Privatization and Economic Growth

This article was downloaded by:[Hiroshima University]On: 25 July 2008Access Details: [subscription number 789277225]Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Journal of Development StudiesPublication details, including instructions for authors and subscription information:http://www.informaworld.com/smpp/title~content=t713395137

Privatisation and economic growth in developingcountriesPaul Cook a; Yuichiro Uchida aa Centre on Regulation and Competition, Institute for Development Policy andManagement, University of Manchester, UK.

Online Publication Date: 01 August 2003

To cite this Article: Cook, Paul and Uchida, Yuichiro (2003) 'Privatisation andeconomic growth in developing countries', Journal of Development Studies, 39:6,121 — 154

To link to this article: DOI: 10.1080/00220380312331293607URL: http://dx.doi.org/10.1080/00220380312331293607

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf

This article maybe used for research, teaching and private study purposes. Any substantial or systematic reproduction,re-distribution, re-selling, loan or sub-licensing, systematic supply or distribution in any form to anyone is expresslyforbidden.

The publisher does not give any warranty express or implied or make any representation that the contents will becomplete or accurate or up to date. The accuracy of any instructions, formulae and drug doses should beindependently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings,demand or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with orarising out of the use of this material.

http://www.informaworld.com/smpp/title~content=t713395137

http://dx.doi.org/10.1080/00220380312331293607

http://www.informaworld.com/terms-and-conditions-of-access.pdf

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

Privatisation and Economic Growth inDeveloping Countries

PAUL COOK and YUICHIRO UCHIDA

This article re-examines the relation between privatisation andeconomic growth. Previous studies that have attempted to measurethis relationship have concluded that privatisation has had asizeable positive effect on economic growth. Our study uses datafor 63 developing countries over the time period 1988–97. It usesthe framework of an extreme-bounds analysis (EBA) to conduct across-country growth regression analysis. Our findings contradictearlier results, but reaffirm the view that effective competition andits regulation may need to accompany privatisation to make apositive impact on economic growth.

I . INTRODUCTION

Privatisation accelerated in the 1990s. In Europe this was sparked by theliberalisation of markets at the European Union level and budgetaryconstraints faced by government [Parker, 1999]. Privatisation particularlygained momentum in the late 1980s and spread to a wide range ofdeveloping economies. Over the last decade a significant proportion ofprivatisation transactions have been in developing economies and haveentailed sales of public utilities. Privatisation transactions for the utilitiessector have accounted for over a third of all transactions in developingeconomies since 1988. In the period 1988–93 the value of sales forinfrastructure industries amounted to US$30 billion, compared to US$78billion for all privatisation transactions in developing economies [WorldBank, 1995a, Kikeri, 1998].

There is now sufficient time lapse since privatisation was firstimplemented in developing countries for the results to be evaluated at themacroeconomic level. Admittedly, privatisation proceeded in the 1980s

Paul Cook and Yuichiro Uchida, Centre on Regulation and Competition, Institute forDevelopment Policy and Management, University of Manchester, UK; correspondence: [email protected].

The Journal of Development Studies, Vol.39, No.6, August 2003, pp.121–154PUBLISHED BY FRANK CASS, LONDON

396jds05.qxd 29/08/03 14:13 Page 121

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

without much knowledge of either its impact or contribution to economicgrowth. Instead greater attention was paid to measuring the extent ofprivatisation [Bennell, 1997]. This situation is changing and more recentcontributions assessing post-privatisation performance are reviewed inMegginson and Netter [2001]. Among those most frequently cited are:Galal, Jones, Tandon and Vogelsang [1994]; Megginson, Nash and VanRandenborgh [1994] and Boubakri and Cosset [1998]. However, thesestudies have largely concentrated on financial measures of performance,and have incorporated industrialised countries and higher incomedeveloping countries in their relatively small databases. In recent yearsthere have also been a few attempts to measure the direct impact ofprivatisation on economic growth in a cross-country context [Plane, 1997;Barnett, 2000]. These studies have concluded that privatisation has had asizeable positive effect on economic growth.

The purpose of this analysis is to re-examine the relation betweenprivatisation and economic growth, using data for 63 developing countriesover the period 1988–97. The next section provides a brief review of thetheoretical links between a change in ownership from public to private andeconomic growth, and of the empirical literature that has examined thisrelationship. The third section discusses the methods adopted to explore therelation between privatisation and economic growth and the data used. Thefourth section discusses the results of the analysis. The final sectionprovides a summary and draws some broad conclusions.

II . THEORY AND EMPIRICAL ANALYSIS

Policy-makers in developing economies have often set a broader agenda forprivatisation than the efficiency and resource allocation objectives that wereimplicit in the policy conditions of structural adjustment programmes. Themotives for privatisation have encompassed improved fiscal equity anddistributional performance, although the importance attached to each hasvaried between and within countries over time [UNCTAD, 1996; Yarrow,1999]. Nevertheless, the link between privatisation and economic growthrelates most directly to the microeconomic theories used to justifyprivatisation. At the heart of the debate are theoretical perspectives on theownership issue drawn from property rights theory, public choice theoryand principal agent analysis [Alchian, 1965; Tullock, 1965; Jensen andMeckling, 1976]. By the end of the 1970s these theories were influencingattitudes towards public ownership among policy-makers in developed anddeveloping countries [Cook and Kirkpatrick, 1988; Martin and Parker,1997]. The key theoretical elements underpinning the argument for achange in ownership from public to private related, first, to the view that

122 THE JOURNAL OF DEVELOPMENT STUDIES

396jds05.qxd 29/08/03 14:13 Page 122

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

public ownership led to the pursuit of objectives that detracted fromeconomic welfare maximisation [Boycko, Shleifer and Vishny, 1996].Secondly, an ownership change could improve economic performance bychanging the mechanisms through which different institutionalarrangements affect the incentives for managing enterprises [Vickers andYarrow, 1988; Laffont and Tirole, 1991; Cook and Fabella, 2002].

These arguments are linked to presumptions concerning the condition ofpublicly-owned enterprises before they are privatised. A typical viewpresented publicly-owned enterprises as overextended and poor performers[Kikeri, Nellis and Shirley, 1994]. In this situation publicly-ownedenterprises crowded out private enterprises in their access to credit anderected statutory barriers to preserve the monopoly status of publicly-ownedenterprises. It was argued that the net effect of a change in ownership frompublic to private would be improved economic efficiency, and over time anincrease in investment.

If privatisation was sufficiently extensive and had efficiency-inducingeffects, the contribution of improved performance could be detected at themacroeconomic level. Privatisation would reduce crowding out and providemore credit to the private sector. It would increase the opportunities forinvestment in newly privatised enterprises by releasing them from thecapital constraints previously faced under public ownership. A change inownership would increase efficiency by introducing changes to thegovernance mechanisms and structure of incentives facing employees.

Attempting to measure the contribution of an ownership change oneconomic growth is complicated by the fact that economic performance islikely to be affected by factors that affect the wider economic environmentin which privatised enterprises operate. Privatisation is often accompaniedin developing countries by changes in economic policies that affecteconomic growth. Significant attention has focussed on the process ofderegulation and the importance of competition and its relation to economicefficiency. Vickers and Yarrow [1988] have argued that competition and itsregulation are crucial for the improvement of efficiency in privatisedenterprises. Unravelling the separate effects of policy changes and degreesof competition is difficult, and partly explains the relative deficiency ofempirical analysis in this area. The other major constraint to thedevelopment of empirical investigations has obviously related to the timeperiod that has elapsed since privatisation. Until recently insufficient datawas available to carry out studies capable of measuring the dynamic effectsof privatisation.

Two recent studies have attempted to measure the impact of privatisationon economic growth in developing countries. The first, by Plane [1997],was used for 35 developing countries covering the period 1984–92. The

123PRIVATISATION AND ECONOMIC GROWTH

396jds05.qxd 29/08/03 14:13 Page 123

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

study, using Probit and Tobit models, found that privatisation positivelyaffected GDP growth and that the effect on growth was more significant foractivities of a public goods type than for other sectors. The study concludesthat on average institutional reform increased economic growth from 0.8 percent to 1.5 per cent between the sub-periods 1984–88 and 1988–92.

The second study was published by the IMF in 2000 [Barnett, 2000].This study investigated the effect of privatisation on real GDP growth,unemployment, and investment. Only twelve developing countries wereused in this 18-country study. The rest consisted of transitional economies.The empirical analysis from this study strongly supported the hypothesisthat privatisation was positively correlated with real GDP growth. Theyfound that privatisation of a one per cent of GDP was associated with anincrease in the real growth rate of 0.5 per cent in period one and 0.4 per centin period two. These periods vary for each country to reflect periods ofactive privatisation, but the precise span of years for the study is notspecified. It has also been suggested that the privatisation variable used inBarnett’s is likely to capture the positive impact of a general regime changetowards better economic policies [Davis, Ossowski, Richardson andBarnett, 2000].

III . METHODOLOGY AND DATA

In most developing countries privatisation has been a new policy instrumentand has typically been implemented sporadically rather than annually. As aresult, this study applies a cross-country regression analysis to examinewhether or not privatisation affects economic growth. Cross-countryregression analysis has been widely used to scrutinise the determinants oflong-run growth. These studies have identified various policy and socio-demographic factors that may contribute to long-run growth. Studies ofthese types and their interpretation have led to debates over the validity ofthese approaches.

It has become apparent that cross-country growth regressions potentiallysuffer a number of statistical and conceptual problems. Harberger [1987]points out that one of the most fundamental statistical problems could becaused by the inclusion of countries that have little in common into the sameregression. In principle, regression analysis necessitates observations thatare drawn from a distinct population. Thus, the inclusion of the countrieswhich are intrinsically different may result in unacceptable levels ofstatistical bias [Levine and Zervos, 1993].

An important conceptual problem relates to the way in which cross-country growth regression analysis is conducted. A typical cross-countrygrowth regression analysis uses data that are averaged over a significantly


396jds05.qxd 29/08/03 14:13 Page 124

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

long period, to permit the coefficients on various variables to be interpretedas elasticities. In turn, these coefficients are used to indicate the percentagegrowth that results from a one percent change in a policy variable. None theless, Levine and Zervos [1993] warn that this type of analysis ought to betreated cautiously since a cross-country regression itself does not providethe complete answer to any causal relationships between policy variablesand growth. Accordingly, they argue that ‘cross-country regressions shouldbe viewed as evaluating the strength of partial correlations, and not asbehavioural relationships that suggest how much growth will change whenpolicies change’ [1993: 426].

In practice cross-country regression analysis has been one of the mostwidely used methods in research on the relationship between policy changesand growth. In order to take advantage of its usefulness, it is imperative tominimise any possibility that distorts results, and to check for the robustnessof results. With these reservations in mind, this paper carries out a cross-country growth regression analysis using the framework of the so-calledextreme-bounds analysis (EBA). The EBA was developed by Leamer[1978, 1983, 1985] and Leamer and Herman [1983] and has been appliedpractically in a number of empirical studies [Cooley and Leroy, 1981; Derviet al., 1990; Levine and Renelt, 1992; Levine and Zervos, 1993].

The EBA can be conducted by applying the following linear ordinaryleast squares (OLS) regression:

Y = β1I + β2M + β3Z + u,

where Y is the GDP per capita growth rate, I is a set of variables that arecommonly included in the regression, M is the policy variable of particularinterest, and Z is a set of up to three variables chosen from a pool of policyvariables. This method follows Levine and Renelt [1992] and is differentfrom the original EBA developed by Leamer [1978], which includes all theZ variables in a regression. It is more practical and has been referred to as amodified version of EBA [Doppelhofer, Miller, and Sala-i-Martin, 2000].

In accordance with the EBA procedure, the combinations of a set of theZ variables are changed to examine whether or not the coefficient on M isstatistically significant and maintains the same sign throughout the process.To conduct the EBA, all possible combinations of regressions are estimatedto obtain β2 and the corresponding standard deviation, σ, for eachregression. Then, the upper or maximum extreme bound is estimated as themaximum value of β2 plus σ2. The lower or minimum extreme bound isdefined as the minimum value of β2 minus σ2. If these maximum andminimum extreme bounds keep the same sign, the result is interpreted asrobust rather than ‘fragile’ (see Appendix 1 for a more detailed account ofthe EBA).


396jds05.qxd 29/08/03 14:13 Page 125

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

Several potential problems relating to EBA have been identified [Sala-i-Martin, 1994]. Significant among these is the so-called reverse datamining problem where the control variables in a regression are drawn fromthe true population. This implies that all data contain some element of error.By seeking different combinations of control variables, it is highly likelythen that there exists a combination that results in a variable of interestbeing statistically insignificant or changing its coefficient sign. As a result,the EBA ‘may be too strong’ [Sala-i-Martin, 1994: 743].

Accordingly, Sala-i-Martin [1997] has attempted to estimate the entiredistribution of β2. He also suggests an alternative method to conduct asensitivity analysis, although the statistical rationale for the approach hasbeen shown to be uncertain [Folster and Henrekson, 2001]. More recently,Doppelhofer, Miller, and Sala-i-Martin [2000] have developed analternative method based on a Bayesian approach. Our analysis, however,continues to use the modified EBA developed by Levine and Renelt [1992]as these later contributions are still in their experimental stage.

Privatisation Variable

The most crucial question is how to measure the magnitude of privatisationin a given country. Plane [1997] has used the cumulative proceeds fromprivatisation during the period 1988–92 as a share of GDP in 1990 and adummy variable based on the information obtained from this indicator. Asimilar indicator is used in this study, although the average GDP during thesample period is used as a weight instead of choosing a particular year. Thisreduces the arbitrariness from choosing a particular year for which thesample countries might experience a variety of external or internal shocks.In addition, the use of a dummy variable for privatisation is droppedsince it cannot convey information on the magnitude of privatisation intothe regression.

The dataset for privatisation is summarised in Table 1. The availabilityof privatisation data largely determines the number of sample countries andthe sample period. As a result, 63 developing countries and the period1988–97 are selected. The dataset shows the highly concentrated pattern ofprivatisation, despite its adoption in a wide range of countries. LatinAmerican countries dominate the table in terms of privatisation as a shareof GDP. The dataset and full list of other variables corresponding to thesecountries are shown in Appendix 2.

To reduce the risk that the privatisation variable reflects other policy andstructural reforms at the macroeconomic or aggregate level, it is necessaryto check if the effect of this variable can be isolated. This can be achievedby examining the relationship between some of the policy variables thatmight be captured by the privatisation variable. The effects of the


396jds05.qxd 29/08/03 14:13 Page 126

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

TA

BL

E 1

CU

MU

LA

TIV

E P

RO

CE

ED

S F

RO

M P

RIV

AT

ISA

TIO

N (

CP

P)

AN

D A

VE

RA

GE

GD

P (

AG

DP

) D

UR

ING

TH

E P

ER

IOD

19

88

–9

7(I

N U

S$

MIL

LIO

NS

)

Cou

ntry

CPP

AG

DP

CPP

/C

ount

ryC

PPA

GD

PC

PP/

AG

DP

(%)

AG

DP

(%)

1A

rgen

tina

2792

1.0

2826

7.3

9.9

33Ta

nzan

ia14

0.8

1503

1.3

0.9

2Pe

ru74

77.5

8705

1.1

8.6

34U

gand

a15

1.6

1642

9.3

0.9

3M

alay

sia

1007

6.3

1186

95.4

8.5

35Z

imba

bwe

197.

322

240.

30.

94

Sing

apor

e45

78.0

5966

9.2

7.7

36H

ondu

ras

93.5

1100

3.2

0.8

5B

eliz

e54

.379

3.0

6.8

37K

enya

235.

128

329.

60.

86

Jam

aica

532.

981

22.0

6.6

38To

go38

.852

20.3

0.7

7Pa

nam

a83

2.8

1336

3.5

6.2

39In

dia

7073

.011

7216

6.0

0.6

8T

rini

dad

& T

obag

o44

8.4

7599

.45.

940

Kor

ea, R

epub

lic25

46.0

4422

76.5

0.6

9Z

ambi

a41

7.2

7865

.95.

341

Gui

nea

45.0

9928

.10.

510

Mex

ico

3335

3.1

6479

00.5

5.1

42Jo

rdan

58.7

1195

9.8

0.5

11Pa

pua

New

Gui

nea

223.

643

97.3

5.1

43T

haila

nd13

78.4

2969

25.5

0.5

12B

oliv

ia88

4.2

1772

0.7

5.0

44T

unis

ia17

1.0

3738

0.8

0.5

13B

razi

l34

559.

487

7419

.53.

945

Om

an60

.113

370.

70.

414

Gha

na87

2.6

2309

8.0

3.8

46E

cuad

or16

9.4

4833

6.9

0.4

15V

enez

uela

5914

.116

8554

.83.

547

Mal

i21

.960

15.3

0.4

16C

olom

bia

5685

.121

5739

.92.

648

Cos

ta R

ica

56.7

1873

5.7

0.3

17B

arba

dos

51.0

2039

.52.

549

Para

guay

42.0

1576

7.2

0.3

18M

oroc

co18

46.7

7730

7.4

2.4

50C

amer

oon

41.1

2377

9.9

0.2

19So

uth

Afr

ica

6064

.425

7829

.92.

451

Gui

nea-

Bis

sau

0.5

235.

20.

220

Cot

e d’

Ivoi

re47

6.3

2087

4.1

2.3

52M

alaw

i10

.855

49.1

0.2

21Sr

i Lan

ka72

5.9

3464

0.0

2.1

53B

angl

ades

h60

.398

712.

60.

122

Nic

arag

ua13

0.2

6520

.62.

054

Bur

kina

Fas

o6.

484

11.6

0.1

23E

gypt

2777

.714

2925

.91.

955

Bur

undi

4.2

4332

.20.

124

Phili

ppin

es38

10.0

2046

78.8

1.9

56G

uate

mal

a43

.433

999.

60.

125

Sene

gal

191.

412

344.

01.

657

Nep

al15

.018

530.

00.

126

Tur

key

3843

.430

6790

.81.

358

Sier

ra L

eone

1.6

2396

.30.

127

Moz

ambi

que

110.

689

32.9

1.2

59U

rugu

ay17

.023

925.

70.

128

Paki

stan

1951

.016

1154

.21.

260

Iran

18.1

2094

17.6

0.0

29C

hile

1484

.112

9306

.11.

161

Mau

rita

nia

1.1

3357

.20.

030

Ben

in56

.556

92.5

1.0

62V

ietn

am2.

672

856.

00.

031

Indo

nesi

a51

62.8

4944

40.0

1.0

63Y

emen

0.8

8544

.50.

032

Nig

eria

763.

587

263.

80.

9

Sour

ce: P

rim

ary

data

fro

m W

orld

Ban

k Pr

ivat

izat

ion

Dat

abas

e an

d Pr

ivat

isat

ion

Inte

rnat

iona

l (v

ario

us i

ssue

s),

and

Wor

ld B

ank

Glo

bal

Dev

elop

men

tN

etw

ork

Gro

wth

Dat

abas

e.

396jds05.qxd 29/08/03 14:13 Page 127

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

privatisation variable may be linked to the size of the government budgetdeficit, since privatisation may have been implemented as a quick solutionto a budgetary problem in developing countries [Yarrow, 1999]. In order torule out this possibility, a simple correlation between the privatisationindicator and the average ratio of government budget deficit to GDP duringthe sample period 1988–97 was examined.

The result, shown in Figure 1, indicates that there is no correlationbetween the two variables (the correlation coefficient is 0.23). Argentina,Peru, and Malaysia implemented sizeable privatisation programmes whilemaintaining relatively low budget deficits. Guinea-Bissau and Mozambiquerecorded considerably high budget deficits with relatively smallprivatisation programmes. It should be emphasised that this simplecorrelation analysis, conducted without lags, does not eliminate thepossibility that some countries might have reduced their budgetary deficitsas a result of privatisation.

Similarly, the privatisation variable may capture the effects of policyreforms pursued through World Bank adjustment programmes. In oursample, 56 of the 63 countries received various adjustment loans withpolicy reform conditions during the period 1985–97. In particular,Argentina, Bangladesh, Bolivia, Ghana, Jamaica, Mauritania, and Zambia


FIGURE 1

THE CORRELATION BETWEEN PRIVATISATION ANDGOVERNMENT BUDGET DEFICIT

Note: PRIV/GDP and BUD/GDP are the average percentages during the period 1988–97.Primary data from World Bank Privatisation Database and World Bank GlobalDevelopment Network Growth Database.

0123456789

10

-30 -25 -20 -15 -10 -5 0 5 10 15

BUD/GDP

PRIV

/GD

P

396jds05.qxd 29/08/03 14:13 Page 128

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

received adjustment loans with explicit conditions to implementprivatisation programmes.

The World Bank states that ‘privatisation became an important feature inthe Bank’s adjustment lending from the mid-1980s. This reflected both theconcurrent ideological shift in favour of private ownership as well as theBank’s long, and painful, experience with failed attempts at reformingpublic enterprises’ [World Bank, 1995b: 56]. Before 1985, eight per cent ofall public enterprise reform-related adjustment loans attached conditionsregarding the implementation of privatisation. Between 1985 and 1989, theproportion increased to 14 per cent and to 32 per cent in the period between1990 and 1994 [World Bank, 1994].

A simple correlation was conducted between privatisation and WorldBank adjustment loans in order to eliminate the possibility that theprivatisation variable captures the macroeconomic effects of adjustmentprogrammes. The magnitude of World Bank adjustment loans is calculatedin the same manner as the privatisation indicator (i.e. the cumulative amountof World Bank adjustment loans during the period 1985–97 divided by theaverage GDP during the same period). The result reported in Figure 2clearly indicates that there is no apparent relationship between privatisationand adjustment loans (the correlation coefficient is 0.03).


FIGURE 2

THE CORRELATION BETWEEN PRIVATISATION ANDWORLD BANK ADJUSTMENT LOANS

Note: PRIV/GDP and SAL/GDP are the average percentage during the period 1988–97. Primarydata from World Bank Privatisation Database, World Bank Global Development NetworkGrowth Database, and World Bank SAL Database.

0

2

4

6

8

10

12

0 2 4 6 8 10 12 14 16

SAL/GDP

PRIV

/GD

P

396jds05.qxd 29/08/03 14:13 Page 129

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

Of course, these results do not automatically confirm that theprivatisation indicator is able to capture the effects of privatisation at themacroeconomic level. It only assures us that the privatisation indicator isnot picking up the effects of other policy reforms. In addition, since ourstudy has significantly increased the number of sample countries comparedwith the previous studies, then the possibility that the privatisation indicatorcaptures the effects of other macroeconomic reforms is considerablyreduced.

IV. EXTREME BOUNDS ANALYSIS (EBA)

The analysis is conducted in two parts. In the first, the control variables areexamined to find appropriate specifications for the I variables. In the secondpart, the results of introducing privatisation are reported.

The Basic Test for the Control Variables (the I variables)

In principle, the I variables are always included in an EBA. These variableshave to be selected with great care. In the context of a cross-country growthregression, the control variables ought to control for initial economic,political, and social conditions of countries. There are several variables thatare routinely included in these types of regressions. They are: (1) log ofinitial GDP per capita; (2) initial life expectancy at birth; (3) averagepopulation growth rate; (4) the ratio of government consumption to GDP;(5) the ratio of gross domestic investment to GDP; and (6) the rate ofsecondary school enrolment or attainment. These variables are sometimescalled ‘Barro-regressors’ and are used widely in cross-country growthregressions [Barro, 1991].

Earlier, Levine and Zervos [1993] have argued that the inclusion of afiscal variable (4) does not necessarily control for initial conditions orpolitical stability but is a contemporaneous economic policy indicator.However, Barro and Sala-i-Martin [1995] have argued that a fiscal variablecan be used as a proxy ‘for political corruption or other aspects of badgovernment, as well as for the direct effects of non-productive publicexpenditures and taxation’ [1995: 434].

The inclusion of investment as a control variable also remainscontroversial. Levine and Renelt [1992] have included the ratio ofinvestment to GDP in their I variables, although a number of studies haveshown that causality runs from growth to investment and not vice versa[Blomstrom, Lipsey and Zejan, 1993; Barro and Sala-i-Martin, 1995]. Morerecently, Temple [1999] argues that the correlation is reverse and robust.Besides the concern over the direction of causality, there is also a questionover what measure of investment to use. In general, gross domestic


396jds05.qxd 29/08/03 14:13 Page 130

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

investment is often used as a proxy for investment. However, Barro andSala-i-Martin [1995] argue that since gross domestic investment includesboth private and public spending, it may be an inappropriate measure ofinvestment. DeLong and Summers [1991] also argue that the use of grossdomestic investment can be ruled out since producer durables are thesignificant component of investment. Since we are unable to resolve thecausality problem, we test for different combinations of the I variables usingboth gross domestic investment and government consumption.

Another specification problem we have encountered relates to theinclusion of the secondary school enrolment ratio. This was not included inour study since it does not capture the quality of education. Barro and Sala-i-Martin [1995] have argued that it is more appropriate to use the secondaryschool attainment ratio, although if we use this variable, the number ofcountries will have to be severely reduced owing to a lack of data. Instead,the initial life expectancy at birth is used. This variable captures variousaspects of human capital development, such as investment in education andincreases in schooling, thereby reflecting the quality of education in general[Ram and Schulz, 1979; Sala-i-Martin, 1997; Kalemili-Ozcan, Ryder andWeil, 2000].

Combinations of variables were used to find the I variables relevant forthis study. This was achieved by examining four different combinations ofselected I variables. The first specification included the log of initial GDPper capita for 1988 (hereafter LGYP), the initial life expectancy at birth for1987 (LIFE), the average population growth during the sample period1988–97 (POP), and the average ratio of GDI to GDP during the sampleperiod (GDI). The second specification included LGYP, LIFE, POP, GDI,and the average ratio of government consumption to GDP during the sampleperiod (GOVC). The third specification included LGYP, POP, LIFE, andGOVC. Finally, the fourth specification consisted of LGYP, LIFE, and POP.In all cases the dependent variable is the average growth rate of GDP percapita during the sample period (GYP).

In order to check the robustness of the basic results, a series of EBAswas conducted by including the Z variables. The following Z variables wereselected: (1) openness, which is calculated as (export + import)/GDP(hereafter OPEN); (2) informal market premium (IMP); (3) the standarddeviation of IMP (SDIMP); (4) the average ratio of FDI to GDP (FDI); (5)the average number of major political crises, revolutions, and coups as aproxy for political instability (STAB); (6) the average inflation rate (INFL);(7) the standard deviation of average inflation rates (SDINF); (8) theaverage ratio of liquid liabilities (M3) to GDP used as a proxy for the levelof financial sector development (M3); (9) the standard deviation of M3(SDM3); (10) the average ratio of total external debt to GDP (DEBT); (11)


396jds05.qxd 29/08/03 14:13 Page 131

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

the average ratio of government budget surplus/deficit to GDP (BUD); and(12) three regional dummies for East Asia, Latin America and Sub-SaharanAfrica (EA, LA, AF, respectively). Some of the I variables are also includedas Z variables. They are GOVC (in the first and fourth specifications) andGDI (in the third and fourth specifications).

These variables, with the exception of (4), are frequently used in cross-country growth regressions. The ratio of FDI to GDP is included on theassumption that FDI may play a significant role in generating positivespillover effects, bringing new technologies and management andmarketing skills that contribute to economic growth in developingcountries. In addition, STAB is included as a Z variable, although it couldbe used as a control variable since there may exist a reverse causality thatruns from political instability to economic growth. In this respect,Londregan and Poole [1990] argue that higher political instability is likelyto be caused by lower economic growth, and Barro [1991] and Alesina andPetrotii [1993] have also found evidence supporting this reverse causality.

The EBAs for the control variables in the four specifications wereconducted by computing a total of 7,140 regressions, and the results areshown in Table 2. Note that any result obtained from a regression thatexhibits a significant level of multicollinearity (that is, a regression showingthat a variance inflation factor (VIF) is greater than 10) was discarded fromthese results.

In general, the inclusion of LGYP is expected to represent the rate ofconvergence across the sample countries. Other studies that have includeddeveloped countries have found evidence of conditional convergence [e.g.,Mankiw, Romer and Weil, 1992; Barro and Sala-i-Martin, 1995]. Anegative and statistically significant coefficient for this variable signifiesevidence of conditional convergence. In our study, the results for LGYPobtained from the first, third, and fourth specifications appear to be negativeand fragile (see the notes in Table 2 for the criterion used to interpret EBAresults).

With respect to the results for the first and second specifications, LIFEappears to be fragile, although robust results were obtained in the third andfourth specifications. Throughout, the sign of the coefficient is positive, andthe robust results imply that this variable captures aspects of human capitaldevelopment reasonably well. A robust negative result was obtained forPOP from the first and forth specifications. In contrast, the result obtainedfrom the second specification was completely insignificant and was fragilefor the third. These contrasting results may suggest that this variable isextremely sensitive to the inclusion or exclusion of other variables, asindicated by Levine and Renelt [1992]. Robust results were obtained inrelation to GDI and GOVC. The signs of the coefficients for GDI are


396jds05.qxd 29/08/03 14:13 2

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08


TABLE 2

THE EBA RESULTS OF THE I VARIABLES (AT 0.05 LEVEL)

Specification I

Variable Coefficient t-statistic Z-variables included EBA result

LGYP Min –0.012 –2.05 STAB, EA, AF FragileMax –0.011 –2.02 STAB, FDI, EABase –0.005 –0.94 –

LIFE Min 0.052 2.01 GOVC, SDIMP, LA FragileMax 0.061 2.23 GOVC, BUD, LABase 0.025 0.98 –

POP Min –1.065 –3.25 FDI, IMP, LA RobustMax –0.650 –2.01 INFL, M3, EABase –0.782** –2.34 –

GDI Min 0.104 2.17 FDI, EA, LA RobustMax 0.219 5.20 IMF, SDM3, DEBTBase 0.203*** 5.03 –

Specification II


LGYP Min – – – InsignificantMax – – –Base –0.003 –0.56 –

LIFE Min 0.051 2.02 FDI, IMP, LA FragileMax 0.066 2.37 STAB, BUD, LABase 0.028 1.12 –

POP Min – – – InsignificantMax – – –Base –0.390 –1.07 –

GDI Min 0.108 2.25 STAB, EA, LA RobustMax 0.209 5.03 BUD, SDM3, DEBTBase 0.196*** 5.03 –

GOVC Min –0.160 –2.98 BUD, SDIMP, LA RobustMax –0.107 –2.02 STAB, IMP, EABase –0.120** –2.29 –

Specification III


LGYP Min –0.013 –2.34 STAB, EA, AF FragileMax –0.011 –2.04 STAB, SDIMP, EABase –0.005 –0.82 –

LIFE Min 0.052 2.01 GDI, SDIMP, LA Robust Max 0.109 3.95 STAB, BUD, LABase 0.068** 2.43 –

POP Min –0.760 –2.04 STAB, FDI, LA FragileMax –0.742 –2.01 FDI, SDIMP, LABase –0.354 –0.82 –

396jds05.qxd 29/08/03 14:13 Page 133

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

positive, whereas they are negative for GOVC, confirming previous studies[e.g., Barro, 1991].

In summary, LGYP exhibits fragile results, and LIFE and POP appear tobe sensitive to different combinations of other variables, resulting inambiguous results. On the other hand, the results for GDI and GOVC arealways robust. Nevertheless, it is difficult to choose a particularspecification for the control variables from these results. Since a continuedsearch for the most suitable combination of the control variables is beyondthe scope of this study and may prove impossible, as some past studies havedemonstrated, then all four specifications are used to conduct an EBA forthe privatisation variable.

The EBA for Privatisation

The EBA proceeded in two steps. First, basic tests were carried out toestimate baselines before a formal EBA was conducted for privatisation.These regressions included the control variables in the four specifications


GOVC Min –0.203 –3.17 STAB, BUD, M3 RobustMax –0.109 –2.04 GDI, SDINF, SDM3Base –0.139** –2.24 –

Specification IV


LGYP Min –0.015 –2.69 STAB, EA, AF FragileMax –0.011 –2.01 MDM3, EA, AFBase –0.008 –1.20 –

LIFE Min 0.054 2.05 GOVC, FDI, M3 Robust Max 0.103 3.77 GOVC, BUD, LABase 0.067** 2.30 –

POP Min –1.216 –3.46 FDI, SDM3, LA RobustMax –0.660 –2.01 GDI, M3, EABase –0.810** –2.04 –

Notes: Dependent variable: GYP. The base results are calculated excluding the M and Zvariables. Note that following Levine and Renelt [1992] (see Table 1, notes (a), p.947),where an EBA indicates a robust result, but the base result is statistically insignificant, theoverall result is interpreted as fragile. ** = statistically significant at 0.05 level, and ***= statistically significant at 0.01 level.

TABLE 2 (cont.)

Specification III (cont.)


396jds05.qxd 29/08/03 14:13 Page 134

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

examined previously and the privatisation variable. The Z variables wereexcluded. The tests showed that the base results for privatisation werestatistically insignificant. Second, the EBA for the privatisation variablewas conducted by this time including three regressors from a pool of the 15Z variables. The results are shown in Table 3. Although the EBA indicatedthat the results were robust, they still have to be interpreted as fragile sincethe earlier basic tests indicated that they were statistically insignificant.Accordingly, at this stage, we have concluded that only a weak or fragilenegative partial correlation between privatisation and economic growthexists.


TABLE 3

THE EBA RESULTS (AT 0.05 LEVEL) FOR THE PRIVATISATION VARIABLE

The I variables based on Specification I (LGYP + LIFE + POP + GDI)


PRIV Min –0.243 –2.25 BUD, FDI, DEBT Fragile Max –0.219 –2.01 GOVC, FDI, IMPBase –0.056 –0.55 – –

The I variables based on Specification II (LGYP + LIFE + POP + GDI + GOVC)


PRIV Min –0.249 –2.35 BUD, FDI, DEBT Fragile Max –0.216 –2.04 BUD, FDI, EABase –0.077 –0.77 – –

The I variables based on Specification III (LGYP + LIFE + POP + GOVC)


PRIV Min –0.347 –2.82 FDI, SDM3, DEBT FragileMax –0.219 –2.01 GDI, FDI, IMPBase –0.090 –0.76 – –

The I variables based on Specification IV (LGYP + LIFE + POP)


PRIV Min –0.345 –2.76 FDI, SDM3, DEBT FragileMax –0.228 –2.02 FDI, SDIMP, LABase –0.067 –0.55 – –

Notes: Dependent variable: GYP. Note that following Levine and Renelt [1992] (see Table 1,notes (a), p.947), in the case that the EBA indicates a robust result, but the base result isstatistically insignificant, the overall result is interpreted as fragile.

396jds05.qxd 29/08/03 14:13 Page 135

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

Two cautionary notes are required at this point. First, it is known that theinclusion or exclusion of certain control variables sometimes eliminates abivariate relationship [Easterly and Rebelo, 1993]. It is thus plausible toargue that our inconclusive results are due to an inadequate choice orcombinations of the control variables. In order to examine this possibility,all possible combinations of the control variables and the privatisationvariable were scrutinised by applying the EBA technique. This exercisereconfirmed our conclusion by showing that no combination of controlvariables exists that would establish a robust partial correlation betweenprivatisation and economic growth.

Second, it is known that cross-country regression analysis is sensitive tothe inclusion or exclusion of a particular country. This may arise due to theexistence of influential observations and/or outliers in the regressions. Thispossibility is also examined by checking whether or not there are anyinfluential observations or outliers in our basic regression models. It should,however, be pointed out that there is no universally accepted indicator thatis able to identify influential observations and outliers. For instance, Donaldand Maddala [1993] argue that the examination of studentised residuals isthe most appropriate method to identify influential observations, and todetect outliers leverage should also be examined. On the other hand,Beckman and Cook [1983] and Fiebig [1992] argue that large studentisedresiduals may not necessarily point to influential observations, and Fiebig[1992] specifically points out that influential observations may actually berelated to small residuals.

In addition, even if influential observations and outliers are identified, itmay prove difficult to deal with them. Influential observations and outliersmay arise as a result of data errors and model misspecifications. Suitablecorrections may be hampered because of poor data, especially fordeveloping countries, and simply deleting the countries in question may stillbe appropriate. On the other hand, influential observations may representnatural variation which provides valuable information, and eliminatingthese observations would clearly be inappropriate. Even if influentialobservations are kept in the regression, however, the potential effect on theresults of eliminating them ought to be noted [Fiebig, 1992]. One way toaccommodate influential observations in the regression is to apply robustregression estimators instead of OLS (see Rousseeuw and Leroy [1987] fora detailed account of robust regression estimators). It should also be notedthat even these detection techniques and robust estimators may fail to detectoutliers [Luke and Schaffer, 2000].

Accepting these limitations, studentised residuals were examined in anattempt to detect influential observations in our basic regression models. Wealso applied some of the influential observation and outlier detection


396jds05.qxd 29/08/03 14:13 Page 136

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

techniques, DFITS [Welsch and Kuh, 1977], Cook’s Distance [Cook, 1977],and Welsch Distance [Welsch, 1982]. These techniques, in essence,encapsulate the information in the leverage versus residual-squared plot intoa single statistic, creating an index that is influenced by the size of theresiduals or outliers and the size of leverage (see Appendix 3 for thecalculation procedures). In addition, we tested for influential observationsand outliers specifically to examine the privatisation variable by applyingDFBETAs [Belsley, Kuh and Welsch, 1980]. The application of DFBETAspermits us to examine the difference between the coefficients of a specificvariable when a particular observation is included or excluded

Table 4 reports the results of the tests for influential observations.Although it is difficult to compare the indices across the differentspecifications and techniques, these tests identified 15 countries as potentialoutliers or a total of 95 observations as influential. Singapore and Malaysiacontributed significantly to the influential observations, 17.9 per cent and14.7 per cent respectively, followed by Oman 13.7 per cent, and to a lesserextent Sierra Leone 8.4 per cent. These four countries generated more than50 percent of the influential observations. In particular, Welsch Distanceindicates that only Malaysia, Oman, and Singapore are potential outliers.More specifically, of those 20 test results shown in Table 4, Singaporeappears as a potential outlier in 17 of them, Malaysia appears in 14 tests,and Sierra Leone is identified in eight tests. Oman was detected asinfluential by 13 test results only in the first, second, and thirdspecifications. In addition, the combination of Malaysia and Singaporeappear in 14 test results, including all the test results in the third and fourthspecifications. Furthermore, the results of DFBETAs suggest that these twocountries have substantial positive effects on the privatisation variable, inparticular in the third and fourth specifications.

After identifying the influential observations and thus the potentialoutliers in our basic regression models, we attempted to re-estimate thebaselines for privatisation in two steps. First, we re-estimated the baselinesfor privatisation by applying two robust regression estimators, the medianleast squares (MLS) and a robust regression estimator using iterativelyreweighted least squares, including all of the influential observations (seeRousseeuw and Leroy [1986] and Luke and Schaffer [2000] for details ofthese estimators). We then re-estimated the baselines using OLS byeliminating all the influential observations. Second, we deleted theinfluential observations one by one according to the higher absolute valueof each of the indicators for influential observations. Following these steps,we could not find any statistically significant results for privatisation in thefirst and second specifications but found a number of statisticallysignificant results for privatisation in the remaining specifications.


396jds05.qxd 29/08/03 14:13 Page 137

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08


TABLE 4

TESTS FOR INFLUENTIAL OBSERVATIONS AND OUTLIERS

Specification I

Studentized DFBETAs DFITS Cook’s Welsch

1 Uganda 2.563 Jamaica –0.543 Oman 1.065 Jordan 0.305 Oman 10.092 Jamaica –2.559 Malaysia 0.437 Singapore 0.862 Sierra Leone 0.231 Singapore 7.6483 Cameroon –2.514 Singapore 0.364 Vietnam 0.789 Jamaica 0.1944 Jordan –2.341 Peru –0.356 Uganda 0.709 Oman 0.1845 Sierra Leone –2.041 Oman –0.290 Malaysia 0.672 Singapore 0.1206 Jordan 0.255 Vietnam 0.0997 Uganda 0.0778 Malaysia 0.075

Specification II


1 Cameroon –2.959 Jamaica –0.510 Oman 2.327 Oman 0.687 Oman 23.692 Oman 2.837 Oman –0.459 Singapore 0.742 Sierra Leone 0.2473 Uganda 2.321 Peru –0.456 Vietnam 0.733 Jordan 0.2354 Sierra Leone –2.287 Malaysia 0.409 Uganda 0.727 Jamaica 0.1565 Jamaica –2.196 Singapore 0.279 Singapore 0.0776 Jordan –2.148 Cameroon 0.254 Vietnam 0.0747 Uganda 0.0708

Specification III


1 Cameroon –2.810 Malaysia 0.756 Oman 1.327 Sierra Leone 0.323 Oman 13.382 Sierra Leone –2.549 Singapore 0.443 Singapore 1.048 Oman 0.285 Singapore 9.0743 Malaysia 2.324 Peru –0.434 Malaysia 0.971 Singapore 0.170 Malaysia 8.2894 Singapore 2.296 Korea –0.366 Korea 0.641 Malaysia 0.1465 Thailand –0.302 Barbados 0.0666 Argentina –0.294 Korea 0.0657 Oman –0.260

Specification IV


1 Singapore 2.616 Malaysia 0.794 Singapore 1.081 Sierra Leone 0.326 Singapore 9.2132 Cameroon –2.411 Singapore 0.537 Malaysia 0.987 Singapore 0.212 Malaysia 8.4023 Malaysia 2.398 Korea –0.366 Korea 0.646 Malaysia 0.1804 Sierra Leone –2.325 Peru –0.340 Barbados 0.1005 Thailand –0.303 Korea 0.0806 Jamaica –0.295 Jamaica 0.0687 Jordan 0.065

396jds05.qxd 29/08/03 14:13 Page 138

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

Accordingly, the first and second specifications were dropped from anyfurther examination.

Correspondingly, we re-estimated the baselines for privatisation byapplying the two robust regression estimation techniques which enabled usto incorporate the influence of the influential observations or potentialoutliers in the regressions without deleting them. The results are reported inTable 5 in the third and fourth columns of each specification. In the thirdspecification, we found statistically significant results for privatisation fromMLS (at the 0.10 level) and Robust Regressions (at the 0.01 level). In thecase of the fourth specification, we could not find statistically significantresults for privatisation from the MLS but found the statistically significantresults from Robust Regression (at 0.01 level). As for the MLS results fromspecification four, pseudo R2 is too low, and the signs of the coefficients,except those for PRIV and POP, changed. In essence, we have found that arobust partial correlation between privatisation and economic growth can beestablished in these specifications, in particular in the third specification, byusing the robust estimators instead of OLS and without deleting theinfluential observations.

The results of excluding all the influential observations using OLS (atotal of 10 models were estimated) are shown from the fifth to the lastcolumn in Table 5. In these specifications, we found statistically significantresults for privatisation at the 0.01 level (shown in column (1) in the thirdspecification and column (6) in the fourth specification) by deleting all theinfluential observations identified by studentised residuals. Similarly, wefound statistically significant results for privatisation in the fourthspecification (column 10) at the 0.05 level and in the third specification(column 5) at the 0.10 level by eliminating all the influential observationsidentified by Welsch Distance. As for Cook’s Distance, we also foundstatistically significant results for privatisation at the 0.05 level (the thirdspecification (column 4)) and at the 0.10 level (the fourth specification(column 9)).

In contrast, the elimination of all the influential observations identifiedby DFBETAs resulted in statistically insignificant results for privatisation inboth specifications. In terms of the deleted influential observationsidentified by DFITS, the results are somewhat mixed. We found astatistically significant result for privatisation at the 0.10 level in the fourthspecification (column 8) and statistically insignificant result in the thirdspecification (column 3). In sum, we tested ten models using OLS in thethird and fourth specifications and found statistically significant results inseven models.

The results of eliminating the influential observations to examinewhether or not these have led to increases in R2 and/or the statistical


396jds05.qxd 29/08/03 14:13 Page 139

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

TA

BL

E 5

TH

E B

AS

EL

INE

RE

SU

LT

S F

OR

PR

IVA

TIS

AT

ION

Sp

ecif

icat

ion

III

OL

SM

LS

R

obus

t R

egre

ssio

n(1

) S

tude

ntis

ed(2

) D

FB

ET

As

(3)

DF

ITS

(4)

Coo

k’s

(5)

Wel

sch

PR

IV–0

.090

–0

.192

*–0

.254

***

–0.2

92**

*–0

.063

–0.1

78–0

.214

**–0

.223

*(–

0.76

)(–

1.89

)(–

2.94

)(–

2.91

)(–

0.46

)(–

1.58

)(–

2.01

)(–

1.96

)L

GY

P–0

.004

–0.0

10*

–0.0

06–0

.002

–0.0

16**

–0.0

16**

–0.0

09–0

.013

**(–

0.64

)(–

1.72

)(–

1.34

)(–

0.37

)(–

2.66

)(–

2.56

)(–

1.50

)(–

2.18

)L

IFE

0.06

9**

0.06

8***

0.03

20.

036

0.08

5***

0.08

9***

0.05

1*0.

085*

**(2

.46)

(2.7

1)(1

.55)

(1.4

4)(3

.57)

(3.6

3)(1

.95)

(3.3

7)P

OP

–0.3

72–0

.648

–0.8

86**

*–0

.727

**–0

.749

*–0

.897

**–1

.156

***

–1.0

12**

(–0.

85)

(–1.

64)

(–2.

79)

(–2.

06)

(–1.

87)

(–2.

22)

(–2.

85)

(–2.

45)

GO

VC

–0

.143

**–0

.133

**–0

.140

***

–0.1

30**

–0.1

56**

–0.1

45**

–0.1

29**

–0.1

49**

(–2.

28)

(–2.

44)

(–3.

08)

(–2.

60)

(–2.

50)

(–2.

48)

(–2.

25)

(–2.

48)

Con

stan

t–0

.207

**–0

.156

**–0

.025

–0.0

77–0

.179

**–0

.191

**–0

.079

–0.1

86**

(–2.

40)

(–1.

99)

(–0.

40)

(–1.

01)

(–2.

48)

(–2.

57)

(–1.

00)

(–2.

43)

Adj

uste

d R

20.

22–

–0.

310.

300.

320.

310.

34P

seud

o R

2–

0.13

––

––

––

No.

of

infl

uent

ial

––

–A

ll (

4)A

ll (

7)A

ll (

4)A

ll (

6)A

ll (

3)ob

s. e

xclu

ded

Sp

ecif

icat

ion

IV

OL

SM

LS

R

obus

t R

egre

ssio

n(6

) S

tude

ntis

ed(7

) D

FB

ET

As

(8)

DF

ITS

(9)

Coo

k’s

(10)

Wel

sch

PR

IV–0

.067

–0.1

06–0

.250

***

–0.2

85**

*–0

.094

–0.2

04*

–0.2

21*

–0.2

49**

(–0.

55)

(–1.

24)

(–2.

61)

(–2.

71)

(–0.

74)

(–1.

73)

(–1.

95)

(–2.

10)

LG

YP

–0.0

070.

002

–0.0

05–0

.006

–0.0

16**

*–0

.014

**–0

.008

–0.0

12*

(–1.

06)

(0.3

2)(–

1.05

)(–

0.99

)(–

2.67

)(–

2.31

)(–

1.27

)(–

1.95

)L

IFE

0.06

7**

–0.0

110.

023

0.03

80.

085*

**0.

082*

**0.

051*

0.07

8***

(2.3

1)(–

0.54

)(0

.99)

(1.4

4)(3

.37)

(3.1

9)(1

.83)

(2.9

7)P

OP

–0.8

34**

–1.3

40**

*–1

.185

***

–1.1

71**

*–0

.934

**–0

.975

***

–1.2

83**

*–1

.112

***

(–2.

08)

(–4.

69)

(–3.

79)

(–3.

60)

(–2.

60)

(–2.

69)

(–3.

15)

(–3.

04)

Con

stan

t–0

.188

**0.

083

–0.0

05–0

.066

–0.1

93**

–0.1

93**

–0.1

00–0

.187

**(–

2.11

)(1

.31)

(–0.

08)

(–0.

83)

(–2.

57)

(–2.

49)

(–1.

18)

(–2.

35)

Adj

uste

d R

20.

17–

–0.

240.

230.

230.

220.

25P

seud

o R

2–

0.09

––

––

––

No.

of

infl

uent

ial

––

–A

ll (

4)A

ll (

6)A

ll (

3)A

ll (

7)A

ll (

2)ob

s. e

xclu

ded

Note

s:D

epen

dent

var

iabl

e: G

YP.

The

mod

els

(1)

- (1

0) w

ere

esti

mat

ed u

sing

OL

S.

* =

sta

tist

ical

ly s

igni

fica

nt a

t 0.

10 l

evel

, *

* =

sta

tist

ical

ly s

igni

fica

nt a

t0.

05 l

evel

, an

d **

* =

sta

tist

ical

ly s

igni

fica

nt a

t 0.

01 l

evel

. t

-sta

tist

ics

are

in p

aren

thes

es.

396jds05.qxd 29/08/03 14:13 Page 140

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

significance of the privatisation variable are shown in Table 6. Statisticallysignificant results were obtained for privatisation in all six models estimatedin the third and fourth specifications. In these models, higher adjusted R2and/or increased statistical significance for privatisation were obtained bydeleting a few countries from a pool consisting of Barbados, Cameroon,Malaysia, Oman, Sierra Leone and Singapore. Malaysia and Singaporeappear to be the most influential observations. In two models (column 11)and (column 16), higher R2 and increased statistical significance forprivatisation were obtained by eliminating these two countries, and in theremaining 4 models (shown in columns 12, 13, 14 and 15), higher R2 andincreased statistical significance were obtained by eliminating a few of thecountries including Malaysia and Singapore. Accordingly, Malaysia andSingapore appear to bias our regression results (see Stokey [1994] andLorgelly and Owen [1999] for a discussion of biased regression resultscaused by the inclusion of certain East Asian countries).

In summary, we obtained a total of 14 new estimates for privatisation,using OLS and deleting the influential observations identified by fivedifferent outlier detection techniques (note that some of the results areidentical: (column 5) and (column 12) in the third specification and (column10) and (column 16) in the fourth specification). In order to conduct theEBAs for these new estimates, we have selected from Tables 5 and 6baseline results for privatisation, which are statistically significant at 0.05level, for each specification according to the highest adjusted R2. This wasdone because the elimination of a subset of some influential observationsfrom our dataset resulted in the largest reduction in the residual sum ofsquares. It is likely that the eliminated observations are outliers [Gentlemanand Wilk, 1975].

All the baseline results were adopted from Table 6. We chose thebaseline results obtained by Cook’s Distance (the third and fourthspecifications) and the studentised residuals (the fourth specification). Themost likely candidates for outliers turned out to be Barbados, Cameroon,Oman, Malaysia, Sierra Leone, and Singapore. On this basis, Table 7 showsEBAs for privatisation by deleting the influential observations or the mostlikely outliers. This time, robust results were obtained from all twospecifications. We also conducted the EBAs for the other statisticallysignificant base results for privatisation at the 0.05 level by deleting theinfluential observations in Tables 5 and 6. The EBA results for privatisationare negative and robust in all the models.

Consequently, a negative but fragile partial correlation betweenprivatisation and economic growth has been found using the full 63-countrysample with OLS estimators, and a robust negative partial correlation hasbeen found by eliminating the possible outliers. We have also found a robust


396jds05.qxd 29/08/03 14:13 Page 141

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08


TABLE 6

THE BASELINE RESULTS FOR PRIVATISATION EXCLUDINGSELECTED INFLUENTIAL OBSERVATIONS

Specification III

Full sample (11) DFBETAs (12) DFITS (13) Cook’s

PRIV –0.090 –0.252** –0.223* –0.256**(–0.76) (–2.16) (–1.96) (–2.40)

LGYP –0.004 –0.009 –0.013** –0.007(–0.64) (–1.52) (–2.18) (–1.14)

LIFE 0.069** 0.078*** 0.085*** 0.047*(2.46) (3.03) (3.37) (1.72)

POP –0.372 –0.761* –1.012** –1.290***(–0.85) (–1.86) (–2.45) (–3.16)

GOVC –0.143** –0.103* –0.149** –0.128**(–2.28) (–1.77) (–2.48) (–2.18)

Constant –0.207** –0.201** –0.186** –0.069(–2.40) (–2.56) (–2.43) (–0.85)

Adjusted R2 0.22 0.28 0.34 0.34Likely outliers – Malaysia, Oman, Sierra Leone,

excluded Singapore Singapore, Oman,Malaysia Singapore,

Malaysia,Barbados

Specification IV

Full sample (14) DFBETAs (15) DFITS (16) Cook’s

PRIV –0.067 –0.271** –0.288** –0.249**(–0.55) (–2.41) (–2.61) (–2.10)

LGYP –0.007 –0.011* –0.005 –0.012*(–1.06) (–1.95) (–0.77) (–1.95)

LIFE 0.067** 0.075*** 0.038* 0.078***(2.31) (3.01) (1.40) (2.97)

POP –0.834** –1.063*** –1.414*** –1.112***(–2.08) (–3.07) (–4.03) (–3.04)

Constant –0.188** –0.179** –0.069 –0.187**(–2.11) (–2.38) (–0.84) (–2.35)

Adjusted R2 0.17 0.27 0.27 0.25Likely outliers – Singapore, Sierra Leone, Singapore,

excluded Cameroon, Singapore, MalaysiaMalaysia Malaysia,

Barbados

Notes: Dependent variable: GYP. These models were estimated using OLS. * = statisticallysignificant at 0.10 level , ** = statistically significant at 0.05 level, and *** = statisticallysignificant at 0.01 level. . t-statistics are in parentheses. The likely outliers excluded areshown in the order of higher absolute values.

396jds05.qxd 29/08/03 14:13 Page 142

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

negative partial correlation between these two variables using the full 63-country sample with the robust regression estimators. These results indicatethat the privatisation variable is sensitive to combinations of controlvariables and the inclusion of certain countries. In other words, in order todetect a significant result for the privatisation variable, great care is requiredin the selection of a specific conditioning information set and countries.

V. CONCLUSION

The analysis has shown that there is a robust partial correlation betweenprivatisation and economic growth, suggesting that privatisation hascontributed negatively to economic growth. This conclusion is contrary tothe results obtained by Plane [1997] and Barnett [2000]. The analysis in thisstudy has used a different approach to that adopted in Plane’s study. Byusing an extreme bound analysis and a larger sample of countries we havebeen able to undertake a more comprehensive and rigorous analysis of the


TABLE 7

THE EBA RESULTS (AT 0.05 LEVEL) FOR THE PRIVATISATION VARIABLEEXCLUDING OUTLIERS

Specification III (LGYP + LIFE + POP + GOVC)

Coeffi- t-sta- Z-variables included Likely outliers EBA resultcient tistics excluded

Cook’s (13) Min –0.357 –3.23 FDI, SDM3, DEBT Sierra Leone, Max –0.209 –2.11 GDI, FDI, LA Oman, RobustBase –0.256** –2.40 – Singapore,

Barbados

Specification IV (LGYP + LIFE + POP)

Coeffi- t-sta- Z-variables included Likely outliers EBA resultcient tistics excluded

Studentised Min –0.373 –3.21 OPEN, FDI, DEBT Singapore, (14) Max –0.203 –2.01 GOVC, GDI, BUD Cameroon, Robust

Base –0.271** –2.41 – Malaysia

Cook’s (15) Min –0.383 –3.34 OPEN, FDI, DEBT Sierra Leone,Max –0.214 –2.02 FDI, EA, LA Sinagpore, RobustBase –0.288** –2.61 – Barbados

Notes: Dependent variable: GYP. The EBA results for PRIV are obtained by including the controlvariables and three regressors from the pool of the Z variables. ** = statistically significantat 0.05 level. The likely outliers excluded are in the order of higher absolute values.

396jds05.qxd 29/08/03 14:13 Page 143

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

relationship between privatisation and economic growth. The analysis hasalso been strengthened by the application of outlier detection techniques.Despite these improvements in methodology, the results obtained from thistype of analysis must continue to be treated with a fair degree of caution.

Are there policy implementations arising from this analysis? Clearly, thefindings do not refute the notion that the wider economic and socio-politicalenvironment may have important effects on economic growth, and on thesuccess of privatisation. The IMF study, using a limited number ofcountries, has suggested that the positive relationship between privatisationand economic growth could be explained by privatisation acting as a proxyin their analysis for a range of structural measures signifying changes in theeconomic environment. We have attempted, with our measure ofprivatisation, to rule out the possibility that this variable captures the effectsof wider policy reforms.

If, as is argued in the theoretical literature, the change in ownership aloneat the microeconomic level is not sufficient to guarantee greater enterpriseefficiency, then other reforms, more directly related to enterprisedevelopment, may indeed play a crucial role. If the success of privatisationis linked to competition and the regulation of competition, then weaknessesin these fields may explain why privatisation is negatively related toeconomic growth in developing countries. Recent reviews of competitionpolicy in developing countries indicate fundamental weaknesses inimplementation [Cook, 2001]. Similarly, regimes for regulating competitionin developing countries have not developed uniformly and with the samedegree of effectiveness across developing countries. The share of utilities inprivatisation has increased significantly in the 1990s in developing countries,resulting in the creation of numerous private sector monopolies and the needfor better regulation. Even in countries in Latin America where regulatorysystems have developed they may not be working effectively [Cook, 1999].In addition, the IMF report evidence of weak regulatory systems in thetelecommunications sector in some developing countries, which may havecontributed to enhancing the proceeds from privatisation [Davis, Ossowski,Richardson and Barnett, 2000]. The combination of weak regulation andhigh proceeds from privatisation, if accompanied by low gains in economicefficiency, may also provide an explanation for the disappointing results asfar as economic growth is concerned.

The policy implications arising from this analysis squarely point to theneed for further research into the policy environment, in particular the rolesplayed by competition and its regulation if the relation betweenprivatisation and economic growth is to be better understood.

final revision accepted July 2002


396jds05.qxd 29/08/03 14:13 Page 144

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

REFERENCES

Alchian, A., 1965, ‘Some Economics of Property Rights’, II Politico, Vol.30, pp.816–29.Alesina, A. and R. Petrotii, 1993, ‘Income Distribution, Political Instability and Investment’,

NBER Working Paper, No.4486.Barnett, S., 2000, ‘Evidence on the Fiscal and Macroeconomic Impact of Privatisation’, IMF

Working Paper, July, Washington, DC: IMF.Barro, R., 1991, ‘Economic Growth in a Cross-Section of Countries’, Quarterly Journal of

Economics, Vol.106, No.2, pp.407–43.Barro, R. and X. Sala-i-Martin, 1995, Economic Growth, New York: McGraw-Hill.Beckman, R. and R. Cook, 1983, ‘Outliers’, Technometrics, Vol.25, pp.119–49.Belsey, D., Kuh, E. and R. Welsch, 1980, Regression Diagnostics, New York: John Wiley.Bennell, P., 1997, ‘Privatisation in Sub-Saharan Africa: Progress and Prospects during the

1990s’, World Development, Vol.25, No.11, pp.1785–803.Blomstrom, M., Lipsey, R. and M. Zejan, 1993, ‘Is Fixed Investment the Key to Economic

Growth?’ NBER Working Paper, No.4436.Boubakri, N. and J. Cosset, 1998, ‘The Financial and Operating Performance of Newly Privatised

Firms: Evidence from Developing Countries’, unpublished draft, Universite Laval.Boycko, M., Shleifer, A. and R. Vichny, 1996, ‘A Theory of Privatisation’, Economic Journal,

Vol.106, pp.309–19.Cook, P., 1999, ‘Privatisation and Utility Regulation in Developing Countries: The Lessons So

Far’, Annals of Public and Cooperative Economics, Vol.70, No.4, pp.549–87.Cook, P., 2001, ‘Competition and its Regulation: Key Issues’, paper presented at an international

workshop on ‘Competition and Regulation’, University of Manchester, 9 Feb. 2001Cook, P. and C. Kirkpatrick, 1988, Privatisation in Less Developed Countries, Brighton:

Weatsheaf.Cook, P. and R. Fabella, 2002, ‘The Welfare and Political Economy Dimensions of Private vs.

State Enterprise’, The Manchester School, Vol.70, No.2, pp.246–61.Cook, R., 1977, ‘Detection of Influential Observations in Linear Regression’, Technometrics,

Vol.19, pp.15–18.Cooley, T. and S. Leroy, 1981, ‘Identification and Estimation of Money Demand’, American

Economic Review, Vol.71, No.5, pp.825–44.Davis, J., Ossowski, R., Richardson, T. and S. Barnett, 2000, ‘Fiscal and Macroeconomic Impact

of Privatisation’, Occasional Paper, No.194, Washington, DC: IMF.DeLong, J. and L. Summers, 1991, ‘Equipment Investment and Economic Growth’, Quarterly

Journal of Economics, Vol.106, No.2, pp.445–502.Deravi, K., Hegji, C. and D. Moberly, 1990, ‘Government Debt and the Demand for Money: An

Extreme Bound Analysis’, Economic Inquiry, Vol.28, No.2, pp.390–401.Donald, S. and G. Maddala, 1993, ‘Identifying Outliers and Influential Observations in

Econometric Models’, in G. Maddala, C. Rao and H. Vinod (eds.), Handbook of Statistics,Vol.2, Amsterdam: Elsevier.

Doppelhofer, G., Miller, R. and X. Sala-i-Martin, 2000, ‘Determinants of Long-Term Growth: ABayesian Averaging of Classical Estimates (BACE) Approach’, OECD Working Paper, No.266.

Easterly, W. and S. Rebelo, 1993, ‘Fiscal Policy and Economic Growth: An EmpiricalInvestigation’, Journal of Monetary Economics, Vol.32, No.2, pp.417–58.

Fiebig, D., 1992, ‘Diagnostic Checking in Practice: The Case of Outliers and Influential Data’,mimeo, University of Sydney.

Folster, S. and M. Henrekson, 2001, ‘Growth Effects of Government Expenditure and Taxationin Rich Countries’, European Economic Review, Vol.45, No.8, pp.1501–20.

Galal, A., Jones, L., Tandon, P. and I. Vogelsang, 1994, Welfare Consequences of Selling PublicEnterprises: An Empirical Analysis, New York: Oxford University Press.

Gentleman, J. and M. Wilk, 1975, ‘Detecting Outliers II: Supplementing the Direct Analysis ofResiduals’, Biometrics, Vol.31, No.2, pp.387–410.

Harberger, A., 1987, ‘Comment’, NBER Macroeconomics Annual, pp.255–8.Jensen, M. and W. Meckling, 1976, ‘Theory of the Firm: Managerial Behavior, Agency Costs,

and Ownership Structure’, Journal of Financial Economics, Vol.3, No.4, pp.305–60.


396jds05.qxd 29/08/03 14:13 Page 145

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

Kalemili-Ozcan, S., Ryder, H. and D. Weil, 2000, ‘Mortality Decline, Human Capital Investment,and Economic Growth’, Journal of Development Economics, Vol.62, No.1, pp.1–23.

Kikeri, S., 1998, ‘Privatisation and Labor: What Happens to Countries When GovernmentsDivest’, World Bank Technical Paper, No.396, Washington, DC: World Bank.

Kikeri, S., Nellis, J. and M. Shirley, 1994, Privatisation: The Lessons of Experience, Washington,DC: World Bank.

Laffont, J. and J. Tirole, 1991, ‘Privatisation and Incentives’, Journal of Law, Economics andOrganisation, Vol.7, No.1, pp.84–105.

Leamer, E., 1978, Specification Searches: Ad Hoc Inference from Non-Experimental Data, NewYork: Wiley.

Leamer, E., 1983, ‘Let’s Take the Con Out of Econometrics’, American Economic Review,Vol.73, No.1, pp.31–43.

Leamer, E., 1985, ‘Sensitivity Analysis Would Help’, American Economic Review, Vol.75, No.3,pp.308–13.

Leamer, E. and L. Herman, 1983, ‘Reporting the Fragility of Regression Estimates’, Review ofEconomics and Statistics, Vol.65, No.2, pp.306–17.

Levine, R. and D. Renelt, 1992, ‘A Sensitivity Analysis of Cross-Country Growth Regressions’,American Economic Review, Vol.82, No.4, pp.942–63.

Levine, R. and S. Zervos, 1993, ‘What We Have Learned About Policy and Growth from Cross-Country Regressions?’ American Economic Review, Vol.83, No.2, pp.426–30.

Londregan, J. and K. Poole, 1990, ‘Poverty, the Coup Trap, and the Seizure of Executive Power’,World Politics, Vol.423, No.2, pp.151–83.

Lorgelly, P. and P. Owen, 1999, ‘The Effect of Female and Male Schooling on Economic Growthin the Barro-Lee Model’, Empirical Economics, Vol.24, No.3, pp.537–57.

Luke, P. and M. Schaffer, 2000, ‘Wage Discrimination in Russia: An Econometric Investigation’,William Davidson Institute Working Paper Series, No.295, University of Michigan BusinessSchool.

Mankiw, N., Romer, D. and D. Weil, 1992, ‘A Contribution to the Empirics of EconomicGrowth’, Quarterly Journal of Economics, Vol.107, No.2, pp.407–37.

Martin, S. and D. Parker, 1997, The Impact of Privatisation: Ownership and CorporatePerformance in the UK, London: Routledge.

Megginson, W., Nash, R., and M. Van Randenborgh, 1994, ‘The Financial and OperatingPerformance of Newly Privatised Firms: An International Empirical Analysis’, Journal ofFinance, Vol.49, No.2, pp.403–52.

Megginson, W and J. Netter, 2001, ‘From State to Market: A Survey of Empirical Studies onPrivatisation’, Journal of Economic Literature, Vol.39, No.2, pp.321–89.

Parker, D., 1999, ‘Privatisation in the European Union: A Critical Assessment of itsDevelopment, Rationale and Consequences’, Economic and Industrial Democracy, Vol.20,pp.9–38.

Plane, P., 1997, ‘Privatisation and Economic Growth: An Empirical Investigation from a Sampleof Developing Market Economies’, Applied Economics, Vol.29, No.2, pp.161–78.

Ram, R and T. Schulz, 1979, ‘Life Span, Health, Savings, and Productivity’, EconomicsDevelopment and Cultural Change, Vol.27, No.3, pp.399–421.

Rousseeuw, P. and A. Leroy, 1987, Robust Regression and Outlier Detection, New York: JohnWiley.

Sala-i-Martin, X., 1994, ‘Cross-Sectional Regressions and the Empirics of Economic Growth’,European Economic Review, Vol.38, No.3/4, pp.739–47.

Sala-i-Martin, X., 1997, ‘I Just Run Two Million Regressions’, American Economic Review,Vol.87, No.2, pp.176–83.

Stokey, N., 1994, ‘Comments on Barro and Lee’, Carnegie-Rochester Conference Series onPublic Policy, Vol.40, pp.47–57.

Temple, J., 1999, ‘The New Growth Evidence’, Journal of Economic Literature, Vol.XXXVII,pp.112–56.

Tullock, G., 1965, The Politics of Bureaucracy, Washington, DC: Public Affairs Press.UNCTAD, 1996, Comparative Experiences with Privatisation: Policy Insights and Lessons

Learned, New York: United Nations.


396jds05.qxd 29/08/03 14:13 Page 146

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

Vickers, J. and G. Yarrow, 1988, Privatisation: An Economic Analysis, London: MIT PressYarrow, G., 1999, ‘Theory of Privatisation, or Why Bureaucrats are still in Business’, World

Development, Vol.27, No.1, pp.157–68.Welsch, R., 1982, ‘Influence Functions and Regression Diagnostics’, in R. Launer and A. Siegel

(eds.), Modern Data Analysis, New York: Academic Press, pp.149–69. Welsch, R. and E. Kuh, 1977, ‘Technical Report’, in Linear Regression Diagnostics, 923-77,

Cambridge, MA: Sloan School of Management, MIT.World Bank, 1994, World Bank Assistance to Privatization in Developing Countries, OED

Report, No.3273, Aug. World Bank, 1995a, Bureaucrats in Business: The Economics and Politics of Reform, Oxford:

Oxford University Press, for World Bank.World Bank, 1995b, Structural and Sectoral Adjustment Lending: World Bank Experience

(1980–1992), Operations Evaluation Department, Washington, DC: World Bank.

APPENDIX 1

Consider the following three variable linear regression:

yt = xtβ + z1tγ1 + z2tγ2 + ut (1)

where β is the coefficient of particular interest to be estimated, x is the variable of particularinterest or the variable that is always included in a regression, z denotes the ‘doubtful’ variables,and γ is the corresponding coefficient to be estimated. Here, the z variables represent specificationuncertainty, and cannot be excluded a priori.

A Bayesian approach assumes that it is possible to define a prior distribution with the priorlocation of such doubtful variables and to specify their covariance matrix, thereby obtaining aposterior distribution for β. In practice, however, it is difficult to specify such a matrix. Leamer(1978) has developed a theorem in which specification of the prior covariance matrix is notrequired, assuming that specification of the prior location and the sample covariance matrix isonly necessary to constrain the posterior means to lie within a certain ellipsoid or the locus ofconstrained estimates.

Suppose that the object is to show that β is large or that the estimation of γ1 and γ2 is difficult.In a conventional approach, four different regressions would be computed using differentcombinations of the control variables, z1 and z2. Then, among the four different results obtained,the most favourable result would be reported. What Leamer has developed is an alternativeprocedure in which the specification search is enlarged and reports consist of the most favourableand least favourable results.

To enlarge the search for more specifications, a composite variable is defined as follows:

wt(θ) = z1t + θz2t, (2)

where θ is a number to be selected and is allowed to take any value.Then, equation (1) can be rewritten as:

yt = xtβ + wt(θ)γ + ut. (3)

The search can be enlarged by including all values of θ. For each value of θ, a differentconstraint of the form, γ2 = θγ1 is imposed, and there exits a least squares estimate of β, β(θ). Bymaximising β(θ) with respect to θ, the most favourable value of θ can be obtained, and the leastfavourable value can be obtained by minimising the value.

Figure A1 shows the theorem graphically. In a conventional approach, four differentregressions are estimated. These are indicated by the four points, (γ1*,γ2*), (γ1*,0), (0,γ2*), (0,0)in the figure. However, there is no reason why the search should be restricted by only estimatingthese four regressions. To enlarge the search, any values for θ can be used and a differentregression is estimated and the corresponding focus coefficient β* (θ) for each value of θ. Thedished line in the figure depicts the ellipse of constrained estimates defined by the set of allpossible values of (γ1,γ2) that are obtained by changing the value of over the real line. Thus, theellipse contains all posterior means for the distribution of (γ1, γ2).


396jds05.qxd 29/08/03 14:13 Page 147

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08


However, the extreme values, max β* (θ) and min β* (θ), may be located in an unlikelyparameter space on the locus of constrained estimates in relation to the data. To resolve thisproblem, define a statistically feasible likelihood ellipse, say at 0.05 level or 95 per cent, whichis depicted in the dashed lines in the figure, within the likelihood ellipse. This is defined by theassumption that all doubtful variables are included in the regression. The intersection of thepoints in the interior of the ellipse of constrained estimates and this likelihood ellipse is the set ofpoints (γ1

*, γ2*). The area defined by the set is shown by the shaded area in the figure.

Thus, the set of all posterior means for the distribution of (γ1*, γ2

*) is here subject to furtherconstraint in this area, eliminating the unlikely parameter space. The parameter pairs in the areacan be obtained by posterior estimates from prior distribution centred at the origin, and themeasure of specification uncertainty is the difference between the extreme values of β* (θ) overthe area.

FIGURE A1

ELLIPSOIDS OF CONSTRAINED LEAST SQUARES POINTS

396jds05.qxd 29/08/03 14:13 Page 148

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

AP

PE

ND

IX 2

PRIV

GY

PL

GY

PL

IFE

POP

CO

NST

AB

OPE

NG

DI

BU

DFD

IIM

PSD

IMP

INFL

SDIN

FM

3SD

M3

DE

BT

Arg

entin

a0.

100.

028.

5871

.01

0.01

0.04

6.00

0.16

0.17

–0.0

10.

010.

220.

245.

6510

.78

0.18

0.05

0.04

Ban

glad

esh

0.00

0.03

7.17

53.3

00.

020.

041.

000.

230.

19–0

.06

0.00

0.80

0.46

0.05

0.03

0.26

0.03

0.02

Bar

bado

s0.

030.

008.

9174

.34

0.00

0.20

0.00

0.98

0.15

–0.0

20.

010.

080.

070.

020.

020.

620.

060.

07B

eliz

e0.

070.

038.

0972

.57

0.03

0.15

0.00

1.16

0.26

–0.0

30.

030.

150.

100.

030.

020.

430.

030.

05B

enin

0.01

0.01

6.90

50.8

50.

030.

111.

000.

580.

16–0

.07

0.00

0.03

0.01

0.07

0.11

0.26

0.03

0.02

Bol

ivia

0.05

0.02

7.42

56.8

10.

020.

133.

000.

480.

15–0

.03

0.03

0.02

0.02

0.12

0.04

0.35

0.11

0.06

Bra

zil

0.04

0.01

8.34

64.7

60.

020.

174.

000.

170.

21–0

.08

0.01

0.29

0.32

10.2

19.

530.

440.

180.

03B

urki

na F

aso

0.00

0.01

6.26

45.8

70.

020.

146.

000.

380.

220.

000.

000.

030.

010.

050.

090.

230.

010.

02B

urun

di0.

00–0

.03

6.33

46.7

10.

020.

126.

000.

350.

13–0

.06

0.00

0.27

0.13

0.10

0.07

0.19

0.02

0.04

Cam

eroo

n0.

00–0

.04

7.19

53.3

10.

030.

112.

000.

400.

16–0

.03

0.00

0.03

0.01

0.04

0.06

0.19

0.04

0.05

Chi

le0.

010.

068.

2972

.67

0.02

0.10

0.00

0.61

0.25

0.03

0.04

0.13

0.06

0.13

0.07

0.41

0.02

0.08

Col

ombi

a0.

030.

028.

0868

.23

0.02

0.13

3.00

0.34

0.20

–0.0

10.

020.

090.

050.

240.

030.

320.

070.

08C

osta

Ric

a0.

000.

028.

1274

.84

0.02

0.16

0.00

0.82

0.26

–0.0

10.

030.

020.

070.

180.

050.

410.

030.

08C

ote

d’Iv

oire

0.02

–0.0

17.

2650

.96

0.03

0.15

1.00

0.68

0.11

–0.1

40.

010.

030.

010.

050.

130.

280.

010.

13E

cuad

or0.

000.

017.

9567

.04

0.02

0.10

2.00

0.57

0.20

–0.0

20.

020.

130.

100.

420.

160.

270.

060.

09E

gypt

, Ara

b R

ep.

0.02

0.02

7.55

61.0

70.

020.

112.

000.

530.

22–0

.04

0.02

0.07

0.19

0.13

0.05

0.87

0.03

0.06

Gha

na0.

040.

026.

7055

.96

0.03

0.11

0.00

0.50

0.18

–0.0

40.

010.

090.

100.

290.

100.

170.

020.

07G

uate

mal

a0.

000.

017.

6659

.69

0.03

0.06

12.0

00.

420.

15–0

.01

0.01

0.10

0.09

0.16

0.11

0.25

0.02

0.03

Gui

nea

0.00

0.02

6.65

42.5

50.

030.

100.

000.

490.

19–0

.07

0.00

0.07

0.05

0.12

0.11

0.07

0.04

0.04

Gui

nea-

Bis

sau

0.00

0.02

6.48

41.5

60.

020.

081.

000.

510.

31–0

.26

0.00

0.45

0.20

0.54

0.24

0.16

0.02

0.05

Hon

dura

s0.

010.

017.

2665

.40

0.03

0.13

3.00

0.75

0.28

–0.0

60.

020.

260.

300.

180.

080.

330.

030.

11In

dia

0.01

0.04

7.09

57.8

50.

020.

1111

.00

0.21

0.23

–0.0

60.

000.

100.

040.

090.

030.

480.

020.

03In

done

sia

0.01

0.06

7.46

60.2

10.

020.

081.

000.

510.

300.

000.

010.

070.

080.

090.

020.

430.

080.

09Ir

an, I

slam

ic R

ep.

0.00

0.02

8.08

64.7

30.

020.

120.

000.

390.

28–0

.02

0.00

2.29

1.19

0.27

0.09

0.53

0.09

0.04

Jam

aica

0.07

0.01

7.80

72.4

10.

010.

142.

001.

190.

31–0

.03

0.03

0.18

0.09

0.28

0.16

0.53

0.03

0.18

Jord

an0.

00–0

.02

8.16

66.7

10.

040.

250.

001.

320.

31–0

.02

0.00

0.05

0.03

0.06

0.06

1.18

0.15

0.13

Ken

ya0.

010.

006.

8057

.55

0.03

0.16

3.00

0.62

0.21

–0.0

50.

000.

140.

160.

130.

070.

470.

050.

10K

orea

, Sou

th0.

010.

078.

6369

.27

0.01

0.10

0.00

0.64

0.36

–0.0

10.

000.

010.

050.

060.

020.

670.

140.

03M

alaw

i0.

000.

016.

2044

.88

0.03

0.16

0.00

0.60

0.19

–0.0

70.

000.

290.

160.

290.

240.

230.

040.

06M

alay

sia

0.08

0.06

8.36

69.5

00.

030.

131.

001.

650.

36–0

.03

0.06

0.01

0.01

0.04

0.01

1.05

0.27

0.10

396jds05.qxd 29/08/03 14:13 Page 149

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

Mal

i0.

000.

016.

2546

.51

0.03

0.13

2.00

0.54

0.20

–0.0

50.

000.

030.

010.

060.

100.

220.

010.

03M

auri

tani

a0.

000.

016.

6749

.46

0.03

0.13

0.00

1.02

0.20

–0.0

20.

010.

700.

570.

060.

030.

230.

050.

12M

exic

o0.

050.

018.

5869

.82

0.02

0.10

0.00

0.44

0.23

0.00

0.02

0.05

0.05

0.31

0.29

0.25

0.07

0.08

Mor

occo

0.02

0.01

7.66

62.0

10.

020.

172.

000.

570.

22–0

.04

0.01

0.05

0.04

0.04

0.02

0.66

0.08

0.10

Moz

ambi

que

0.01

0.03

6.63

43.4

70.

020.

136.

000.

520.

17–0

.27

0.01

0.25

0.20

0.43

0.14

0.33

0.05

0.06

Nep

al0.

000.

036.

9152

.02

0.03

0.09

1.00

0.46

0.22

–0.0

80.

000.

300.

110.

100.

040.

350.

030.

02N

icar

agua

0.02

–0.0

17.

2762

.17

0.03

0.22

7.00

0.80

0.23

–0.1

50.

020.

410.

6331

.11

45.4

40.

360.

160.

13N

iger

ia0.

010.

026.

8548

.31

0.03

0.13

3.00

0.74

0.18

–0.1

00.

040.

640.

500.

360.

240.

200.

050.

09O

man

0.00

0.01

8.89

67.9

60.

040.

330.

000.

860.

16–0

.09

0.01

0.02

0.01

0.00

0.09

0.31

0.02

0.07

Paki

stan

0.01

0.02

7.22

58.2

40.

030.

146.

000.

360.

16–0

.08

0.01

0.06

0.03

0.10

0.02

0.44

0.03

0.05

Pana

ma

0.06

0.01

7.94

71.7

40.

020.

1711

.00

0.73

0.19

0.02

0.02

0.00

0.00

0.02

0.02

0.52

0.13

0.07

Papu

a N

ew G

uine

a0.

050.

017.

3953

.93

0.02

0.20

9.00

0.97

0.25

–0.0

30.

040.

100.

060.

050.

060.

340.

030.

14Pa

ragu

ay0.

000.

017.

6167

.60

0.03

0.08

0.00

0.82

0.23

0.01

0.02

0.23

0.25

0.20

0.09

0.26

0.05

0.05

Peru

0.09

0.00

7.91

64.3

90.

020.

0812

.00

0.26

0.21

–0.0

20.

020.

360.

4910

.13

20.1

10.

190.

040.

03Ph

ilipp

ines

0.02

0.01

7.42

64.0

00.

020.

1116

.00

0.72

0.23

–0.0

40.

020.

050.

020.

090.

030.

460.

120.

07Se

nega

l0.

020.

007.

0748

.25

0.03

0.14

0.00

0.62

0.15

–0.0

10.

010.

030.

010.

040.

240.

230.

010.

06Si

erra

Leo

ne0.

00–0

.05

6.79

36.8

10.

020.

101.

000.

410.

07–0

.06

0.00

0.88

0.93

0.48

0.27

0.13

0.03

0.06

Sing

apor

e0.

080.

079.

2473

.59

0.02

0.09

0.00

3.68

0.35

0.12

0.10

0.02

0.01

0.04

0.02

1.18

0.04

0.11

Sout

h A

fric

a0.

02–0

.01

8.12

60.4

30.

020.

201.

000.

480.

17–0

.05

0.01

0.04

0.02

0.12

0.03

0.46

0.02

0.03

Sri L

anka

0.02

0.04

7.61

70.7

00.

010.

1012

.00

0.73

0.24

–0.1

00.

010.

130.

100.

110.

030.

400.

020.

04Ta

nzan

ia0.

010.

006.

2851

.01

0.03

0.17

0.00

0.51

0.22

–0.0

90.

010.

300.

230.

260.

080.

240.

030.

04T

haila

nd0.

000.

078.

0067

.49

0.01

0.10

5.00

0.80

0.39

0.02

0.02

0.03

0.01

0.05

0.01

0.79

0.07

0.06

Togo

0.01

0.00

6.46

51.4

10.

030.

138.

000.

750.

16–0

.05

0.00

0.03

0.01

0.07

0.11

0.32

0.05

0.04

Tri

nida

d an

d 0.

060.

009.

0070

.39

0.01

0.13

2.00

0.83

0.16

–0.0

20.

050.

230.

180.

070.

070.

530.

050.

10To

bago

Tun

isia

0.00

0.02

7.90

65.6

30.

020.

160.

000.

890.

26–0

.04

0.02

0.05

0.05

0.06

0.02

0.50

0.02

0.10

Tur

key

0.01

0.02

8.14

64.2

50.

020.

115.

000.

390.

24–0

.05

0.00

0.03

0.04

0.75

0.15

0.29

0.04

0.06

Uga

nda

0.01

0.04

6.27

48.3

40.

030.

092.

000.

300.

14–0

.07

0.01

0.38

0.40

0.48

0.60

0.10

0.01

0.04

Uru

guay

0.00

0.02

8.45

72.1

10.

010.

132.

000.

430.

13–0

.03

0.00

0.12

0.08

0.58

0.28

0.49

0.09

0.06

Ven

ezue

la0.

040.

008.

8270

.53

0.02

0.08

8.00

0.54

0.18

–0.0

20.

020.

190.

360.

500.

310.

350.

080.

08

AP

PE

ND

IX 2

(con

t.)

PRIV

GY

PL

GY

PL

IFE

POP

CO

NST

AB

OPE

NG

DI

BU

DFD

IIM

PSD

IMP

INFL

SDIN

FM

3SD

M3

DE

BT

396jds05.qxd 29/08/03 14:13 Page 150

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

AP

PE

ND

IX 2

(con

t.)

PRIV

GY

PL

GY

PL

IFE

POP

CO

NST

AB

OPE

NG

DI

BU

DFD

IIM

PSD

IMP

INFL

SDIN

FM

3SD

M3

DE

BT

Vie

tnam

0.00

0.05

6.79

65.8

00.

020.

080.

000.

690.

210.

000.

040.

460.

730.

701.

230.

220.

010.

03Y

emen

0.00

0.00

7.54

51.4

90.

040.

190.

000.

610.

20–0

.08

0.04

0.17

0.07

0.25

0.16

0.51

0.07

0.51

Zam

bia

0.05

–0.0

16.

6349

.63

0.03

0.17

1.00

0.72

0.13

–0.1

10.

020.

690.

680.

770.

500.

220.

070.

17Z

imba

bwe

0.01

0.01

6.99

56.9

10.

030.

190.

000.

620.

21–0

.09

0.00

0.29

0.16

0.21

0.06

0.40

0.05

0.08

Mea

n0.

020.

027.

5159

.83

0.02

0.13

3.06

0.67

0.21

–0.0

50.

020.

220.

181.

091.

480.

390.

060.

08St

anda

rd D

evia

tion

0.03

0.02

0.80

9.95

0.01

0.05

3.88

0.47

0.06

0.06

0.02

0.33

0.24

4.28

6.40

0.23

0.05

0.07

Not

es:

PRIV

= t

he m

agni

tude

of

priv

atis

atio

n as

the

rat

io o

f cu

mul

ativ

e pr

ocee

ds f

rom

pri

vatis

atio

n du

ring

the

sam

ple

peri

od 1

988-

97 t

o av

erag

e G

DP

duri

ng t

he s

ame

peri

od;

prim

ary

data

for

the

pro

ceed

s ar

e ta

ken

from

Wor

ld B

ank

Priv

atiz

atio

n D

atab

ase

and

Priv

atis

atio

n In

tern

atio

nal

(var

ious

issu

es);

and

the

data

for

GD

P ar

e ta

ken

from

Wor

ld B

ank

Glo

bal D

evel

opm

ent N

etw

ork

Gro

wth

Dat

abas

e.G

YP

= t

he a

vera

ge r

eal

GD

P pe

r ca

pita

(in

con

stan

t U

S do

llars

, int

erna

tiona

l pr

ices

, bas

e ye

ar 1

985)

gro

wth

rat

e du

ring

the

sam

ple

peri

od;

and

the

data

are

take

n fr

om W

orld

Ban

k G

loba

l Dev

elop

men

t Net

wor

k G

row

th D

atab

ase.

LG

YP

=th

e lo

g of

ini

tial

real

GD

P pe

r ca

pita

in

1988

cap

ita (

in c

onst

ant

US

dolla

rs, i

nter

natio

nal

pric

es, b

ase

year

198

5);

and

prim

ary

data

are

take

n fr

om W

orld

Ban

k G

loba

l Dev

elop

men

t Net

wor

k G

row

th D

atab

ase

POP

= t

he a

vera

ge p

opul

atio

n gr

owth

rat

e du

ring

the

sam

ple

peri

od;

and

the

data

are

tak

en f

rom

Wor

ld B

ank

Glo

bal

Dev

elop

men

t N

etw

ork

Gro

wth

Dat

abas

eL

IFE

= t

he i

nitia

l lif

e ex

pect

ancy

at

birt

h in

198

7 (t

he d

ata

in 1

988

are

not

avai

labl

e);

and

the

prim

ary

data

are

tak

en f

rom

Wor

ld B

ank

Glo

bal

Dev

elop

men

t Net

wor

k G

row

th D

atab

ase

STA

B=

pol

itica

l in

stab

ility

cal

cula

ted

as t

he a

vera

ge n

umbe

r of

maj

or p

oliti

cal

cris

es, r

evol

utio

ns, a

nd c

oups

dur

ing

the

sam

ple

peri

od;

and

the

prim

ary

data

are

take

n fr

om W

orld

Ban

k G

loba

l Dev

elop

men

t Net

wor

k G

row

th D

atab

ase

BU

D=

the

aver

age

ratio

of

cent

ral g

over

nmen

t bud

get s

urpl

us/d

efic

it to

GD

P (c

alcu

late

d in

nat

iona

l cur

renc

y, c

onst

ant p

rice

) du

ring

the

sam

ple

peri

od; a

nd th

e pr

imar

y da

ta a

re ta

ken

from

Wor

ld B

ank

Glo

bal D

evel

opm

ent N

etw

ork

Gro

wth

Dat

abas

eO

PEN

= o

penn

ess

of a

cou

ntry

cal

cula

ted

as (

expo

rt-i

mpo

rt)/

GD

P; a

nd th

e pr

imar

y da

ta a

re ta

ken

from

Wor

ld B

ank

Glo

bal D

evel

opm

ent N

etw

ork

Gro

wth

Dat

abas

eC

ON

=th

e av

erag

e ra

tio o

f go

vern

men

t con

sum

ptio

n to

GD

P (c

alcu

late

d in

nat

iona

l cur

renc

y, c

onst

ant p

rice

) du

ring

the

sam

ple

peri

od; a

nd th

epr

imar

y da

ta a

re ta

ken

from

Wor

ld B

ank

Wor

ld D

evel

opm

ent I

ndic

ator

sG

DI

= t

he a

vera

ge r

atio

of

gros

s do

mes

tic in

vest

men

t to

GD

P (c

alcu

late

d in

nat

iona

l cur

renc

y, c

onst

ant p

rice

) du

ring

the

sam

ple

peri

od; a

nd th

epr

imar

y da

ta a

re ta

ken

from

Wor

ld B

ank

Wor

ld G

loba

l Dev

elop

men

t Net

wor

k G

row

th D

atab

ase

396jds05.qxd 29/08/03 14:13 Page 151

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08


AP

PE

ND

IX 2

(con

t.)

FDI

= t

he a

vera

ge r

atio

of

fore

ign

dire

ct i

nves

tmen

t (n

et i

nflo

ws)

to

GD

P (c

alcu

late

d in

cur

rent

US

dolla

rs)

duri

ng t

he s

ampl

e pe

riod

; an

d th

epr

imar

y da

ta a

re ta

ken

from

Wor

ld B

ank

Wor

ld D

evel

opm

ent I

ndic

ator

sIN

FL=

the

ave

rage

infl

atio

n ra

tes

(GD

P de

flat

or)

duri

ng th

e sa

mpl

e pe

riod

; and

the

prim

ary

data

are

take

n fr

om W

orld

Ban

k W

orld

Dev

elop

men

tIn

dica

tors

SDIN

F=

the

sta

ndar

d de

viat

ion

of I

NFL

IMP

= l

og(1

+ in

form

al m

arke

t pre

miu

m);

info

rmal

mar

ket p

rem

ium

is c

alcu

late

d us

ing

the

follo

win

g fo

rmul

a: (

Para

llel E

xcha

nge

Rat

e/O

ffic

ial

Exc

hang

e R

ate

-1)*

100;

the

prim

ary

data

are

take

n fr

om W

orld

Ban

k G

loba

l Dev

elop

men

t Net

wor

k G

row

th D

atab

ase

SDIM

P=

the

sta

ndar

d de

viat

ion

of I

MP

M3

= t

he r

atio

of

liqui

d lia

bilit

ies

(M3)

to G

DP,

the

prim

ary

data

are

take

n fr

om W

orld

Ban

k G

loba

l Dev

elop

men

t Net

wor

k G

row

th D

atab

ase

SDM

3=

the

stan

dard

dev

iatio

n of

M3

DE

BT

=th

e ra

tio o

f to

tal e

xter

nal d

ebt t

o G

DP,

the

prim

ary

data

are

take

n fr

om W

orld

Ban

k G

loba

l Dev

elop

men

t Net

wor

k G

row

th D

atab

ase.

* W

orld

Ban

k G

loba

l Dev

elop

men

t Net

wor

k G

row

th D

atab

ase

is a

vaila

ble

from

ww

w.w

orld

bank

.org

/res

earc

h/gr

owth

/GD

Nda

ta.h

tm

396jds05.qxd 29/08/03 14:13 Page 152

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08


APPENDIX 3

Influential Observations/Outlier Detection Techniques:In general outliers are categorised into two types. The first is characterised as the

substantially large residual arising from the difference between the actual Yi and the fitted Y∧

i. Thesecond is that Xi is far from the cluster of Xi’s, resulting in the direction of the fitted line varying,although this is not indicated by a high residual. Detecting more than one outlier in a dataset is,in practice, difficult owing to the so-called masking and swamping effects [Luke and Wilk, 2000].The masking effect occurs when a subset of outliers is not identified due to the presence ofanother subset of observations. The swamping effect occurs when normal observations aremisidentified as outliers due to the presence of another subset of observations.

One way to detect influential observations/outliers is to analyse the size of residuals from theregression estimation. This can be done by computing studentised residuals. The studentisedresidual can be computed as:

SRi

= ei

/ (s(i)

),

where hi is the ith diagonal element of the hat matrix H, s is the variance from sample residuals,and ei is the ordinary residual. (i) is the standard error obtained by excluding the ith observation,and then s(i) is calculated as:

s(i)2 = ,

where k is the number of explanatory variables and n is the number of observations in theregression.

DFITS, Cook’s Distance, and Welsch Distance summarise the information in the leverageversus residual-squared plot into a single statistics by creating an index that is influenced by thesize of the residuals or outliers and the size of leverage.

DEFITS can be calculated by:

DFITSi = ,

where ri are the studentised residuals and hi are the size of leverage. Cook’s Distance is calculated by:

CDi = 1/k • si2/s2 • DFITSi

2 ,

where s is the root mean square error of the regression, and si is the root mean square error whenthe ith observation is eliminated.

Welsch Distance can be estimated by:

WDi = DFITSi ,

Then, outliers or the observations that ought to be examined further are defined as:

DFITS > 2 , , Cook’s Distance > 4/n, and Welsch Distance > 3 .knk

ih

n

−−

1

1

i

ii

h

hr

−1

−−

−−1

ˆ22

kn

ekns i

1

)1/()( 22

−−−−−

kn

heskn ii

ih−1

396jds05.qxd 29/08/03 14:14 Page 153

Dow

nloa

ded

By:

[Hiro

shim

a U

nive

rsity

] At:

07:1

6 25

Jul

y 20

08

DFBETAs measure the effect on each coefficient by testing β∧

j – β∧

(i)j , which is defined as:

DFBETAsij = ,

where is the (X′ X)–1jj is the jth diagonal element of (X′ X)–1. The observations that ought to be

examined further are defined as DFBETAs > .n/2

1')( )(

)(ˆˆ

−

−

jji

jj

XXs

iββ


396jds05.qxd 29/08/03 14:14 Page 154

Privatization and Economic Growth

Documents

Transcript of Privatization and Economic Growth