense, ethnic divisions identified in this paper have

been well documented within the Israeli culture itself, and represent a key dimension of

social fragmentation. The results for ethnic diversity in Israel may more easily generalize

to other countries, which have their own culturally defined ethnic boundaries.

Using disaggregated expenditure data for over 100 Local Authorities in Israel, ranging

5

in population size from 5,000 to over 500,000, I replicate the negative coefficient found on

measures of diversity using a pooled sample. These cross-sectional results are similar to

those found in earlier studies. Based on the empirical approaches used in earlier studies,

one would conclude that social fragmentation, both religious and ethnic, is a significant

hindrance to the provision of public goods. In this dimension, there is nothing special about

either the Israeli data or experience.

The unique features of the migration waves from the Soviet Union allow one to go further,

however. First, utilizing a before and after lens, I find only limited evidence that diversity

has a negative effect on public goods expenditures. These results represents a significant

step forward, as the migration wave provides a large shock to diversity at the local level in

Israel, and once one accounts for unobservable fixed effects, there is no consistent evidence

that diversity reduces the provision of public goods. This result is robust to a variety of

measures of public goods (common to previous literature) and both religious and ethnic

fragmentation.

While the shock was uncorrelated with local conditions in Israel, settlement patterns

may be responding to unobserved local conditions. To address this possibility, I instrument

for the change in diversity using the predicted change in diversity based on the settlement

patterns from the 1970’s migration experience. This specification confirms that for local

public expenditures, there is only very limited evidence that religious diversity matters, and

no evidence that ethnic diversity matters.

Next, I turn to the revenues side of the local budget. At the aggregate level, total

revenues and total expenditures respond very similarly to social diversity. The pooled

analysis suggests that there is a strong negative effect of diversity on both sides of the

local government budget ledger. When changes in diversity are instrumented, however,

social fragmentation is not significant. When I dig deeper into sources of revenue, I find a

significant impact of social diversity on locally raised revenue, which is offset by national

government transfers. Furthermore, the mechanisms used to generate local revenue respond

6

differently to different types of social fragmentation.

The results presented here suggest that the usual implementation for measuring the

impact of diversity is flawed, and that theory may need to be re-evaluated in light of this

finding. The methodology used here could be applied in similar contexts to evaluate the

robustness of the results. The connection between diversity and collective action is an

important issue for public policy, and while there have been repeated attempts to understand

the role of diversity, empirical studies thus far have mainly documented negative correlations,

and provided suggestive evidence about the causal force. The present study exploits a

natural experiment that significantly altered the diversity of a country in a short time

horizon to study the causal mechanism leading from diversity to the provision of public

goods.

The paper is organized as follows. In Section 2, related literature is discussed, before

turning to motivating theory in Section 3. Background of the shock is presented in Section 4,

and the identification strategy is presented in Section 5. The data is described in Section 6,

while local government and social fragmentation in Israel is discussed in Section 7. Section 8

contains the empirical design and implementation. In Section 9, results are shown. Section

10 concludes.

1.2 Related Literature

A large body of literature grappling with the question of diversity and public goods

exists, and a complete discussion of the related literature is beyond the scope of the present

paper. Alesina and Ferrara (2005) provide a fuller treatment in their recent survey on

ethnic diversity and economic performance. What distinguishes the present work from the

previous literature is the novel attempt to empirically identify causality, and the application

of this strategy to a variety of public goods covering most local government expenditures

and sources of revenue within a country. A further contribution is that a priori no stand

7

is taken on what the relevant measure(s) of diversity should be. Rather, multiple mea-

sures of diversity common to the literature can be studied simultaneously to evaluate their

importance.

The correlation between diversity and growth outcomes has driven much of the inter-

est in the literature. Easterly and Levine (1997) jump-started the literature by looking

at ethnic-linguistic fractionalization (as measured by language) and cross-country growth

patterns. They found a strong negative correlation, particularly for African countries. A

number of papers followed up this observation by carefully documenting micro observations

within African countries and villages, again finding similar negative correlations between

ethnic fractionalization and public goods provisions. Easterly (2001) updates the analysis

to include measures of institutional quality, and finds that good institutions counterbal-

ance the negative effect of ethnic diversity, though it raises further questions about the

endogeneity of institutions and ethnic fragmentation.

The macro literature captured additional interest by tying the diversity of nations to the

generosity of their welfare systems. In the most well-known study, Alesina et al. (2001) argue

that one of the key determinants of the lack of development of a European-style welfare state

in the U.S. can be found in the racial diversity of the country. All-encompassing welfare

states require social cohesion from cradle to grave, which is easier to maintain if a country

is more homogeneous. In this view, the difference between welfare systems in the U.S. and

Sweden can be explained by the greater homogeneity of the latter.

There are three major challenges for an empirical study of the impact of diversity on the

local provision of public goods. First, a significant number of local governments is needed to

convincingly study the allocation decisions for local public goods. Second, a cross-sectional

approach can be informative, but it ultimately lacks conviction as unaccounted for factors

may be driving the results. Adding a time-series dimension would help inference, but since

diversity changes very little over time in most areas, it is empirically challenging to identify

impacts. The general strategy is to increase the time horizon of study, but then the force of

8

the panel - the ability to control for time-invariant unobserved factors - is reduced. Third,

even when diversity changes over time within localities, these changes may be correlated with

unobserved factors, confounding inference. The migration event in Israel provides an ideal

setting in that it can account for all three challenges: significant cross-sectional variation,

significant changes in diversity over time, and a plausibly exogenous interpretation of these

changes in diversity.

The paper closest to the present study is Alesina et al. (1999) who study ethnic diversity

and public goods in a cross-section of U.S. cities. This provides the clearest parallel since the

focus is on the provision of local public goods, which is more likely to be subject to sorting

as first hypothesized by Tiebout (1956). The authors attempt to control for unobserved

forces driving the negative correlation observed in the cross-section by using a panel, but

due to data limitations, they are unable to convincingly argue that the observed correlation

should be interpreted causally. Even with the appropriate time-series data, however, there

is still concern that unobserved factors that change over time could be biasing the results,

and a panel approach by itself would still be insufficient for identifying causality.

A related paper by Alesina et al. (2004) attempts to tackle both causality and endogenous

sorting behavior in studying the optimal size and characteristics of political boundaries in

the presence of diverse communities. While they don’t look at the variety of public goods

provisions as in Alesina et al. (1999), they do study changes in school jurisdictions over time.

To attempt to identify causality, they use migration phases during World War I and World

War II of blacks from the South to the North. Again, however, sample size is an issue, as

the authors are only able to study a small number political jurisdictions. Furthermore, push

and pull factors in the migration phase are difficult to control and could be influencing both

settlement patterns and political jurisdictions.

From a methodological perspective, the closest paper is Boustan (2010). Building off

an identification strategy proposed by Card (2001), she uses the black migration experience

in the U.S. after World War I to disentangle “white flight” from other causes of white

9

suburbanization. Using a conceptually related IV of predicted migration patterns based

on historical settlement patterns, she is able to disentangle the impact of black migrants

into a neighborhood from other forces driving white residents away (such as changes in

housing prices). Her constructed instruments depend upon predictions for both push and

pull forces, whereas here I use the collapse of the Soviet Union for push exogeneity and

a previous settlement pattern for pull exogeneity. The logic of the identification strategy,

along with the use of a migration episode to identify economic outcomes is similar, although

actual implementation and the content of the study are different. Related network-based

instruments have been employed by Munshi (2003) to study labor market outcomes for

migrants.

Local government in Israel has been studied previously in a number of different dimen-

sions. Brender (2005) studies religious segregation in Israeli-Arab localities, and looks at the

impact of segregation on local tax revenue raised through property taxes collected. Justman

and Spivak (2004) look at changes in socio-economic measures of well-being in local author-

ities over a similar time period. The interaction of fiscal behavior and local government

elections was studied in Brender (2003), while Navon (2006) looks at the evolution of local

government budgets over time in response to changes in budget deficits.

Finally, the immigration wave in Israel after the collapse of the Soviet Union has been

used before to study economic outcomes. Friedberg (2001) studies the impact of immigration

on the labor market. The excellent data collected by the Israeli government on immigrants

allowed her to use information about the occupations of immigrants while in Russia to

estimate the impact of immigrants on native wages. Gandal et al. (2004) use the migration

experience to test the Rybczynski Theorem by looking at the change in production structure

in response to the migration wave, which included a significant number of highly educated

migrants. The immigration experience could be used to further disentangle a variety of

economic questions including the impact of migration on production and trade.

10

1.3 Theory

There are a variety of theories linking diversity and economic outcomes, most of which

emphasize a few main mechanisms. The most common mechanism suggested is simply one

of preferences. Ethnic diversity enters individual preferences directly, as people prefer to

be around others from their "own group," and diversity in terms of inclusion of members

from another group lowers utility. This theory is in some sense a tautology since it fails

to characterize what constitutes an "own group," or explain group formation. Individuals

prefer people like them. Homogeneity, appropriately defined, is good.

A second mechanism emphasizes the difficulty of collective action, particularly issues of

monitoring and enforcement when free-riding behavior is possible. This mechanism assumes

nothing about the taste for diversity, but if there are market imperfections, it may be less

costly to coordinate with people who share a type.2 Affiliation with a group can expand

the range of possible punishments, while increasing benefits from cooperative behavior. The

expansive literature on collective action often emphasizes the relative costs and benefits of

group coordination, though it too often ignores questions of group formation or even what

the salient borders are for constituting a group.

While the first two mechanisms emphasize the costs of diversity, a third competing

mechanism emphasizes potential benefits from diversity. Nearly all of the empirical litera-

ture confirms the negative relationship between diversity and various economic outcomes,

and hence the majority of the theoretical literature has been devoted to explaining why

this negative correlation exists. There are, however, reasons to think that diversity can

improve economic outcomes, and most economic models build in some benefit from diver-

sity, unwittingly or otherwise - capital and labor in a production function being a simple

example.2Miguel and Gugerty (2005) argue that social sanctions within groups are easier to impose than across

ethnic groups. They explore this mechanism empirically and find that in more ethnically fragmented areas,communities impose fewer sanctions on parents who fail to contribute to local school funding.

11

Monopolistic competition models use standard Dixit-Stiglitz preferences and production

functions to capture the positive aspect of diversity. Having a variety of inputs or a variety

of consumption goods increases economic performance in these models, and more diversity

is always preferred in the basic structure. It is natural to assume that integrating variety

also comes with costs, leading to a conclusion that an optimal amount of diversity exists.

Related models of firm organization or innovation emphasize that the variety and diver-

sity of ideas can improve economic outcomes, although there are additional costs associated

with incubating this diversity. In general, a more nuanced model that captures both the

potential benefits as well as the costs of diversity seems preferable to assumptions about

tastes for diversity.

To help set expectations and interpretations of the empirical results, consider the fol-

lowing simple stylized model proposed by Alesina and Ferrara (2005) in their review of

the literature. The output produced in the economy depends upon the total number of

individuals in the economy, different types of individuals, and amount of inputs used:

Y = Nf(x;K) (1.3.1)

where N is total population, x is a fixed factor of input, and K is the number of differ-

ent types in the economy. Assume a standard CRS production function with diminishing

returns to a factor, fx > 0, fxx < 0, fK > 0,and fKK < 0. The benefit from diversity is

captured in the positive first derivative of f with respect to k. Finally, assume that there is

a complementarity in production, fxK > 0.

Individual preferences depend on consumption, both of private and public goods. The

utility derived from consumption of the public good depends upon the number of types as

well as the amount of the public good consumed. The dependence on type captures either

of the first two mechanisms mentioned above. It could simply reflect a dislike of having to

share with a different type. It could also reflect the fact each type has an ideal public goods

preference, but an increase in the number of types involved in producing the public good

12

results in an increase in the expected difference between own preference and group outcome,

an interpretation first suggested in Alesina and Spolaore (1997).

The utility function is given by:

Ui = u(ci) + v(g,K) (1.3.2)

The allocation between public and private goods depends on the tax rate in the economy

such that

g = t ∗ y

= t ∗Nf(x;K) (1.3.3)

In a social planner problem, the optimal allocation solves the following problem:

maxN [u(ci) + v(g,K)] (1.3.4)

subject to

Nc+ g = Nf(x,K)

g = tNf(x;K)

which yields a solution characterized by:

Nvg(g∗, K) = ug(c

∗i ) (1.3.5)

This equation states that the optimal allocation balances the marginal benefits from taxation

(increased consumption of the public good) with the marginal costs of taxation (reduced

private consumption).

Given this equilibrium, the question of how the optimal taxation, and hence public goods

13

provision, changes with diversity can be explored further. Applying the implicit function

theorem yields the following result:

sign{ dt

dK} = sign{tN2vggfk +NvgK − (1− t)uccfk} (1.3.6)

The result, which holds N constant while focusing on just the impact of increased diversity

on public goods provision, suggests that the LHS is in general ambiguous. While most of the

empirical results find the sign to be negative, the trade-offs discussed above leave open the

possibility that an increase in diversity can increase or decrease public goods expenditure.

The key trade-off here is between the magnitude of the marginal benefit of public good

consumption, which declines with social fragmentation, and the increase in productivity as

a result of increased variety in production. The interpretation of the negative correlation

observed in the data in light of the above theory is that the disutility of sharing public

consumption with those different from you outweighs the gains in productivity from having

greater variety in production, resulting in a reallocation of consumption away from public

goods towards private goods.

Furthermore, while the theory usually assumes the marginal impact of diversity on pro-

duction is positive, or that the marginal impact of diversity on public consumption is neg-

ative, it is possible to be agnostic about the direction of impacts. The rationale for these

assumptions was made to match the observed negative correlation in the data. For our

present purposes, it is sufficient that diversity impacts both productivity and consumption

of public goods, without having to make further assumptions on the theory. With this basic

model in mind, we can now turn to the experiment.

1.4 Background

Starting from its founding in 1948, Israel has repeatedly experienced significant waves of

immigration. Prior to statehood, migrants from Europe and the Arab world were common,

14

and these trends continued throughout the modern history of Israel. By the 1980’s, however,

immigration had slowed significantly. Around one thousand immigrants arrived each month

throughout the 1980’s.

This relatively consistent trend was broken sharply at the end of the decade, beginning

with the collapse of the Soviet Union in the Fall of 1989. Mass migration followed, with

the peak of monthly immigration topping 36,000 in 1990 (see Figure 1.1). Over a two-year

period, 1990 and 1991, the population of Israel increased by nearly seven percent (see Figure

1.2). By the end of 1991, immigration settled down to around 5-10 thousand per month,

which continued for most of the rest of the decade. Over the first half of the decade, over

600,000 immigrants from the former Soviet Union arrived, which resulted in an increase

of the population by over twelve percent. This represents a truly remarkable immigration

experience, in both size and swiftness.

This mass migration can be directly linked to the lifting of emigration restrictions in

the Soviet Union, which, when coupled with uncertain and unstable political conditions,

led many Russian Jews to emigrate. Israel was a likely destination for a variety of reasons,

including the lack of restrictions placed on new immigration. Aliyah, or the legal right

of return, gives eligible immigrants certain political rights, including assisted settlement,

automatic citizenship, and all the rights associated with citizenship.

In a period of marked uncertainty, access to Israel for migrants was highly appealing.

The United States, for example, changed their immigration policy towards the Former

Soviet Union (FSU) in response to the political collapse. Prior to 1990, Soviet emigrants

were accorded refugee status and migration, if possible, was less restricted. Starting in

1990, a standard quota approach to immigration was used for the Former Soviet Union.

This severely limited immigration to the U.S. In addition, the U.S. enacted a policy that

targeted “family reunification” and prioritized immigrants who already had close relatives

living in the U.S. While 200,000 Soviet emigrants moved to Israel in 1990, only 35,000

emigrated to the United States.

15

For many countries, immigrants are actively excluded from the political process, either

through explicit restrictions or informal barriers. This was not the case in Israel, where

immigrants immediately had the right to vote, and political levers were in place from earlier

Soviet settlement experiences. The right to vote is granted to every resident of a Local

Authority, regardless of citizenship, so long as they are listed in the population registry and

are 18 years or older in the election year (Elazar and Kalchheim (1988)). This political access

is important for uncovering the impact of diversity and immigration on local public goods. If

part of the observed negative relationship between immigration waves and declining welfare

states is because of political participation, it would be a mistake to attribute this impact to

changing diversity rather than political mechanisms. Immediate access to political levers in

Israel is important when studying changes in local government expenditures and revenues,

since immigrants need to be able to participate in the political process for this to be a

meaningful outcome to measure.

While voting participation at the local level in Israel has been trending down over time,

the most significant change in participation came in 1978, when local and national elections

were decoupled. After a sharp fall in local government participation in 1978, voting was flat

for most of the next two decades, before falling sharply to fifty percent in 2003. Immigration

doesn’t appear to be a significant contributor to these trends. In 1989, voting participation

in local governments was about fifty-nine percent, compared to fifty-six percent in 1993, but

trends were generally flat in the 1980’s and 1990’s as can be seen in Figure 1.3.

While the migration wave in response to the collapse of the Soviet Union was unique

in scope and speed, it wasn’t entirely unprecedented. A similar Soviet migration wave to

Israel had occurred in the early 1970’s. Following the Six Days War in 1967, the Soviet

Union and Israel cut diplomatic relations. In response to domestic repression, Soviet Jewish

dissidents organized a hijacking of a plane headed to Sweden, in what become known as the

Dymshits–Kuznetsov Hijacking Affair. The authorities in the Soviet Union responded to

this incident by harshly cracking down on Jewish dissidents. As international condemnation

16

grew, the Soviet Union relaxed emigration rules and allowed significant numbers of Soviet

Jews to emigrate to Israel.

As can be seen in Figure 1.1, there was a spike in monthly immigration starting in the

early 1970’s, ending around 1975. While the magnitude of this immigration experience

is swamped by the 1990’s experience, it was a significant immigration wave at the time,

relative to the total population of Israel. The earlier immigration episode represented about

a four percent increase in the population.

This initial Soviet immigration wave created settlement patterns that were relevant for

the 1990’s immigration experience. Most of the immigrants coming in the 1970’s could

not speak the official language. Fewer than twenty-five percent of the immigrants had

any previous experience with Hebrew, and actual fluency was significantly lower. The

immigrants also had different culinary and liturgical practices, and the communities they

set up in the 1970’s would create network effects that attracted immigrants in the 1990’s,

who similarly lacked proficiency in Hebrew. These network effects form the basis of the

instrumental approach employed below.

The Soviet Jewish immigrants were distinct from native Israelis in a number of important

ways. In particular, many of the Russians Jews were highly educated with significant work

experience. On average, the typical Russian migrant was more highly educated than the

native Israeli. After the migration experience, Israel would have one of the highest PhD per

capita ratios in the world. Besides labor market integration, there were significant differences

in linguistic and religious characteristics. While Jewish, most of the new immigrants were

significantly less religiously oriented than the natives. In addition, the unfamiliar languages

increased barriers to the integration of the new migrants into society.

While Israel has traditionally had a welcoming immigration policy for foreign Jews,

the size and speed of this particular immigration phase was challenging for both natives

and migrants, who had to learn to integrate culturally, politically, and economically. This

process of integration had profound impacts on many aspects of life for both groups and,

17

given the size of the adjustment required, provides an ideal natural experiment for exploring

the impact of changing diversity on the provision of local public goods.

1.5 Identification

The search for an exogenous source of variation in diversity centers on the migration

wave, in response to the political collapse in the Fall of 1989, from countries that were for-

merly part of the Soviet Union. The collapse of the Soviet Union was swift and unexpected,

and the migration episode of Soviet Jews to Israel that followed was astonishingly large over

a short period of time.

There are two key aspects to the migration phase that make it a useful natural experiment

for studying the impact of diversity on public goods expenditure. First, the push side of

the migration - emigration from the Former Soviet Union - was exogenous to the local

conditions in Israel. Second, settlement patterns of migrants in Israel were influenced by

historical settlement patterns, which can be used to deal with the worry that settlement

responded to unobserved characteristics of the local authorities at the time of settlement.

After the collapse of the Soviet Union, there were few countries, including the U.S., will-

ing to accept Soviet emigrants in large numbers. Israel, with its right of return policy, was

willing to accept unlimited numbers of Soviet Jews, providing fast entry and settlement.

An additional useful feature of the experiment is that because of the size of the migration

episode, areas were differentially impacted such that measured diversity in some local po-

litical jurisdictions increased, while in other cases measured diversity decreased. The same

basic shock altered measures of diversity in different directions, providing an additional

dimension along which to measure the impact of diversity on public goods provision.

Push side exogeneity is clear, as the number of immigrants in Israel prior to the political

collapse of the Soviet Union was small, but increased dramatically in the Fall of 1989 as the

Soviet Union crumbled. In other examples of large mass migrations, one might be concerned

18

about push side exogeneity, since it is usually something about local characteristics in the

landing country that drive the migration wave. In this example, there were significant num-

bers of emigrants trying to leave the Soviet Union, and there were limited landing options.

Put another way, the collapse of the Soviet Union was exogenous from the perspective of

local conditions in Israel.

The more pressing issue for identification is a concern that settlement is not random, but

rather responding to unobservable characteristics at the locality level. These unobservables

could significantly bias the results. I address this issue in two ways. First, using a panel

of localities over time, time-invariant characteristics are accounted for in the analysis. A

panel approach to estimating the impact of diversity on the provision of public goods has

been used in previous studies, but there are reasons to think the approach in Israel’s case

is likely to produce better inference. Since diversity changes so little within a country over

time (in enough localities to ensure statistical validity), panels usually stretch over decades.

The longer the time horizon, however, the less likely that important unobservables are time-

invariant. In Israel’s case, the migration shock represents a significant change to diversity

over a short period of time, making it more likely that important unobservables fall into the

time-invariant category.

While time-invariant factors are accounted for with locality fixed effects, there is a con-

cern that idiosyncratic forces may be driving both settlement patterns and public good

expenditures. For example, the mayor of a locality that had traditionally been hospitable

to immigrants, fearing that the area can’t handle a large influx of migrants, decides to take

steps to minimize immigration flows. This would represent an idiosyncratic change in local-

ity behavior since the area had previously been hospitable to immigrants, and could lead to

bias in the estimation of the impact of diversity on public goods expenditure.

To deal with this kind of concern, I construct instruments for actual changes in diversity

using the migration networks from the 1970’s. The basic idea is to ask what diversity

would have looked like if migrants in 1990 had followed the settlement patterns in 1970.

19

Settlement patterns are highly correlated because of the strong network effects based on

shared linguistic, culinary, and cultural characteristics. The settlement patterns of the

1970’s, however, are unlikely to be correlated with idiosyncratic settlement decisions in 1990,

except to the degree that there are time-invariant factors in both periods. The impact of

these time-invariant forces is accounted for using fixed effects. Predicted changes in diversity

would then be valid instruments for actual changes in diversity. Combining the migration

shock and predicted changes in diversity comprises the strategy to causally identify the

impact of diversity on the provision of public goods at the local level.

Besides the empirical difficulty of untangling causation from correlation, there is also a

confusion over the relevant cleavage in society that exacerbates collective action problems.

Studies have looked at a variety of measures of fragmentation, but these often vary with

the environment. Race is typically emphasized in the United States, while ethno-linguistic

diversity has been emphasized in cross-country studies. Additional studies have empha-

sized socio-economic cleavages through the prism of education and age. All of these studies

emphasize one particular form of diversity without being able to justify why that particu-

lar dimension should be the salient fracture. A benefit of the exploitation of the natural

experiment under consideration in this paper is that different channels of diversity can be

examined simultaneously to see which, if any, truly influence the provision of public goods.

At the most basic level, the introduction of foreigners increased diversity along the

native/foreign cleavage. Studying a large increase in immigrants then captures a first cut at

the meaning of diversity, a nebulous concept in practice. Moving beyond the foreign/native

divide, I focus on two dimensions of social fracture in Israeli society - religious and ethnic

diversity. Education, economic status, population density, and age structure represent other

competing dimensions of diversity in society, each of which was altered by the immigration

experience. Rather than having to guess what measure of fragmentation is truly salient, we

can allow the data to speak for itself.

For all of these reasons - push side exogeneity, large migration flows, pre-existing Soviet

20

migrant network, dimensions of social fragmentation - the Soviet immigration experience

in Israel is an ideal setting in which to identify the impact of changes in diversity (along a

variety of dimensions) on public goods provisions at the local level.

1.6 Data

The data used in this paper draws on a comprehensive source of information about local

public budgets. Local Authorities in Israel are the primary governmental structure for local

issues. These authorities raise revenues from within their geographic areas through taxes

and fees, while also receiving grants from the national government. Furthermore, Local Au-

thorities have access to credit markets and can borrow (or lend) to cover revenue shortfalls.

The budget is composed of two components, the Ordinary Budget and the Extraordinary

Budget. The Ordinary Budget includes expenditures for four main purposes: General Ad-

ministration, Local Services, State Services, and Establishments. The Ordinary budget has

three broad sources: Own Revenue, Transferred Income, and Government Participation.

The Local Authority has significant discretion over how they allocate expenditures across

services, although they receive some restrictions about usage from national government

grants.

The data is drawn from three main sources: 1) Local Authorities in Israel, Financial

Data, 2) Local Authorities in Israel, Physical Data, and 3) Labor Force Survey (LFS). This

data was collected annually for the years 1985 until 1993.

To construct geographically disaggregated demographic variables, the annual LFS was

employed. The LFS collects data on a variety of dimensions, including information on

household demographics, education, occupation, and labor market participation. In trying

to hew to the previous literature, I follow the convention of capturing diversity using a

measure of fractionalization.3 A Herfindahl index is used to measure fracture in a society3While Bossert et al. (2011) derive a theoretically consistent measure of fractionalization, I follow the

standard in the literature by using the Herfindahl Index so that the results here are as comparable toprevious studies as possible. Esteban and Ray (2011) derive a theoretically consistent measure of conflict

21

based on group shares,

Fractionalization = 1− Σis2i (1.6.1)

where si is the share of the group over the total population. Fractionalization variables are

created for the two main measures of diversity, religion and ethnicity. 4

Since Israel has been a country with significant immigration, data collection on immigra-

tion issues is excellent. The LFS collects data on the country of origin as well as country of

origin of the father for each Jewish resident. I construct a measure of ethnic diversity using

information on the country of origin of the father, which is available for immigrant house-

holds as well as domestic households. The LFS organizes countries into larger geographic

groupings, which form the basis of the ethnic fractionalization shares.

Israel is a nation made up primarily of Jews, but religious fractionalization is a source

of conflict within the country. The religious groups identified in the LFS are Jewish, Chris-

tian, Muslim, Druze, and Other. In the data, about eighty-six percent of the population

is identified as Jewish. When fractionalization measures are developed at the local level,

the average measure of fractionalization is 0.09, with a standard deviation of 0.17. There is

significant variation in religious fractionalization, ranging from 0 to 0.73. So while there are

areas that are completely homogeneous, there is significant variation in religious fractional-

ization across localities. For comparative purposes, in Alesina et al. (1999), the fraction of

“white” in the 1990 census in the U.S. was 0.79, and the average fractionalization measure

was 0.27. The standard deviation in racial diversity across municipalities was 0.17, with

a range between 0.01 to 0.73. While the Israel experience has a lower average measure

of fractionalization, and has a significant number of highly homogeneous localities, there

is also significant variation in religious fragmentation within Israel, comparable to racial

fragmentation in the U.S. statistically.

based on indexes of polarization, fractionalization, and income inequality.4The Herfindahl index is interpreted as the probability that two randomly drawn individuals from the

unit of observation (in this case, the local authority) belong to two different social groups.

22

Control variables on education, socio-economic status, and age are also constructed,

as these have been shown to be important in previous studies. Descriptive statistics on

measures of diversity are presented in Table 1.1. The one exceptional measure is the share of

immigrants, which is on average thirty-five percent, and the variation is also quite dramatic.

In the most extreme observation, the immigrant share in a locality is seventy-six percent.

On the public finance side, I focus on expenditure variables that compare to previous

measures of public goods used in the literature so as to replicate the results of the previous

literature on local public goods as closely as possible. The data includes complete informa-

tion on both the ordinary and extraordinary budget, at various levels of disaggregation. For

the primary analysis, I focus on total expenditure as well as education and welfare spending.

On the revenues side, I focus on total revenues, own revenues (and the sources of these own

revenues), and targeted government participation.

The financial data were drawn from audits undertaken on the local authorities on an

annual basis from 1985 to 1993. This nine-year panel allows for a comparison of pre-shock

trends across localities, though for the preferred specification, I take a before and after

approach, using data from 1989 and 1993. Summary statistics are shown in Table 1.2. The

average locality spending per capita over the panel is 1,100 new shekels, which is about $550

(compared to $876 per capita in Alesina et al. (1999)).

In addition to the financial data, I incorporate information from two additional sources,

the Audit of Local Authorities, Physical Data and the Immigrant Absorption Survey, 1972.

The Physical Data audit includes information on population size, area of locality, munici-

pality status, and other locality characteristics. The immigrant survey is a comprehensive

three-year panel on immigrants who arrived from the Soviet Union in 1972 and 1973, pro-

viding information on settlement behavior as well as household demographic information.

Over the time period being considered, there are slightly more than 100 local authorities

in the sample, ranging in size from just around 5,000 inhabitants to over 500,000. This

provides ample variation in which to analyze the impact of local public goods as the size of

23

local authorities changes. Furthermore, these localities represent about eighty-five percent

of the population in 1989, with the rest of the country living in local councils (for which

matched data was not available) or associations with fewer than 5,000 people in each area.

These places tend to be more remote and agricultural-based.

1.7 Social Cleavage and Role of Local Government in

Israel

1.7.1 Social Cleavage

It is necessary to identify social cleavages that are important within the culture to

understand the impact of social diversity on fiscal redistribution. While certain dimensions

of racial or ethnic categories may matter in one society, they may be completely unimportant

in another. For the purposes of making general statements about the impact of diversity,

one needs to identify the aspects of a culture that represent real division, especially those

aspects which the society itself identifies as important. Religion, ethnicity, and class have

constituted the most significant cleavages in Israeli society (Ben Rafael and Sharot (1991)).

Religious fragmentation in Israel has long been a significant source of social disruption,

and rightly represents an area of investigation. The ethnic dimension of social fragmentation

is less well understood outside of Israel, but the society itself has identified ethnic divisions

as an important source of conflict. Jewish ethnic identity is strongly tied to geographic

branching. For centuries prior to the establishment of Israel as a nation-state, Jews migrated

throughout the world and melded into different cultures. As part of this process, traditions

and religious practices evolved in dialogue with foreign cultures. When Israel was founded

as a nation in 1948, the waves of immigrants that followed brought back with them different

traditions, languages, tastes, and liturgical interpretations.

The primary ethnic division is between the Ashkenazim and the Mizrahim. The Ashke-

24

nazim, which is literally translated as German but has come to more broadly encompass

Jews from Western Europe, is considered the dominant ethnic group. The Ashkenazi were

the driving political force during the founding of the state, and controlled the levers of power

starting in 1948. Mizrahim is used to signify Jews who fall outside of this ethnic tradition.

Mizrahim literally translates as Eastern, and it is used to describe Jews who emigrated from

predominantly Muslim cultures. While the early ethnic contestations in Israel were neatly

categorized with this dichotomy, there are additional competing ethnic divisions, including

the Beta Israel (Ethiopian Jews), Soviet and Eastern European Jews, and Jews from North

America. It is common to distinguish a Jew in Israel using a country of origin adjective - a

Syrian Jew or an American Jew, for example. Geographic branching plays an important role

in distinguishing ethnic divisions in Israel, and forms the basis of the ethnic fragmentation

variable constructed here.

The salience of these ethnic divisions can be see throughout the history of Israel. Follow-

ing the founding of the country, immigrants in the 1950s were evenly split between Ashkenazi

and Mizrahim, but the Mizrahim tended to settle in peripheral and less productive areas.

These settlement patterns have been attributed to the fact that the Ashkenazim controlled

the levers of political power from the founding of the country (Smooha (1993)).

These early settlement patterns and frictions manifested themselves in social discord.

In 1959, ethnic tensions spilled over in the form of the Wadi Salib Riots, which pitted the

Mizrahim against the Ashkenazi over issues of economic resources, particularly affordable

housing. The riots eventually resulted in more public spending on public housing and

improved access to public goods for many Mizrahim.

Tensions boiled over again in the early 1970’s. A small but vocal group of Mizrahim

started the Black Panther movement, demanding increased political and economic rights

for the Mizrahim and other disadvantaged groups in Israel. This Black Panther movement

forged ties with the Black Panther movement in the United States, as well as anti-apartheid

organizations in South Africa. It was the first Jewish movement to explicitly compare the

25

plight of the Mizrahim to Arabs in Israel. In response to resulting riots, the government

redirected resources towards impoverished areas, with increased public housing support

again significant.

Political expression of social fragmentation was solidified in the 1977 election, as the

Likud party won power for the first time in Israel’s history. The electoral shift that swept

the center-right party into power has been attributed to the changing voting patterns and

increased political power of the Mizrahim. The election marked a shift in party affiliation,

with explicit ethnic party identification emerging. The ethnic political party became impor-

tant in the 1980’s, particularly the Shas party, which was the first political party in Israel

to explicitly identify with an ethnic group. In the 1988 Knesset elections, about eighty per-

cent of Eastern Jews voted for Likud, Shas, or smaller parties on the right, while a similar

proportion of Ashkenazi voted for Labor or parties on the left. (Smooha (1993))

While outsiders tend to think of religious fragmentation as the only social fracture in

Israel, as the preceding has suggested, ethnic divisions within Israel are strong, with in-

group identification well defined. These ethnic differences are reflected in political voting

behaviors and social unrest. Ethnic diversity is an important fracture in Israeli society, and

the large migration wave from the Soviet Union following its political collapse exacerbated

these ethnic social cleavages.

1.7.2 Local government

Local government is Israel is made up of three different types of administrative units:

Municipalities, Local Councils, and Regional Councils. Municipalities govern larger urban

areas, usually with over 20,000 residents. Local Councils are made up of smaller urban

areas, with around 5,000-20,000 residents. Regional Councils are smaller administrative

units, usually governing agricultural communities and small settlements. The data covers

all Municipalities and most Local Councils over 5,000 residents. The Local Authorities

covered in the sample account for about eighty-five percent of the population in 1989.

26

The legal status of local governments and their relation with government ministries is

based upon the Municipal Corporations Ordinance 5724-1964. Local Government is autho-

rized to operate in six primary areas, including legislation, taxation, financial management,

joint activities with other bodies, and more general powers. The Ministry of Foreign Affairs

describes the scope of local government in Israel as “while not completely independent in

any of these areas, a local authority is able to act on behalf of local interests within each of

them according to the wishes of the elected representatives of the local constituency.”5

Local Authorities serve a variety of functions. Functions include the developing and

planning of local infrastructure, sanitation, parks, education, welfare, culture, and envi-

ronmental protection. The Local Authorities are responsible for primary and secondary

education, although some education is provided by local non-profits with aid from the local

government. In the realm of welfare services, local government targets needy populations,

such as the elderly and disabled.

Local government has three main sources of revenue for the ordinary budget: locally-

generated income, government participation, and loans. While Local Authorities had tra-

ditionally had trouble financing and balancing their budgets, starting with reforms in

1981 greater financial accountability was demanded of the Local Authorities, and locally-

generated income increased substantially. Locally-generated income comes primarily in the

form of local taxes and payments for services.

Government participation takes the form of general grants and targeted grants. General

grants are not tied to any specific expenditures, and can be used however the Local Authority

sees fit. Targeted grants go to specific expenditures, such as welfare. The Ministry of Labor

and Social Welfare sets certain standards, which are funded with ear-marks. However,

the Local Authority is able to authorize higher welfare standards if local interests demand

it, and these higher standards are funded out of locally generated income. The Ministry

of Education sets standards on the curriculum, while the Local Authorities decide on the5The Ministry of Foreign Affairs, http://www.mfa.gov.il/MFA/Government/Branches+of+Government/

Executive/Israeli+Democracy+-+How+does+it+work.htm

27

implementation of education, including the hiring of teachers, administration, and building

of schools.

Finally, local governments can secure loans to finance investment projects, such as wa-

ter treatment, sanitation, education and cultural facilities, as well as general development

projects to support local interests. On occasion, loans are also given to balance budget

shortfalls. After Local Authorities were reformed in 1981, increased powers were delegated

to the Local Authorities, consistent with the view that local government is better able to

meet the needs of the electorate in many social arenas. The Ministry of Foreign Affairs

sums it up: “Studies show that local authorities generally succeed in fulfilling their duties

and in completing projects which they initiate, even though many approvals are involved

in the process. The influence of the local authority is relatively wide in many areas, even

when the central government controls the purse strings or other factors.”

1.8 Empirical Design and Implementation

The majority of work on diversity and provision of public goods depends on cross-section

analysis, which is plagued by omitted variable bias. While studies try to include as many

control variables as seem appropriate, these are limited by data and an awareness of channels

through which provisions of public goods work. The omitted variables problem (OVP) will

be a concern in any cross-sectional analysis on the impact of diversity.

While the OVP is well-known, providing a sufficient solution is far more challenging.

Adding more control variables is unlikely to yield more convincing results. Where possible,

incorporating a time dimension to the analysis might help to alleviate concerns over omitted

variables that are invariant over time. While a reasonable approach, progress has been

hampered by the fact that diversity changes little over time within a country, and even

when there are significant changes to diversity, the forces driving these changes are likely

correlated with changes in local public goods.

28

In the Alesina et al. (1999) study of U.S. municipalities for example, the authors use a

single cross-section in 1990 for the primary analysis. They attempt to incorporate a time-

dimension, but are limited by the data. Using 1970 or 1980 census data would fail to capture

enough significant changes in measures of diversity to make the analysis worthwhile (since

diversity cannot be separately identified from other time-invariant factors). Instead, they

incorporate data from 1960, but only for a limited number of areas and a more restricted

measure of racial diversity (the census tracked fewer racial categories in 1960). Of course,

while looking over a thirty year time horizon solves the problem of racial variation, there

are no doubt significant unobserved forces changing over the same period, which confounds

inferences about the impact of social diversity. These kinds of issues with data and variation

plague most studies that incorporate a time dimension.

To push forward then requires, at a minimum, good data on local government revenues

and expenditures as well as significant changes to diversity over time. Furthermore, since

a fixed effects specification can only control for unobservables that are time-invariant, the

best hope for providing a causal interpretation on the impact of diversity on local public

goods requires a treatment policy. The approach employed in this paper relies on just such

a natural experiment.

The collapse of the Soviet Union and emigration to Israel provides sufficient variation

in diversity over time to make the analysis meaningful. In addition, the exogenous shock

of migration from the collapse of the Soviet Union is uncorrelated with local conditions in

Israel. The last component of identification exploits pre-existing settlement patterns from

an earlier Soviet emigration episode to construct instruments for changes in social diversity.

1.8.1 Estimation

The standard approach in the literature would estimate the following:

public goodi = β0 + β1[Diversityi] +Xiβ2 + �i (1.8.1)

29

where public goodi is a measure of the provision of public goods in region i, Diversityi is a

measure of diversity in region i, and Xi is a set of controls. The omitted variables problem

arises if there are relevant variables not included in the set of controls, leading to biased

estimates of β1. If these omitted variables are time-invariant, then adding a time-dimension

to the analysis can provide unbiased estimation of β1. Suppose the provision of public goods

depends on unobserved time-invariant factors and a common time-trend:

public goodit = β0 + β1[Diversityit] +Xitβ2 + ziγ + δt + �it (1.8.2)

In this case, taking differences yields the following estimating equation:

∆public goodit = β1[∆Diversityit] +∆Xitβ2 +∆δt +∆�it (1.8.3)

which yields unbiased estimates of the impact of diversity on the provision of public goods

under the assumption that unobserved factors are time-invariant. The traditional challenge

with a difference-estimator is that diversity is not separately identifiable from other time-

invariant factors, that is, the rank condition is not satisfied.

As is standard in empirical approaches that use a differences estimator, assuming that

the parallel trends assumption is met, a simple difference estimation with no controls would

be sufficient. However, the migration treatment under study alters not only the composition

of diversity, but also the composition of the community along a number of other dimensions.

I include control variables that have been identified as important by the cross-section lit-

erature, but with the added benefit that these controls are also experiencing significant

variations from the migration shock. After isolating the relevant treatment variables, the

identifying assumption is that the parallel trends assumption is valid in all other dimensions.

For baseline results, the entire nine year panel is used. When a differences-specification

is used, this reduces the panel length to eight years. While estimates of diversity are

unbiased under the rank and exogeneity assumptions, efficient inference requires additional

30

assumptions. There are two polar assumptions one could make about serial correlation in

error terms at this point. The fixed effects model (Equation (1.8.2)) requires that there is

no serial correlation in the error terms within localities for efficient inference. This seems

unlikely to be true. The differences estimator (Equation (1.8.3)) assumes error terms follow

a random walk, the polar opposite assumption about the error terms compared to Equation

(1.8.2) - extreme dependence in the error terms. This is also seems unlikely to be true.

To deal with this concern, and to avoid having to take a stand on the temporal structure

of the error terms, I focus on a “before and after” approach. With T > 2, concerns about

biased standard errors (Bertrand et al. (2004)) hampers inference. When the time period

is two, the fixed effects and first-differences estimators are the same, since there is no serial

correlation in the error terms by construction. For the preferred specification, I focus on

before the shock (1989) and after the shock (1993). Limitations in access to data from

which diversity measures are constructed precludes studying longer changes in diversity at

present.

The main concern with either Equation (1.8.2) or Equation (1.8.3) in the present context

is unobserved idiosyncratic behavior in a locality in response to the immigration wave. The

way to see this is to consider the following possibility: a mayor of a local authority that

was previously hospitable to migrants observes the large influx of immigrants, and decides

to put into place unobserved barriers to migration to that locality. This is a problem for

identification because the locality had previously been hospitable to immigrants, but is no

longer hospitable. If the preference for immigrants was unchanged, it would be wiped out

by the fixed effect and would not impact the estimation of diversity. The possibility for

these unobserved, time-varying locality effects suggests the need to instrument for changes

in diversity. In terms of identification, the concern is that the error terms are correlated with

changes in diversity, even after conditioning on locality fixed effects and control variables.

The settlement patterns from the 1970’s emigration experience form the basis of the

construction of the instruments. The central idea is that settlement patterns of Russians in

31

the 1970’s and 1990 are correlated through migration networks, but the settlement patterns

of the 1970’s are uncorrelated with any locality innovations in 1990. The settlement patterns

may be driven by unobservables that are invariant over time, but these are accounted for

with the locality fixed effects. Using 1970’s settlement patterns rather than the actual 1990’s

settlement patterns, I construct a measure of predicted diversity, and I use the resulting

predicted change in diversity to instrument for the actual change in diversity. I estimate

the following specification:

∆public goodit = β1[∆Diversityit] +∆Xitβ2 +∆δt +∆�it (1.8.4)

and instrument for ∆Diversityit with ∆PredictedDiversityit.

The instruments are constructed using the settlement patterns from the 1970’s, and

predicting the number of Soviet immigrants in each locality instead of the actual number

of Soviet immigrants observed in each locality. Using the predicted number of Soviet im-

migrants in place of actual immigrants, I recalculate religious and ethnic fragmentation for

each locality. For areas that received no Soviet immigrants in either the 1970’s or 1990’s,

there is no difference between predicted and actual changes in diversity.6

One view of the experiment under consideration is a short-run response of diversity on

public goods budgets via the political mechanism. This is a specific answer to untangling how

public goods expenditures (revenues) respond to diversity, and the short-run analysis can be

thought of as studying a more general case with limited mobility. A related, but separate,

question of how diversity impacts local government would consider long-run adjustment

mechanisms, particularly sorting. Tiebout (1956) argues that at the local level, it is labor

mobility and sorting which could drive all of the adjustment in response to changes in

preferences for public expenditures. In a world with zero or small fixed costs of mobility,

the sorting mechanism may be the appropriate one to consider. When transport costs are6Since control variables could have similar worries, and as a robustness check, I also use the settlement

patterns to predict changes in control variables as well. The results are unchanged using either constructionof controls.

32

high, the level of expenditure is more likely to be the margin of adjustment. For the Israeli

case, where linguistic and cultural barriers are significant, the assumptions of high mobility

costs (and hence a focus on political adjustment rather than geographic adjustment) seems

reasonable. In future work, I consider long-run adjustment mechanisms, including internal

sorting and long-run political adjustment.

1.9 Results

1.9.1 Preliminary Results

As a first glance at the data, consider the trends in per capita spending by localities over

the entire sample, from 1985 until 1993. One major worry is that pre-trends in Israel could

be driving the results. There is good reason to think that the pre-trends are not influencing

the results since the collapse of the Soviet Union was exogenous from the perspective of

the local conditions in Israel, but it is possible that settlement patterns were influenced by

diverging pre-trends.

To see that this is not the case, I split the sample into those localities that received

significant migration and those that received limited migration. The first thing to notice in

Figure 1.4 is that “high treatment” localities do in fact look different than “low treatment”

localities.7 This fits with a view that immigrants are not settling randomly. However,

while there are level differences in per capita spending, it is equally important that there

are no apparent differences in pre-trends between areas that received immigrants. This

is not terribly surprising given the nature of the immigration shock, but it is reassuring

nonetheless that the ultimate outcomes of interest are not being driven by trends that drive

both migration patterns and future local government spending.7Note that the blip in the expenditure data in 1991 is not the result of the immigration wave driving down

per capita expenditures, but rather reflects changes in accounting practices. The Local Authorities Auditswitched from a fiscal year to a calendar year, meaning there was only nine months worth of expendituredata in the 1991 audit.

33

Settlement patterns of Soviet immigrants are not random, but rather exhibit certain

patterns in the data. In Figure 1.5, the share of Soviet immigrants is plotted against

the initial immigrant share (all immigrants) of the locality. It is important to note that

there is a slight positive relationship between initial immigrant share and Soviet settlement

patterns, which suggests that immigrants locate in places that already have significant levels

of immigrants. There is also significant variation in Soviet settlement shares compared to

initial immigrant shares, so that immigrant share by itself cannot explain Soviet settlement

patterns.

Besides settling in areas with significant immigrant communities, Soviet immigrants tend

to settle in larger areas. Figure 1.6 plots the share of Soviet settlement by initial population

in 1989. It is apparent from the figure that Soviet immigrants tend to settle in larger areas,

but there is still significant variation in settlement patterns, even after controlling for initial

population size.

While there are obvious patterns in settlement behavior for Soviet immigrants, there

is less evidence of settlement patterns driven by initial fragmentation. In Figure 1.7, the

share of Soviet immigrants is plotted against initial religious fragmentation. For low and

medium levels of religious fragmentation, there is significant variation in the patterns of

Soviet settlement. There does appear to be some bias in the settlement patterns, as the

most religiously fragmented areas do not receive much Soviet immigration. Figure 1.8 tells

a similar story for Soviet settlement and initial ethnic fragmentation. There is a positive

relationship between initial ethnic fragmentation and Soviet settlement patterns, but there

is also significant variation in settlement patterns such that initial ethnic fragmentation does

not explain settlement patterns.

Having taken a glimpse at the raw data, let us turn to the baseline results using the entire

panel. Focusing first on total spending per capita and religious fragmentation, columns (1)

and (4) of Table 1.3 replicate the standard approach taken in the previous literature using

pooled OLS. The coefficient on religious fragmentation is large, negative, and statistically

34

significant. While the magnitude of the coefficient drops with the inclusion of controls, the

general observation that religious fragmentation is negatively correlated with local govern-

ment spending is confirmed. These results suggest that Israel is similar to other countries

that have been studied in that a robust negative relationship is observed in the cross-section

between social diversity and the provision of public goods.

Once we move to columns (2) and (5) of Table 1.3, however, we see that this result is

not robust to changes in religious diversity over time. These columns include locality fixed

effects, and once these time-invariant effects are included, the impact of religious diversity

falls to 0. This result would not be surprising in other panel contexts if religious diversity

was not changing much over time, because the zero coefficient could be rationalized as not

independently identifiable. However, that observation is not valid in the Israeli context,

which experienced significant changes in religious diversity over this time period. Finally,

columns (3) and (6) interact religious diversity with a shock-indicator to separate out the

pre and post-shock experiences. Column (6) suggests there was a different response to

diversity after the Soviet migration wave, as the interaction term is positive and statistically

significant. This result is explored further in the pre and post-shock specification.

Similar patterns emerge when education and welfare spending are considered (Tables 1.4

and 1.5 respectively). In both cases, columns (1) and (4) find the coefficients on religious

diversity are large, negative, and statistically significant, consistent with previous findings.

With only this information, one would conclude that religious diversity has a negative impact

on the provision of public expenditures. Once again, however, columns (2) and (5) suggest

this statement is too bold. For both welfare and education, the fixed effects specification

is statistically insignificant (and positive). Furthermore, for the welfare regressions, column

(6) reports a positive and statistically significant coefficient on the interaction term between

religious diversity and a post-shock indicator, suggesting once again that the local response

to changes in diversity may behave differently before and after the shock.8

8Similar results hold for other measures of public goods (culture, sanitation, public property, and water)and are available upon request.

35

Having considered religious fragmentation, next consider the impact of ethnic diversity

on the provision of public goods. Looking first at the size of local government, total per

capita spending appears to respond negatively to ethnic fragmentation in the pooled OLS

specification (Columns (1) and (4) of Table 1.6). As with religious fragmentation, one would

conclude based on this analysis alone that ethnic diversity is an important determinant of

the provision of public goods. Following the same basic patterns as before, columns (2) and

(5) tell a very different story. The coefficient on ethnic diversity is positive, but statistically

insignificant. Column (3) suggests that there is a differential response before and after

the shock, although the inclusion of controls in column (6) finds no statistical difference in

behavior.

Looking next at educational spending per capita and welfare spending per capita, a

very similar story emerges (Tables 1.7 and 1.8, respectively). Once again, the pooled cross-

sectional evidence would point to a large negative impact of diversity on the provision of

public goods. This is consistent with the evidence presented for religious fragmentation.

The inclusion of controls in column (4) reduces the magnitude slightly, but the negative

and statistically significant coefficient remains. When locality fixed effects are included, the

impact of ethnic fragmentation on education disappears (Table 1.7). The impact of ethnic

fragmentation on welfare per capita spending is positive, but statistically insignificant (Table

1.8). The interaction of ethnic fragmentation and a shock indicator in columns (3) and (6)

suggest that the impact of ethnic diversity before and after the shock are different, which

will be explored further in the next section.

The two main points to take away from these baseline results are that first, Israel looks

similar to other countries in terms of the impact of social diversity and the provision of

public goods when the standard approach is utilized. The standard approach attempts to

deal with OVP by including a battery of control variables. The inclusion of reasonable

controls does not change the basic cross-sectional result that religious and ethnic diversity

are important social fractures in Israeli society. The second main result is that the use of

36

locality fixed effects during a time of mass migration reduces the importance of ethnic and

religious diversity. The migration shock was exogenous to local conditions, and provides

sufficient variation in social diversity to separately identify diversity from time-invariant

factors. Under these conditions, there is no evidence that supports the view that ethnic or

religious diversity negatively impacts the provision of local public goods. To explore this

issue further, I turn to a pre- and post-shock analysis and attempt to address concerns over

nonrandom settlement patterns of migrants.

1.9.2 Pre and Post-Shock Approach

While the panel evidence is suggestive, there are potentially confounding issues for infer-

ence. The primary worry is that there is persistence over time in the measures of diversity

(especially pre-shock, where there was limited variation within a locality), and more gener-

ally, there is potentially serial correlation in the error terms. In the panel approach in the

previous section, a robust variance matrix was employed, which is valid in the presence of

any type of serial correlation or heteroskedasticity so long as the number of years in the

panel, T, is small relative to the observational units, N (Wooldridge (2002)).

An alternative approach, which has been suggested by Bertrand et al. (2004), is to

collapse the panel to two years, before and after the event under study. This has the benefit

of eliminating serial correlation concerns in the data, without having to take a stand on

estimators. Since there is a clear shock experience in Israel, I focus on differences between

1989 and 1993. This has an additional benefit in the current context since both years were

election years for Local Authorities. This reduces the possibility of picking up political

economy effects unrelated to diversity (e.g., spending run-ups in election years, spending

declines after elections), and provides a cleaner test of the political channel mechanism

underlying the theory. Since there are municipal elections in both years, local politicians

should be responding to the new social conditions, and the observed expenditure changes

should reflect these underlying political needs. In other countries, elections so close in

37

time to the migrant experience may not pick up voting patterns of immigrants (instead,

capturing the political exclusion of migrants), but these concerns are mitigated in Israel

given the unique characteristics of the state (since new immigrants are immediately granted

political rights and access).

In addition to collapsing the panel to two years, the issue of migrant settlement needs

to be addressed. While the flow of migrants is exogenous from the perspective of Israel,

the actual settlement patterns within Israel are not exogenous. Some of the forces driving

landing patterns can be accounted for, but there is still the possibility that landing patterns

are being driven by unobserved forces also correlated with public expenditure decisions,

which would bias the estimation. The IV strategy discussed previously is now implemented.

The impact on total expenditures per capita can be seen in the first panel of Table 1.9.

Column (1) repeats the pooled cross-sectional estimation, but now only for years 1989 and

1993. Once again, the coefficient on religious fragmentation is negative and statistically

significant. The inclusion of controls reduces the impact by about half (not reported), but

it is still negative and statistically significant. Once again, however, the conclusion that

diversity negatively impacts the provision of public goods is premature. Turning to column

(2), with the inclusion of locality fixed effects, the coefficient drops by about half relative

to columns (1), and the coefficient is no longer statistically different from zero. Column

(3) instruments for diversity using predicted diversity. This lowers the estimated coefficient

still further, suggesting there was some sorting based on unobservables, but it appears not

to be a major contributor to the estimated effect.

The exercise is repeated for educational spending (middle panel, Table 1.9) and welfare

spending (last panel, Table 1.9), both on a per capita basis. For educational spending,

column (4) show a strong negative coefficient on religious diversity, which is robust to the

inclusions of controls. The inclusion of locality fixed effects reduces the magnitude of the

coefficient and the statistical significance. The IV estimates in column (6) find the impact

of religious diversity to be essentially zero.

38

Welfare spending per capita (last panel, Table 1.9) has slightly different patterns. The

pooled cross-section is again negative, but the inclusion of controls (column (7)) reduces the

significance of religious diversity (although the point estimate is still very large). However,

the inclusion of fixed effects results in a large and statistically significant negative effect. The

IV specification confirms that the point estimate is quite large and statistically significant.

There is some evidence then that cross-sectional evidence can’t be considered “suggestive” in

general, but welfare spending per capita in particular does respond negatively to increased

religious fragmentation.

Ethnic fragmentation analysis reveals similar patterns. Focusing first on total spending

per capita (Table 1.10), the standard approach would once again conclude that ethnic

fragmentation negatively impacts the provision of public goods. The inclusion of controls

in column (1) reduces the magnitude of the coefficient slightly, but the effect is still large,

negative, and statistically significant. The inclusion of locality fixed effects reduces the size

of the effect, and in the case of column (2), eliminates the significance. The estimated

impact using the IV specification in column (3) is essentially zero. There is some evidence

of sorting on unobservable characteristics since the estimated coefficient falls dramatically

between columns (2) and (3). Based on the most preferred specification, the impact of

ethnic diversity on total expenditures per capita is zero.

The evidence on education spending and ethnic fragmentation is less clear-cut. As can

be seen in the second panel in Table 1.10, the coefficient on ethnic fragmentation is negative

and statistically significant without controls (not reported), and negative but not significant

in column (4), once controls are included. The inclusion of fixed effects in column (5) and

then instrumenting in column (6) does not change the results. The estimated coefficient is

negative, but insignificant as the standard errors are large.

The third panel of Table 1.10 shows the results for welfare spending per capita. As with

previous public goods, the coefficient on ethnic fragmentation is negative and statistically

significant, and is also robust to the addition of controls. Including fixed effects in columns

39

(8) lowers the estimated coefficient on ethnic fragmentation to zero. When ethnic fragmen-

tation is instrumented for, the coefficient is positive but insignificant. The standard errors

are quite large, but there is no evidence that increased ethnic diversity results in lower

welfare spending per capita.

Taking stock of the entire body of evidence, when looking across localities in a moment

in time, it appears that religious and ethnic fragmentation matters significantly for public

goods. Qualitatively similar negative relationships have been found in many different coun-

tries, and across cities within countries. The Israeli experience is similar to these earlier

studies. What distinguishes the Israeli experience is that we need not just rely on controls to

deal with the omitted variables problem. Looking before and after the collapse of the Soviet

Union and the resulting waves of immigrants to Israel, I find only very limited evidence

that religious fragmentation leads to lower provision of public goods, and no evidence that

ethnic fragmentation leads to lower expenditures on public goods. With the exception of

welfare spending and religious fragmentation, the best estimates of the impact of ethnic and

religious diversity on public expenditures is zero. Welfare spending does seem to respond

negatively to increased religious diversity, and is larger than the cross-sectional estimation.

Overall, what seems to matter most for explaining the provision of public goods is institu-

tional factors (captured by time-invariant fixed effects), and an older, more simple story -

the number of people. Locality population is consistently significant in the IV specifications.

This suggests that the number, rather than the type, of people is what matters for collective

action and the provision of public goods.

1.9.3 Revenues

The expenditure analysis might be misleading if the financing is coming from national

sources rather than local sources. While the decisions to spend locally reflect local prefer-

ences, some government financing can be specifically targeted to meet national needs, and

hence the inference on local expenditures may be conflated with national aims that are

40

independent or even contrary to local interests.

To study this further, I turn next to the revenue side of the budget. Revenue in the

ordinary budget is made up of three broad components - own revenue, government par-

ticipation, and transferred income. As transferred income makes up a small component, I

focus on government participation and local raised revenues. There is significant variation

across localities in the contribution of locally raised revenue as a share of total revenues,

and there is significant variation in the sources of local revenue from taxes and other sources

(including fees for service, licenses, etc.).

Using the same analytic framework as with expenditures, Table 1.11 reports the impact

of religious fragmentation on total revenues, own revenues, and targeted government grants.

In the left panel of the table, the cross-sectional evidence suggests that religious fragmen-

tation has a negative impact of total revenues per capita, with a similar magnitude as was

found with expenditures per capita (column 4). The inclusion of locality fixed effects and

instrumenting for changes in diversity lowers the point estimate by about half, which is not

statistically significant (but also not statistically different from the pooled estimate). These

patterns are consistent with total expenditures analyzed above.

When total revenues is broken up into own revenues and targeted grants, some interesting

patterns emerge. The pooled estimates for the impact of religious fragmentation on own

revenues is negative and statistically significant, and the magnitude nearly doubles when

fixed effects and instruments are used. Targeted grants have a negative point estimate

in the pooled analysis, but become positive and statistically significant once instrumented

for. This evidence suggests that in the case of religious fragmentation, the total impact

on revenues per capita is masking two opposing forces. At the local level, own revenue

responds negatively to increased religious fragmentation, but this is counter-balanced by

targeted grants that respond positively to increased religious diversity.

The role of targeted grants raises an issue as to whether this is a mechanism that defuses

or exacerbates social fragmentation. On the one hand, targeted governmental transfers could

41

reflect a governmental response to minimize the negative local impact on own revenues,

thereby neutralizing the impact of local revenue decisions. On the other hand, targeted

transfers could exacerbate the impact of social diversity by targeting specific populations

at the expense of other social groups. The appearance of a trade-off at the aggregate level

could be masking significant changes at more disaggregated levels.

Digging deeper into the mechanisms through which own revenue is raised, I find that

increased religious fragmentation has different impacts on sources of revenue. Table 1.13

breaks up own revenues per capita into revenues raised through taxes and those raised

through other mechanisms such as licenses and fees for services provided. Analysis for total

own revenue was negative in the cross-section and roughly twice as large using instruments

and fixed effects. At disaggregated levels, the pooled estimate for local taxes and local fees

are similar in magnitude to each other, and statistically significant.

When looking at changes within localities over time, however, religious fragmentation

has a strong negative effect on other sources of income per capita. It is this channel of

raising revenue that seems to be most affected from the changes in religious fragmentation,

rather than local tax revenues. The point estimate for local tax revenues is slightly smaller

than the pooled estimate, although not statistically distinguishable. The suggestion here is

that in response to an increase in religious diversity, local taxes don’t change directly, but

that possibly less observable ways of de-funding local government are utilized instead.

Turning next to ethnic fragmentation, Table 1.12 shows that the pooled estimate for

the impact of total revenues is negative and statistically significant, but once locality fixed

effects are accounted for and changes in diversity are instrumented, the impact of ethnic

fragmentation is essentially zero. The estimated magnitudes and patterns are very similar

to those found in the expenditures analysis.

Breaking this down further, for locally raised revenue, the pooled estimated of ethnic

fragmentation is positive and statistically significant. The inclusion of locality fixed effects

reduces the estimated effect, but once the change in diversity is instrumented for the es-

42

timated coefficient is similar in magnitude (though not statistically distinguishable from

the pooled estimate). Targeted grants respond negatively to increased ethnic fragmenta-

tion, although the estimate is not significant in the cross-section once controls are included.

The point estimate using locality fixed effects and instruments is positive, and statistically

distinct from the pooled estimate.

The disaggregated patterns suggest that ethnic fragmentation has a positive impact on

own revenues, but this effect is hidden at the aggregate level because of offsetting government

participation. For ethnic fragmentation, targeted government grants also respond positively

to ethnic fragmentation, contrary to the evidence in the pooled cross section. It appears that

general government grants contribute to the zero estimated effect of ethnic fragmentation

on total revenues.

Exploring the impact of ethnic fragmentation on own revenues further, I look at local

tax revenues and other forms of local revenue separately in Table 1.14. The point estimate

for the impact of ethnic fragmentation on Other Income is essentially zero in column (3).

Compare this to the estimated impact of ethnic fragmentation on local tax revenues, which

is positive and significant (and of similar magnitude) in both the pooled and instrumented

estimates.

Contrary to religious fragmentation, local taxes are the mechanism of adjustment in

response to changes in ethnic fragmentation, and there is a positive impact on increased

ethnic fragmentation leading to higher local tax revenues per capita. Other sources of local

income don’t appear to respond to increases in ethnic fragmentation, although they respond

dramatically to changes in religious fragmentation.

The results presented for local government revenues are similar at the aggregate level to

local government expenditures studied in the previous section. These aggregate similarities

give way to interesting differences in the source of income. For religious fragmentation, the

lack of an effect at the aggregate level is masking a strong negative response of local revenue

to an increase in religious fragmentation, with an increase in targeted government revenue

43

(which may or may not be exacerbating social fragmentation depending on the role of

targeted grants). The channel of adjustment for own revenue is coming from fees and licenses

rather than taxes. For ethnic fragmentation, similar aggregate revenue patterns obscure

the fact that increased ethnic fragmentation actually leads to an increase in local revenue

collection. Government grants offset this positive local effect on own revenue. Contrary

to religious fragmentation, it is local tax revenue that plays an integral role in explaining

the response of local revenues to changes in ethnic fragmentation. The results suggest that

social fragmentation does impact local government behavior, but the mechanisms are far

more nuanced than the standard theory suggests.

1.10 Conclusions

There has been extensive discussion of the role that diversity plays in facilitating collec-

tive action, guided by the empirical observation that diversity and public goods provision

are negatively correlated. In this paper, I argue that there are good reasons to be wary of

a causal interpretation of higher diversity leading to lower public goods provision. While

observed negative correlations are robust in the sense that they have been replicated in a

variety of settings, the strategy to identify causality has been hampered by poor data, small

samples, and limited experimental validity. This paper attempts to address this pressing

issue by exploring a natural experiment and utilizing unique instrumental variables.

Consistent with previous literature, I find a large, negative and statistically significant

effect of religious and ethnic fragmentation on public goods expenditure. Based on this

observation, one would conclude, as have previous studies, that social diversity hampers the

fiscal redistributive policy at the local level. However, in the current context, progress can

be made by exploiting a large migration wave and the resulting changes in social diversity.

The collapse of the Soviet Union in 1989 ushered in a period of significant migration to

Israel, with Soviet Jews emigrating by the hundreds of thousands. Over a two-year pe-

44

riod, the total population of Israel increased by close to seven percent. I utilize the size,

swiftness, and unexpected nature of this migration shock to study the response of public

goods expenditures at the level of the local government. I find limited evidence that social

fragmentation negatively impacts public goods expenditure. For religious diversity, neither

total expenditure nor education expenditure responds negatively to increased fragmenta-

tion. I do find evidence that religious fragmentation negatively impacts welfare spending.

For ethnic fragmentation, I find no evidence to support the view that diversity negatively

impacts public goods expenditure. Instrumenting for changes in diversity using predicted

changes in diversity based on 1970’s settlement patterns confirms these results.

When I consider revenues of local government, interesting patterns emerge. Total rev-

enues have similar patterns as total expenditures, with strong cross-sectional negative effects

that disappear when estimated using changes over time and instrumenting for changes in

diversity. When I break revenues up into locally raised revenues and targeted government

transfers, I find that the estimated aggregate impact is made up of conflicting disaggregated

effects. For religious fragmentation, locally raised revenue responds negatively while govern-

ment transfers respond positively. These two forces cancel each other out at the aggregate

level. When sources of locally raised revenues are analyzed further, it is sources other than

taxes that are most strongly negatively affected by increased religious diversity. For ethnic

fragmentation, locally raised revenues respond positively to ethnic fragmentation, and this

positive effect is driven by the positive impact of ethnic fragmentation on local tax revenue.

I find no effect of ethnic fragmentation on other sources of locally raised revenues.

Besides the migration shock, Israel is an excellent setting for studying the impact of

diversity on public goods because its history as a nation of immigrants has led it to collect

extensive information on migration and the countries of origin of its citizens. This infor-

mation is useful for constructing disaggregated geographic measures of ethnic and religious

fragmentation. Under its right of return policy, Israel immediately grants full voting rights

to immigrants, giving them a political channel through which to operate - an important

45

mechanism that may be lacking in other migration contexts. Finally, previous migration

episodes generated migrant networks that help mitigate concerns that settlement patterns

were driven by unobserved factors which also drove public expenditure decisions.

Differences between the cross-sectional and instrumented approaches have a number of

possible explanations. The first possibility is that the cross-sectional approach is not the

right way to measure the impact of diversity on public goods provision since there are

local characteristics that matter for both social fragmentation of a locality and the size of

the local government. I find some evidence that suggests these local time-invariant factors

are important for explaining spatial variation in local government spending and diversity.

Furthermore, the fact that immigrants had a political outlet helps to explain the inability

to find significant effects leading from diversity to the provision of public goods. As has

been suggested by Easterly (2001) in cross-country analysis, access to good institutions - in

this case comprehensive voting rights for immigrants in Israel - could help to mitigate any

negative effects from increased diversity. Unlike other countries, eligible immigrants have

minimal barriers to political participation, and this political access may help to explain

observed spending behavior.

The results on the revenues side of the ledger suggest that local preferences may be

offset by national government behavior. For revenues, the failure to find aggregate effects

for total revenue masks the reality that local and national sources of revenue offset each

other. For religious fragmentation, targeted grants increased with religious diversity while

local revenues declined. While these effects statistically offset each other at the aggregate

level, it may be the case that targeted grants actually exacerbate social tension if the funding

is targeted towards particular groups at the expense of others. Furthermore, locally sourced

revenues respond differently to types of social fragmentation. Religious fragmentation has

a significant impact on sources of revenue other than taxes including licenses and fees for

service, while increased ethnic fragmentation works through local taxes. This suggests that

the revenue generating mechanism, not just the level of revenue collected, may respond to

46

social fragmentation.

In future work, long-run adjustments will be studied more carefully. Here, I focused on

the short-run, where mobility costs are high and political channels are the mechanism by

which adjustment takes place. In the longer run, internal sorting may play an important

role for adjustment if households move to areas that better reflect their political preferences

(Tiebout Sorting). Furthermore, fiscal adjustment in the long-run may differ systematically

from the short-run fiscal adjustment studied here. Extending the dataset to include long

differences and collecting data on internal migration patterns will help to address both of

these issues.

The results here suggest that diversity may not be the hindrance to collective action

as is often assumed. In particular, for public expenditure, the evidence suggests an older

and simpler story of collective action failure in which collective action is harder to maintain

as the number of individuals increases, while the type of individuals plays an insignificant

role. For public revenue, diversity interacts with both national and local revenue gener-

ating decisions, which have offsetting effects. In addition, the mechanism used to raise

local revenue, not just the level of revenue, responds to social diversity. While the results

focus on the short-run adjustment in public goods expenditures and revenues, there is a

longer-run adjustment of internal sorting that may help to reconcile the observed negative

correlation with the results presented here. Future work should consider these longer term

adjustments in experimentally valid environments like the one discussed here. Finally, to

evaluate the robustness of the results, the methodology employed here could be applied to

similar migration experiences.

47

1.11 Figures and Tables

0

10000

20000

30000

40000

1/31/70

1/31/71

1/31/72

1/31/73

1/31/74

1/31/75

1/31/76

1/31/77

1/31/78

1/31/79

1/31/80

1/31/81

1/31/82

1/31/83

1/31/84

1/31/85

1/31/86

1/31/87

1/31/88

1/31/89

1/31/90

1/31/91

1/31/92

1/31/93

1/31/94

1/31/95

1/31/96

1/31/97

1/31/98

1/31/99

1/31/00

1/31/01

1/31/02

1/31/03

1/31/04

1/31/05

1/31/06

1/31/07

1/31/08

1/31/09

1/31/10

Figure 1.1: Israeli Immigration by Month, 1970-2010

Figure 1.2: Israeli Population Growth, 1970-2010

48

Figure 1.3: Voter Participation, Municipal and Knesset Elections, 1949-2003Note: Voter participation in local and national elections for selected elections. After 1973, when local andnational elections were decoupled, the nearest national election is used as a point of comparison. Nationalelection year in brackets.

Figure 1.4: Total per capita spending in High and Low Migration Intensity LocalitiesNote: Localities were split into two samples based on the number of immigrants received from 1990 until1993. Total spending per capita was then calculated for these low and high treatment localities for eachyear in the sample, 1985-1993.

49

Figure 1.5: Soviet Settlement by Initial Immigrant Share

Figure 1.6: Soviet Settlement by Initial Population

50

Figure 1.7: Soviet Settlement by Initial Religious Fragmentation

Figure 1.8: Soviet Settlement by Initial Ethnic Fragmentation

51

Mean Std Dev Min Max NImmigrant Share 0.353 0.250 0.000 0.759 930

Religion Fragmentation 0.098 0.173 0.000 0.730 930Ethnic Fragmentation 0.693 0.117 0.146 0.848 594

Share Under 17 0.381 0.080 0.000 0.656 930Share Over 65 0.105 0.063 0.000 0.368 930

Share Post-Secondary Education 0.175 0.115 0.000 0.679 930Skilled Occupation Ratio 0.310 0.134 0.000 0.823 930

Table 1.1: Summary Statistics, Demographic

Expenditures N Mean Std Dev Min MaxTotal per capita 953 1103.201 728.281 71.874 4787.941Education Share 953 0.308 0.093 0.110 0.653Welfare Share 952 0.090 0.050 0.000 0.244Culture Share 949 0.061 0.031 0.002 0.182

Sanitation Share 953 0.090 0.033 0.018 0.228Public Property Share 920 0.006 0.012 0.000 0.123

Water Share 951 0.081 0.033 0.000 0.229

Table 1.2: Summary Statistics, Expenditures

52

(1) (2) (3) (4) (5) (6)

Religious Fragmentation -0.848*** -0.0338 -0.216* -0.293*** -0.0213 -0.175(0.0892) (0.125) (0.128) (0.0840) (0.122) (0.126)

Rel. Frag X Post Shock 0.303*** 0.258***(0.0609) (0.0599)

Under 17 Share -1.147*** -0.238 -0.185(0.315) (0.146) (0.145)

Over 65 Share 2.012*** -0.406* -0.320(0.420) (0.224) (0.227)

Post-Secondary Share -0.708*** -0.514*** -0.490***(0.220) (0.148) (0.151)

Skilled Industry Share -1.199*** -0.0511 -0.0522(0.192) (0.108) (0.106)

ln (Population) -0.0314** -0.556*** -0.532***(0.0159) (0.125) (0.128)

Observations 929 929 929 929 929 929R-squared 0.684 0.978 0.979 0.787 0.980 0.980

Year Dummies Yes Yes Yes Yes Yes YesLocality FE No Yes Yes No Yes Yes

Table 1.3: Ln (Total Spending Per Capita), Religious FragmentationNote: Robust Standard Errors in Brackets. * indicates significant at 10%; ** significant at 5%; ***significant at 1%. Regressions were weighted by Labor Force Survey (LFS) observations. The interactionterm includes an indicator for all shock years, 1990-1993.

53

(1) (2) (3) (4) (5) (6)

Religious Fragmentation -1.068*** 0.0175 -0.0344 -0.784*** 0.0222 -0.00738(0.101) (0.107) (0.108) (0.0989) (0.106) (0.106)

Rel. Frag X Post Shock 0.0865 0.0495(0.0693) (0.0695)

Under 17 Share -1.280*** -0.120 -0.110(0.321) (0.162) (0.163)

Over 65 Share 1.115** -0.189 -0.173(0.518) (0.240) (0.245)

Post-Secondary Share -0.714*** -0.198 -0.194(0.229) (0.191) (0.192)

Skilled Industry Share -0.708*** -0.0762 -0.0764(0.185) (0.105) (0.105)

ln (Population) -0.0649*** -0.588*** -0.583***(0.0192) (0.133) (0.134)

Observations 929 929 929 929 929 929R-squared 0.739 0.976 0.977 0.786 0.978 0.978

Year Dummies Yes Yes Yes Yes Yes YesLocality FE No Yes Yes No Yes Yes

Table 1.4: Ln (Education Spending Per Capita), Religious FragmentationNote: Robust Standard Errors in Brackets. * indicates significant at 10%; ** significant at 5%; ***significant at 1%. Regressions were weighted by Labor Force Survey (LFS) observations. The interactionterm includes an indicator for all shock years, 1990-1993.

54

(1) (2) (3) (4) (5) (6)

Religious Fragmentation -1.406*** 0.197 -0.284 -0.682*** 0.244 -0.196(0.197) (0.330) (0.332) (0.200) (0.328) (0.336)

Rel. Frag X Post Shock 0.750*** 0.691***(0.179) (0.185)

Under 17 Share -1.961*** -0.301 -0.160(0.663) (0.427) (0.426)

Over 65 Share 2.163*** -0.484 -0.261(0.565) (0.480) (0.485)

Post-Secondary Share -0.284 -1.143*** -1.083***(0.352) (0.302) (0.309)

Skilled Industry Share -0.884* 0.0329 0.0217(0.477) (0.349) (0.342)

ln (Population) 0.0227 -0.665*** -0.600***(0.0179) (0.178) (0.189)

Observations 916 916 916 916 916 916R-squared 0.445 0.923 0.925 0.543 0.925 0.927

Year Dummies Yes Yes Yes Yes Yes YesLocality FE No Yes Yes No Yes Yes

Table 1.5: Ln (Welfare Spending Per Capita), Religious FragmentationNote: Robust Standard Errors in Brackets. * indicates significant at 10%; ** significant at 5%; ***significant at 1%. Regressions were weighted by Labor Force Survey (LFS) observations. The interactionterm includes an indicator for all shock years, 1990-1993.

55

(1) (2) (3) (4) (5) (6)

Ethnic Fragmentation -1.779*** 0.310 0.343 -1.382*** 0.232 0.285(0.260) (0.379) (0.387) (0.334) (0.413) (0.413)

Eth. Frag X Post Shock -0.243 -0.466*(0.246) (0.241)

Under 17 Share 1.794*** 0.426 0.448*(0.416) (0.261) (0.264)

Over 65 Share 2.634*** 0.0280 -0.154(0.478) (0.368) (0.362)

Post-Secondary Share -0.421 -0.506** -0.595**(0.264) (0.236) (0.256)

Skilled Industry Share 0.243 0.0289 -0.0572(0.359) (0.207) (0.206)

Immigrant Share 0.501* 0.00320 0.139(0.268) (0.174) (0.198)

ln (Population) 0.0112 -0.931*** -0.991***(0.0183) (0.139) (0.141)

Observations 593 593 593 593 593 593R-squared 0.796 0.970 0.970 0.821 0.974 0.975

Year Dummies Yes Yes Yes Yes Yes YesLocality FE No Yes Yes No Yes Yes

Table 1.6: Ln (Total Spending Per Capita), Ethnic FragmentationNote: Robust Standard Errors in Brackets. * indicates significant at 10%; ** significant at 5%; ***significant at 1%. Regressions were weighted by Labor Force Survey (LFS) observations. The interactionterm includes an indicator for all shock years, 1990-1993. The sample includes 66 Local Authorities thatare predominantly Jewish (non-Jewish population less than 5%).

56

(1) (2) (3) (4) (5) (6)

Ethnic Fragmentation -1.550*** -0.0316 -0.132 -1.333*** -0.0450 -0.121(0.263) (0.161) (0.148) (0.191) (0.145) (0.143)

Eth. Frag X Post Shock 0.760*** 0.657***(0.135) (0.129)

Under 17 Share -0.151 0.174 0.142(0.436) (0.215) (0.210)

Over 65 Share 1.421** -0.273 -0.0161(0.595) (0.313) (0.294)

Post-Secondary Share -1.102*** -0.151 -0.0260(0.280) (0.227) (0.209)

Skilled Industry Share -0.997*** -0.153 -0.0313(0.343) (0.167) (0.165)

Immigrant Share -0.554** 0.302 0.111(0.268) (0.253) (0.201)

ln (Population) -0.0253 -0.610*** -0.525***(0.0216) (0.143) (0.149)

Observations 593 593 593 593 593 593R-squared 0.777 0.979 0.981 0.808 0.981 0.982

Year Dummies Yes Yes Yes Yes Yes YesLocality FE No Yes Yes No Yes Yes

Table 1.7: Ln (Education Spending Per Capita), Ethnic FragmentationNote: Robust Standard Errors in Brackets. * indicates significant at 10%; ** significant at 5%; ***significant at 1%. Regressions were weighted by Labor Force Survey (LFS) observations. The interactionterm includes an indicator for all shock years, 1990-1993. The sample includes 66 Local Authorities thatare predominantly Jewish (non-Jewish population less than 5%).

57

(1) (2) (3) (4) (5) (6)

Ethnic Fragmentation -1.779*** 0.310 0.343 -1.382*** 0.232 0.285(0.260) (0.379) (0.387) (0.334) (0.413) (0.413)

Eth. Frag X Post Shock -0.243 -0.466*(0.246) (0.241)

Under 17 Share 1.794*** 0.426 0.448*(0.416) (0.261) (0.264)

Over 65 Share 2.634*** 0.0280 -0.154(0.478) (0.368) (0.362)

Post-Secondary Share -0.421 -0.506** -0.595**(0.264) (0.236) (0.256)

Skilled Industry Share 0.243 0.0289 -0.0572(0.359) (0.207) (0.206)

Immigrant Share 0.501* 0.00320 0.139(0.268) (0.174) (0.198)

ln (Population) 0.0112 -0.931*** -0.991***(0.0183) (0.139) (0.141)

Observations 593 593 593 593 593 593R-squared 0.796 0.970 0.970 0.821 0.974 0.975

Year Dummies Yes Yes Yes Yes Yes YesLocality FE No Yes Yes No Yes Yes

Table 1.8: Ln (Welfare Spending Per Capita), Ethnic FragmentationNote: Robust Standard Errors in Brackets. * indicates significant at 10%; ** significant at 5%; ***significant at 1%. Regressions were weighted by Labor Force Survey (LFS) observations. The interactionterm includes an indicator for all shock years, 1990-1993. The sample includes 66 Local Authorities thatare predominantly Jewish (non-Jewish population less than 5%).

58

Tabl

e1.

9:E

xpen

ditu

res

per

capi

ta(R

elig

ious

Frag

men

tatio

n)

ln(P

erC

apit

aTo

talS

pend

ing)

ln(P

erC

apit

aE

duca

tion

Spen

ding

)ln

(Per

Cap

ita

Wel

fare

Spen

ding

)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Rel

igio

usFr

agm

enta

tion

-0.4

10**

-0.2

42-0

.225

-0.5

48**

*-0

.123

-0.0

818

-0.9

10-1

.807

*-1

.826

**(0

.170

)(0

.255

)(0

.229

)(0

.184

)(0

.205

)(0

.237

)(0

.566

)(1

.005

)(0

.917

)U

nder

17Sh

are

-1.9

75**

*-0

.114

-0.1

80-1

.305

**-0

.197

-0.1

41-4

.391

***

-0.0

164

-0.3

81(0

.481

)(0

.376

)(0

.326

)(0

.554

)(0

.337

)(0

.337

)(1

.264

)(1

.747

)(1

.364

)O

ver

65Sh

are

0.22

5-0

.790

-0.6

52-0

.604

-0.4

36-0

.308

1.00

2-2

.532

-2.3

91(0

.565

)(0

.481

)(0

.499

)(0

.590

)(0

.588

)(0

.516

)(1

.363

)(2

.802

)(2

.050

)Po

st-S

econ

dary

Shar

e0.

475

-0.1

73-0

.196

0.68

6**

-0.0

784

-0.2

890.

280

-2.4

95*

-2.5

57*

(0.3

22)

(0.3

41)

(0.3

57)

(0.3

34)

(0.4

18)

(0.3

69)

(0.6

94)

(1.4

88)

(1.4

75)

Skill

edIn

dust

rySh

are

-0.6

13**

0.25

40.

149

-0.1

86-0

.124

-0.2

44-1

.279

0.21

2-0

.109

(0.3

07)

(0.2

57)

(0.2

12)

(0.2

95)

(0.2

35)

(0.2

19)

(0.8

50)

(1.1

68)

(0.8

43)

ln(P

opul

atio

n)-0

.057

0**

-1.2

79**

*-1

.272

***

-0.0

668*

*-0

.989

***

-0.9

66**

*0.

120*

*-1

.758

**-1

.743

**(0

.025

6)(0

.164

)(0

.201

)(0

.028

9)(0

.153

)(0

.208

)(0

.057

1)(0

.677

)(0

.798

)

Obs

erva

tion

s20

420

420

420

420

420

420

220

220

2R

-squ

ared

0.60

70.

936

0.93

50.

545

0.92

40.

924

0.42

50.

656

0.65

5Y

ear

Dum

mie

sY

esY

esY

esY

esY

esY

esY

esY

esY

esLo

calit

yFE

No

Yes

Yes

No

Yes

Yes

No

Yes

Yes

IVs

No

No

Yes

No

No

Yes

No

No

Yes

Not

e:R

obus

tSt

anda

rdE

rror

sin

Bra

cket

s.*

indi

cate

ssi

gnifi

cant

at10

%;*

*si

gnifi

cant

at5%

;***

sign

ifica

ntat

1%.

Reg

ress

ions

wer

ew

eigh

ted

byLa

bor

Forc

eSu

rvey

(LFS

)ob

serv

atio

ns.

The

sam

ple

incl

udes

data

from

1989

and

1993

.T

hein

stru

men

tsus

edin

colu

mns

(3),

(6),

and

(9)

are

base

don

pred

icte

dse

ttle

men

tpa

tter

ns,a

sde

scri

bed

inth

epa

per.

59

Tabl

e1.

10:

Exp

endi

ture

sPe

rC

apita

(Eth

nic

Frag

men

tatio

n)

ln(P

erC

apit

aTo

talS

pend

ing)

ln(P

erC

apit

aE

duca

tion

Spen

ding

)ln

(Per

Cap

ita

Wel

fare

Spen

ding

)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Eth

nic

Frag

men

tation

-0.6

78**

*-0

.177

-0.0

331

-0.4

30-0

.255

-0.3

57-0

.959

*0.

0452

0.44

3(0

.202

)(0

.183

)(0

.213

)(0

.267

)(0

.220

)(0

.341

)(0

.522

)(0

.819

)(0

.521

)U

nder

17Sh

are

-0.3

900.

0887

0.01

25-0

.341

0.29

20.

324

0.82

50.

363

0.30

4(0

.474

)(0

.260

)(0

.307

)(0

.581

)(0

.355

)(0

.493

)(0

.747

)(0

.569

)(0

.753

)O

ver

65Sh

are

0.24

7-0

.449

-0.2

32-0

.542

-0.5

80-0

.726

0.25

5-0

.483

0.16

5(0

.512

)(0

.467

)(0

.552

)(0

.606

)(0

.818

)(0

.887

)(0

.739

)(1

.136

)(1

.354

)Po

st-S

econ

dary

Shar

e-0

.404

*-0

.064

70.

0242

-0.1

13-0

.060

6-0

.331

-0.4

85-0

.144

-0.3

11(0

.241

)(0

.327

)(0

.336

)(0

.343

)(0

.531

)(0

.540

)(0

.428

)(0

.818

)(0

.825

)Sk

illed

Indu

stry

Shar

e0.

305

-0.1

45-0

.043

90.

613

0.07

48-0

.025

90.

501

0.61

30.

550

(0.3

75)

(0.2

15)

(0.2

36)

(0.4

95)

(0.3

67)

(0.3

79)

(0.5

53)

(0.4

49)

(0.5

79)

Shar

eIm

mig

rant

-0.9

79**

*0.

141

0.29

1-0

.841

**0.

479

0.41

10.

854*

-0.2

90-0

.245

(0.2

55)

(0.2

68)

(0.3

09)

(0.3

35)

(0.3

75)

(0.4

96)

(0.4

50)

(0.6

68)

(0.7

58)

ln(P

opul

atio

n)-0

.058

7**

-0.8

33**

*-0

.861

***

-0.0

539*

-0.7

20**

*-0

.681

***

-0.0

267

-0.7

77-0

.768

**(0

.025

4)(0

.158

)(0

.142

)(0

.030

3)(0

.198

)(0

.229

)(0

.029

3)(0

.546

)(0

.349

)

Obs

erva

tion

s13

213

213

213

213

213

213

213

213

2R

-squ

ared

0.71

40.

975

0.97

40.

604

0.93

70.

936

0.62

40.

877

0.87

4Y

ear

Dum

mie

sY

esY

esY

esY

esY

esY

esY

esY

esY

esLo

calit

yFE

No

Yes

Yes

No

Yes

Yes

No

Yes

Yes

IVs

No

No

Yes

No

No

Yes

No

No

Yes

Not

e:R

obus

tSt

anda

rdE

rror

sin

Bra

cket

s.*

indi

cate

ssi

gnifi

cant

at10

%;*

*si

gnifi

cant

at5%

;***

sign

ifica

ntat

1%.

Reg

ress

ions

wer

ew

eigh

ted

byLa

bor

Forc

eSu

rvey

(LFS

)ob

serv

atio

ns.

The

sam

ple

incl

udes

data

from

1989

and

1993

.T

hein

stru

men

tsus

edin

colu

mns

(3),

(6),

and

(9)

are

base

don

pred

icte

dse

ttle

men

tpa

tter

ns,

asde

scri

bed

inth

epa

per.

The

sam

ple

incl

udes

66Lo

calA

utho

riti

esth

atar

epr

edom

inan

tly

Jew

ish

(non

-Jew

ish

popu

lation

less

than

5%).

60

Tabl

e1.

11:

Sour

ces

ofR

even

uePe

rC

apita

,Rel

igio

usFr

agm

enta

tion

Tota

lRev

enue

spe

rca

pita

Ow

nR

even

ues

per

capi

taTa

rget

edG

rant

spe

rca

pita

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Rel

igio

usFr

agm

enta

tion

-0.4

20**

*-0

.264

-0.2

30-0

.528

***

-0.8

76**

*-0

.907

***

-0.3

080.

586

0.59

7**

(0.1

60)

(0.4

27)

(0.2

24)

(0.1

99)

(0.2

85)

(0.2

38)

(0.2

31)

(0.6

68)

(0.2

79)

Und

er17

Shar

e-2

.003

***

-0.3

77-0

.485

-3.4

44**

*-0

.788

*-0

.844

**-0

.651

0.21

80.

187

(0.4

26)

(0.3

93)

(0.3

19)

(0.4

38)

(0.3

99)

(0.3

38)

(0.5

73)

(0.4

53)

(0.3

97)

Ove

r65

Shar

e0.

105

-0.5

22-0

.401

0.43

2-0

.448

-0.5

32-0

.702

0.64

80.

586

(0.5

11)

(0.5

95)

(0.4

88)

(0.7

30)

(0.5

48)

(0.5

17)

(0.6

51)

(0.6

43)

(0.6

07)

Post

-Sec

onda

rySh

are

0.54

2**

-0.0

199

0.03

291.

189*

**0.

650*

*0.

920*

*0.

0871

-0.3

49-0

.467

(0.2

74)

(0.3

83)

(0.3

50)

(0.3

66)

(0.3

20)

(0.3

70)

(0.4

04)

(0.4

50)

(0.4

35)

Skill

edIn

dust

rySh

are

-0.5

83**

0.13

40.

0730

-1.0

82**

*0.

120

0.13

8-0

.120

-0.2

54-0

.388

(0.2

78)

(0.2

87)

(0.2

07)

(0.3

61)

(0.2

37)

(0.2

20)

(0.3

78)

(0.3

99)

(0.2

58)

ln(P

opul

atio

n)-0

.031

3-1

.007

***

-1.0

08**

*0.

148*

**-0

.329

*-0

.362

*-0

.022

0-0

.682

***

-0.6

78**

*(0

.023

7)(0

.153

)(0

.197

)(0

.031

9)(0

.197

)(0

.209

)(0

.031

2)(0

.245

)(0

.245

)

Obs

erva

tion

s20

420

420

420

420

420

420

420

420

4R

-squ

ared

0.63

40.

930

0.93

00.

741

0.89

10.

890

0.46

80.

918

0.91

7Y

ear

Dum

mie

sye

sye

sye

sye

sye

sye

sye

sye

sye

sLo

calit

yFE

noye

sye

sno

yes

yes

noye

sye

sIV

sno

noye

sno

noye

sno

noye

s

Not

e:R

obus

tSt

anda

rdE

rror

sin

Bra

cket

s.*

indi

cate

ssi

gnifi

cant

at10

%;*

*si

gnifi

cant

at5%

;***

sign

ifica

ntat

1%.

Reg

ress

ions

wer

ew

eigh

ted

byLa

bor

Forc

eSu

rvey

(LFS

)ob

serv

atio

ns.

The

sam

ple

incl

udes

data

from

1989

and

1993

.T

hein

stru

men

tsus

edin

colu

mns

(3),

(6),

and

(9)

are

base

don

pred

icte

dse

ttle

men

tpa

tter

ns,a

sde

scri

bed

inth

epa

per.

61

Tabl

e1.

12:

Sour

ces

ofR

even

uePe

rC

apita

,Eth

nic

Frag

men

tatio

n

Tota

lRev

enue

spe

rca

pita

Ow

nR

even

ues

per

capi

taTa

rget

edG

rant

spe

rca

pita

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Eth

nic

Frag

men

tation

-0.5

16**

*-0

.039

70.

0564

0.67

2***

0.27

60.

508

-0.4

330.

146

0.22

9(0

.173

)(0

.126

)(0

.169

)(0

.191

)(0

.234

)(0

.325

)(0

.348

)(0

.370

)(0

.279

)U

nder

17Sh

are

-0.9

53**

-0.2

49-0

.348

-2.6

68**

*-0

.674

-0.7

90*

0.40

60.

0499

-0.0

554

(0.4

18)

(0.2

62)

(0.2

44)

(0.6

26)

(0.6

16)

(0.4

69)

(0.6

17)

(0.4

16)

(0.4

03)

Ove

r65

Shar

e-0

.282

-0.4

50-0

.221

-0.0

658

-0.4

21-0

.135

-1.8

28**

*0.

575

0.25

7(0

.491

)(0

.418

)(0

.438

)(0

.868

)(0

.748

)(0

.844

)(0

.652

)(0

.710

)(0

.725

)Po

st-S

econ

dary

Shar

e-0

.043

60.

184

0.35

30.

353

0.72

81.

080*

*-0

.567

-0.2

40-0

.562

(0.2

52)

(0.2

75)

(0.2

67)

(0.3

98)

(0.4

62)

(0.5

14)

(0.4

18)

(0.4

25)

(0.4

42)

Skill

edIn

dust

rySh

are

0.52

50.

232

0.36

1*0.

217

0.00

413

-0.0

0755

0.49

30.

412

0.41

8(0

.367

)(0

.155

)(0

.187

)(0

.578

)(0

.385

)(0

.361

)(0

.610

)(0

.348

)(0

.310

)Sh

are

Imm

igra

nt-0

.789

***

-0.0

643

0.05

22-1

.118

***

-0.4

61-0

.360

0.48

5-0

.268

-0.2

14(0

.255

)(0

.179

)(0

.245

)(0

.370

)(0

.311

)(0

.472

)(0

.397

)(0

.356

)(0

.406

)ln

(Pop

ulat

ion)

-0.0

394

-0.6

17**

*-0

.638

***

0.09

39**

*-0

.586

***

-0.6

41**

*-0

.033

2-0

.184

-0.1

90(0

.024

5)(0

.118

)(0

.113

)(0

.033

2)(0

.189

)(0

.218

)(0

.030

0)(0

.248

)(0

.187

)

Obs

erva

tion

s13

213

213

213

213

213

213

213

213

2R

-squ

ared

0.69

20.

983

0.98

20.

672

0.93

50.

932

0.65

40.

967

0.96

6Y

ear

Dum

mie

sye

sye

sye

sye

sye

sye

sye

sye

sye

sLo

calit

yFE

noye

sye

sno

yes

yes

noye

sye

sIV

sno

noye

sno

noye

sno

noye

s

Not

e:R

obus

tSt

anda

rdE

rror

sin

Bra

cket

s.*

indi

cate

ssi

gnifi

cant

at10

%;*

*si

gnifi

cant

at5%

;***

sign

ifica

ntat

1%.

Reg

ress

ions

wer

ew

eigh

ted

byLa

bor

Forc

eSu

rvey

(LFS

)ob

serv

atio

ns.

The

sam

ple

incl

udes

data

from

1989

and

1993

.T

hein

stru

men

tsus

edin

colu

mns

(3),

(6),

and

(9)

are

base

don

pred

icte

dse

ttle

men

tpa

tter

ns,

asde

scri

bed

inth

epa

per.

The

sam

ple

incl

udes

66Lo

calA

utho

riti

esth

atar

epr

edom

inan

tly

Jew

ish

(non

-Jew

ish

popu

lation

less

than

5%).

62

Table 1.13: Own Revenue Sources Per Capita, Religious Fragmentation

Other Income per capita Local Tax Revenues per capita

(1) (2) (3) (4) (5) (6)Religious Fragmentation -0.431* -1.264*** -1.260*** -0.595*** -0.436 -0.488

(0.247) (0.289) (0.300) (0.189) (0.442) (0.352)Under 17 Share -3.210*** -0.834 -0.968** -3.586*** -0.700 -0.697

(0.476) (0.520) (0.427) (0.510) (0.554) (0.501)Over 65 Share 0.710 0.874 1.100* 0.0267 -1.676** -2.010***

(0.772) (0.809) (0.653) (0.827) (0.682) (0.766)Post-Secondary Share 0.882** -0.608 -0.353 1.413*** 1.625*** 1.918***

(0.371) (0.469) (0.467) (0.478) (0.494) (0.548)Skilled Industry Share -0.981** -0.0544 -0.0536 -1.251*** 0.140 0.176

(0.447) (0.295) (0.277) (0.380) (0.409) (0.325)ln (Population) 0.0841** -0.401 -0.414 0.209*** -0.273 -0.324

(0.0328) (0.248) (0.263) (0.0394) (0.285) (0.309)

Observations 204 204 204 204 204 204R-squared 0.631 0.861 0.860 0.747 0.775 0.774

Year Dummies yes yes yes yes yes yesLocality FE no yes yes no yes yes

IVs no no yes no no yes

Note: Robust Standard Errors in Brackets. * indicates significant at 10%; ** significant at 5%; ***significant at 1%. Regressions were weighted by Labor Force Survey (LFS) observations. The sampleincludes data from 1989 and 1993. The instruments used in columns (3), (6), and (9) are based on predictedsettlement patterns, as described in the paper.

63

Table 1.14: Own Revenue Sources Per Capita, Ethnic Fragmentation

Other Income per capita Local Tax Revenues per capita

(1) (2) (3) (4) (5) (6)Ethnic Fragmentation 0.351 -0.112 0.0983 0.992*** 0.631** 0.896**

(0.304) (0.404) (0.421) (0.235) (0.271) (0.409)Under 17 Share -1.881*** 0.707 0.575 -3.106*** -1.750** -1.843***

(0.686) (0.748) (0.607) (0.787) (0.782) (0.591)Over 65 Share 0.462 1.689 2.450** -0.656 -2.116** -2.203**

(0.881) (1.151) (1.093) (1.124) (0.819) (1.063)Post-Secondary Share -0.134 -0.0515 0.346 0.708 1.100** 1.419**

(0.404) (0.710) (0.665) (0.517) (0.527) (0.647)Skilled Industry Share -0.124 -0.469 -0.333 0.469 0.468 0.341

(0.614) (0.429) (0.467) (0.739) (0.580) (0.455)Share Immigrant -1.495*** -0.416 -0.322 -0.759 -0.454 -0.329

(0.384) (0.454) (0.611) (0.466) (0.376) (0.595)ln (Population) 0.0476 -0.495* -0.507* 0.136*** -0.622*** -0.718***

(0.0336) (0.279) (0.282) (0.0435) (0.229) (0.274)

Observations 132 132 132 132 132 132R-squared 0.543 0.893 0.888 0.624 0.908 0.905

Year Dummies yes yes yes yes yes yesLocality FE no yes yes no yes yes

IVs no no yes no no yes

Note: Robust Standard Errors in Brackets. * indicates significant at 10%; ** significant at 5%; *** signif-icant at 1%. Regressions were weighted by Labor Force Survey (LFS) observations. The sample includesdata from 1989 and 1993. The instruments used in columns (3), (6), and (9) are based on predicted settle-ment patterns, as described in the paper. The sample includes 66 Local Authorities that are predominantlyJewish (non-Jewish population less than 5%).

Chapter 2

Capacity Constrained Exporters: Micro

Evidence and Macro Implications1

2.1 Introduction

Standard intermediate microeconomics courses teach that short run marginal cost is

increasing with output due to fixed factors in production. In practice, most theory mod-

els in international trade assume that firms face constant marginal cost. To the extent

that the model is used to study relatively short run consequences, these models may be

ignoring important features. However, unless there exists strong evidence to suggest that

the assumption is anything other than innocuous, there is little reason to give up the con-

stant marginal cost assumption, not least because its simplifying nature greatly enhances

modeling tractability.

This paper questions the validity of this simplifying assumption. First, we demonstrate

robust evidence for the presence of increasing marginal cost and identify its main sources. We

show that financial as well as physical capacity constraints give rise to increasing marginal

costs. Next, we build a structural model with constrained exporters to quantify aggregate1This chapter coauthored with JaeBin Ahn (IMF). We are especially grateful to Eric Verhoogen for

sharing the dataset. The views expressed in this paper are those of the authors and should not be attributedto the International Monetary Fund, its Executive Board, or its management.

64

65

implications of the presence of capacity constrained firms. We find that the presence of

constrained firms can reduce aggregate output responses to external demand shocks, and

raise aggregate price level substantially.

Our study begins from the notion that domestic sales of firms with constant marginal

cost are predicted to be independent of their export sales, whereas firms with increasing

marginal cost would face a trade-off between domestic and export sales. For example, when

a firm increases export sales in response to a positive external demand shock, it will incur

an increase in marginal cost, which in turn makes it optimal for the firm to reduce domestic

sales. On the other hand, constant marginal cost implies that external demand shocks will

not affect the level of marginal cost, keeping domestic sales unchanged.

Exploring Indonesian firm-level domestic and export sales data, our reduced form ap-

proach delivers robust findings that exporting firms in general face strong trade-offs between

domestic and export sales. To identify the sources of such trade-offs, we investigate if the

degree of export-domestic sales trade-offs varies systematically with firms’ characteristics.

The underlying idea is that if increasing marginal cost prevails, we should observe a negative

relationship between export and domestic sales. Furthermore, we expect stronger patterns

in the data for firms that are capacity constrained, as these firms face the steepest cost of

increasing production.

We confirm the idea by showing that such patterns exist in the data, once firm pro-

ductivity changes are accounted for, and that these trade-offs are mostly driven by both

physical and financial constraints. We use a capacity realization variable as a proxy for

physical capacity constraints, and employ various financial capacity constraints measures

developed in the literature. The coefficient estimates suggest that (physically and finan-

cially) unconstrained firms exhibit no or a very weak negative correlation between export

and domestic sales growth, whereas being physically or financially constrained adds about

a .2 percentage point reduction in domestic sales for each one percentage point growth in

export sales.

66

Having demonstrated the robustness of this trade-off in the data, we turn next to quan-

tifying the aggregate implication. We develop a structural form estimation process, and

perform counterfactual exercises. Our contributions in the structural approach are two-

fold. First, we build off the static portion of the seminal structural trade model of Aw et al.

(2011). In particular, we consider capacity constrained firms explicitly, and thus relax the

independent markets assumption for these firms. The novelty in the estimation process lies

in exploiting the exporter’s optimality condition that the marginal revenue in each market

is equalized. As part of the process, we are able to recover firm level demand curves, which

in turn enable us to back out firm level price and quantity sold in each market.

The subsequent counterfactual exercises constitute the second major contribution of

the paper, providing quantitative implications of capacity constrained firms. Intuitively,

increasing marginal cost would reduce firm output responses to external demand shocks

via offsetting movements in domestic sales. In addition, capacity constraints lead firms to

charge a higher price than would otherwise be optimal. Our structural estimates suggest

that the presence of such capacity constrained firms can (1) reduce aggregate output re-

sponses to external demand shocks by around 30%, and (2) raise the aggregate price level by

around 23%. These counterfactual results suggest that capacity constrained firms generate

important policy implications.2

Related Literature The point of departure for this paper comes from the standard

models of international trade that have followed from the seminal works of Krugman (1979),

Krugman (1980), and Melitz (2003). The key feature of those models for the present pur-

poses is the assumption of constant marginal costs, which allows domestic and foreign2There is an important distinction between capacity constraints that are a direct consequence of optimal

firm investment decisions and capacity constraints that are outside the control of the firm. In the presenceof demand uncertainty, firms optimally choose their ex ante capacity level, which ex post may be bindingafter the realization of demand shocks. There is little for policy to do in this regard since capacity level ischosen optimally given available information. However, financial constraints, which are beyond the controlof the firm, would limit the ability of firms to choose the optimal level of physical capacity, leaving morescope for policy interventions. Our findings on the importance of financial constraints in addition to physicalcapacity constraints are especially noteworthy from this perspective.

67

markets to be treated as independent markets in the analysis. This property was made ex-

plicit in the recent structural approaches to international trade, simplifying the estimation

process substantially (Das et al. (2007); Aw et al. (2011)).

We demonstrate that the assumption of constant marginal costs, and hence final market

independence, is not supported in the data, and that the assumption is not innocuous. We

augment the static decision problem of Aw et al. (2011) to consider capacity constrained

firms, thereby allowing inter-market dependence for these firms.

There is an emerging literature that explores the relationship between domestic and

export sales as evidence for the presence of increasing marginal cost. Blum et al. (2011) find

a negative correlation between domestic and export sales growth from Chilean firm level

data, which they conjecture is being driven by physical capacity constraints. Soderbery

(2011) finds a similar pattern of export and domestic sales when looking at firm level data

from Thailand, and uses a similar measure of capacity utilization as here to document the

existence of physically constrained firms, but unlike our paper, does not consider financial

dimensions, which are more likely to be beyond the control of individual plants.

Berman et al. (2011) on the other hand, find the opposite pattern from French firm

level data when they instrument for export sales growth using information on the number

and location of export markets. They interpret this finding as evidence of complementarity

between exogenous changes in foreign demand and domestic sales.

Related papers focus on firm level output volatility, which we document and quantify

structurally. Based on a similar observation from French firms covered in the Amadeus

database, Vannoorenberghe (2012) further explores firm level output volatility, and con-

cludes that constant marginal cost assumptions may be inappropriate. Nguyen and Schaur

(2011) also study the effects of increasing marginal cost on firm level volatility using Danish

firm level data. Our paper differs from these papers in that we explore sources of increasing

marginal cost, and develop a structural estimation model to quantify aggregate implications.

Our reduced form approach resembles the strategy used in Fazzari et al. (1988). They

68

start from the theoretical notion that, in the presence of imperfect financial markets, credit

constrained firms’ investment will be sensitive to their cash-flow. Higher cash-flow sensitivity

of investment for credit constrained firms in the data serves as supporting evidence for

imperfect financial markets. In a similar vein, we draw out the implications of constant

marginal costs for export and domestic sales, and find an interrelationship as evidence for

increasing marginal costs.

Our finding can serve as direct micro-evidence that justifies the modelling strategy in

several recent papers that consider decreasing returns to scale production or borrowing

constraints to explain salient features of new exporter dynamics (Ruhl and Willis (2008);

Kohn et al. (2012); Rho and Rodrigue (2012)) or patterns of foreign aquisitions (Spearot

(Forthcoming)).3

This paper is also close to the literature that studies credit constraints and international

trade. Previous studies focus on export fixed costs financing, and thus extensive margin

effects of credit constraints (Chaney (2005); Manova (2011)). Indeed, there is abundant

evidence that credit constrained firms are less likely to become exporters (Muûls (2008),

among others). Our paper complements the literature by studying the intensive margin, and

showing that credit constraints affect incumbent exporters as well through the marginal cost

channel. This is also consistent with trade finance literature that studies intensive margin

adjustments during the great trade collapse (e.g., Ahn (2011); Paravisini et al. (2011)).

One of aggregate implications of capacity constrained firms discussed in this paper offers

an alternative explanation for the short-run trade elasticity puzzle. Ruhl (2008) considers

an extensive margin adjustment in response to temporary and permanent shocks to explain

low short-run trade elasticity and high long-run trade elasticity. Arkolakis et al. (2011)

introduces switching frictions from customers’ side to generate staggered short-run trade

dynamics. Our finding suggests that export cannot fully respond to external demand shocks3The structural estimation process in Rho and Rodrigue (2012), in particular, is closely related to our

paper. Unlike their approach that imposes and estimates increasing marginal costs across all firms, weseparate out constrained and unconstrained firms based on our reduced-form evidence. Also, their focus ison firm-dynamics, while we explore static issues.

69

due to inherent capacity constraints at the firm level.

The other aggregate implication of capacity constraints relates to the finance and mis-

allocation literature (e.g., Buera and Shin (2010); Buera et al. (2011); Midrigan and Xu

(2010); Buera and Moll (2012)). Compared to the literature that studies TFP losses from

misallocation induced by financial frictions in a dynamic model, we present static welfare

losses from financial constraints via higher aggregate price levels.

In sum, our paper is the first to identify multiple sources of increasing marginal costs,

both physical and financial, to incorporate these micro frictions into a structural estimation

framework, and to use this procedure to quantify aggregate implications.

The remainder of the paper proceeds as follows: Section 2 illustrates background theo-

retical discussion, and Section 3 describes the Indonesian firm level data used in this paper.

Section 4 reports empirical findings from the reduced form approach, and Section 5 develops

a structural estimation process, and provides quantifying example to gauge the macroeco-

nomic implications. Section 6 concludes the paper.

2.2 Illustrative Theory

This section aims to provide a simple theoretical framework to contrast different predic-

tions on the relationship between domestic and export sales movements, depending on the

underlying characteristics of marginal cost curve. A particular emphasis should be made

on the fact that such predictions neither hinge on any specific model structure, nor require

sophisticated theory models. For each type of marginal cost curve considered below, we

begin by finding optimal sales quantity in each market, and then track the subsequent op-

timal sales decision in response to positive external demand shocks. It is important to note

that since the area under each marginal revenue curve corresponds to sales revenues in each

market, sales revenues are expected to move in the same way as quantities sold in each

70

market in what follows.4

Constant marginal cost When a firm’s marginal cost is constant, independent of

the total amount of goods produced, the optimal output for each individual (segmented)

market is independent of all other markets. In other words, when demand conditions in one

market change, the firm would adjust sales in that particular market, leaving sales in all

other markets unchanged.5 This is illustrated in Figure 2.1.

Initially, the firm’s optimal operating point in each market is determined by the usual op-

timality condition that marginal revenue in each market equals marginal cost (i.e., MRD =

MRF = MC∗). For given domestic and export demand curves, this condition gives the

optimal output for the domestic market, Q∗D, and the optimal export volume, Q∗

F , with

total output being given by Q∗ = Q∗D + Q∗

F . Now, suppose the firm experiences a positive

foreign demand shock, which shifts up both the export demand curve and the marginal

revenue curve in the export market. In response, the optimal export volume increases from

Q∗F to Q∗∗

F at which point the optimality condition in the export market is satisfied with

the new marginal revenue curve (i.e., MR�F (Q

∗∗F ) = MC∗). Since the marginal cost and

the domestic marginal revenue curves are unchanged, the optimal output for the domestic

market is unchanged at Q∗d. In sum, constant marginal cost technology predicts that, other

things equal, exports respond to export demand shocks, but domestic sales are unaffected

at the firm level.

Increasing marginal cost When a firm’s marginal cost increases with the total

amount of goods produced, optimal outputs for each segmented market are no longer in-

dependent of each other. When demand conditions in one market change, the firm would

adjust the sales in that market. This, in turn, alters the marginal cost, which would affect4More precisely, this will be valid as long as the price elasticity of demand is greater than 1. This will be

relevant for our empirical exercises below since our plant-level dataset contains information on sales revenuerather than quantity sold.

5This property is implicit in all trade models with constant marginal costs including Krugman (1979),Krugman (1980), and Melitz (2003), and explicitly assumed in structural applications such as Das et al.(2007) and Aw et al. (2011)).

71

the optimal production decision in the other market. The situation with increasing marginal

cost is illustrated in Figure 2.2.

At the initial equilibrium with Q∗D, Q∗

F and Q∗ = Q∗D + Q∗

F , the firm satisfies the

optimality condition by equating marginal revenue from each market with marginal cost

(i.e., MRD(Q∗D) = MRF (Q∗

F ) = MC(Q∗)). Now, suppose again that there occurs a positive

export demand shock, which shifts up the marginal revenue curve in the export market. The

firm responds to positive export demand shocks by raising export sales because of higher

marginal revenue relative to the current marginal cost level in the export market. However,

as the firm produces more to meet the increased export sales, it incurs an increase in

marginal cost due to the nature of increasing marginal cost. This means that, for unchanged

domestic market conditions, the firm would incur losses by keeping domestic sales at Q∗D,

since marginal cost exceeds marginal revenue at this point in domestic market. The firm’s

optimal response is then to decrease domestic sales to recover the optimality condition in

the domestic market. As a result, in the new equilibrium, the firm equates marginal revenue

in each market to marginal cost with higher export sales, lower domestic sales, and higher

marginal cost than before (i.e., Q∗∗F > Q∗

F , Q∗∗D < Q∗

D, and Q∗∗ > Q∗). Therefore, increasing

marginal cost technology predicts that firm level export and domestic sales would respond

to export demand shocks in opposing ways.

Infinite marginal cost (Capacity constraints) In Figure 2.3, we propose a special

example of increasing marginal cost technology, namely, infinite marginal cost. This can

be understood as a combination of the two earlier cases in that a firm operates normally

with constant marginal cost technology, but faces capacity constraints at a certain level of

production beyond which production becomes infeasible.6

Marginal cost is constant up to the output level Q∗, and it jumps to an infinite level

beyond this point, implying that the firm’s production capacity is such that the firm’s6We present this special case here because our empirical section below shows this is the closest to the

patterns observed in the data.

72

maximum feasible output level is Q∗. Depending on market conditions, such a capacity

constraint may or may not be binding. A firm without any capacity constraint would find

it optimal to produce Q�D and Q�

F in the domestic and the export markets, respectively, as

shown in the constant marginal cost case earlier. However, if the sum of these output levels

exceeds the maximum capacity (e.g., Q�D+ Q�

F > Q∗), the capacity constraint is binding,

and the firm cannot attain the first best outcome. Instead, the firm needs to find sub-

optimal points, Q∗D and Q∗

F , which satisfy (i) MRD(Q∗D) = MRF (Q∗

F ) > MC∗ and (ii)

Q∗ = Q∗D +Q∗

F . We focus on this latter case with the binding capacity constraint, because

it reduces to the earlier constant marginal cost case otherwise. Now, suppose that there

occurs a positive export demand shock as before. As the firm decides to export more in

response to positive demand shocks abroad, the capacity constraint forces the firm to face a

trade-off between export and domestic sales, to keep total output at the maximum feasible

level, Q∗. Furthermore, the new equilibrium needs to satisfy the sub-optimality condition

at which marginal revenue from each market is equalized but exceeds the level of marginal

cost (i.e., MRD(Q∗∗D ) = MRF (Q∗∗

F ) > MC∗). Consequently, the new equilibrium features an

increase in export sales and a decrease in domestic sales with total output unchanged (i.e.,

Q∗∗F > Q∗

F , Q∗∗D < Q∗

D, and Q∗∗D +Q∗∗

F = Q∗). As was true with the more general case above,

we conclude that the presence of capacity constraints, unlike constant marginal cost, leads

to a negative correlation between export and domestic sales at the firm level in response to

market-specific demand shocks.

Sources of export-domestic sales trade-offs The most common rationale for in-

creasing marginal cost is the presence of fixed factors in production. For example, when

a firm cannot freely change the capital stock in the short run, the usual Cobb-Douglas

production technology leads to an increasing marginal cost (e.g., as modeled in Blum et

al. (2011)). Even when factors are flexible to adjust, still it is often increasingly costly as

exemplified by overtime pay for labor.

Regarding capacity constraints, we can think of various factors, which may come from

73

physical or financial dimension. Any incumbent production line or plant itself has maximum

capacity it can produce, and since it takes time to expand the production facility, it is natural

to expect a firm to face a physical capacity constraint. In addition, financial institutions

often set a line of credit to each borrower, beyond which a borrower has to pay a prohibitive

premium. Existing collateral value or credit history may also act as a natural borrowing

limit for each firm, which will in turn limit the maximum feasible production level.7

An alternative source of capacity constraints comes from managerial ability constraints,

often referred to as a span of control problem a la Lucas (1978). Simply put, an en-

trepreneur’s managerial skill exhibits decreasing returns to scale of the whole operation

such that as the entrepreneur devotes her time and efforts in expanding export markets, the

firm would start losing its domestic market share because she cannot spend as much time

and effort on the domestic operation as before, and vice versa.

So far, we have proceeded as if the patterns of correlation between domestic and export

sales growth are sufficient to verify the characteristics of marginal cost technology. The

reality is more complicated because, unlike our simple comparative statics analysis, domestic

demand shocks may arrive simultaneously with export demand shocks. To the extent that

domestic demand shocks are negatively correlated with export demand shocks, negative

trade-offs between export and domestic sales may arise even with constant marginal cost

curve. In other words, if foreign and domestic demand shocks are negatively correlated, it

would bias the data towards our interpretation incorrectly. Although this is not very likely

according to the literature on business cycle co-movements, we acknowledge that it is not

entirely implausible.8 On the other hand, if they are positively correlated, it would bias the

results against finding negative tradeoffs. In the empirical section below, we will present

systemic evidence that our findings are not simply driven by such negatively correlated

demand shocks.7It is worth noting that increasing fixed costs of reaching new (foreign) customers as in Arkolakis (2010)

will generate export-domestic sales tradeoffs only if firms face financial constraints.8Bilateral or multilateral trade liberalization may generate such patterns, affecting domestic and export

sales in opposing ways.

74

Although our theory holds most tightly when a firm produces and sells an identical

product for two segmented markets (i.e., domestic and export markets), it is valid in more

general cases as well. For example, multi-products firms even with a dedicated export market

product line9 will face such tradeoffs by reallocating resources when they face capacity

constraints. However, export-domestic sales trade-offs may occur in multi-products firms

not necessarily due to increasing marginal costs but rather as a result of extensive margin

adjustments (Bernard et al. (2010)).

Exchange rate movements would work against finding evidence for export-domestic sales

tradeoffs. In the case of producer currency pricing, effective marginal costs for exporting

should be multiplied by exchange rate. Then, currency depreciation will lower effective

marginal costs for exporting, leading to increases in exporting. At the same time, it will make

imported goods relatively expensive to domestic goods, shifting up the domestic demand

curve and hence generating higher domestic sales. In the case of local currency pricing,

export sales may change in domestic currency unit via valuation effect, but since domestic

sales will not respond to exchange rate movements, this tend to generate no relationship

between domestic and export sales.

Lastly, it is important to note that firm productivity evolves over time. In fact, pro-

ductivity growth, negative or positive, would affect export and domestic sales in the same

direction. Even with increasing marginal cost, if a firm’s productivity improves, the marginal

cost curve would shift right in Figure 2.2, and the relevant marginal cost level goes down

in Figure 2.2, possibly leading to increases in both domestic and export sales in response to

positive export demand shocks. This force would work against finding evidence for export-

domestic sales trade-offs.

Aggregate implication The presence of increasing marginal cost is a firm level micro

phenomenon, and it will have direct impacts on the firm level export-domestic sales rela-

tionship. Once aggregated, however, it also has an important macroeconomic implication.9A good example is the VW plant in Mexico (Verhoogen (2008))

75

Since external demand shocks induce adverse movements in domestic sales for exporters

with increasing marginal cost, aggregate output responses to external demand shocks will

depend critically on the share of firms with increasing marginal cost, as well as the de-

gree of these costs, in the economy. For example, total output in the economy populated

primarily by constant marginal cost exporters becomes very sensitive to external demand

shocks, whereas an economy with mostly increasing marginal cost exporters reduces output

volatility in response to external demand shocks due to offsetting movements in domestic

sales.

Furthermore, when increasing marginal cost takes the particular form of capacity con-

straints, as described in Figure 2.3, its direct consequence is that the price charged by such

constrained firms is higher than the optimal price that would have been charged in the ab-

sence of any constraints. The wedge between actual and optimal prices can then be used to

measure welfare losses caused by capacity constraints. Our structural section will quantify

both of these implications.

2.3 Data

The data is drawn from a well-used plant level dataset collected by the Indonesia Central

Bureau of Statistics (BPS).10 The survey includes all medium and large manufacturing

plants with more than 20 employees starting from 1975, however information on exporting

wasn’t included in the questionnaire until 1990. We choose to start our analysis in 1990 for

this reason, leaving us with a seven year panel.

The dataset is quite rich, with information on sector of main product, type of owner-

ship (public, private, and foreign), output, exports, assets, disaggregated inputs (including

energy, raw materials, and labor), and a variety of other measures that give a complete

portrait of firm boundaries, production and sales decisions.11 There are over 300 five-digit10Other studies that employed the same dataset include Blalock and Gertler (2004), Blalock and Gertler

(2008), Mobarak and Purbasari (2006), Amiti and Konings (2007), and Sethupathy (2008) among others.11Specifically, the dataset records export sales as the percentage of total output. Instead of taking the re-

76

ISIC manufacturing industries in the dataset. For our structural estimation, we will focus

on the largest exporting industry, manufacturing of wood, and wood and cork products

(ISIC 331).12

The Annual Manufacturing Survey (SI) is designed to record all registered manufacturing

plants. The BPS submits a questionnaire each year, and when the questionnaires are not

returned, field agents visit the plant to ensure compliance or verify the plant is no longer in

operation. The survey is conducted at the plant level. An additional survey is sent to the

head office of each multi-plant firm. Our data does not allow us to distinguish between single

and multi-plant firms. The BPS suggests that about 5% of plants are part of a multi-plant

firm. For the rest of the paper, we will use plant and firm interchangeably. Government

laws require that the data collected will only be used for statistical purposes and will not

be disclosed to tax authorities. This suggests the financial data is reasonably well reported.

Using an industry level wholesale price index published by the BPS, we deflate our measures

of sales, materials, and capital used in the analysis, which effectively removes industry level

inflationary trends. Admittedly, this will not be able to remove firm level prices, and thus

we do not interpret deflated sales as quantities sold.13

The Indonesian dataset is particularly useful for our purposes because it contains in-

formation on both physical and financial capacity constraints, allowing us to disentangle

these two possible sources of increasing marginal costs. The questionnaire asks specifically

about capacity utilization, which forms the basis of our measure of physical capacity con-

maining output, (total output-export), as domestic sales, we consider inventory adjustments by substractingchanges in inventory holdings from the remainder, (total output-export).

12This industry can be considered highly differentiated according to Broda and Weinstein (2006) withdemand elasticity around 2 (SITC Rev3. code 244-248).

13This gives rise to potential biases in productivity estimates. As De Loecker (2011) pointed out, however,productivity growth measures will not be biased under the assumption that input variation is not correlatedwith the price deviation when every firm’s price relative to the industry price index does not change overtime. This is one reason why we run growth regressions with productivity growth measures in our analysisbelow. The other reason is due to the fact that our export sales information comes from the "percentageof total outputs" that is exported. To the extent that the information is subject to reporting errors, itis possible that such reporting errors generate systemic negative correlation between domestic and exportsales. However, if we believe reporting errors are persistent over time at firm level, growth measures willnot be affected by such reporting errors.

77

straints. Our primary measure of physical capacity constraints is 100% capacity utilization,

which maps most closely to the infinite marginal cost case in our theoretical model. We try

alternative cut-off values of capacity constrained firms for robustness.

We also construct measures of financial constraints based on financial information of the

firm, including cash flow and assets, access to foreign loans and foreign ownership status.

Specifically, we construct a financial distress measure as the cash-flow/asset ratio, and define

financially constrained firms as the bottom 50% of firms ranked by this measure. Alterna-

tively, we assign foreign owned firms as unconstrained, and domestic firms as constrained

firms, with a threshold level of 50% in the share of stocks held by foreigners. We define the

third financial capacity constraint measure based on outstanding foreign loans. Figure 2.4

reports cross-correlation between physical and financial capacity constraints measures.

The cleaned dataset includes a little over 100,000 observations, including 3,241 unique

plants. Our primary analysis focuses on the firm level yearly growth in export and domestic

sales, thereby restricing our samples to firms that export in consecutive years. This leaves

us with 7,540 observations. Figure 2.5 provides a brief description of our primary sample,

continuing exporters, in comparison to all exporters and all non-exporters. On average,

continuing exporters are bigger in terms of total output, domestic sales, and export sales.

They tend to have a larger share of foreign owned stocks and larger foreign loans, whereas

cash-flow/asset ratio is lower for these firms. Capacity utilization does not seem to vary

significantly across different groups.

Among continuing exporters, there are 608 observations with physical capacity con-

straints, which consist of 8% of total sample. The first financial constraint measure classifies,

by construction, 50% of sample observations into constrained firms. On the other hand, the

other two measures include about 91-93% of samples as constrained firms. By comparing

continuing exporters in Figure 2.5 to each column in Figure 2.6, physically constrained firms,

on average, tend to sell more both domestically and abroad, whereas financially constrained

firms sell less in both markets.

78

2.4 Reduced Form Evidence

In this section, we follow a reduced form approach to identify the presence of increasing

marginal cost as well as the sources. Specifically, we explore the relationship between firm

level export and domestic sales growth. Our theoretical discussion in Section 2 suggests that

we should observe no clear relationship between changes in export and domestic sales growth

when firms have constant marginal cost technology, whereas the presence of increasing

marginal cost technology would result in a negative correlation between them.

Figure 2.7 reports correlation patterns between export and domestic sales growth. Col-

umn 1 shows almost zero correlation between export and domestic sales growth. This may

suggest that constant marginal cost technology prevails the economy, and is not a partic-

ularly bad assumption. However, it is important to note that this simple correlation does

not account for productivity growth, which affects export and domestic sales in the same

direction: improvement in productivity shifts down the marginal cost curve, which in turn

raises optimal output in both domestic and export markets. Failing to control for produc-

tivity growth, thus, amounts to a typical omitted variable problem, resulting in upward

bias. Column 2 confirms this idea. After controlling for productivity growth (measured

as labor productivity), the data reveals a strong negative correlation between export and

domestic sales growth: a 1 percentage point increase in export growth is associated with .13

percentage point decrease in domestic sales. Adding sector-year level (column 3) does not

change the result, reflecting that the observed negative correlation is not driven by particu-

lar sector-year level variations such as tariff changes. Adding firm level fixed effects (column

4) confirms that it is indeed the within-firm phenomenon consistent with our comparative

statics illustrated in section 2. The result is not sensitive to the choice of productivity

measures (column 5 and 6).14 We take this as suggestive evidence that marginal cost is, on14Specifically, we estimate TFP in the following two ways. First, we regress log (value added) on log

(capital) and log (labor) for each industry in year t, and estimate the industry-year level capital and laborshare. Then, TFP is calculated as the firm-level residual, which can be interpreted as the deviation fromindustry-year mean. Second, we apply the methodology of Levinsohn and Petrin (2003) with raw material

79

average, increasing rather than constant.

It is only suggestive because there are alternative explanations consistent with the ob-

served correlation patterns in Figure 2.7 as discussed in section 2. Our preferred strategy to

verify specific sources of increasing marginal cost is to control for capacity constraints explic-

itly on top of the basic simple correlation analysis. The idea is that if it is indeed increasing

marginal cost that drives the observed patterns, we expect to find even stronger patterns

for capacity constrained firms, because they are more likely to face increasing marginal cost.

The corresponding specification is given as:15

∆ ln(domestic sales)ist = α + β1∆ ln(export)ist + β2(capacity constraint)ist

+ β3∆ ln(export)ist ∗ (capacity constraint)ist

+ β4∆ ln(productivity)ist + FEst + FEi + εist

for firm i in industry s in year t, where FE stands for fixed effects. The capacity constraint

is a dummy variable with 1 for constrained firms and 0 otherwise. Our main focus is on the

coefficient of the interaction term, β3. β3 < 0 implies that constrained firms show a stronger

negative correlation between export and domestic sales growth, supporting the increasing

marginal cost story. As discussed in section 2, capacity constraints can come from either

physical or financial dimension. In what follows, we will control for these two types of

capacity constraints separately.

We begin with physical capacity constraints. As a proxy for physical capacity con-

straints, we employ the capacity utilization variable, and treat those firms that report 100%

of capacity realization as physically constrained, and all other firms as unconstrained.16 Col-

and labor as freely varying inputs, and electricity and fuels usage as well as capital as proxies for productivity.15Throughout the section, we will report main coefficients β1,β3,β4 only, but all the regressions include

a constant term and capacity constraint dummies as well.16Soderbery (2011) also employs a similar capacity utilization variable from Thai data as a proxy for

physical capacity constraints.

80

umn 1 in Table 2.8 confirms that physical capacity constraints are indeed a relevant source

of increasing marginal cost. The size of the coefficients is such that firms with no such

physical constraint reduces domestic sales by .16 percentage point per every 1 percentage

point growth in export sales, whereas physically constrained firms contract domestic sales

by .36 percentage point (.16+.20). In other words, those firms that are subject to physical

capacity constraints tend to exhibit a tradeoff between export and domestic sales more than

twice as strong as unconstrained firms.

To check if financial capacity constraint also matters, we use three different measures

of financial capacity constraints. We construct a financial distress measure as the cash-

flow/asset ratio, and define financially constrained firms as the bottom 50% of firms ranked

by this measure. Cash-flow/asset ratio is one of the most popular proxies for financial

constraints in corporate finance literature (Kaplan and Zingales (1997); Whited and Wu

(2006); Lin et al. (2011)). Alternatively, we assign foreign owned firms as unconstrained,

and domestic firms as constrained firms, with a threshold level of 50% in the share of

stocks held by foreigners. There is numerous evidence that foreign owned firms are least

likely to face credit constraints (e.g., Manova et al. (2011) for China, and Blalock et al.

(2008) for Indonesia among others). The last measure builds on the idea that those firms

that have access to an extra source of financing, notably foreign loans, are less likely to

be financially constrained (Fanelli et al. (2002)). Accordingly, we define the third financial

capacity constraint measure based on outstanding foreign loans.

Column 2 in Figure 2.8 shows that it is only financially distressed firms that exhibit

a negative correlation between export and domestic sales growth. As we include both

physical and financial capacity constraints in the regression, column 3 reports that export-

doemstic sales trade-offs are entirely driven by either physically or financially constrained

firms. When we use foreign/domestic ownership as a proxy for financial constraints, the

results look very similar. Such a negative relationship between export and domestic sales

disappears for foreign firms, and it is only domestic firms that exhibit export-domestic

81

sales trade-offs (column 4). Adding physical capacity constraint in column 5, we find that

domestic firms with physical constraints contract domestic sales by .4 percentage points for

every 1 percentage point reduction in exports. Domestic firms without physical constraints

or foreign firms with physical constraints reduces domestic sales by .2 percentage points

for every 1 percentage point growth in exports. Foreign firms without any constraints do

not face any trade-offs between export and domestic sales. Access to foreign loans, as an

alternative proxy for firms’ financial constraints, gives basically the same result except that

the financial constraint effect becomes statistically insignificant (column 6 and 7). Figure 2.8

thus suggests that the observed negative tradeoffs come mainly from physical and financial

capacity constraints.

To check if the results are robust to alternative productivity measures, we repeat baseline

regressions with different measures of productivity. Figure 2.9 summarizes the regression

results when we use TFP instead of labor productivity. They are very similar to the ones

with labor productivity in Table 2.8, confirming that unconstrained firms show no clear

trade-off patterns, whereas physically or financially constrained firms experience a strong

negative correlation between export and domestic sales growth. The results with an alter-

native measure of TFP (following Levinsohn and Petrin (2003)), summarized in Figure 2.10,

are very similar to the ones presented in Figure 2.9.

As additional robustness checks, we apply different criteria for physical and financial

capacity constraints. For physical capacity constraints, we relax the threshold level of 100%

capacity utilization to 70%, and Figure 2.11 summarizes the regression results with the

new 70% threshold. Similarly, alternative definitions of financial constraints are used in

regressions reported in Figure 2.12. Column 1 and 2 show the results with a lower threshold

level of the bottom 10% in the financial distress measure, and column 3 and 4 are the results

with a new foreign/domestic ownership criterion of 20% share of stocks held by foreigners.

Both measures effectively tighten up the measure of financial capacity constraints. Overall,

the results are robust to alternative capacity constraints measures.

82

Lastly, we consider the inventory adjustment process. The motivating idea is that if a

firm faces a tradeoff between domestic and export sales due to increasing marginal costs,

the firm will first turn to inventory holdings prior to substituting domestic sales for exports

sales (and vice versa). When a firm with increasing marginal cost faces positive foreign

demand shocks, for example, the firm would increase exports at the expense of decreased

domestic sales, but this adjustment will only take place after the firm runs down existing

inventory stocks. The firm would prefer to meet an increase in foreign demand through

inventory adjustment before incurring an increase in marginal costs by producing more.

Therefore, if the observed negative correlation between export and domestic sales indeed

reflects the presence of increasing marginal cost, we should expect such export-domestic

sales trade-offs to prevail especially for those firms that actually reduced their inventory

holdings. To check this, we add an inventory adjustment dummy that takes 1 for firms with

a decrease in inventory holdings and 0 otherwise, and interact this indicator with export

sales growth. The results reported in Figure 2.13 support this view. Column 1 shows

that firms that reduced inventory stocks indeed experienced a .22 percentage point larger

reduction in domestic sales for a percentage point increase in exports. The result is robust

to the inclusion of financial and physical capacity constraints (Columns 2-5).

In sum, we have shown that the underlying negative correlation between export sales

growth and domestic sales growth is robust to a variety of measures of productivity, and

it is stronger for financially or physically constrained firms, which is also robust to alter-

native measures of constraints. This reduces concerns that the results are driven by a

negative correlation between domestic and export demand, because it is hard to explain

why the negative correlation between domestic and export demand is stronger for capacity

constrained firms. We have also shown that inventory adjustment behavior is consistent

with our increasing marginal cost view of firm production. More importantly, our results

show that unconstrained firms do not exhibit any such negative correlation. We take the

results as evidence for the presence of capacity constrained firms in the economy. It has yet

83

to be shown that it is economically important. We turn next to quantifying the effect of

constrained firms in the aggregate.

2.5 Structural Form Approach

We develop a structural form analysis to quantify the aggregate implication of the pres-

ence of increasing marginal cost in the economy. In addition to providing quantitative im-

plications, our contribution from this section includes a methodological one that identifies

firm level price and quantity sold in each market separately.

Specifically, our estimation framework builds heavily on the static part of the innovative

structural trade model in Aw et al. (2011). Based on our findings from the reduced form

approach, we modify their model by taking into account the presence of increasing marginal

cost explicitly. We categorize firms into two groups: capacity constrained and unconstrained.

Capacity constrained firms include those firms that used 100% of capacity or the firms with

cash-flow/asset ratio being bottom 50%. All other firms are classified as unconstrained

firms. Further, we assume that constrained firms face infinite marginal cost as described in

Figure 2.3 in section 2 at firm specific capacity constraint level, qtotit , which is assumed to

be always binding. Consequently, we allow constrained exporters to face inter-dependent

markets (i.e., export-domestic sales trade-offs). Then, we exploit optimality conditions for

unconstrained exporters, and the sub-optimality condition for constrained exporters, which

enables us to identify firm level demand curve in each market, and hence firm level price and

quantity in each market. Subsequent counterfactual exercises suggest the substantial role

of capacity constraints in dampening the aggregate output sensitivity to demand shocks.

For the following estimation procedure, we pick one industry with ISIC code 331(Man-

ufacture of wood and wood and cork products, except furniture), the largest exporting

industry in Indonesia by volume.

84

2.5.1 Structural Framework

We assume that domestic and export markets are segmented, each of which is governed

by CES demand function. Specifically, domestic demand function faced by each firm i at

time t is given as:

qdit = Φdt

�pdit

�−σd ⇐⇒ pdit =�Φd

t

� 1σd

�qdit�− 1

σd , (2.5.1)

where σd is the elasticity of substitution in domestic market. The aggregate demand level in

domestic market at each time t, Φdt , determines the position of the demand curve common

to every firm. For a set of firms without any capacity constraint (i.e., constant marginal

cost), the optimal price is simply the markup over its marginal cost:

pjit =σj

σj − 1MCit, (2.5.2)

for j = D for domestic goods and F for export goods. Therefore, the level of marginal cost

becomes the sole factor determining firm specific domestic sales along the common demand

curve for this set of firms. Regarding the export demand curve, we allow idiosyncratic

export demand shifters17, zexit , on top of the common aggregate export demand level, Φext ,

leading to firm specific export demand curve given as:

qexit = Φext zexit (pexit )

−σex ⇐⇒ pexit = (Φext zexit )

1σex (qexit )

− 1σex , (2.5.3)

and unconstrained firms achieve the optimal export sales with the optimal price given in

equation (2.5.2).

Following Aw et al. (2011), we assume that marginal cost is independent of total output

level (i.e., constant marginal cost), and is a function of firm’s own capital level, kit, industry-

wide factor prices, wt, and its own unobservable productivity level, ωit:17Without this term, the model will predict a constant export-to-domestic sales ratio across firms, which

is not supported in the data.

85

ln (MCit) = β0 + βk ln (kit) + βw ln (wt)− ωit (2.5.4)

Since the optimal price is the markup over marginal cost for a set of unconstrained firms as

shown in equation (2.5.2), total variable cost, which is simply the marginal cost times the

total output, is expressed as:

TV Cit = qditMCit + qexit MCit =σd − 1

σdrdit +

σex − 1

σexrexit , (2.5.5)

for unconstrained firms, where rdit and rexit are domestic sales revenue and export sales rev-

enue, respectively.

Also, the optimal pricing rule in (2.5.2) allows us to express the domestic revenue of

unconstrained firms as:

rdit = pditqdit = Φd

t

�σd

σd − 1MCit

�1−σd

, (2.5.6)

and similarly for export sales of these firms as:

rexit = pexit qexit = Φex

t zexit

�σex

σex − 1MCit

�1−σex

, (2.5.7)

In fact, the optimal price in equation (2.5.2) is the outcome of the optimality condition that

equates marginal cost with marginal revenue. This means that unconstrained firms satisfy

the optimality condition in each market at the same time as below:

MRdit = MRex

it = MCit (2.5.8)

Unlike unconstrained firms, however, capacity constrained firms cannot produce more

than a certain level of output, beyond which actual marginal cost becomes infinite. Under

our assumption that the constraint is always binding, constrained firms cannot achieve

the optimality condition above, and instead operate at the sub-optimal point at which the

86

following condition holds:

MRdit = MRex

it > MCit (2.5.9)

Since equation (2.5.2) is not valid for constrained firms, equation (2.5.5), (2.5.6), and (2.5.7)

will not hold for constrained firms. In what follows, we first derive estimation procedures

for unconstrained firms, before turning to constrained firms.

2.5.2 Structural Estimation

Unconstrained exporters In order to take the theoretical framework from the previ-

ous section to the data, we begin by estimating the elasticity of substitution in each market

using equation (2.5.5):

TV Cit =

�σd − 1

σd

�rdit +

�σex − 1

σex

�rexit + eit, (2.5.10)

Total variable cost on left hand side of equation (2.5.10) comes from the data as the sum of

intermediate input costs and total labor payment. Admittedly, parts of labor payment are

associated with fixed overhead costs, and therefore, it is at best a proxy for total variable

cost with measurement error eit. Domestic sales and export sales revenue on right hand

side are taken directly from the data. Running a simple OLS regression gives coefficient

estimates from which we can back out elasticities σd and σex.

Next, we turn to the optimality condition that marginal revenue in each market is

equalized. Domestic sales revenue in equation (2.5.6) can be expressed alternatively as:

rdit = pditqdit =

�Φd

t

� 1σd

�qdit�σd−1

σd , (2.5.11)

by converting price as a function of quantity as expressed in demand equation (2.5.1). We

can write down export sales revenue in a similar way:

87

rexit = pexit qexit = (Φex

t zexit )1

σex (qexit )σex−1σex , (2.5.12)

and the optimality condition that equates marginal revenue across each market becomes:

MRdit

MRexit

=

�σd−1σd

� �qdit�−1

σd�Φd

t

� 1σd

�σex−1σex

�(qexit )

−1σex (Φex

t )1

σex (zexit )1

σex

= 1 (2.5.13)

Then, we replace the quantity of domestic sales as a function of domestic sales revenue and

aggregate demand from equation (2.5.11), and similarly for the quantity of export sales, to

get:

MRdit

MRexit

=

�σd−1σd

� �rdit� −1

σd−1�Φd

t

� 1σd−1

�σex−1σex

�(rexit )

−1σex−1 (Φex

t )1

σex−1 (zexit )1

σex−1

= 1 (2.5.14)

As long as we have recovered firm-level export demand shifters zexit , taking domestic sales

and export sales from the data, and using the estimated elasticities, this is essentially solving

the equation with unknown parameter Kt for each year t, where

Kt =

�Φd

t

� 1σd−1

(Φext )

1σex−1

(2.5.15)

That is, the first part of the optimality condition (i.e., equalizing marginal revenue in each

market) pins down a quasi-ratio between aggregate demand in domestic and export market.

In order to estimate firm-level export demand shifters zexit , we now exploit the second part of

the optimality condition (i.e., marginal revenue equals marginal cost) expressed in equation

(2.5.6) and (2.5.7) with specific marginal cost structure given in equation (2.5.4).

Substituting equation (2.5.4) for marginal cost in equation (2.5.6) and (2.5.7), domestic

sales in equation (2.5.6) is rewritten in log as:

88

ln�rdit�

= (1− σd) ln

�σd

σd − 1

�+ ln

�Φd

t

�

+(1− σd) (β0 + βk ln (kit) + βw ln (wt)− ωit)

Rearraging constant and time specific terms, it is reduced to:

ln�rdit�= γd

0 +�

γdtDt + (1− σd) (βk ln (kit)− ωit) (2.5.16)

with time dummy Dt.

A key issue in estimating equation (2.5.16) is that firm’s productivity ωit is not observ-

able to us, and especially when productivity levels are correlated with capital level, simple

regression yields biased estimates. In spirit of Olley and Pakes (1996) and Levinsohn and

Petrin (2003), and following Aw et al. (2011), we assume that the term composed of capital

and productivity can be proxied by cubic function of capital, material costs, and fuels usage:

(1− σd) (βk ln (kit)− ωit) = h (kit,mit, nit) + vit, (2.5.17)

and consequently, we estimate the following equation:

ln�rdit�= γd

0 +�

γdtDt + h (kit,mit, nit) + vit, (2.5.18)

with error term vit orginating from the cubic function proxy procedure.

Likewise, export sales in equation (2.5.7) is rewritten in log as:

ln (rexit ) = (1− σex) ln

�σex

σex − 1

�+ ln (Φex

t ) + ln (zexit )

+ (1− σex) (β0 + βk ln (kit) + βw ln (wt)− ωit) ,

89

and rearranging terms gives:

ln (rexit ) = γex0 +

�γext Dt + (1− σex) (βk ln (kit)− ωit) + ln (zexit ) (2.5.19)

Since equation (2.5.17) gives the following relationship:

(1− σex) (βk ln (kit)− ωit) =(1− σex)

(1− σd)(h (kit,mit, nit) + vit) , (2.5.20)

plugging equation (2.5.20) into equation (2.5.19) yields the estimation equation for export

sales:

ln (rexit )−(1− σex)

(1− σd)(h (kit,mit, nit) + vit) = γex

0 +�

γext Dt + ln (zexit ) (2.5.21)

that enables us to recover firm specific export demand shifters as residuals from the above

regression with intercepts and time dummies. Note that we have obtained the estimate of

(h (kit,mit, nit) + vit) from the regression of equation (2.5.18) above.

Having recovered firm specific export demand shifters zexit , we are able to solve the

equation (2.5.14) and get the quasi-ratio in (2.5.15). Still, however, domestic and export

market aggregate demand levels are not identified separately, and we need to take one

last step of normalization. Our strategy is to back out each of aggregate demand levels

separately, by setting the mean of log marginal costs to zero.

In practice, we plug price equation in (2.5.2) into equation (2.5.13) after using the fact

that quantity is revenue divided by price (i.e., qjit = rjit/pjit):

MRdit

MRexit

=

�σd−1σd

�σd−1σd

�rdit�−1

σd�Φd

t

� 1σd

�σex−1σex

�σex−1σex

(rexit )−1σex (Φex

t )1

σex (zexit )1

σex

(MCit)�

1σd

− 1σex

�

= 1 (2.5.22)

90

Taking logarithm on the above equation, and using the solution of the equation (2.5.14)

provided in (2.5.15), we can get rid of domestic aggregate demand term, Φdt , and keep

export aggregate demand, Φext , as the only unknown parameter:

ln

��σd − 1

σd

�σd−1σd �

rdit�−1

σd

�− ln

��σex − 1

σex

�σex−1σex

(rexit )−1σex (zexit )

1σex

�+

�σd − 1

σd

�lnKt

=

�1

σex− σd − 1

σex − 1

1

σd

�lnΦex

t +

�1

σd− 1

σex

�ln (MCit) (2.5.23)

Again, we take domestic sales and export sales from the data, and use estimated elasticities,

recovered export market shifters as well as the quasi demand ratio in equation (2.5.15).

Running the regression of LHS in equation (2.5.23) with time dummies, we can recover

aggregate export demand level in each year t, Φext , and we can also back out aggregate

domestic demand level, Φdt , from equation (2.5.15). Note that these are the normalized

estimates with the mean of ln (MCit) being zero. Lastly, from equation (2.5.11) and its

export sales equivalent in (2.5.12), we can find out each firm’s price and quantity sold in

each market separately.

Constrained exporters Most of the above equations do not hold for the group of

constrained firms because those equations are mostly derived from the fact that optimal

price equals markup over marginal cost, which is not true for constrained firms. A notable

exception is equation (2.5.14), because constrained firms also maximize their profits by

equating marginal revenue from each market as in equation (2.5.9). In addition, although

we employed only unconstrained firms to get the results, the estimated elasticities as well as

aggregate demand levels are common to both unconstrained and constrained firms. Thus,

by inputting appropriate values in equation (2.5.14) for constrained firms, we can recover

idiosyncratic export demand shifters, zexit , for each of these firms as in:

91

MRdit

MRexit

= 1 ⇒

�σd−1σd

� �rdit� −1

σd−1�Φd

t

� 1σd−1

�σex−1σex

�(rexit )

−1σex−1 (Φex

t )1

σex−1

= (zexit )1

σex−1 (2.5.24)

Now that we know everything about firm level demand curve for this group of firms, we can

find out each firm’s price and quantity sold in each market separately from equation (2.5.11)

and (2.5.12). This automatically gives us information on each of these firms’ actual capacity

constraint because, from our assumption, their output constraint is always binding:

qtotit = qdit + qexit (2.5.25)

Summary Below, we summarize the structural estimation process:

For Unconstrained Exporters:

(a) Run a regression in equation (2.5.10), and get σd and σex

(b) Run a regression in equation (2.5.18),

and get estimated values of h (kit,mit, nit) + vit

(c) Plug the estimated values in step (a) and (b) into equation (2.5.21),

run a regression,and recover zexit from residuals

(d) Substitute the estimated values in step (a) and (c) into equation (2.5.14),

and get the solution Kt in equation (2.5.15)

(e) Use the estimated values in step (a), (c), and (d), run a regression

in equation (2.5.23), and recover Φext and Φd

t

(f) Get firm level price and quantity using equation (2.5.11) and (2.5.12)

and values from step (a), (c) and (e)

For Constrained Exporters:

92

(g) Use the values from step (a) and (e), and get zexit from equation (2.5.24)

(h) Get firm level price and quantity using equation (2.5.11), (2.5.12)

and values from step (a), (e) and (g)

Since non-exporters share the same domestic aggregate demand level and the elasticity

of substitution with exporters, we can also back out their domestic price and quantity

sold from equation (2.5.9). Constrained non-exporters are assumed to face the binding

constraint: qtotit = qdit.

Table 2.1 reports key parameter estimates from the structural estimation procedure.18

2.5.3 Counterfactuals I

We perform counterfactual analysis to study the effects of positive export market de-

mand shocks on total revenue at industry level as well as firm level. Our underlying assump-

tion is that unconstrained firms can adjust output freely at its own constant marginal cost,

whereas constrained firms always face binding constraints at total output qtotit found in equa-

tion (2.5.25). Our counterfactual scenario is to imagine one percent increase in aggregate

export market demand, Φext , leaving aggregate domestic market demand, Φd

t , unchanged,

and calculate hypothetical firm level responses. We do not account for extensive margin ad-

justments (i.e., switching to or out of exporting), and consider intensive margin adjustments

of incumbent exporters only.19

For unconstrained firms, it is quite simple to get new optimal total sales, because do-

mestic sales would not change at all, while exports will increase exactly by one percent. For

constrained firms, however, we need to find out new sub-optimal domestic sales quantity

and exports quantity that still satisfy the sub-optimality condition in equation (2.5.9) with18The estimates of σd−1

σdand σd−1

σdare 0.573 and 0.551 with standard errors 0.04 and 0.03, respectively.

19To be able to account for extensive margin adjustments, we will need to go through fixed and/or sunkcost estimation, which is beyond the scope of this paper.

93

new aggregate export demand level and the capacity constraint in equation (2.5.25) at the

same time. This counterfactual result is reported in Table 2.14.

If we aggregate domestic sales and exports by constrained firms and unconstrained firms

separately, we can see that domestic sales stay the same level, but exports increase by

one percent for unconstrained firms, as we expected. For constrained firms, however, the

results indicate that domestic sales decrease by around .41%, while export sales increases

by around .53%. In terms of total sales, actual domestic sales/export ratio is such that it

increase by around .78% for unconstrained firms, but only by around .38% for constrained

firms. This results in only about .56% increase in total aggregate sales in response to 1%

positive demand shock in export markets. Noting that the industry would have experienced

about .78% increases in total sales if there were no constrained firms, this implies that the

presence of capacity constrained firms reduces the aggregate sales responses by around 30%

(from .78% to .56%). Looking at aggregate export responses, we find that the presence of

capacity constrained firms reduces the aggregate export responses by around 27% (from 1%

to .73%). This suggests the potential role of capacity constraints in explaining the short-run

trade elasticity puzzle as described in Ruhl (2008) among others.

We can do the same exercise by introducing 1% negative demand shocks, of which

results are reported in Table 2.15. This is exactly the mirror image of the earlier case with

positive export demand shocks, and we find that the presence of capacity constrained firms

again reduces the aggregate sales responses by around 30% (from -.78% to -.57%), and the

aggregate export responses by 26% (from -1% to -.74%).

Consequently, the industry’s overall output sensitivity to external demand shocks is

dampened by 30% due to the presence of capacity constrained firms: the industry cannot

reap the full benefits from positive external demand shocks, but can avoid from being fully

hit by negative external demand shocks.

94

2.5.4 Counterfactuals II

The presence of capacity constraints has a second significant impact on aggregate out-

comes. Welfare is directly affected by the existence of capacity constrained firms, who charge

higher prices than their unconstrained counterparts, thereby increasing the aggregate price

index and thus lowering welfare. It is straightforward that we can calculate the welfare

losses from capacity constraints, by comparing actual prices charged by constrained firms

with hypothetical prices that would have been charged by these firms if they had not been

constrained.20 Our structural estimation process provides actual prices, but hypothetical

prices are not available. Since firms would charge the optimal price as markup over marginal

cost when they are not constrained, we need to estimate firm level marginal cost, which we

have not pursued in this paper. Instead, we make an assumption that constrained firms’

marginal cost distribution is identical to the marginal cost distribution of unconstrained

firms. Note that we do know unconstrained firms’ marginal costs because their marginal

costs should equal marginal revenues, which are easily recovered from equation (2.5.14) with

estimated parameters.

In practice, we let constrained firms pick marginal cost draws randomly from the empir-

ical distribution function of unconstrained firms’ marginal costs, subject to the condition

that constrained firms’ marginal revenue is greater than the drawn marginal cost level (see

equation (2.5.9). With marginal cost draws picked, we can calculate constrained firms’ op-

timal prices that would have been charged had it not been for capacity constraints. Then,

we can construct a hypothetical domestic price index by adding unconstrained firms’ actual,

optimal prices. We repeat the procedure 100 times, and compare the hypothetical domestic

price index with the actual domestic price index. Our result suggests that domestic price

index would have been lowered by 47% without capacity constraints. When domestic goods

consumption share is given by 1/2, this implies that capacity constraints result in about20Again, we do not consider extensive margin adjustment effects, and assume that all incumbent firms

stay in the domestic market in the absence of capacity constraints.

95

23% welfare losses.21

Alternatively, imagine an economy with resource misallocation (e.g. arising from finan-

cial frictions) such that more efficient firms are capacity constrained. Specifically, we assume

that constrained firms’ marginal cost distribution follows the bottom 10% of unconstrained

firms’ marginal cost distribution. Repeating the procedure under this misallocation assump-

tion, we find that the domestic price index would have been 71% lower without capacity

constraints, implying welfare losses of 35% due to the presence of capacity constraints.

This suggests that the combination of capacity constraints and resource misallocation has

significant implications for the economy (additional welfare losses of 12%).

2.6 Conclusion

In this paper, we show that the assumption of constant marginal cost technology, which

is implicit or explicit in most theory models of international trade, has predictions about

firm level foreign and domestic sales which are inconsistent with the data. We utilize a

reduced form approach to demonstrate a strong negative relationship between export and

domestic sales growth rates. We show that once productivity is properly accounted for,

a significant trade-off at the firm level is apparent. This is evidence against the standard

constant marginal cost view.

Furthermore, we explore the sources of this increasing marginal cost technology, and

find that physically and financially constrained firms have significant and large negative

correlations between export and domestic sales. Financial constraints are shown to be at

least as important as physical capacity constraints in contributing to the observed trade-off.

This suggests that a constant marginal cost view is inappropriate for internationally inte-

grated firms, and that short-run firm constraints could be quite significant for understanding

aggregate outcomes.

Next, we attempt to quantify the importance of this micro level friction for aggregate21The underlying model for this section is provided in Appendix.

96

fluctuations. Starting with the recent structural work of Aw et al. (2011)), we modify and

advance this framework to include capacity constrained firms. Having derived the necessary

identifying moments, we structurally estimate the impact of capacity constrained firms for

macroeconomic fluctuation. Focusing on the largest exporting industry in Indonesia, we

find that the presence of capacity constrained firm could reduce aggregate output responses

to external demand shocks by around 30%. In addition, we show that capacity constraints

could result in welfare losses by about 23%. These counterfactual estimates suggest that the

existence of capacity constrained firms do indeed have significant aggregate consequences.

In future work, we seek to extend our framework to a dynamic setting, where we can

structurally estimate the impact of capacity constrained firms along the extensive margin,

including the recovery of sunk costs associated with exporting.

97

2.7 Figures and Tables

Figure 2.1: Constant Marginal Cost and Production

Figure 2.2: Increasing Marginal Cost and Production

98

Figure 2.3: Infinite Marginal Cost and (Sub) Optimal Production

Figure 2.4: Cross Correlation of Constraint Measures

99

Figure 2.5: Summary Statistics

Figure 2.6: Summary Statistics for Constrained Firms

Figure 2.7: Domestic and Export Sales Tradeoffs

100

Figure 2.8: Capacity Constraints and Domestic-Export Sales Trade Offs

Figure 2.9: Robustness Check with Productivity as TFP

101

Figure 2.10: Robustness Check with Productivity as Levinsohn and Petrin Methodology

102

Figure 2.11: Robustness Check with Alternative Physical Capacity Constraint Measure

103

Figure 2.12: Robustness Check with Alternative Financial Capacity Constraints Measure

104

Figure 2.13: Robustness Check with Inventory Adjustments

σd = 2.35 σex = 2.2Φd

1990 = 1, 104, 561 Φex1990 = 2, 491, 660

Φd1991 = 1, 057, 013 Φex

1991 = 3, 723, 407Φd

1992 = 1, 415, 523 Φex1992 = 3, 760, 982

Φd1993 = 1, 100, 565 Φex

1993 = 4, 125, 749Φd

1994 = 1, 055, 139 Φex1994 = 5, 333, 864

Φd1995 = 1, 162, 663 Φex

1995 = 3, 917, 337Φd

1996 = 1, 126, 510 Φex1996 = 4, 123, 405

Table 2.1: Implied Parameter Values

105

Figure 2.14: One Percent Positive External Demand Shock

Figure 2.15: One Percent Negative External Demand Shock

Chapter 3

The Import Elasticity Puzzle: An Aggregation

Puzzle?1

3.1 Introduction

Recent interest in the collapse of international trade and its determinants has revived a

longer standing interest in the excess volatility of trade relative to GDP. While the decline

in world production was significant during the financial crisis of 2008-2009, the fall in world

trade was even more startling. And yet, while the magnitudes of the declines were large,

the decline in trade relative to the decline in output was not particularly unique. Rather,

the excess sensitivity of trade to production was consistent with longer run world trends.2

As was first noted by Houthakker and Magee (1969) and replicated by subsequent studies,

trade has historically been more volatile than GDP. In their influential study, Houthakker

and Magee found that the income elasticity of U.S. imports was approximately 1.5, far

exceeding the theoretical prediction of an income elasticity of 1. This enduring feature

of the data has come to be known as the “Elasticity Puzzle”. If anything, the puzzle has

become more puzzling over time. Freund (2009) estimates the world income elasticity to be

1.77 from 1960 to 2006, and over 3 for the period 1990 to 2006.1A special thanks to Eric Verhoogen for providing access to the data.2See, for example, Freund (2009).

106

107

There have been a variety of attempts to understand and explain excess trade sensitivity,

but as of yet, no compelling explanation of observed trade and production patterns has been

offered. In the present study, I take a new approach and start from the micro perspective

of internationally engaged firms. I show that the theory as it has been articulated is best

understood as a description of firm behavior, and therefore a firm level empirical approach is

the most appropriate setting for testing the validity of the theory. I show that the theoretical

predictions at the firm level are equivalent to the aggregate predictions - namely, that the

income elasticity of imports is predicted to be one.

Turning to the data, first I show that aggregate national accounts data for Indonesia dis-

plays the traditional excess sensitivity of trade, as has been found in previous countries and

contexts. Using annual data from 1959-2010, the estimated income elasticity of imports

is over 2. Looking at the decade from 1990 to 1999, the time frame consistent with the

available micro data, the income elasticity of imports is estimated to be 1.635. Changing

the frequency of observation, evidence from quarterly data provides similar estimates. Fur-

thermore, the income elasticity of exports follows patterns found in previous studies, with

nearly identical estimates for annual data (1.68 vs 1.635), and lower estimates relative to

imports at a quarterly frequency (1.23 vs. 1.47). Based on this macro approach, Indonesia

like other countries, appears to suffer from the excess sensitivity of trade puzzle.

Having documented the macro puzzle, I then move to firm level data to test this predic-

tion directly. Guided by the theoretical section, I derive a theoretically consistent estimating

equation which can be taken to the data directly. Using imported intermediate inputs for

all large and medium manufacturing firms in Indonesia, the estimated income elasticity of

imports is 1 - precisely as predicted by the standard theory. Furthermore, the import price

elasticity and the domestic price index elasticity have equal and opposite signs, an auxil-

iary prediction of the CES model of important demand. Importing behavior within firms

matches up precisely with what the theory would predict.

While the micro evidence resolves one aspect of the puzzle - namely that there is nothing

108

puzzling at the firm level - it raises an alternative puzzle of why the macro and micro evidence

are different. I consider this new evidence in light of existing competing theories, and provide

a discussion of additional channels to explore to account for both data perspectives. Future

consideration of the trade elasticity puzzle will need to account for the presence of a macro

puzzle and the absence of a micro puzzle.

The paper is organized as follows. In Section 2, the original trade elasticity puzzle is

clarified, while section 3 discusses related literature. Section 4 lays out the theory for trade

elasticity estimation, while section 5 derives an empirically implementable import demand

estimating equation. Section 6 describes the macro and micro data sources. Section 7

documents the traditional macro puzzle for Indonesia, while Section 8 examines importing

behavior at the firm level. Section 9 provides a discussion of current theories in light of the

presented macro and micro evidence. Section 10 concludes.

3.2 What’s so puzzling?

Since it was first noted by Houthakker and Magee (1969), the puzzling feature of exces-

sively sensitive trade has been the source of much interest among international economists.

The puzzle can be understood as a tension between testable implications of theoretical mod-

els and the predictive power of those models. This tension in economic science has most

famously been discussed in Friedman (1953).

Under the assumptions of a constant income elasticity and optimizing behavior, the

income elasticity of imports is predicted to be one. When this simple model is taken to the

aggregate data, income elasticity is estimated to be well above 1, and one can thoroughly

reject the hypothesis that income elasticity is equal to one. There is an additional implication

that a constant income elasticity estimate above one would ultimately predict that the

GDP share of imports will exceed one. Nonetheless, a constant elasticity model performs

extremely well in terms of predictive power, as has been demonstrated by Marquez (2002a).

109

Taking the model and the resounding empirical rejection seriously, one approach is to

drop the assumption of a constant income elasticity. Dropping this assumption implies

that income elasticity is not known a priori, although there are some theoretical conditions

imposed on the behavior of varying income elasticity parameters. Using an Almost Ideal

demand system, for example, allows income elasticity to vary over time, is straight-forward

to implement empirically, and has nice theoretically proprieties. All of these benefits, how-

ever, are blunted by the observation that variable elasticity models have extremely poor

predictive power.

The tension highlighted by Friedman (1953) is in full force for this elasticity puzzle,

as a model that has been thoroughly rejected is used precisely because it has practical

explanatory power, while more theoretically consistent models are rarely used because of

their inferior predictive ability.

To resolve the puzzle, there have been two common approaches. The first approach

attempts to augment the variable elasticity puzzle to improve predictive power. The second

approach attempts to find a theoretically consistent constant elasticity formulation that

maintains the explanatory power of the model, which essentially reduces to articulating

a constant elasticity formulation that results in an estimated income elasticity of imports

equal to one. In this paper, I take the second tack, matching theory and data as closely as

possible by focusing on firm level demand for imports.

3.3 Related Literature

Interest in the relative movements of trade to production goes back to at least Houthakker

and Magee (1969). Their basic finding has been confirmed in many countries over many dif-

ferent time periods.3Given the robustness of these findings, there have been many attempts

to understand the sources driving the income elasticity of imports.3For an exhaustive listing of such studies, see Marquez (2002a).

110

Feenstra and Shiells (1994) argue that an important component of observed excess sen-

sitivity can be explained by the introduction of new varieties of products, which biases

measures of import prices and import demand. In Marquez (2002b), immigration plays

an important role in understanding excess sensitivity, as migrants bring with them home-

country habits which distort import patterns as a result.

In more recent contributions, Engel and Wang (2011) argue that the key to understand-

ing the import puzzle is in the characteristics of goods, with a particular emphasis on durable

goods. They build a model that incorporates durable and nondurable goods, and show that

the cyclical nature of durable goods can help to explain the volatility of imports.

There has been a renewed interest in the excess volatility of trade because of the so-

called “Great Trade Collapse”. Ahn et al. (2011b) argue that the financial collapse, and in

particular, the collapse in trade finance contributed to the excessively large decline in trade

during this time. Ahn (2011) provides a theoretical motivation for why trade finance can

lead to excessively volatile trade.

In one of the few papers to consider the sensitivity of trade using firm level data, Amiti

and Weinstein (2011) argue that the health of banks are a significant determinant of export

movements. Using a novel data set that links Japanese manufacturing firms with parent

banks, they exploit variation in the health of banks to study the financial channel impact

on exports. They find large and significant effects of bank health of the volatility of exports,

though only small effects of bank health of domestic sales. While they are interested in

understanding how the financial channel could contribute to the understanding of the Great

Trade Collapse, they don’t directly consider the excess sensitivity of trade to production at

the firm level, which is the central question of the present work.

In response to the trade finance story, Levchenko et al. (2010) argue that the explanation

for the great trade collapse (and excessively sensitive trade in general) comes from the

composition of goods. First, they argue that the collapse in international trade in 2008-09

was not particularly unusually given the size of the world output decline. Furthermore, they

111

argue that trade is more volatile than production because of the composition of goods that

are typically traded. Imports tend to be overrepresented by goods that are move volatile

over the business cycle, and this compositional story can explain much of the observed excess

sensitivity of trade.

Alternative explanations for the Great Trade collapse and the observed excessive volatil-

ity of trade have been offered. Alessandria et al. (2010) argue that the role of inventories

as a buffer against demand shocks helps to reconcile the observed patterns of trade and

production, while Bems et al. (2010) argue that demand linkages via intermediate inputs

explain the collapse of trade relative to production. This is related to the larger question

of the impact of vertical specialization and linkages on economic outcomes (see Yi (2003)

for example). The role of intermediary or “pure” trading firms has been explored further in

Ahn et al. (2011a).

The present paper deviates from the previous literature by emphasizing a tight link be-

tween theory and empirics. Import demand behavior is ultimately driven by firm decisions,

and therefore a proper evaluation of the theory should rely on firm observations. The focus

on imported intermediate inputs in the present paper aligns most closely with the typical

articulation of the theory and matches the available firm level data.

3.4 Theory

The import elasticity puzzle as articulated above comes from a tension between the

successful predictive power of a model that is theoretically inconsistent. In this section,

I lay out the standard theory behind the constant elasticity approach to import demand.

Using this theoretical framework, in the following section, I derive the specific estimating

equation used in the analysis. For comparative purposes, I outline the standard varying

elasticity framework proposed as an alternative approach to import demand estimation.

112

Constant Elasticities Consider a simple model of firm level demand for intermediate

inputs to production, under the assumption that foreign and domestic varieties of interme-

diate inputs are imperfect substitutes. The firm chooses foreign and domestic varieties of

inputs so as to minimize its cost function subject to a given level of output produced:

C(pf , pd, Qy) = minqf ,qd{pf ∗ qf + pd ∗ qd} (3.4.1)

subject to f(qf , qd) ≥ Qy (3.4.2)

where pf and pd represents the price of foreign and domestic imports, respectively, qf and qd

represent the quantity of foreign and domestic inputs, and Qy is the total firm level output

of the final product.

Given this objective function, the firm demand for imports that minimizes costs is given

by;

qf =∂C(pf , pd, Qy)

∂pm(3.4.3)

So far, we have put minimal structure on the optimization problem of the firm. If we

further assume that firm production technology exhibits constant returns to scale, then we

can rewrite the cost function to depend only upon relative prices:

C(pf , pd, Qy) = C(pf , pd) ∗Qy (3.4.4)

We can rewrite the demand for imports and show that it is homogeneous of degree 1 in Qy:

qf =∂C(pf , pd, Qy)

∂pf=

∂[C(pf , pd) ∗Qy]

∂pf= Cpm(pf , pd) ∗Qy (3.4.5)

where Cpm(.) is the partial derivative of the cost function with respect to the price of imports.

The demand for imports under these minimal assumptions implies that the output elasticity

of import demand equals 1, and furthermore, if Qy is associated with firm income, then the

113

income elasticity of imports is 1.

Given this minimal set of restrictions and assumptions, the theory predicts exactly what

the income elasticity of imports should be: unity. This suggests that it is not necessary

to estimate income elasticity of imports with data since theory has a specific prediction

about the value. As was mentioned previously, however, studies have consistently estimated

the income elasticity to be well above 1. This rejection of an implication of the model

wouldn’t by itself be a puzzle, except for the fact that the constant elasticity formulation

has exceptional predictive power, which more theoretically consistent formulations (varying

elasticity) lack.

Varying Elasticity With very specific assumptions about production technology, opti-

mizing behavior implies that the income elasticity of imports should be equal to 1, and hence,

doesn’t need to be estimated. Alternatively, one could argue that such specific demands on

production technology are not warranted, in which case, income elasticity of imports is not

known a priori, and must be estimated in the data.

Following Barten (1964) and Theil (1965), one can show that for any utility function,

differentiating the first-order conditions associated with that utility function yields the fol-

lowing expression;

ωft ∗ d lnqft = µ(yt,pftpdt

) ∗ d lnyt + π(yt,pftpdt

) ∗ d lnpftpdt

(3.4.6)

where ωft =pftqft

pdtqdt+pftqftis the budget share of imports, y = Y

P (pf ,pd)is real income, µ(.) is

the marginal budget share, and π(.) is the compensated price effect. Income elasticity is

given by

ηf,t =µ(yt,

pftpdt

)

ωft(3.4.7)

while the compensated price elasticity is given by:

114

�f,t =π(yt,

pftpdt

)

ωft(3.4.8)

In this formulation, income and price elasticities are not assumed to be constant, but rather

respond to relative prices and income. These unknown parameters can then be estimated.

To take the theory to the data, however, requires further assumptions. A common

approach is to treat µ and π as constants:

ωft ∗ d lnqft = µ ∗ d lnyt + π ∗ d lnpftpdt

+ urt (3.4.9)

where urt is the approximation error of the solution of the first-order conditions. This

formulation is referred to as the Rotterdam formulation in the literature.

An alternative approach is to find an exact solution for an approximation to the utility

function, as opposed to an approximate solution to any utility function as in the Rotterdam

formulation. Deaton and Muellbauer (1980) use a PIGLOG formulation to derive an Almost

Ideal Demand System (AIDS):

ln wft = δ ∗ ln yt + γ ∗ ln pftpdt

+ uat (3.4.10)

where uat is a residual introduced by the approximation of the utility function.

Using this more flexible approach, income elasticity of imports is µωft

under the Rot-

terdam formulation and 1 + δωft

under the AIDS formulation. The necessary requirement

for constant income elasticities would either be constant import shares (ωft is constant),

or exactly offsetting changes in parameters (µ or δ) as import shares change. A simple

inspection of country or firm import shares suggests that the first possibility is inconsistent

with the data. Therefore, the assumption of constant elasticities in models comes down to

an assumption that structural parameters vary in systematic ways with changes in import

shares.

The general tension between theory and empirical work has been highlighted by Marquez

115

(2002a). The simple theoretical models outlined here present challenges when the theory is

connected with the data. On one hand, the constant elasticity formulation has very clear and

testable hypotheses. The income elasticity, if constant, will be equal to 1. This hypothesis

has been soundly rejected in the data. Nonetheless, these constant elasticity models have

been used widely because while the data rejects the models implications, these models have

been very successful in terms of predictive and explanatory power. On the other hand,

models that allow for variable elasticity are consistent with both theory and data (and are

readily implementable), but have performed poorly in terms of predictive power. Therein

lies the import elasticity puzzle.

Having outlined the basic theory, I follow the previous literature in searching for a

constant elasticity formulation that satisfies the motivating theoretical model. My point of

departure consists in the novelty of testing the theory at the appropriate unit of observation

- the firm.

3.5 Empirical Implementation

As the theory outlined above suggests, very limited assumptions are required to derive

the basic result of a constant elasticity model. So long as firms behave optimally, foreign

and domestic varieties are imperfect substitutes, and technology exhibits constant returns

to scale, one can derive the key insight that a constant income elasticity of demand must

equal one. Extending this general model to a specific estimating equation is the goal of the

present section.

Assume that an individual plant minimizes costs subject to production technology, which

is a function of foreign and domestic varieties of inputs that are combined using a CES

aggregator. The firm solves the following program:

minqft,qdt{pdtqdt + pftqft} subject to (3.5.1)

116

Qyt ≤ [ρ1� q

�−1�

dt + (1− ρ)1� q

�−1�

ft ]�

�−1 (3.5.2)

where pit is the price of intermediate input produced in country i at time t, � is an elasticity

of substitution, and ρ is a weighting measure that captures the importance of domestic

varieties in production.

Solving for firm demand for intermediate inputs,

qft = (1− ρ)(pftPt

)−�Qyt (3.5.3)

where Pt is the standard CES price index,

Pt = [ρ(pdt)1−� + (1− ρ)(pft)

1−�]1

1−� (3.5.4)

To get to our estimating equation, taking logs yields:

ln qft = ln(1− ρ)− �ln(pftPt

) + ln Qyt (3.5.5)

The model predicts that the income elasticity of intermediate inputs should be 1, and that

the price elasticity of imports should have equal and offsetting effects, given by �.

3.6 Data

The data is drawn from a well-used plant level data set collected by the Indonesia Central

Bureau of Statistics (BPS).4 The survey includes all medium and large manufacturing plants

with more than 20 employees starting from 1975, however information on importing wasn’t

included in the questionnaire until 1990. While the East Asian Financial crisis introduces

some worries about the reliability of collected data, periods of financial crises have been of4Other studies that employed the same data set include Ahn and McQuoid (2012), Blalock and Gertler

(2004), Mobarak and Purbasari (2006), Amiti and Konings (2007), Blalock and Gertler (2008), and Sethu-pathy (2008) among others.

117

particular interest for understanding trade elasticities, and these years are included in the

analysis.

The data set is quite rich, with information on sector of main product, type of owner-

ship (public, private, and foreign), domestic output, imports, exports, assets, disaggregated

inputs (including energy, raw materials, and labor), and a variety of other measures that

give a complete portrait of firm boundaries, production and sales decisions. There are over

300 five-digit ISIC manufacturing industries in the data set.

The Annual Manufacturing Survey (SI) is designed to record all registered manufacturing

plants. The BPS submits a questionnaire each year, and completion of the survey is legally

required. When the questionnaires are not returned, field agents visit the plant to ensure

compliance or verify the plant is no longer in operation. The survey is conducted at the

plant level. An additional survey is sent to the head office of each multi-plant firm. Our data

does not allow us to distinguish between single and multi-plant firms. The BPS suggests

that about 5% of plants are part of a multi-plant firm. For this reason, I will use plant and

firm interchangeably. Government laws require that the data collected will only be used

for statistical purposes and will not be disclosed to tax authorities. Safeguards are put in

place to keep plant-specific financial information from being obtained by tax authorities or

competitors, which suggests that financial data is reasonable well reported.

Using a wholesale price index published by the BPS, I deflate our measures of sales and

materials used in the analysis to express values in real terms. The construction of import

prices is done at the 3-digit industry (89 3-digit manufacturing industries) level using an

input-output table supplied by the BPS, and industry import prices from national trade

flows data. Domestic price indices were also constructed at the 3-digit industry level based

on information supplied by the BPS.

The complete data set includes nearly 200,000 observations over a ten year period from

1990 to 1999, including 37,405 unique plant observations. Of this group, there are 38,831

plant-year observations with positive importing, from a total of 9,169 unique plants. This

118

forms the sample upon which the analysis is based. Summary Statistics are provided in

Table 3.1.

The aggregate data is taken from national accounts. The annual data covers the period

1959-2010, and is drawn from the International Financial Statistics (IFS) database provided

by the IMF. The quarterly data was provided by the Bank of Indonesia, and covers 1993

Q1 to 2003 Q4.

3.7 Macro Puzzle

I now turn to the aggregate data to show that Indonesia, like so many other countries

that have been studied previously, shows a pattern of excess sensitivity of trade. After

documenting the excess sensitivity of the macro data, in the next section I will turn to

examining the micro data.

Looking first at annual data from 1959-2010, one can see in Figure 3.1, imports and

exports are far more sensitive than domestic production. This phenomenon is not a tempo-

rary or short run phenomenon, as it spans the entire fifty years included in the data. Nor

is there any evidence that the phenomenon is becoming less important over time. In fact,

while the shocks in the early 1970s were particularly large, trade volatility since the 1980s

has been remarkably persistent.

Numerically, I estimate the income elasticity of imports to be 2.10 for the entire sample,

and the income elasticity of exports to be 2.11, as can be seen in Table 3.2 . When I restrict

the sample to post-1980, the income elasticity of imports falls slightly to 1.71, which is

statistically distinct from the entire sample estimate. For exports, the income elasticity is

slightly less sensitive, with an estimate of 1.53. Finally, when I consider only the years

included in the micro data, I estimate the income elasticity of imports to be 1.64 and the

income elasticity of exports to be 1.68. Based on the annual data analyzed here, Indonesia

displays the classic trade elasticity puzzle, with a consistent income elasticity well above

119

one.

The evidence presented thus far is for annual data, which is the frequency at which the

micro firm level data is available. Nonetheless, by looking at shorter frequencies for the

macro data, it becomes apparent that observed excess volatility is not a feature of the time

horizon of the data. I next look at quarterly data from 1993-2002, which covers most of the

time series available for the micro data.5

Figure 3.2 reaffirms that excess sensitivity of trade observed at annual frequencies is

also prevalent at quarterly frequencies. While there is more quarter to quarter volatility in

production relative to annual growth, trade is even more volatile, and consistently demon-

strates excess volatility relative to production throughout the time period being considered.

Numerical results are reported in Table 3.3. From the period 1993 until the end of 1999, the

income elasticity of imports is 1.94, and the income elasticity of exports is 1.23. Over the

longer time series available (1993-2002), the estimated income elasticity of imports is 1.47

and 1.23 for exports. The difference in income elasticity for imports and exports, though

not always statistically significant here, has been observed in earlier studies (see Marquez

(2002a) for references). The higher quarterly frequency tells a qualitatively similar story to

the annual frequency data.

This is not the first study to note that Indonesia suffers from the trade sensitivity

puzzle at the aggregate level. Marquez (2002a) looks at a number of East Asian countries

separately, and he estimates that the income elasticity of imports for Indonesia from 1980 to

1997 is 1.39, roughly consistent with the famous estimate for the U.S. found in Houthakker

and Magee (1969). When I restrict the annual data to the same time window, I estimate

the elasticity of income to be 1.47.

Overall, at the aggregate level, Indonesia displays qualitatively and quantitatively similar

behavior to that which has been found in other countries. Indonesia appears to suffer from

excess trade sensitivity at the macro level. In the next section, I pivot to firm level data to5At present, there is not useful quarterly macro data with necessary price deflators available to include

1990-1993, though there is micro firm level data covering this time period.

120

see if excess trade sensitivity is also a feature of the micro data.

3.8 Micro Evidence

To investigate the puzzle further, I turn next to a firm level investigation. While previous

research has attempted to understand the elasticity puzzle through an aggregate lens, this

paper provides the first evidence at the micro level. By drilling down to the appropriate

theoretical agent and empirical observation, we will be able to see just how deep the elasticity

puzzle runs.

The micro evidence focuses on the manufacturing sector in Indonesian, and in particular,

manufacturing production and imported intermediate inputs. While the micro data does

not align exactly with the macro data studied in the previous section, it does provide a

relevant lens through which to evaluate the underlying theory and can illuminate possibly

explanations of the elasticity puzzle.

Starting with pooled firm observations in column (1) in Table 3.4, the estimated income

elasticity of imports is 1.08, and it is tightly estimated with a standard error of 0.003.

Column (1) controls for both domestic industry prices as well as import prices. The estimate

is significantly different from 1, as with the macro data, though not as large as was found in

the previous section. The inclusion of appropriate import and domestic price indices may

help to explain part of the difference in estimates.

To check the robustness of the estimated income elasticity of imports, I include year

and industry fixed effects, respectively, in columns (2) and (3), and then both in column

(4). It is notable that while the estimated price elasticities of imports are significantly

different once year and industry effects are included, the estimated income elasticity hardly

changes. In column (4), the estimated income elasticity of imports is 1.06, and again it is

tightly estimated with a standard of 0.004. Once more, the estimated elasticity of imports is

statistically distinct from 1, although it is much smaller than what was found in the macro

121

section above.

While the pooled data is insightful, a proper test of the theory would be to focus on the

internal response of firms, rather than the pooled response. Column (5) does this through

the inclusion of plant fixed effects. It is dramatic that once plant level fixed effects are

included, the estimated income elasticity falls to 1.006, and is statistically indistinguishable

from one, although it is again tightly estimated with a standard error of 0.007. The results

are entirely consistent with the theory outlined above, and tell a very different story than

the macro evidence presented earlier. The estimated import and domestic price elasticities

are also consistent with the theory, as it cannot be rejected that the magnitudes of the

estimates are different.

The micro evidence therefore suggests that firm level behavior, at least as it relates

to imported intermediate inputs to production, is wholly consistent with the theoretical

structure presented above. There is no import elasticity puzzle present at the firm level,

which is surprising given that the macro evidence is similar to previously studied countries

and does exhibit the “puzzling” behavior. The results presented here have both constructive

and destructive implications.

On the constructive side, the firm behavioral estimation is perfectly in line with that

predicted by theory. There is no micro (plant) import elasticity puzzle. This implies that

the search for an explanation of macro level behaviors shouldn’t focus on micro frictions.

The results presented here rule out a within firm explanation of the macro volatility.

On the destructive side, the import elasticity puzzle becomes more puzzling since it

disappears at a sufficiently disaggregated unit of observation. The challenge in resolving

the import elasticity puzzle now requires an additional explanation of the lack of an import

elasticity puzzle at the plant level.

122

3.9 Discussion

The twin observations of excessive trade sensitivity at the macro level and the theoret-

ically consistent unit elasticity of income at the micro level requires deeper consideration.

At a basic level, this may simply be a story of aggregation bias. Much of the field of macro

focuses on understanding why results at the aggregate level can appear to be very different

from the micro-economic behavior. Work now moves to understanding the “aggregation

puzzle” of trade elasticities. Some potential explanations to explore in future work are now

considered.

Composition As has been suggested in previous studies, one explanation of the observed

excess sensitivity of trade might be purely compositional. That is, those goods which

are the most volatile over the business cycle may be those that are disproportionately

traded internationally. The previous literature has emphasized the influence of both final

consumption goods and durable goods.

In the present study, firm level import demand of intermediate inputs don’t show excess

sensitivity, but imported intermediates to production may simply be less volatile than final

consumption goods (for example). If the driving force behind the observed excess sensitivity

are types of goods other than intermediate manufacturing imports, this would alter how we

understand the macro sensitivity puzzle. Since most models of import demand are implicitly

or explicitly about intermediates to production, it is important that the excess sensitivity

puzzle is not observed at this level. Excess sensitivity in final consumption goods, for

example, may be an important area for further economic research, but this is a very different

explanation of macro sensitivity than intermediate input volatility.

The firm level data used in the present study is poorly equipped to say much about

the compositional aspect of the aggregation puzzle. Firm import demand of manufacturing

intermediate inputs show no such excess sensitivity. Alas, the data does not include infor-

123

mation on imported final goods (or services) which might be more volatile in terms of trade

than production.6

Durable and non-durable goods, which have been another proposed dimension of the

compositional effect, can be explored further using this firm level data. Comparing differ-

ences between durable and non-durable industries using the micro data may shed light on

the relative importance of this channel for understanding the macro elasticities observation.

There is, however, an unanswered question as to whether it is final demand for durables, as

opposed to durable intermediates, that matters for generating excess sensitivity. Using in-

dustry classifications for the main product as well as input-output matrices for intermediate

inputs, the current data set will allow for a deeper investigation of this potential channel.

Fragmentation A second proposed explanation for the excess sensitivity of trade has

been increased fragmentation of the production process. In this view, the fragmentation

of production has increased trade as the production process is broken down into smaller

units and re-located across space (and borders). For an illustrative example, holding output

constant, the fragmentation of the production process would lead to growth in international

trade with no effect on production (by assumption). Increasing fragmentation could then

be a key driver of the excess sensitivity puzzle.

The evidence presented here is not supportive of this explanation. The focus on inter-

mediate input demand in the fragmentation hypothesis is not supported by the data. There

is no within-firm evidence of excess sensitivity of imports. Instead, while the fragmentation

process may be important across countries, there is no evidence that within-firm behavior

is consistent with this type of fragmentation story. Firms appear to import intermediate

varieties and produce output at constant budget shares.

An alternative possibility, related to the fragmentation hypothesis, is that fragmentation

shows up across firms within a country, but not within firms. That is, rather than an6Indonesian GDP by sectors are not presently available electronically for the years under consideration

in this paper.

124

intensive margin adjustment generating excess sensitivity of trade, an extensive margin

adjustment might be a source of excess sensitivity. Firms may enter and exit the import

market, which could generate excessive trade elasticities at higher levels of aggregation.

The preliminary evidence presented here suggests that trade elasticities are higher when

measured across firms rather than within firms. While the estimated income elasticity

using pooled firm data was well below the macro estimates, it was higher than the within

firm estimates. This suggests that the extensive margin has some role to play in explaining

the “aggregation puzzle” of estimated trade elasticities.

Furthermore, the current estimation procedure has trouble dealing with zeros in the

data. Zeros at the firm level are a challenge for estimating import elasticity at the micro

level, but not a problem at higher levels of aggregation. In future work, accounting for firm

switching into and out of the import market could help reconcile macro and micro trade

elasticity estimates.

Gross Value versus Value-added Related to the fragmentation explanation is the ob-

servation that GDP and trade are distinction concepts, as GDP is measure of value-added

in production while trade measures gross flows. For simplicity, imagine a good produced

in one country that is exported back and forth multiple times between two countries. This

would show up in the data as significant gross flows of trade, while the value added in pro-

duction would be relatively minor. Gross trade flows can be misleading when intermediate

inputs are a significant component of international trade and there is little value added of

that trade (e.g. Johnson and Noguera (2012)). The suggestion is that the comparison of

value-added production and gross trade flows is inappropriate. The current data will allow

for a greater investigation of the importance of gross versus net flows since firm level data

includes not only information on revenues and production, but also on value-added during

production.

125

3.10 Conclusion

While there has been an extensive literature investigating the import elasticity puzzle,

this paper is the first to directly study the phenomenon at both the macro and micro levels.

First, I discuss the history of the trade elasticity puzzle, and why in fact this feature of the

data is so puzzling.

Second, I document that the import elasticity puzzle is apparent in Indonesia when using

national accounts data. Income elasticity of imports is estimated to be above 1.5 for the

fifty years worth of data starting in 1959. This estimate is robust to different time periods

and at both higher and lower frequencies of data (quarterly and annual). Indonesia, like

many other countries studied previously, suffers from an excess sensitivity of trade.

Next, I consider import demand at the level of the firm. The theoretical underpinning

of the excess sensitivity of trade is most appropriately applied to the firm, particularly for

intermediate inputs to production. Most models that study the import elasticity puzzle

either explicitly or implicitly consider the importing behavior of a firm, making the firm

the natural unit of observation for evaluating the theory. I estimate the income elasticity of

imports for manufacturing firms in Indonesia to be 1, precisely as predicted by the model,

and significantly different from the macro estimates. There is no apparent import elasticity

puzzle at the firm level.

Finally, I evaluate current theories that attempt to explain the excess sensitivity of trade

in light of this new evidence. The puzzle appears to be more accurately described as an

aggregation puzzle. At present, neither fragmentation nor compositional stories are con-

vincing explanations of this aggregation puzzle. Further work will explore these arguments

more thoroughly given the new micro evidence.

While the results presented here pose an intriguing question as to why Indonesia shows

an elasticity puzzle at the aggregate level, but no such puzzle at the micro level, it may

be the case that this behavior is unique to this particular data set. Future work should

126

verify that the apparent aggregation puzzle identified here is robust to additional countries.

Indonesia displays the classic import elasticity puzzle at the macro level, but there may be

idiosyncratic microeconomic reasons why no import elasticity puzzle is observed at the firm

level. Verifying this aggregation puzzle in more countries would provide useful information

about the underlying causes of excessively sensitive trade.

127

3.11 Figures and Tables

Figure 3.1: Real GDP and Trade Growth, 1959-2010 (percent changes year to year)Source: International Financial Statistics (IFS)

Figure 3.2: Real GDP and Trade Growth, 1993Q1-2003Q4 (percent changes quarter toquarter)

Source: Bank of Indonesia

128

Values (in 1,000 Rupiah) Obs Mean Std. Dev Min MaxTotal Output 38831 36426.74 192319 6.02 1.22E+07

Intermediate Inputs 38831 22378.86 106592 0.48 7856573Total Raw Materials 38831 18335.86 93672 0.15 7032410

Domestic Raw Materials 38831 9057.29 61595 0.00 5241484Imported Raw Materials 38831 9278.58 48373 0.01 2208281

Export Sales 38831 7667.15 57702 0.00 3997247

Table 3.1: Summary Statistics, All Importing Firms

Annual Frequency 1959-2010 Post-1980 1990-1999

Income Elasticity of Imports 2.10 1.71 1.64(0.047) (0.098) (0.153)

Income Elasticity of Exports 2.11 1.54 1.68(0.064) (0.094) (0.132)

N 53 31 10

Table 3.2: Income Elasticity, Annual Frequency, 1959-2010Note: Regressions use the log of real gdp as the explanatory variable. Standard errors are given in paran-thesis. Data comes from the International Financial Statistics (IMF) database.

Quarterly Frequency 1993-2002 1993-1999

Income Elasticity of Imports 1.47 1.940.322 0.360

Income Elasticity of Exports 1.24 1.230.232 0.310

N 40 28

Table 3.3: Income Elasticity, Quarterly Frequency, 1993-2002Note: Regressions use the log of real gdp as the explanatory variable. Standard errors are given in paran-thesis. Data comes from the Bank of Indonesia.

129

Elasticities (1) (2) (3) (4) (5)

Income 1.08 1.08 1.06 1.06 1.006(0.003) (0.004) (0.004) (0.004) (0.007)

Import Prices -0.74 -0.66 -0.96 -0.97 -1.00(0.008) (0.01) (0.02) (0.05) (0.03)

Domestic Industry Prices 0.84 1.23 1.02 1.05 0.94(0.02) (0.03) (0.04) (0.08) (0.05)

Year FE No Yes No Yes YesIndustry FE No No Yes Yes Yes

Plant FE No No No No YesN 38,831 38,831 38,831 38,831 38,831

Table 3.4: Import Demand Elasticities, Plant Level, 1990-1999 (Annual)

References

Afonso, Gara, Anna Kovner, and Antoinette Schoar, “Stressed, Not Frozen: The

Federal Funds Market in the Financial Crisis,” The Journal of Finance, 2011, 66 (4),

1109–1139.

Ahn, JaeBin, “A Theory of Domestic and International Trade Finance,” IMF Working

Paper, 2011, (11/262). 2.1, 3.3

, Amit K. Khandelwal, and Shang-Jin Wei, “The role of intermediaries in facilitating

trade,” Journal of International Economics, 2011, 84 (1), 73 – 85. 3.3

and Alexander McQuoid, “Capacity Constrained Exporters: Micro Evidence and

Macro Implications,” 2012. 4

, Mary Amiti, and David Weinstein, “Trade Finance and the Great Trade Collapse,”

American Economic Review, 2011, 101 (3), 298–302. 3.3

Alesina, Alberto and Eliana La Ferrara, “Participation in Heterogeneous Communi-

ties,” The Quarterly Journal of Economics, 2000, 115 (3), pp. 847–904.

and , “Ethnic Diversity and Economic Performance,” Journal of Economic Literature,

2005, 43 (3), pp. 762–800. 1.2, 1.3

and Enrico Spolaore, “On the Number and Size of Nations,” Quarterly Journal of

Economics, 1997, 112 (4), 1027–1056. 1.3

130

131

, Arnaud Devleeschauwer, William Easterly, Sergio Kurlat, and Romain

Wacziarg, “Fractionalization,” Journal of Economic Growth, 2003, 8 (2), pp. 155–194.

, Edward Glaeser, and Bruce Sacerdote, “Why Doesn’t the United States Have a

European-Style Welfare State?,” Brookings Papers on Economic Activity, 2001, (2), pp.

187–254. 1.2

, Reza Baqir, and Caroline Hoxby, “Political Jurisdictions in Heterogeneous Com-

munities,” Journal of Political Economy, 2004, 112 (2), pp. 348–396. 1.2

, , and William Easterly, “Public Goods and Ethnic Divisions,” The Quarterly

Journal of Economics, 1999, 114 (4), pp. 1243–1284. 1.2, 1.6, 1.8

Alessandria, George, Joseph Kaboski, and Virgiliu Midrigan, “The Great Trade

Collapse of 2008-09: An Inventory Adjustment?,” IMF Economic Review, 2010, 58, 254–

294. 3.3

Amiti, Mary and David E. Weinstein, “Exports and Financial Shocks,” The Quarterly

Journal of Economics, 2011, 126 (4), 1841–1877. 3.3

and Jozef Konings, “Trade Liberalization, Intermediate Inputs, and Productivity:

Evidence from Indonesia,” The American Economic Review, 2007, 97 (5), pp. 1611–1638.

10, 4

Arkolakis, Costas, “Market Penetration Costs and the New Consumers Margin in Inter-

national Trade,” Journal of Political Economy, 2010, 118 (6), 1151 – 1199. 7

, Jonathan Eaton, and Samuel Kortum, “Staggered Adjustment and Trade Dynam-

ics,” November 2011. 2.1

Aw, Bee Yan, Mark Roberts, and Daniel Xu, “R & D Investment, Exporting, and

Productivity Dynamics,” American Economic Review, 2011, 101 (4), 1312–44. 2.1, 5, 2.5,

2.5.1, 2.5.2, 2.6

132

Barten, A. P., “Consumer Demand Functions under Conditions of Almost Additive Pref-

erences,” Econometrica, 1964, 32 (1/2), pp. 1–38. 3.4

Bems, Rudolfs, Robert Johnson, and Kei-Mu Yi, “Demand Spillovers and the Col-

lapse of Trade in the Global Recession,” IMF Economic Review, 2010, 58, 295–326. 3.3

Berman, Nicolas, Antoine Berthou, and Jerome Hericourt, “Export Dynamics and

Sales at Home,” August 2011. 2.1

Bernard, Andrew B., Stephen J. Redding, and Peter K. Schott, “Multiple-Product

Firms and Product Switching,” American Economic Review, March 2010, 100 (1), 70–97.

2.2

Bertrand, Marianne, Esther Duflo, and Sendhil Mullainathan, “How Much Should

We Trust Differences-in-Differences Estimates?,” Quarterly Journal of Economics, 2004,

119 (1), 249–275. 1.8.1, 1.9.2

Blalock, Garrick and Paul J. Gertler, “Learning from exporting revisited in a less

developed setting,” Journal of Development Economics, 2004, 75 (2), 397 – 416. 10, 4

and , “Welfare gains from Foreign Direct Investment through technology transfer to

local suppliers,” Journal of International Economics, 2008, 74 (2), 402 – 421. 10, 4

, , and David I. Levine, “Financial constraints on investment in an emerging market

crisis,” Journal of Monetary Economics, 2008, 55 (3), 568 – 591. 2.4

Blum, Bernardo, Sebastain Claro, and Ignatius Horstmann, “Occasional vs Peren-

nial Exporters: The Impact of Capacity on Export Mode,” May 2011. 2.1, 2.2

Bossert, Walter, Conchita D’Ambrosio, and Eliana La Ferrara, “A Generalized

Index of Fractionalization,” Economica, 2011, 78 (312), 723–750. 3

Boustan, Leah Platt, “Was Postwar Suburbanization "White Flight"? Evidence from the

Black Migration,” The Quarterly Journal of Economics, 2010, 125 (1), 417–443. 1.2

133

Brender, Adi, “The effect of fiscal performance on local government election results in

Israel: 1989–1998,” Journal of Public Economics, 2003, 87 (9-10), 2187 – 2205. 1.2

, “Ethnic Segregation and the Quality of Local Government in the Minority’s Localities:

Local Tax Collection in the Israeli-Arab Municipalities as a Case-Study,” SSRN eLibrary,

2005. 1.2

Broda, Christian and David Weinstein, “Globalization and Gains from Variety,” Quar-

terly Journal of Economics, 2006, 121 (2). 12

Buera, Francisco J. and Benjamin Moll, “Aggregate Implications of a Credit Crunch,”

Working Paper 17775, National Bureau of Economic Research January 2012. 2.1

and Yongseok Shin, “Financial Frictions and the Persistence of History: A Quantitative

Exploration,” Working Paper 16400, National Bureau of Economic Research September

2010. 2.1

, Joseph P. Kaboski, and Yongseok Shin, “Finance and Development: A Tale of

Two Sectors,” American Economic Review, 2011, 101 (5), 1964–2002. 2.1

Card, David, “The Impact of the Mariel Boatlift on the Miami Labor Market,” Industrial

and Labor Relations Review, 1990, 43 (2), pp. 245–257.

, “Immigrant Inflows, Native Outflows, and the Local Labor Market Impacts of Higher

Immigration,” Journal of Labor Economics, 2001, 19 (1), pp. 22–64. 1.2

Cascio, Elizabeth and Ethan Lewis, “Cracks in the Melting Pot: Immigration, Social

Choice, and Segregation,” 2010.

Chaney, Thomas, “Liquidity Constrained Exporters,” July 2005. 2.1

Das, Sanghamitra, Mark J. Roberts, and James R. Tybout, “Market Entry Costs,

Producer Heterogeneity, and Export Dynamics,” Econometrica, 2007, 75 (3), 837–873.

2.1, 5

134

Deaton, Angus and John Muellbauer, “An Almost Ideal Demand System,” The Amer-

ican Economic Review, 1980, 70 (3), pp. 312–326. 3.4

E., Jr. Lucas Robert, “On the Size Distribution of Business Firms,” The Bell Journal of

Economics, 1978, 9 (2), pp. 508–523. 2.2

Easterly, William, “Can Institutions Resolve Ethnic Conflict?,” Economic Development

and Cultural Change, 2001, 49 (4), pp. 687–706. 1.2, 1.10

and Ross Levine, “Africa’s Growth Tragedy: Policies and Ethnic Divisions,” The

Quarterly Journal of Economics, 1997, 112 (4), pp. 1203–1250. 1.2

Elazar, Daniel Judah and Chaim Kalchheim, Local Government in Israel, Lanham:

University Press of America, 1988. 1.4

Engel, Charles and Jian Wang, “International trade in durable goods: Understanding

volatility, cyclicality, and elasticities,” Journal of International Economics, 2011, 83 (1),

37 – 52. 3.3

Esteban, Joan and Debraj Ray, “Linking Conflict to Inequality and Polarization,” The

American Economic Review, 2011, 101 (4), 1345–1374. 3

Fanelli, Jose Maria, Ricardo N. Bebczuk, and Juan J. Pradelli, “Determinants and

Consequences of Financial Constraints Facing Firms in Argentina,” RES Working Papers

3147, Inter-American Development Bank, Research Department 2002. 2.4

Fazzari, Steven M., R. Glenn Hubbard, Bruce C. Petersen, Alan S. Blinder,

and James M. Poterba, “Financing Constraints and Corporate Investment,” Brookings

Papers on Economic Activity, 1988, 1988 (1), pp. 141–206. 2.1

Feenstra, Robert and Clinton Shiells, “Bias in U.S. Import Prices and Demand,” 1994.

3.3

135

Freund, Caroline, “The Trade Response to Global Downturns: Historical Evidence,” Pol-

icy Research Working Paper Series 5015, The World Bank August 2009. 3.1, 2

Friedberg, Rachel M., “The Impact of Mass Migration on the Israeli Labor Market,”

Quarterly Journal of Economics, 2011/09/27 2001, 116 (4), 1373–1408. 1.2

Friedman, Milton, Essays in Positive Economics, Chicago: University of Chicago Press,

1953. 3.2

Gandal, Neil, Gordon H. Hanson, and Matthew J. Slaughter, “Technology, Trade,

and Adjustment to Immigration in Israel,” European Economic Review, 2004, 48 (2), 403

– 428. 1.2

Ganim, As’ad, Ethnic Politics in Israel: The Margins and the Ashkenazi Center, Vol. 16,

London: Routledge, 2010.

Glaeser, Edward L., David I. Laibson, José A. Scheinkman, and Christine L.

Soutter, “Measuring Trust,” Quarterly Journal of Economics, 2000, 115 (3), 811–846.

, Jose A. Scheinkman, and Andrei Shleifer, “Economic Growth in a Cross-section

of Cities,” Journal of Monetary Economics, 1995, 36 (1), 117 – 143.

Houthakker, H. S. and Stephen P. Magee, “Income and Price Elasticities in World

Trade,” The Review of Economics and Statistics, 1969, 51 (2), pp. 111–125. 3.1, 3.2, 3.3,

3.7

Hoxby, Caroline M., “Does Competition Among Public Schools Benefit Students and

Taxpayers?,” The American Economic Review, 2000, 90 (5), pp. 1209–1238.

Israel Social Sciences Data Center at the Hebrew University, Labor Force Survey

1985 No. 0560 Israeli Central Bureau of Statistics Jerusalem, Israel 1986.

, Labor Force Survey 1986 No. 0559 Israeli Central Bureau of Statistics Jerusalem, Israel

1987.

136

, Labor Force Survey 1987 No. 0558 Israeli Central Bureau of Statistics Jerusalem, Israel

1988.

, Labor Force Survey 1988 No. 0557 Israeli Central Bureau of Statistics Jerusalem, Israel

1989.

, Labor Force Survey 1989 No. 0556 Israeli Central Bureau of Statistics Jerusalem, Israel

1990.

, Labor Force Survey 1990 No. 0555 Israeli Central Bureau of Statistics Jerusalem, Israel

1991.

, Labor Force Survey 1991 No. 0554 Israeli Central Bureau of Statistics Jerusalem, Israel

1992.

, Immigrant Absorption Survey 1972-1973 No. 0333 Israeli Central Bureau of Statistics

Jerusalem, Israel 1993.

, Labor Force Survey 1992 No. 0553 Israeli Central Bureau of Statistics Jerusalem, Israel

1993.

, Labor Force Survey 1993 No. 0552 Israeli Central Bureau of Statistics Jerusalem, Israel

1994.

, Employment Survey of USSR Immigrants Who Arrived in Israel Between Oct. - Dec.

1990 No. 0527 Israeli Central Bureau of Statistics Jerusalem, Israel 1995.

Israeli Central Bureau of Statistics, Local Authorities In Israel, Financial Data

(1984/1985) Special Series No. 809 Jerusalem, Israel 1987.

, Local Authorities In Israel, Physical Data (1984/1985) Special Series No. 798 Jerusalem,

Israel 1987.

137

, Local Authorities In Israel, Financial Data (1985/1986) Special Series No. 830

Jerusalem, Israel 1988.

, Local Authorities In Israel, Physical Data (1985/1986) Special Series No. 816 Jerusalem,

Israel 1988.

, Local Authorities In Israel, Financial Data (1986/1987) Special Series No. 860

Jerusalem, Israel 1989.

, Local Authorities In Israel, Physical Data (1986/1987) Special Series No. 841 Jerusalem,

Israel 1989.

, Local Authorities In Israel, Financial Data (1987/1988) Special Series No. 887

Jerusalem, Israel 1990.

, Local Authorities In Israel, Physical Data (1987/1988) Special Series No. 872 Jerusalem,

Israel 1990.

, Local Authorities In Israel, Physical Data (1988/89) Special Series No. 890 Jerusalem,

Israel 1991.

, Local Authorities In Israel, Financial Data (1988/1989) Special Series No. 904

Jerusalem, Israel 1992.

, Local Authorities In Israel, Financial Data (1989/1990) Special Series No. 922

Jerusalem, Israel 1992.

, Local Authorities In Israel, Physical Data (1990) Special Series No. 908 Jerusalem, Israel

1992.

, Local Authorities In Israel, Financial Data (1990/1991) Special Series No. 950

Jerusalem, Israel 1993.

138

, Local Authorities In Israel, Physical Data (1991) Special Series No. 931 Jerusalem, Israel

1993.

, Local Authorities In Israel, Financial Data (1991) Special Series No. 986 Jerusalem,

Israel 1994.

, Local Authorities In Israel, Physical Data (1992) Special Series No. 966 Jerusalem, Israel

1994.

, Local Authorities In Israel, Financial Data (1992) Special Series No. 1014 Jerusalem,

Israel 1995.

, Local Authorities In Israel, Physical Data (1993) Special Series No. 992 Jerusalem, Israel

1995.

, Local Authorities In Israel, Financial Data (1993) Special Series No. 1050 Jerusalem,

Israel 1996.

Johnson, Robert C. and Guillermo Noguera, “Accounting for Intermediates: Produc-

tion Sharing and Trade in Value Added,” Journal of International Economics, 2012, 86

(2), 224 – 236. 3.9

Justman, Moshe and Avia Spivak, “Socioeconomic Dynamics of Local Authorities in

Israel,” Israel Economic Review, 2004, 2 (1), 1–27. 1.2

Kaplan, Steven N. and Luigi Zingales, “Do Investment-Cash Flow Sensitivities Provide

Useful Measures of Financing Constraints?,” The Quarterly Journal of Economics, 1997,

112 (1), pp. 169–215. 2.4

Kohn, David, Fernando Leibovici, and Michal Szkup, “Financial Frictions and New

Exporter Dynamics,” January 2012. 2.1

Krugman, Paul, “Scale Economies, Product Differentiation, and the Pattern of Trade,”

The American Economic Review, 1980, 70 (5), pp. 950–959. 2.1, 5

139

Krugman, Paul R., “Increasing returns, monopolistic competition, and international

trade,” Journal of International Economics, 1979, 9 (4), 469 – 479. 2.1, 5

Levchenko, Andrei, Logan Lewis, and Linda Tesar, “The Collapse of International

Trade During the 2008-2009 Crisis: In Search of the Smoking Gun,” 2010. 3.3

Levinsohn, James and Amil Petrin, “Estimating Production Functions Using Inputs to

Control for Unobservables,” The Review of Economic Studies, 2003, 70 (2), 317–341. 14,

2.4, 2.5.2

Lin, Chen, Yue Ma, and Yuhai Xuan, “Ownership structure and financial constraints:

Evidence from a structural estimation,” Journal of Financial Economics, 2011, 102 (2),

416 – 431. 2.4

Loecker, Jan De, “Recovering markups from production data,” International Journal of

Industrial Organization, May 2011, 29 (3), 350–355. 13

Manova, Kalina, “Credit Constraints, Heterogeneous Firms, and International Trade,”

October 2011. 2.1

, Shang-Jin Wei, and Zhiwei Zhang, “Firm Exports and Multinational Activity Under

Credit Constraints,” March 2011. 2.4

Marquez, Jaime, Estimating Trade Elasticities, Boston, Mass. USA: Kluwer Academic

Publishers, 2002. 3.2, 3, 3.4, 3.7

, “The Puzzling Income Elasticity of US Imports,” 2002. 3.3

Melitz, Marc J., “The Impact of Trade on Intra-Industry Reallocations and Aggregate

Industry Productivity,” Econometrica, 2003, 71 (6), pp. 1695–1725. 2.1, 5

Midrigan, Virgiliu and Daniel Yi Xu, “Finance and Misallocation: Evidence from

Plant-level Data,” Working Paper 15647, National Bureau of Economic Research January

2010. 2.1

140

Miguel, Edward and Mary Kay Gugerty, “Ethnic Diversity, Social Sanctions, and

Public Goods in Kenya,” Journal of Public Economics, 2005, 89 (11-12), 2325 – 2368. 2

Mobarak, Ahmed Mushfiq and Denni Puspa Purbasari, “Corrupt Protection For

Sale to Firms: Evidence from Indonesia,” April 2006. 10, 4

Munshi, Kaivan, “Networks in the Modern Economy: Mexican Migrants in the U. S.

Labor Market,” The Quarterly Journal of Economics, 2003, 118 (2), pp. 549–599. 1.2

Muûls, Mirabelle, “Exporters and Credit Constraints: A Firm-Level Approach,” Septem-

ber 2008. 2.1

Navon, Guy, “Budgetary Dynamics in the Local Authorities in Israel,” Israel Economic

Review, 2006, 4 (2), 19–52. 1.2

Nguyen, Daniel and Georg Schaur, “Cost Linkages Trnsmit Volatility Across Markets,”

May 2011. 2.1

Olley, G. Steven and Ariel Pakes, “The Dynamics of Productivity in the Telecommu-

nications Equipment Industry,” Econometrica, 1996, 64 (6), pp. 1263–1297. 2.5.2

Paravisini, Daniel, Veronica Rappoport, Philipp Schnabl, and Daniel Wolfenzon,

“Dissecting the Effect of Credit Supply on Trade: Evidence from Matched Credit-Export

Data,” 2011. 2.1

Poterba, James M., “Demographic Change, Intergenerational Linkages, and Public Edu-

cation,” The American Economic Review, 1998, 88 (2), pp. 315–320.

Rafael, Eliezer Ben and Stephen Sharot, Ethnicity, Religion, and Class in Israeli

Society, Cambridge: Cambridge University Press, 1991. 1.7.1

Rho, Young-Woo and Joel Rodrigue, “Firm-Level Investment and Export Dynamics,”

January 2012. 2.1, 3

141

Ruhl, Kim and Jonathan Willis, “New Exporter Dynamics,” November 2008. 2.1

Ruhl, Kim J., “The International Elasticity Puzzle,” 2008. 2.1, 2.5.3

Schmelz, U. O., Sergio Della Pergola, and Uri Avner, Ethnic Differences Among

Israeli Jews: A New Look, Vol. No. 22, Jerusalem: Institute of Contemporary Jewry,

Hebrew University of Jerusalem, 1991.

Sethupathy, Guru, “Does Exporting Lead to Productivity Spillovers in Horizontal or

Vertical Industries? Evidence from Indonesia,” August 2008. 10, 4

Shapira, Anita, Israeli Identity in Transition, Westport, Conn.: Praeger Publishers, 2004.

Smooha, Sammy, Israeli Democracy Under Stress, Boulder: Lynne Rienner Publishers,

1993. 1.7.1

Soderbery, Anson, “Market Size, Structure, and Access: Trade with Capacity Con-

straints,” October 2011. 2.1, 16

Spearot, Alan, “Firm Heterogeneity, New Investment, and Acquisitions,” Journal of In-

dustrial Economics, Forthcoming. 2.1

Theil, Henri, “The Information Approach to Demand Analysis,” Econometrica, 1965, 33

(1), pp. 67–87. 3.4

Tiebout, Charles M., “A Pure Theory of Local Expenditures,” Journal of Political Econ-

omy, 1956, 64 (5), pp. 416–424. 1.2, 1.8.1

Urquiola, Miguel, “Does School Choice Lead to Sorting? Evidence from Tiebout Varia-

tion,” The American Economic Review, 2005, 95 (4), pp. 1310–1326.

Vannoorenberghe, G., “Firm-level volatility and exports,” Journal of International Eco-

nomics, 2012, 1. 2.1

142

Verhoogen, Eric A., “Trade, Quality Upgrading, and Wage Inequality in the Mexican

Manufacturing Sector,” The Quarterly Journal of Economics, 2008, 123 (2), 489–530. 9

Vigdor, Jacob L., “Community Composition and Collective Action: Analyzing Initial

Mail Response to the 2000 Census,” Review of Economics and Statistics, 2004, 86 (1),

303–312.

Whited, Toni M., “Debt, Liquidity Constraints, and Corporate Investment: Evidence

from Panel Data,” The Journal of Finance, 1992, 47 (4), pp. 1425–1460.

and Guojun Wu, “Financial Constraints Risk,” Review of Financial Studies, 2006, 19

(2), 531–559. 2.4

Wooldridge, Jeffrey M., Econometric Analysis of Cross Section and Panel Data, Cam-

bridge, Mass.: MIT Press, 2002. 1.9.2

Yi, Kei-Mu, “Can Vertical Specialization Explain the Growth of World Trade?,” Journal

of Political Economy, 2003, 111 (1), pp. 52–102. 3.3

Appendix A

Appendix: Capacity Constrained Exporters:

Micro Evidence and Macro Implications

A.1 Underlying Model for Welfare Loss Evaluation

This section provides an underlying model framework that is used to quantify the welfare

loss from capacity constraints in Section 2.5.4. We consider a particular upper-tier utility

function:

U = CαdC

1−αimp ,

which has the corresponding total aggregate price index expressed as:

P = Pαd P

1−αimp ,

where Pαd is the aggregate price index for domestic goods and P 1−α

imp is the aggregate price

index for imported goods, defined respectively as:

Pd = [�

i�imp

(pdi )1−σd ]

11−σd ],

and

143

144

Pimp =

��

i∈imp

�pdi�1−σd

� 11−σd

This utility system implies that a constant fraction, α, of total spending is devoted to

domestic goods, irrespective of relative price level of domestic goods to imported goods.1

We can further expand the aggregate price index for domestic goods by distinguishing

the goods produced by constrained firms from those by unconstrained firms:

Pd =

��

i∈dom

�pdi�1−σd

� 11−σd

=

��

i∈unconstrained

�pdi�1−σd +

�

i∈constrained

�pdi�1−σd

� 11−σd

=

��

i∈unconstrained

�σd

σd − 1MCi

�1−σd

+�

i∈constrained

�pdi�1−σd

� 11−σd

The last expression reflects that unconstrained firms charge optimal prices, which is simply

the markup over marginal costs, whereas constrained firms do not. We also construct a

hypothetical domestic price index that would have been obtained if constrained firms could

have charged optimal prices:

P hypd =

��

i∈unconstrained

�σd

σd − 1MCi

�1−σd

+�

i∈constrained

�σd

σd − 1MCi

�1−σd� 1

1−σd

Welfare loss from capacity constrained domestic producers is then calculated by comparing

the hypothetical and the actual domestic goods price index, weighted by the domestic goods

consumption share α : −d lnP = α�lnP hyp

d − lnPd

�

1This in turn implies that the aggregate demand level for domestic goods, Φdt , in equation (2.5.1) is

expressed as Φdt = αRd

t (Pd)σd , where Rd

t is the total spending in the domestic economy.