 

 

LIBRARY

Michigan State

Unlverslty

 

 

PLACE IN RETURN BOX
to remove this checkout from your record.
TO AVOID FINES return on or before date due.

 

DATE DUE

MTE DUE

DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1M chIRCJDﬂOmpGS—p.“

THREE I
PROE

THREE ESSAYS ON LATIN AMERICAN DEVELOPMENT ISSUES:
PRODUCTIVITY GROWTH, INTERNATIONAL TRADE AND
VIOLENT CRIME
By

Pablo F ajnzylber

A DISSERTATION
Submitted to
Michigan State University
in partial fulﬁllment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY
Department of Economics

1998

THREE I
PROD

The ﬁrst C
ofproductimy gr.
Aim providing 5(
melted and ICVI
W635 and proc
mine this relax
the mimation of c
measures of opem

Dareshwar (1993 )

complement. Our

WWW was r
Wintneg were OPI

The Seconc'
3nd pTOdUCliVitV g.
In W

3mm“

c0“Slder

ABSTRACT
THREE ESSAYS ON LATIN AMERICAN DEVELOPMENT ISSUES:
PRODUCTIVITY GROWTH, INTERNATIONAL TRADE AND
VIOLENT CRIME
By

Pablo Fajnzylber

The ﬁrst essay studies the inﬂuence of openness to international trade on the rates
of productivity growth of 18 Latin American countries during the period 1950-1995.
After providing some background on the macro and trade policies of the countries
involved, and reviewing the theoretical and empirical literature on the link between trade
openness and productivity growth, we apply three types of empirical methodologies to
examine this relationship: a growth accounting analysis, a study of structural breaks, and
the estimation of dynamic panel data regressions of productivity growth on several
measures of openness. The data comes from the databases prepared by Nehru and
Dareshwar (1993) and Easterly, Loayza and Montiel (1997), which we update and
complement. Our main ﬁndings are that, on average, the growth of total factor
productivity was relatively faster during the periods in which the Latin American
countries were open to international trade, but also that the pace of physical capital
accumulation was relatively slower during these periods.

The second essay studies the relationship between openness to international trade
and productivity growth at the industry level, focusing on ﬁve Latin American countries
during the period 1970-1994. After describing the economic performance of the

industries considered, we estimate the effect of different measures of openness on the

youth of labor am
ECLAClNlDO's
relationship betwee
he Latin American

The third es
for the period 1970
Surveys. to anal yze
model of the incent
sections and panel I
rates. deterrence ef}
inertia is significan'

and robbery rates.

Pablo F ajnzylber

growth of labor and total factor productivity at the industry level. The data comes from
ECLAC/UNIDO’s PADI database. The results reject the hypothesis of a general positive
relationship between openness and productivity grth at the industry level, at least for
the Latin American countries considered.

The third essay uses a new data set of crime rates for a large sample of countries
for the period 1970-1994, based on information from the United Nations World Crime
Surveys, to analyze the determinants of national homicide and robbery rates. A simple
model of the incentives to commit crimes is proposed and estimated using both cross-
sections and panel data. The results show that increases in income inequality raise crime
rates, deterrence effects are signiﬁcant, crime tends to be counter-cyclical, and criminal

inertia is signiﬁcant even after controlling for other potential determinants of homicide

and robbery rates.

FOI Carmelita a;

For Carmelita and Gabriela

I am
and to the Fe
studies at .\l
to my thesis
during the p:
Norman L02
Bank on the
this dissem:
Heather. Hi:
key COmPOI}
Would like 1
her low and

Various SIEp

ACKNOWLEDGMENTS

I am most grateful to the Secretariat of Science and Technology of Brazil (CNPq),
and to the Federal University of Minas Gerais (Brazil), who sponsored my doctoral
studies at Michigan State University. I also would like to express my deepest appreciation
to my thesis advisor, Professor Rowena Pecchenino, for her advice and encouragement
during the preparation of this dissertation. I am also very grateful to Daniel Lederrnan and
Norman Loayza, with whom I collaborated in a research project conducted at the World
Bank on the determinants of criminal behavior worldwide, which led to the third essay of
this dissertation. The ﬁ'iendship of Adriana, Anita, Bryan, Carlos, Denise, Doroteia,
Heather, Hirokatsu, Joy, Luiz, Nelson, Ramon, Rosane, and Sandro, among others, was a
key component of my standard of leaving in East Lansing. Last, but certainly not least, I
would like to acknowledge my biggest appreciation to Carmelita, my wife, not only for
her love and support, but also for her patience in listening to the oral accounts of the

various steps that led to the completion of this dissertation.

LIST OF TABLES

 

LIST OF FTGL'RES
UST OT ABBREV

CHAPTERT
TRADE LIBERA
GROWTH IN LI‘
Introduction .....
Macroeconomic
Trade Liberalila
Trade and Grow
Methodology an
Grosxth
Structur
Regress
3mm and (
References ... .
Appendix A: T
Appendix B: \

CHAPTER 2
OPENNESS T
AMIERICAN
1mmfluction .
IOm 1mm“
GIOWIh and 5
Empirical Sp
Conomemc
STilliiitiOn R

TABLE OF CONTENTS

LIST OF TABLES .................................................................................... viii
LIST OF FIGURES .................................................................................. xi
LIST OF ABBREVIATIONS ....................................................................... xii
CHAPTER 1
TRADE LIBERALIZATION AND TOTAL FACTOR PRODUCTIVITY
GROWTH IN LATIN AMERICA: 1950-1995 .......................................... 1
Introduction .................................................................................... 1
Macroeconomic Policy and Trade Liberalization in Latin America ................... 3
Trade Liberalization and Growth: Old and New Theory ................................ 8
Trade and Growth: The Empirical Evidence .............................................. 15
Methodology and Results ................................................................... 27
Growth Accounting ................................................................... 27
Structural Breaks ....................................................................... 36
Regression Analysis ................................................................... 41
Summary and Conclusions ................................................................... 50
References ...................................................................................... 61
Appendix A: Tables and Figures ............................................................ 68
Appendix B: Updating the Nehru and Dareshwar (1993) Data Base 88
CHAPTER 2
OPENNESS TO TRADE AND PRODUCTIVITY GROWTH IN LATIN
AMERICAN INDUSTRIES: 1970-1994 ........................................................ 91
Introduction ..................................................................................... 91
F rom Import Substitution to Trade Liberalization: Historical Background ........... 97
Growth and Structural Change in Industry: 1970-94 ..................................... 101
Empirical Speciﬁcation ......................................................................... 110
Econometric Methodology .................................................................. 119
Estimation Results .............................................................................. 125
Regressions with Trade Flows-Based Openness Variables 127
Regressions with Trade Policy-Based Openness Variables 129
Summary and Conclusions ................................................................... 133
References 139
Appendix A: Tables and Figures ........................................................... 143
Appendix B: Additional Tables ............................................................. 170
CHAPTER 3
WHAT CAUSES VIOLENT CRIME? ........................................................ 178
Introduction .................................................................................... l 78
Literature Review ............................................................................. 181

vi

A Simple. Reduce
The Data ..........
National I
Explanator
Empirical lmplent
Cross-Sec:|
Panel Reg?
Conclusions ......
References .......
Appendix A: Data
Appendix B: Tab. '
Appendix C: Add:

 

 

 

A Simple, Reduced-Form Model of Criminal Behavior .................................. 189

The Data ......................................................................................... 194
National Crime Rates ................................................................ 195
Explanatory Variables ................................................................ 197

Empirical Implementation .................................................................... 201
Cross-Sectional Regressions ....................................................... 203
Panel Regressions .................................................................... 209

Conclusions 220

References 223

Appendix A: Data Description and Sources ............................................... 226

Appendix B: Tables and Figures ........................................................... 228

Appendix C: Additional Tables ............................................................. 243

vii

Table 1.1: Int"

Table 1.2: 81:
Selected C on

Table 1.3: Gr
Table 1.4: To
Table 1.5: Fe

Table 1.6: CI
OTthe Econm

Table 1.7: Tc
Employment

Table 1.8: St:

Table 1.9 C”
TOT CapaCTI'V

Table 110: C
for Capacity
6 ECOnom}

Tablet”:

\.

Table 1.12;

rn

Table 1-14 C
WOrkeL
Table 1.1 s; (

Table 116 (
Including Cc

Tablel.]7.(

LIST OF TABLES

Table 1.1: Indicators of Trade Regimes Before and After Reform ....................... 69
Table 1.2: Bilateral Real Exchange Rate Relative to the US. Dollar in

Selected Countries, 1980/1993 ............................................................. 70
Table 1.3: Growth Decomposition: 1950/1995 .......................................... 71
Table 1.4: Total Factor Productivity Growth by Sub-periods: 1950/1995 ............ 72
Table 1.5: Periods of Openness: 1950/ 1995 .............................................. 73
Table 1.6: Changes in Growth Rates: Periods of Openness and Closedness

of the Economy, 1950/1995 .................................................................. 74
Table 1.7: Total Factor Productivity Growth Adjusting for

Employment, 1980/1995 ..................................................................... 75
Table 1.8: Structural Trend Breaks in GDP, 1950/1995 ................................. 76

Table 1.9: Growth Rates of Total Factor Productivity Adjusting
for Capacity Utilization: 1950/ 1995 ...................................................... 77

Table 1.10: Changes in Growth Rates of Capital Stocks, Adjusting
for Capacity Utilization: Periods of Openness and Closedness of

the Economy, 1950/1995 .................................................................... 78
Table 1.11: Structural Trend Break Tests in Import-Output Ratios ................... 79
Table 1.12: Structural Trend Break Tests in Export-Output Ratios .................... 80

Table 1.13: Structural Trend Breaks in Total Factor Productivity, 1950/1995 ...... 81

Table 1.14: Structural Trend Breaks in GDP and Capital Stock

per Worker, 1950/ 1995 ........................................................................ 82
Table 1.15: GMM Regressions of Growth in GDP per Worker ........................ 83
Table 1.16: GMM Regressions of Growth in GDP per Worker

Including Control Variables .................................................................... 84
Table 1.17: GMM Regressions of Growth in Capital per Worker ..................... 85

viii

Table 1.18: GM
Including Cont!

Table 2.1: Com

Table 3-33 Com
and lmPOm‘ 19

Table 3.3: EX?“

Table 2.4: R3105

Table 3.5: DeCO
in 28 Industrial

Table 2.6: G.\1.\
on Trade Flows-

Table 2.7: G.\1.\
on Trade F lows-

Table 2.8: G.\l.\
on Trade Flows-

Table 2.9: G.\1\
on Trade F lows-

Table 2.10: CM
on Trade Flows;

Table 1.18: GMM Regressions of Growth in Capital per Worker

Including Control Variables .................................................................. 86
Table 2.1: Composition of Merchandise Exports, 1970/1994 .......................... 144
Table 2.2: Composition of Industrial Value Added, Exports

and Imports, 1970/1994 ....................................................................... 145
Table 2.3: Export and Import Coefﬁcients in Manufacturing

Output, 1970/1994 ........................................................................... 147
Table 2.4: Rates of Growth of Industrial Labor Productivity, 1970/1994 ............. 149
Table 2.5: Decomposition of the Rates of Labor Productivity Grth

in 28 Industrial Sectors ...................................................................... 151
Table 2.6: GMM Regressions of Industry Labor Productivity Growth

on Trade F lows-based Openness Variables: Argentina, 1970/ 1992 ................... 153
Table 2.7: GMM Regressions of Industry Labor Productivity Growth

on Trade Flows-based Openness Variables: Brazil, 1970/1994 ........................ 154
Table 2.8: GMM Regressions of Industry Labor Productivity Growth

on Trade F lows-based Openness Variables: Chile, 1970/ 1994 ........................ 155
Table 2.9: GMM Regressions of Industry Labor Productivity Growth

on Trade Flows-based Openness Variables: Colombia, 1970/1994 .................... 156
Table 2.10: GMM Regressions of Industry Labor Productivity Growth

on Trade Flows-based Openness Variables: Mexico, 1970/ 1994 ....................... 157
Table 2.11: Summary of Regression Results from Tables 6 to 10 ..................... 158

Table 2.12: GMM Regressions of Industry Labor Productivity Grth
on Trade Policy-based Openness Variables: Argentina, 1986/1993 ................... 159

Table 2.13: GMM Regressions of Industry Labor Productivity Growth
on Trade Policy-based Openness Variables: Brazil, 1986/1994 ........................ 160

Table 2.14: GMM Regressions of Industry Labor Productivity Growth
on Trade Policy-based Openness Variables: Mexico, 1986/1994 161

Table 2.15: Summary of Regression Results from Tables 12 to 14 .................... 162

ix

Table 2A1: C las
Factor Intensity ..

Table 2A2: Cod
of the lnternation

Table 2A3: Coc
of Industry Labo
Variables ........

Table 2A4: Tat
Coellicients of ‘
Mexico. 1986311

Table 1A5: Bi-
10 Trade. and th

Table 2A6: Bi
to Trade. and LI

Table EAT: BS
10 Trade. and 1

Table 31 SUIT

Table 3.2: 0L5
Rate, 1970'19c

Table 33 01.:
R316. 1970:19(

Table 34: GM

Table 3-.5 GM

by C°umrs (A

Table 2.A.1: Classiﬁcation of Manufacturing Industries According to
Factor Intensity ............................................................................... 171

Table 2.A.2: Codes and Descriptions of Sectors at the 3-digit Level
of the International Standard Industrial Classiﬁcation .................................. 172

Table 2.A.3: Codes and Descriptions of Sectors used in the Regressions
of Industry Labor Productivity Growth on Trade Policy-based Openness
Variables ......................................................................................... 173

Table 2.A.4: Tariff and Non Tariff Barriers to Trade: Averages and
Coefﬁcients of Variation across 17 Sectors in Argentina, Brasil and

Mexico, 1986/1993 ........................................................................................ 174
Table 2.A.5: Bivariate Correlations between Tariffs, Non-Tariff Barriers
to Trade, and the Growth of Exports and Imports: Argentina, 1986/1993 .......... 175
Table 2.A.6: Bivariate Correlations between Tariffs, Non-Tariff Barriers
to Trade, and the Growth of Exports and Imports: Brazil, 1986/1993 ................ 176
Table 2.A.7: Bivariate Correlations between Tariffs, Non-Tariff Barriers
to Trade, and the Growth of Exports and Imports: Mexico, 1986/1993 .............. 177
Table 3.1: Summary Statistics for Crime, Convictions and Police Rates ............. 229

Table 3.2: OLS Cross-Sectional Regressions of Intentional Homicide

Rate, 1970/1994 .............................................................................. 230
Table 3.3: OLS Cross-Sectional Regressions of Intentional Robbery

Rate, 1970/ 1994 .............................................................................. 232
Table 3.4: GMM Panel Regressions of Intentional Homicide Rate ................... 234
Table 3.5: GMM Panel Regressions of Robbery Rates ................................. 236
Table 3.A.1: Description and Source of the Variables .................................... 244

Table 3.A.2: Summary Statistics of Intentional Homicide Rates
by Country (Annual Data) ................................................................... 249

Figure 1.1: Open (
Figure 1.1: Manul
Figure 2.2: Labor
Figure 3.3: Labor

Figure 2.4: Labor
1971’] 199-1

Figure 2.5: Labor
1970 199-1

Figure 2.6: Labo
19‘0'1994........

Figure 2.7: Labr‘
1970 1994

F1{lure 3.1: [mi

Ficus 3.2: It]
’POPUlation-tte

Figure 3.3; He
19701994

Figure 3.4; Me
1970 1994 ,
FlE‘a‘re 3

y 521m

Figure 3.6: In
theCan'bbea,

LIST OF FIGURES

Figure 1.1: Open Countries as Percent of Total, 1950/1995 ............................. 87
Figure 2.1: Manufacturing Share in GDP, 1970/1994 ................................... 163
Figure 2.2: Labor Productivity in Manufacturing, 1970/1994 .......................... 164
Figure 2.3: Labor Productivity in Semi-Manufactures, 1970/1994 .................... 165
Figure 2.4: Labor Productivity in Other Traditional Industries,

1970/1994 ............................................................................................................... 166
Figure 2.5: Labor Productivity in Basic-Inputs Industries,

1970/1994 ............................................................................................................... 167
Figure 2.6: Labor Productivity in New Labor-Intensive Industries,

1970/ 1994 ............................................................................................................... 168
Figure 2.7: Labor Productivity in New Capital-Intensive Industries,

1970/1994 ............................................................................................................... 169
Figure 3.1: Underlying Determinants of Criminal Activities ................................... 237

Figure 3.2: The World: Intentional Homicide Rate
(population-weighted average) ................................................................................. 238

Figure 3.3: Median Intentional Homicide Rates by Income Groups,
1970/1994 ............................................................................................................... 239

Figure 3.4: Median Intentional Homicide Rates by Regions,
1970/1994 ............................................................................................................... 240

Figure 3.5: Intentional Homicide Rates in South-America and Mexico,
1970/1994 ............................................................................................................... 241

Figure 3.6: Intentional Homicide Rates in Central-America and
the Caribbean, 1970/1994 ......................................................................................... 242

xi

TAP: Economically
ECLAC: Economic

GDP: Gross Domes

 

GNP: Gross Nation.
GMM: Generalized
131C: lntemational 5
.\TB: Non-TarilT B
PADT: Piogram for
TF1: Total Factor

LXCTAD: L'nited

WDQ; United y

LIST OF ABBREVIATIONS

EAP: Economically Active Population.
ECLAC: Economic Commission for Latin America and the Caribbean.
GDP: Gross Domestic Product.

GNP: Gross National Product.

GMM: Generalized Method of Moments.

ISIC: International Standard Industrial Classiﬁcation.

NTB: Non-Tariff Barrier.

PADI: Program for the Analysis of Industrial Dynamics.

TFP: Total Factor Productivity.

UNCTAD: United Nations Conference on Trade and Development.

UNIDO: United Nations Industrial Development Organization.

xii

 

TRADE LlB

 

1- introduction
The impact .

oftheoretical and c1

 

muted stronger a1
“RmbbpiAtu
decades have Witne
“WMMMM
Plocess of dism an
interruptions. sine
reform packages .
Cans. have been

adTUSUIIenL

Pmbm
Wmmmm,
Eula] effects or

research b‘ SILK
‘Wilona,

”Winning Exe

Chapter 1:
TRADE LIBERALIZATION AND TOTAL FACTOR PRODUCTIVITY

GROWTH IN LATIN AMERICA: 1950-1995

1- Introduction

The impact of trade liberalization on economic growth has long been the subject
of theoretical and empirical debate. Recent developments in trade and growth theory have
provided stronger analytical foundations for the arguments on the dynamic effects of
“opening-up”. At the same time, the issue has gained increased attention as the last two
decades have witnessed an unprecedented movement towards economic integration
among nations. In Latin America, in particular, most countries have engaged in a rapid
process of dismantling the protectionist policies that had prevailed, with some
interruptions, since the 19303. The new trade policies have usually been the hallmark of
reform packages encompassing a broad range of market-oriented policies and, in many
cases, have been implemented in the context of aggressive programs of macroeconomic
adjustment.

Partly because of the relatively short period of time that has elapsed since the
implementation of the new policies, few studies have dealt with the measurement of their
actual effects on economic growth. The present paper attempts to contribute to this
research by studying the inﬂuence of openness to international trade on the rates of total
factor productivity (TFP) growth of 18 Latin American countries during the period 1950-
1995. To this end, we perform three types of analysis. The ﬁrst one is a growth

accounting exercise in which the contributions to GDP growth that are associated with

the gronth of. respc
context we examir.1
overall grouth peril
particular. how it h.;
For the categorizati
and Warner (1995)
18 countries liberal ;
WWW) episodes
data that we use cor
Population and outrl
1995.
The second
StmClllral breaks in
richer the rates of
ten for the PIESCnce
per Worker, and an i

sSupF: Tests propo

l:

 

 

 

the growth of, respectively, TFP, capital stocks and the labor force are calculated. In this
context, we examine how the relative importance of TFP grth in the explanation of the
overall growth performance of the countries considered has evolved over time and, in
particular, how it has changed after the implementation of the trade liberalization reforms.
For the categorization and timing of the latter, we follow the criteria suggested by Sachs
and Warner (1995) for the characterization of an economy as “open”. By these criteria, all
18 countries liberalized their trade regimes in the last decade, while 10 of them had
temporary episodes of “openness” in the previous decades (mainly in the 19503). The
data that we use comes from the data base on physical capital stocks, working-age
population and output constructed by Nehru and Dareshwar (1993), which we update to
1995.

The second approach that we follow is that of testing for the existence of
structural breaks in the series of import-output and export-output ratios and examining
whether the rates of TFP growth have increased or decreased after the breaks. We also
test for the presence of structural breaks in the series of GDP per worker, capital stocks
per worker, and an index of TFP. The econometric procedure that we use is based on the
“SupF,” tests proposed by Vogelsang (1994), as implemented by Ben David and Papell
(1997). These tests have the advantage of being general enough to allow for the presence
of unit roots, polynomial trends, and serial correlation.

Finally, the third methodology that we adopt is that of estimating dynamic panel
data regressions of TFP growth on several indicators of openness to international trade.
The econometric technique that we use is based on the Generalized Method of Moments

(GMM) estimator proposed by Blundell and Bond (1997), and controls for the existence

of joint endogeneity
speciﬁc efTects. ThoE
data from Easterly L

The rest oft

macroeconomic poi

 

prtttides some bacl.‘
the theoretical issm

Section 4 provides 1

 

and methodologies

2- Macroeconomic

In Latin Ar
only one compone
encompassing pris

We Policies have

of joint endogeneity in the explanatory variables, as well as for the presence of country-
speciﬁc effects. The variables representative of openness are constructed on the basis of
data ﬁ'om Easterly et a1. (1997) and the World Bank data bases.

The rest of this paper is organized as follows. Section 2 comments on the
macroeconomic policy context in which the new trade policies have been put in place and
provides some background on the extent and speed of these reforms. Section 3 discusses
the theoretical issues involved in the analysis of the trade and growth relationship.
Section 4 provides some previous empirical evidence on this issue. In section 5 the data
and methodologies used in the paper are described and the results of our empirical

exercises are presented. Section 6 offers a summary of results and concluding remarks.

2- Macroeconomic Policy and Trade Liberalization in Latin America

In Latin America and elsewhere, the recent process of trade liberalization has been
only one component of a broader movement toward market-oriented reforms
encompassing privatization and ﬁnancial liberalization. Somewhat paradoxically, the new
trade policies have been implemented in the context of intense macroeconomic
instability, and have often been adopted in conjunction with stabilization packages.
Whether the two sets of policies are jointly sustainable is still an open question and the
importance of the short-run achievements cannot be underscored. In the words of Rodrik,
“the success of reforms will depend less on the direct consequences of the new trade
policies than on the resolution of the macroeconomic difﬁculties in which these countries

are presently engulfed” (1992: 102).

That the tra

 

stabilization pacla,

the "structural" reti

   
  
  
  

the stabilization et‘: .
competition provid
tool in the battle at
potentially associat
competitiveness of
liberalization can a
etlects on the exte1
political...the othc
lndeed. th.
American countn
domesuc price it
real allllfeciatior.
imF‘OFlS 1ban on
dFT-‘TeClEttiOn, (0

.0

That the trade reforms may contribute, to some extent, to the success of the
stabilization packages is not open to discussion. The radical shift in policies involved in
the "structural" reforms has played an important role in strengthening the credibility of
the stabilization efforts (Rodrik, 1995: 2965). Moreover, the increased foreign
competition provided by import liberalization has been considered a potentially useful
tool in the battle against inﬂation. Finally, the gains in technical efficiency that are
potentially associated with increased openness can, at some point, improve the
competitiveness of the export sector. However, it is clear that in the short run trade
liberalization can also complicate the picture of macroeconomic adjustment through its
effects on the external balance. As stated by Dombusch, “one problem for trade reform is
political...the other comes from the exchange rate” (1992: 81).

Indeed, the anti-inﬂationary policies that have been applied in many Latin
American countries have rested on the use of the exchange rate as an "anchor" of the
domestic price level. This implies the nominal stability of the exchange rate and even its
real appreciation. Trade liberalization, on the other hand, invariably has a faster impact on
imports than on exports, usually leading, in the absence of a compensating exchange rate
depreciation, to the occurrence of large trade deﬁcits. Quoting Dombusch one more time,
"if reserves are not available and depreciation is impractical, the only realistic option for
trade policy is to approach liberalization more gradually" (1992: 82).

In practice, Latin American countries have made very rapid advances in the

liberalization of their trade regimes‘. Only a decade ago, Latin America was considered to

 

1 Ironically, a much more gradual approach was taken by the Asian countries whose
success stories provided much of the motivation for the adoption, elsewhere, of outward-

o1

have the most disto
however. dramatic
recently been descr
liberalization foun '
95).
Table 1.1 ill

That Stands out is t1
indicated by the re
31mm all countri.
the Years am pm
Fist-reform leVe'
lChile), the degr
Elle Educed tarit
rectum“ of the

completely e] in

have the most distorted external sector of the world (Edwards, 1995: 115). Since 1985,
however, dramatic changes in trade regimes have occurred in the region, which has
recently been described by a World Bank study as "rapidly moving toward the level of
liberalization found in the East Asian newly industrializing countries” (Dean et al., 1994:
95).

Table 1.1 illustrates the extent of the trade reforms undertaken’. The ﬁrst feature
that stands out is the drastic reduction in the average level of nominal protection. This is
indicated by the reduction in tariff rates, which now average less than 20 percent in
almost all countries. This represents a sharp decrease ﬁom the corresponding ﬁgures for
the years that preceded the reforms, which were usually two or three times higher than the
post-reform level. Secondly, even though only one country displays a uniform tariff rate
(Chile), the degree of dispersion of the import tariffs has been reduced dramatically, as
the reduced tariff ranges illustrate. A third characteristic of the reforms is the abrupt
reduction of the coverage of non-tariff barriers, which in some cases have been

completely eliminated. Finally, there is evidence that export taxes and restrictions have

 

oriented development strategies. Quoting Rodrik, "with regard to liberalizing trade
restrictions, for example, it is clear that East Asian countries did not go nearly as far as
some Latin American countries have done recently, and that whatever was accomplished
took place a lot more gradually" (1995: 2944).

2 It is worth noting that most of the data in this Table comes from a study by Alam and
Rajapatirana (1993), who focus on the trade reforms in Latin America during the 19805 —
the exception is the data for the coverage of non-tariff barriers, which was taken mostly
from Edwards (1995). The years that Alam and Rajapatirana (1993) assume to have been
the ﬁrst years of the reforms are not always the same as those that we consider below in
our estimation exercise. However, they do provide a good indication of the policy
changes that occurred in Latin America in the last decade or so. The reform years
considered by Alarn and Rajapatirana are as follows: 1988 for Argentina, 1985 for
Bolivia, 1987 for Brazil, 1985 for Chile, 1985 for Colombia, 1986 for Costa Rica, 1989

been reduced or e11

135‘). Consistent vi :

 

 

American countrie~
has increased in all
Regarding 1
countries exchang .I
cases. this was the
91' Alarm and Rajap
aSSOCiated with. sig
depreciations. hoot
etpenencing a sig
region has expen'e
As explair
two £33013“. "Firs
\
5)::9U3dor, 198
i lco’ 1989 To

3111113165 Of 1}]
blind in Bum} E

been reduced or eliminated in several countries (Dean et al. 1994: 77, and Edwards, 1995:
125). Consistent with the changes in trade policy, the trade intensity of the Latin
American countries — deﬁned as the ratio of real imports plus real exports to real GDP —
has increased in all but one country (Honduras)3.

Regarding the evolution of the exchange rate, Table 1.2 shows that most
countries’ exchange rates considerably depreciated between 1980 and 1987. In many
cases, this was the result of policies aimed at increasing the incentive to export. As shown
by Alam and Raj apatirana, in the 1980s “the trade reforms were always preceded by, or
associated with, signiﬁcant depreciation of the real exchange rate (1993: 11).” These
depreciations, however, were not always sustained after 1990, as several countries began
experiencing a signiﬁcant real appreciation of their currencies. Not surprisingly, the
region has experienced growing trade and current-account deﬁcits‘.

As explained by Edwards, the appreciation of the exchange rates was the result of

two factors: “ﬁrst, many countries used exchange rate policy as an anti-inﬂationary tool,

 

for Ecuador, 1986 for Guatemala, 1986 for Honduras, 1982 for Jamaica, 1985 for
Mexico, 1989 for Paraguay, 1989 for Peru, 1987 for Uruguay, and 1989 for Venezuela.

3 Estimates of the structure-adjusted trade intensity of Latin American countries can be
found in Burki and Perry (1997: 30-33). This indicator is obtained by correcting the ratio
of trade to GDP for certain structural characteristics that determine a country's volume of
trade, such as size and transport costs. As such, it is expected to reﬂect the level of trade
explained by trade policy. The estimates show that "the average (structure-adjusted) trade
intensity for the region has risen signiﬁcantly in the 19903", allowing Latin America to
approach the corresponding average for the OECD, but still lagging far behind the
average of the Asian newly industrializing countries (Burki and Perry, 1997: 33).

4 Figures from ECLAC (1996) show that between 1990-91 and 1992-94 Latin America
and the Caribbean’s current account deﬁcit increased from 1.1% to 3.2% of GDP. In the
latter period, the ﬁgures were above 5% in nine countries (Bolivia, Costa Rica,
Guatemala, Haiti, Honduras, Mexico, Nicaragua, Paraguay and Peru). In the trade
account, the region evolved ﬁ'om a surplus of 1.4% of GDP in 1990-91, to a deﬁcit of
1.4% of GDP in 1992-94 (ECLAC, 1996: 26).

and second. mliSS-i
abundant (1995: l.
conflict between 5‘
associated vsith ex
the L'.S. economy
As the Me
real exchange rate
foreign portfolio i
Wainability of r]
Circ“Instances ca]
5 Well as for the
mtlatiom- purpc
accelerating the g
Edwards. While c
that the disappoir
Tears of the TETOr
dithcult. As Eda
in the Way in Whi

heart of expm 6x

and, second, massive capital inﬂows into Latin America made foreign exchange too
abundant (1995: 137).” We have already referred to the ﬁrst factor as a potential source of
conﬂict between stabilization and trade policies. The second factor is to a great extent
associated with external conditions, among which are the relatively low interest rates in
the US. economy.

As the Mexican currency crisis of December 1994 has shown, the combination of
real exchange rate appreciation, large current account deﬁcits and strong dependency on
foreign portfolio investments can have very explosive consequences, and put in risk the
sustainability of the whole process of economic reform’. The avoidance of these critical
circumstances calls for the very prudent management of the current and capital accounts
as well as for the use of some restraint in the utilization of the exchange rate for anti-
inﬂationary purposes. But the Mexican experience also highlights the importance of
accelerating the gains in productivity that the reforms can potentially bring about.
Edwards, while commenting on the lessons to be drawn from the Mexican crisis, shows
that. the disappointing performance of aggregate productivity growth during the early
years of the reforms in this country made the handling of its external problems more
difﬁcult. As Edwards stated, “productivity gains are a fundamentally important element
in the way in which the overall external sector develops. Productivity growth is at the

heart of export expansion and thus contributes to keeping the current account in balance

 

’ As stated by Calvo (1996), “the December 20, 1994, devaluation brought the economy
down like a house of cards. Output fell by more than 7 percent in 1995, the current
account deﬁcit sharply swung from about 8 percent of GDP in 1994 to zero, and investors
turned their noses away from high-yield Mexican public debt even though the
international community had plunked about $50 billion in a rescue package (1996: 1)”.

(1995: 302')".6 “he
improvements in p

1994 — is the main

3- Trade Liberalizz

 

The exister
the potential of the
inthe economics p
comparative adVar
311mm? emCl€Tlt
based 0“ the assur
magnitude of the .
QUOIing ROdrik‘ ..
under lLS9111 neoc]
percentage mints

The lheOr
Other hand‘ ha\e

(1995: 302)”.6 Whether trade liberalization is capable of bringing about rapid
improvements in productivity — even though it might have failed to so in Mexico before

1994 — is the main question that the following sections attempt to address.

3- Trade Liberalization and Growth: Old and New Theory

The existence of net beneﬁts arising from trade liberalization and, in particular,
the potential of the latter to generate growth effects, have long been controversial issues
in the economics profession. In a tradition that comes from Ricardo’s theory of
comparative advantage, economic theorists have usually emphasized the static gains in
allocative efﬁciency arising from freer trade. The theory for these once-and-for-all gains,
based on the assumption of perfect competition, has long been understood and tested. The
magnitude of the corresponding beneﬁts, however, appears to be relatively small.
Quoting Rodrik, “reasonable estimates of the welfare cost of relative-price distortions
under usual neoclassical assumptions rarely produce numbers in excess of a couple of
percentage points of GNP (1995: 2932).”7

The theoretical arguments for the dynamic gains from trade liberalization, on the
other hand, have, until recently, been stated in less formal terms. This explains, at least in

part, the fact that the issue has remained a controversial one. Probably one of the ﬁrst to

 

6 Edwards (1995: 298) also quotes a World Bank study —Trends in Developing Economies
1994—that in September 1994 had pointed out this problem: “productivity growth has so
far been insufficient to offset the loss of external competitiveness implied by the peso
appreciation. . .with current account deﬁcits of over $20 billion supported by even higher

levels of foreign capital inﬂows, Mexico is vulnerable to foreign capital volatility” (p.
331)

defend the existe
Smith. In the 11':
effect on the exp
increases the ext
In other words.
grooth in prodt
incentives for t]
Interest
defended “in“;

Substitution b3

defend the existence of a positive link between international trade and growth was Adam
Smith. In the Wealth of Nations, this author argues that international trade, through its
effect on the expansion of markets, opens new possibilities for the division of labor,
increases the extent of specialization and promotes improvements in technical efﬁciency".
In other words, in Smith’s optimistic view of development, international trade propagates
growth in productivity through the exploitation of economies of scale and the creation of
incentives for the development of new productive technologies.

Interestingly, dynamic gains were also at the core of the arguments of those that
defended “inward-oriented strategies”. Quoting Pack (1988), “early proponents of import
substitution based their policies partially on infant industry arguments and the rapid
growth in productivity they expected during the stage when industrial skills were created
and modern technology mastered. Their main assumption was that the period of
protection would be utilized to increase technical efﬁciency and move towards
internationally competitive prices” (1988: 348). In fact, as stated by Krueger (1997), “in
the 1950’s and 1960’s, the neoclassical argument for an open trade regime was rejected
on the grounds that it was ‘static’ and ignored ‘dynamic considerations’”(1997: 10).

Furthermore, as accounted by this author, starting in the late 1960’s and 1970’s, critics of

 

7 An example is given by Haberger (1959), who estimated the welfare cost of protection
in Chile to be 2.5 percent of GNP, as opposed to 10 percent for domestic distortions — see
Dombusch (1992: 74).

' When referring to the impact of the discovery of America on the European economy, for
example, Adam Smith states: “By opening a new and inexhaustible market to all the
commodities of Europe, it gave occasion to new divisions of labour and improvement of
art, which, in the narrow circle of the ancient commerce, could never have taken place for
want of a market to take off the greater part of their produce. The productive powers of
labour were improved, and its produce increased in all different countries of Europe and
together with it the real revenue and wealth of its inhabitants (1976: 448).”

the import-subst
of the use of trac
generated as a b.

There m
trade regimes or
entrepreneurial I
technologies — l
D115 type of arg
oriented econon
new of academ
155V” In the 1;
application of n:
mOdehng 0f ime

Oftechmlogical

the import-substitution strategies concentrated on static issues, such as the sub.optimality
of the use of trade policy for development purposes and the rent-seeking activities
generated as a by-product of protection (Krueger, 1997: 5).

There were also writers that, in the tradition of Adam Smith, advocated liberal
trade regimes on the basis of dynamic considerations, such as their potential to spur
entrepreneurial effort, explore economies of scale and promote the adoption of modern
technologies — beside the gains from specialization according to comparative advantage’.
This type of argument, however, was more prevalent among policy- and empirically-
oriented economists. As shown by Rodrik, “this rationale for trade was hidden from the
view of academic economists by the intellectual appeal of the Ricardian outlook (1992b:
155).”'° In the last decade, nonetheless, this situation has changed thanks to the
application of new modeling tools to trade and growth issues. Indeed, the formal
modeling of international trade in imperfectly competitive markets, and the incorporation

of technological change as an endogenous process in models of equilibrium growth have

 

9 Bela Balassa is probably the best exponent of this literature. While commenting on the
TFP growth performance of countries with outward- and inward-oriented development
strategies, this author asserts that “outward orientation leads to the efﬁcient use not only
of existing resources, but also of increments in resources, permits the exploitation of
economies of scale, and provides the stick and carrot of competition that gives
inducement for technological change” (1993: 47). This vision is already present in a 1970
paper in which the author argues against “the evidence that the static cost of protection
would be outweighed by the dynamic beneﬁts of the inward-looking strategy. Rather, the
continued sheltering of domestic industry from foreign competition and disincentives to
exporting involve a dynamic cost to the national economy in the form of opportunities
forgone for improvements in productivity” (1989: 243).

'° In the same spirit, Edwards asserts that “for a long time it was argued that the
theoretical underpinnings of the proposition that freer trade enhances growth were weak.
While the theory was clear regarding the static gains from free trade, the generalization of
these results to a dynamic equilibrium growth setting presented some problems. Only

10

Proxided m?W in:
however. the lite

In the “6'
is c(planned b." 1
shown by GrOSSI
at a rate that is if
relations" (199::
the outcome 0f e
research and deV
trade can affect g

Grossmar
grouth where tec
differentiated prc
context the auth<
relationship. Firs

knowledge. reduc

Countries. Seconc

provided new insights into the trade and growth relationship. Even with these new tools,
however, the literature has been unable to reach unambiguous and general conclusions.

In the neoclassical growth model proposed by Solow (1956), steady-state growth
is explained by technological change, which is treated as an exogenous process. As
shown by Grossman, in this framework, "long run growth in an open economy proceeds
at a rate that is independent of its trade policies or the nature of its international economic
relations" (1992: 10). The recent attempts to model growth and technological progress as
the outcome of economic forces -— either through learning by doing or by investments in
research and development (R&D) — have shed light on several channels through which
trade can affect growth.

Grossman and Helpman (1991), for example, consider models of R&D-driven
growth where technological progress occurs either through the introduction of new
differentiated products or through the quality upgrading of existing products. In this
context, the authors discuss four different mechanisms underlying the trade growth
relationship. Firstly, they assume that trade may facilitate the international diffusion of
knowledge, reducing the cost of product development and accelerating growth in all
countries. Secondly, trade may favor growth through the reduction in research
redundancy that is brought about by the integration of world commodity markets. A third
mechanism at work is the increase in the size of the market in which ﬁrms operate. This
has ambiguous effects on growth, as it causes an increase in sales and proﬁts for a given

market structure — and, thus, an incentive for new product development and growth — but

 

recently with the renewed interest on growth theory, and the resulting ‘endogenous’
growth models, new developments in this direction have been made (1992: 32).”

11

also an increase in
technology. The n
spillovers are natit
countries can be e
same negative eff.
disadvantage in R
Kmanan (1987 t.
and gTOMh. Final
10changes in thei
grown. Specitica
on gromh dEpend
Production of trad
economy. A Simil

Batiz and ROmEr .

also an increase in competition that may induce a reduction in the investments in
technology. The net effect from these two forces depends on the extent to which research
spillovers are national in scope: if international knowledge ﬂows are not perfect, smaller
countries can be expected to see their share of the world market decline over time. The
same negative effect of trade on growth can occur when a country begins with a
disadvantage in R&D and technology spillovers are national in reach. In this setting, as in
Krugman (1987), history matters in the determination of dynamic comparative advantage
and growth. Finally, when countries are dissimilar in their factor endowments, trade leads
to changes in their intersectoral specialization and consequently in their aggregate rates of
growth. Speciﬁcally, openness to international trade can have positive or negative effects
on growth depending on whether it causes a reallocation of resources towards the
production of traditional goods, high technology goods or the R&D sector of each
economy. A similar decomposition of the growth effects of trade is proposed by Rivera-
Batiz and Romer (1991), who also ﬁnd that "allocation effects can increase or decrease
the rate of growth" (p. 973). These effects are expected to be larger when the differences
in the trading partners' endowments are bigger — such as in the case of North-South trade.
Lucas (1988, section 5 and 1993) has also emphasized the sectoral composition of
output in his explanation of the trade and growth relationship. Lucas proposes a multi-
good model where learning-by-doing is the engine of growth. As in Krugman (1987), it is
assumed that different goods are associated with different "learning rates", so that the mix
of goods produced in a particular country determines its rate of growth. Trade plays the
role of determining, through comparative advantage, the sectoral mix of production and

hence the aggregate rate of growth. Lucas (1993) also assumes that the sectoral rates of

12

learning are decrez
etolution of the ct
"growth miracles"
of demand and sut
a large exporter. .1
between the mix (
could widen over
EIOMh episode" 1

As shovm
and 1993) belong
“1&1 "Stress the ur
the gmmh rates 1
gromh rates acrc
techn(logical ch;
Protided that it i:

no ' -
ldlﬁlme Intern

could have tesrat

learning are decreasing over time, so that growth can only be sustained by the permanent
evolution of the economy's production structure. In this context, the occurrence of

" growth miracles" — such as Korea's — requires the creation of a gap between the structure
of demand and supply in the economy, which can only be possible if the country becomes
a large exporter. As stated by the author: "Korea needed to open a large difference
between the mix of goods produced and the mix of goods consumed, a difference that
could widen over time. Thus, a large volume of trade is essential to a learning-based
growth episode" (Lucas, 1993: 269).

As shown by Feenstra (1996), the models proposed by Lucas (1988, section 5,
and 1993) belong to a class of learning-by-doing and human capital accumulation models
that "stress the unequal growth rates of economies, as motivated by the wide disparity in
the growth rates of actual countries" (1996: 229)”. However, Feenstra shows that uneven
growth rates across countries can also be obtained in models of "endogenous
technological change" such as those proposed by Grossman and Helpman (1991),
provided that it is assumed that R&D knowledge diffuses freely within borders but does
not diffuse internationally”. An important point stressed by F eenstra — and one that
could have testable implications -— is that without this hypothesis, the models of R&D-
driven growth predict that trade leads to convergence in growth rates, even when the

allocative effects discussed above are involved. In the latter case, however, convergence

 

” Examples of models of this class are Krugman (1987), Young (1991), Azariadis and
Drazen (1990), and Stokey (1991).

'2 Evidence against the hypothesis of international diffusion of knowledge can be found in
the recent papers by Bowen et a1. (1987) and Trefﬂer (1995), who show that the
Hecksher-Ohlin model of trade is not supported by empirical evidence due to uniform

13

may occur towal
correSponding c:
to which Feenstt
controversy ove
focused on the c
exploration of \a
1996: 252).

To sumr
310th relations
DaIlOn into a Veg
also unleashes f

Countries d0 nor

may occur towards a rate that does not necessarily exceed the autarky growth rates of the
corresponding countries. In any case, it is worth noting that the concept of convergence
to which Feenstra refers is qualitatively different from the one addressed in the recent
controversy over convergence”. Indeed, as shown by Feenstra, most of this literature has
focused on the convergence in the level of output, while "there has been much less
exploration of whether the growth rates of countries differ systematically" (Feenstra,
1996:252)

To summarize the contributions of the new growth literature on the trade and
growth relationship, it may be useful to quote Helpman (1992): "The integration of a
nation into a world trading system unleashes powerful forces that speed up growth. But it
also unleashes forces that are harmful to growth. The former dominate, however, when
countries do not differ too much in terms of resource composition, and knowledge ﬂows
freely across national borders... When knowledge accumulation is localized, however,
history can extract powerful effects on the evolution of trade patterns and growth rates.
Under these circumstances small initial differences in knowledge capital can translate
into large long-um differences in sectoral structures, trade patterns and growth rates"

(1992: 265).

 

technological differences across countries - a fact that had already been discussed by
Minhas (1962).

'3 For an account of this debate, see Durlauf (1996) and the papers included in the
corresponding issue of the Economic Journal.

14

4- Trade and Grt

The relat
number of empir
on the dynamic l
emphasis on the
recently. after th
new models to p
SlJb.lect has been
methodologies v
“ﬂotation of grc
Openness. ran gj n
some studies ha,
time dimension c
Sontces 0f Variatj
the bnis of “her
be said that eVEn
memOdOlOgies‘ I]
trade. Some i mm

The pTOblem Ofd

4- Trade and Growth: The Empirical Evidence

The relationship between growth and trade liberalization has been the subject of a
number of empirical studies. The motivation has been, in many cases, to provide evidence
on the dynamic beneﬁts or costs of different strategies of development, usually with an
emphasis on the debate over the inward- versus outward-oriented approaches. More
recently, after the resurgence in interest in growth theory, and due to the failure of the
new models to predict unambiguous effects of trade on growth, the empirical work on the
subject has been seen as a way of “to help resolve the debate” (Harrison, 1996: 420). The
methodologies vary from the use of growth accounting techniques to the econometric
estimation of grth equations. There is also considerable variety of measures of trade
openness, ranging from policy indicators to indicators of trade performance. Furthermore,
some studies have used cross-sectional analysis while others have concentrated on the
time dimension of the series involved. Only a few studies have taken advantage of both
sources of variation, using panel data techniques. Finally, a distinction can be made on
the basis of whether the units of analysis are ﬁrms, industries or countries. Overall, it can
be said that even though there is a great variety in conceptual approaches and empirical
methodologies, most studies ﬁnd a positive relationship between growth and openness to
trade. Some important methodological problems, however, plague most of the studies.
The problem of determining the direction of causality between trade and growth and,
more generally, the possible endogeneity of the measures of openness in most
econometric studies are probably the best examples.

Case studies carried out at the ﬁrm-level in less developed countries have

provided some evidence on the type of technical change underlying productivity

15

increases in count:
Especially in Latir
technological char
strategies. As dess
Cqument and us
undertaken. panic
It is not clear. hot
generalized to m,
hand a“310g0us :
“here a more Ou‘
b“ CORSidel-able l
“PIECluded the n
Purchasable 0n t
lieetuing “as m
Moi-{name See

draw this I3136 0

increases in countries with inward- and outward oriented development strategies.
Especially in Latin America and India, it has been shown that signiﬁcant indigenous
technological change has taken place even in the context of intensive import-substituting
strategies. As described by Pack (1992), “rather than simply purchasing foreign
equipment and using it according to prevailing norms, an indigenous effort was
undertaken, particularly in large ﬁrms, that changed the method of production” (p. 22)”.
It is not clear, however, to what extent to which the learning obtained in this process was
generalized to the majority of the ﬁrms in the corresponding industries. On the other
hand, analogous studies of ﬁrms in East Asian newly industrializing countries (N ICs)
where a more outward-oriented regime has prevailed — although sometimes accompanied
by considerable government intervention" -— show that their approach to industrialization
“precluded the need for unique, site- and material-speciﬁc innovations that were not
purchasable on the world market” (Pack, 1992: 24). In these countries, technology
licensing was much more common than indigenous research and their impressive growth
performance seems to suggest that this was a winning strategy. However, it is difficult to
draw this type of conclusion exclusively from ﬁrm case studies, which brings us to
review, at least selectively, the cross-industry and cross-country studies on the subject“.
In a study that covers 21 industries in 17 countries, Nishimizu and Page (1991)

regress the average growth of TFP on the growth of exports, imports and domestic

 

” As stressed by this author, “the documentation of this indigenous technical change is
intrinsically interesting and provides a good antidote to the view implicit in international
trade theory and rnicroeconomics, of a uniform international technology costlessly
available to everyone” (Pack, 1992: 22).

'5 See, on this matter, World Bank (1993) and Rodrik (1994).

16

demand. control
policy regimes"
TFP gromh in t
policies in gene:
particular" ( p. 2
import penetrati
1973 ). Neither r
causation: TFP 1
indUSU'lal compt-
indexes. Similar
midi of 4 devel
indusm'eg

A differe
SPeCiﬁc 30"8mn
L'sing data {mm

author Shows the

coma“ With t.

plOdUCti‘ityh (p.

Incentives, l1 aye

o

5113111115 0f CapiL

TVbOUI( exlen
"the 1992)“
SI

demand, controlling for the effect of restrictive trade policies and non-market oriented
policy regimes". Their main result is that “[E]xport growth is positively correlated with
TFP growth in the industrial sector, but only in economies that follow market-oriented
policies in general and that do no resort extensively to quantitative import restrictions in
particular” (p. 256). Nevertheless, the authors also ﬁnd a negative relationship between
import penetration and TFP growth in the period following the ﬁrst oil shock (after
1973). Neither result, it should be stressed, provides insight into the direction of
causation: TFP performance could well be the cause and not the effect of the levels of
industrial competitiveness, as reﬂected in the export performance and import penetration
indexes. Similar results, however, are obtained by Nishimizu and Robinson (1984) in a
study of 4 developing countries, and by Bonelli (1992) who analyzes data on Brazilian
industries.

A different approach is adopted by Lee (1996), who focuses on the effect of
speciﬁc government policies on the productivity performance of 38 Korean industries.
Using data from a four-period panel covering the period ﬁ'om 1963 through 1983, the
author shows that “trade protections, such as tariffs and import restrictions, are negatively
correlated with the growth rates of value added, capital stock, and total factor
productivity” (p. 402). Another ﬁnding is that industrial policies, as expressed in tax
incentives, have a positive effect on output growth but that this occurs through the

stimulus of capital accumulation and not TFP growth.

 

'6 For more extensive reviews of this literature, see Pack (1988), Havrylyshin (1990),
Tybout (1992) and Rodrik (1995).

'7 The study includes countries with different levels of development, over periods that
vary somewhat between the late 19503 and early 19805.

17

  
  
  

A few stu i
American countri
(1993). for examp
the period 1 76 ’l

and capital had no

most indusm'es del
Costs and previous

OkS (1994
Chilean lﬂdUStries
Striking ditferencC

slight ”Coven. of

A few studies have analyzed the industrial productivity performance of Latin
American countries after the implementation of market oriented reforms. Agacino et al.
(1993), for example, show that in the case of Chile the initial reaction to the reforms (in
the period 1976/ 1981) was a more intense use of the factors of production — both labor
and capital had negative growth rates in this period — which was reﬂected in a positive
rate of growth of TFP. This increase in productivity, however, is attributed by the
authors to an increase in productive efﬁciency and not to technological change. In the
19803, on the other hand, Chilean industry displayed a negative rate of TFP growth, as
most industries decreased their capital/labor ratios in a context of relatively low labor
costs and previous ﬁnancial stress.

Oks (1994) looks at the post-reforms productivity performance of Mexican and
Chilean industries. The author reviews several studies and points out that, even with
striking differences in the ﬁgures for productivity, there seems to be agreement on a
slight recovery of productivity growth in Mexico after 1987 after having experienced
negative growth in 1985/1988. Oks also ﬁnds very small rates of TFP growth in the case
of Chile. As an explanation for these results, the author suggests that the real depreciation
of the capital stock associated with the probable acceleration of the rate of obsolescence —
due to modernization — may be underestimated in the data: “productivity just doesn’t
show up because existing measurements of capital do not capture adequately the real
depreciation” (1994: p. 60).

In a recent research project led by James Tybout at the World Bank, the
relationships between trade liberalization, technical efﬁciency, price-cost markups and

industry rationalization have been studied for a sample of semi-industrialized countries,

18

mnmmmmt
shows a positiv
level of protect
for Turkey. C o:
the trade re forr
economies of s
encounter the .
rationalizatior
foreign Compt
Plant size, an.

Helle
indlsmaiizal
pr 0POSitiOn 1
m5061mm} v
oﬁentatiOn .
SIUdies fOUI

the gene Sp

using both plant- and industry-level data". In Chile, Mexico and Turkey, this research
shows a positive relationship between TF P growth and both the reductions in the sectoral
level of protection and the increases in import penetration. Similarly, studies carried-out
for Turkey, Cote d’Ivoire and Mexico conclude that price-cost margins were reduced by
the trade reforms of the 808. With regard to the effect of the latter on the exploitation of
economies of scale, however, several of the papers produced within the project fail to
encounter the expected positive relationship between liberalization and industry
rationalization. As summarized by Tybout (1992), “it appears that exposure to increased
foreign competition is not closely linked with entry patterns, tends to induce reductions in
plant size, and may cause some improvements in technical efﬁciency” (p. 207).

Helleiner (1994), summarizing 14 country studies on trade policy and
industrialization, asserts that “the case studies [. . .] offer very weak, if any, support for the
proposition that either import liberalization or export expansion are particularly
associated with overall productivity growth” (p. 30). Furthermore, “the role of trade
orientation of individual industries was mixed” (p. 31) as, depending on the country, the
studies found either positive or negative relationships between sectoral TFP grth and
the corresponding levels of protection against imports and the rates of export growth.

Edwards (1995), on the other hand, presents data on the change in aggregate
growth of TFP after the liberalization of the trade regimes in 6 Latin American countries.

With the exceptions of Mexico and, to a lesser extent, of Bolivia, considerable increases

 

'3 See Tybout (1991, 1992), Roberts and Tybout (1996: chapter 1) and Rodrik (1995:
2970-2971) for a summary of the project’s results.

19

are encountered '.
consistent with re
the rate of TF P gr
find signs of 3 rec
with the findings
experience after t’
the evidence clea~
lp. l8). Neverthe.
Bosworth (1994 t
than in 1950-73 l
these differences
the treatment of t
Another (3
cross‘mtlntry g“
PETfonnanCe‘ and
ROdrik (1995), “1
higher growl“, ll

Uncovered by Pri

 

are encountered". Mexico presents a slight decline in aggregate TFP, a result that is
consistent with results presented by Lefort and Solimano (1994). These authors ﬁnd that
the rate of TFP growth was negative in the period 1982-1991 (p. 29). However, they also
ﬁnd signs of a recovery of GDP and TFP growth since 1988 — a result that is consistent
with the ﬁndings of Oks (1994) at the industry level. In their analysis of the Chilean
experience after the reforms initiated in 1974, Lefort and Solimano (1994) frnd that “all
the evidence clearly shows an acceleration in the rate of growth of TF P after the reforms”
(p. 18). Nevertheless, this result is contradicted by the evidence presented by Marfén and
Bosworth (1994), who ﬁnd that TF P growth was on average lower in the period 1973-89
than in 1950-73 (respectively 0.21 percent and 1.05 percent). It is possible, however, that
these differences in results are due to the use of different methodologies with regard to
the treatment of the cyclical changes in growth and the deﬁnition of capital.

Another group of studies has been concerned with the econometric estimation of
cross-country growth equations in which some measure of trade policy, of trade
performance, and/or of price distortions, are used as explanatory variables. As shown by
Rodrik (1995), “these studies generally conclude that openness has been conducive to
higher growth” (p. 293 8). One important problem with this type of work has been
uncovered by Pritchett (1996), who analyzes the relationship between different empirical
proxies for trade policy stance and ﬁnds that “the alternative objective measures of trade

policy examined are completely uncorrelated across countries” (p. 308) and produce

 

'9 The other countries are Argentina, Chile, Costa Rica and Uruguay.

20

"entirely di if:
robustness of

The rr.
direct admini:
procedure im;
trivial probler.
not clear whet
of simple aver
Since in many

“entirely different country rankings” (p. 329)”. These results point to the lack of
robustness of the studies that use only some particular measure of trade policy stance.

The most natural way to go in measuring trade openness is probably the use of
direct administrative measures of trade policy. In aggregate studies, however, this
procedure implies the calculation of average indexes of trade policy, which is not a
trivial problem. In the case of tariff and non tariff barriers to imports, for example, it is
not clear whether the application of a weighting system is a better procedure than the use
of simple averages. The latter may bias the measure of the actual restrictions upwards,
since in many cases the highest barriers apply to products that are not traded at all. But
for the same reason, the use of weights based on trade ﬁgures may cause an under-
estimation of the barriers to trade, since the products with the highest restrictions are also
the least traded exactly because of the government policies. An alternative that has been
used, among others, by the World Bank in the 1987 World Development Report, is to
construct subjective indexes of trade orientation. These, however, have been criticized for
their lack of international comparability. For these reasons, many cross-country studies
have avoided the use of direct policy measures and have made use of indicators of trade
performance or price distortions.

One of the ﬁrst attempts at measuring the effect of outward orientation on growth

in a cross-country setting was Michaely (1977), who found a signiﬁcantly positive

 

2° Four types of empirical measures of policy orientation across countries are examined
by Pritchett (1996). These are: “(a) the share of trade (or imports) in GDP (adjusted for
country structural characteristics or factor endowments), (b) the average tariff and
coverage ratio of nontariff barriers (NTBs), (c) measures of the deviation of countries’
actual trade pattern from the pattern predicted from a model of resource-based
comparative advantage and (d) a measure of price distortions” (p. 308).

21

correlation betwer
(1983) proposed :
exports sectors. 5
this result could l
of exports multir
capital. L'sing a:
model. As show
Slimming varia
the existence of
the level of inc:
the conIfibutior

As She:
in eXpons Can
iItVolved. The

trade Shares‘ 2

correlation between the rate of growth of export shares of GDP, and output growth. Feder
(1983) proposed a model where the export sector generates positive externalities on non-
exports sectors, so that its expansion has a positive effect on growth. Feder showed that
this result could be empirically tested by regressing output growth on the rate of growth
of exports multiplied by the export share in output, and the growth rates of labor and
capital. Using a sample of 31 countries, Feder (1983) found evidence supporting his
model. As shown by Edwards (1993), many studies followed this line of research,
estimating variations of Feder’s regression. Among their ﬁndings, it is worth mentioning
the existence of different relationships between exports and GDP growth depending on
the level of income of the countries involved, and the existence of diminishing returns in
the contribution of exports to output growth.

As shown by Edwards (1992), the above studies implicitly assume that the growth
in exports can be used as an indicator of the type of trade regime in the countries
involved. The same assumption underlies the studies that use trade shares, or changes in
trade shares, as openness indicators — as do Helliwell and Chung (1991) and Helliwell
(1994), for example, who also ﬁnd a signiﬁcant positive impact of trade on growth“. As
stated by Harrison (1996), “one problem with this approach, however, is that trade flows

are at best an imperfect proxy for trade policy. Other factors, such as country size or

 

2' Helliwell (1994), in a study of 19 industrial countries during the period from 1963 to
1989, regresses TFP growth on both the level and the ﬁrst diference of the ratio of total
trade to GDP. He also uses, as explanatory variables, the log of GDP as a measure of
scale, and the ratio of the current level of efficiency in the United States to the preceding
year’s efﬁciency level in each country, as a way of testing for convergence in
productivity. The author ﬁnds evidence that both the level and the rate of change of the
trade-output ratio have a positive effect on productivity growth, which he interprets as

22

foreign capital in
deviation of actu
and transport cos
criticized for the
An alter“.
proposed by Le
an empirical He
ratios for 53 co
GDP per Capitz
IIttmures is th
Yanks certain (
tesult is Show
indicators of.
crossﬁmmr:
indexes 1de
Hart
the inCQntiV.
compaﬁson
neverthelCS
and Don;1r

Pariryasa

foreign capital inﬂows, also affect trade”. This has led to the use, by some authors, of the
deviation of actual from predicted trade ﬂows, based on variables such as country size
and transport costs (Syrquin and Chenery, 1989). These measures, however have been
criticized for the absence of an underlying theoretical model to predict trade ﬂows.

An alternative that certainly has stronger theoretical foundations has been
proposed by Leamer (1988), who constructs measures of openness from the residuals in
an empirical Hecksher-Ohlin model estimated to explain trade ﬂows and trade intensity
ratios for 53 countries. Edwards (1992), using these measures, ﬁnds that the growth in
GDP per capita is positively associated with trade openness. A problem with Learner’s
measures is that, as shown by Rodrik (1995), it has “serious shortcomings in the way it
ranks certain countries” (p. 2939). However, in Edwards’ study, the above mentioned
result is shown to be robust to the replacement of Leamer’s indexes by alternative
indicators of trade orientation. In fact, similar ﬁndings are reported by the author in a
cross-country study that focuses in the grth of TFP and uses 9 different openness
indexes (Edwards, 1997).

Harrison (1996) suggests that the ideal measure of the impact of trade policy on
the incentives for exporting and import-competing industries would be based on “price
comparisons between goods sold in domestic and international markets” (p. 421). These,
nevertheless, are not available most of the time. One possibility, pursued by Barro (1991)
and Dollar (1991), is to use the deviation of the local price level from purchasing power

parity as a measure of outward/inward orientation. Both authors ﬁnd that these measures

 

suggesting “that the level of openness may have effects on both the level and the rate of
growth of productivity” (p. 265).

23

   
  
  
   
  

of openness raise
relative domestic
the other hand. it:
by regressing the
index is “highly L
countries“ (p, 544
that in many case
of countries. and
Well” (Rodrik. 1.
price Compaﬁgor
a OllEOPOlistic
to. 425),

The 3P?
Comprehengiy e
OpenneSs as an
whether mese
defends the Us
UnObsen,ed CC
OVEr time fOr ‘

Indicators for

of openness raise GDP growth per capita. Barro (1991) actually concentrates on the
relative domestic prices of the investment goods to international prices. Dollar (1991), on
the other hand, uses lO-year averages and controls for the countries’ factor endowments
by regressing the deviation in price levels on national income. The author ﬁnds that his
index is “highly correlated with the per capita GDP growth in a large sample of 95
countries” (p. 540). Dollar’s methodology, however, has been criticized on the grounds
that in many cases it captures “the exchange rate (and therefore macroeconomic) stance
of countries, and miss out on micro price distortions when exchange rates are managed
well” (Rodrik, 1995: 2940)”. Furthermore, as stressed by Harrison (1996), “international
price comparisons cannot disentangle the impact of domestic market imperfections (such
as oligopolistic marketing channels for imported goods) ﬁ'om trade policy interventions”
(p. 425).

The approach adopted by Harisson (1996), in one of the most recent and
comprehensive studies on the subject, is to “gather as many different measures of
openness as are available for a cross-section of developing countries over time, and test
whether these measures generally yield the same results” (p. 425). Indeed, this author
defends the use of panel data techniques in order to control for the existence of
unobserved country-speciﬁc effects and to account for the changes that have occurred
over time for the same countries. The seven measures selected — which do not include

indicators for which data is not available over time, such as Leamer’s indexes and the

 

2’ Interestingly, Rodrik (1995) gives a very positive evaluation of the relatively similar
methodology used by Barro, which focuses in the deviation in the price level of
investment goods: “perhaps the most credible of the cross-country regression studies are

24

   

and ovenaluatior

Harrison
- including the 1;
different measur.
exDated to affer
regres5'10“). A r:
is the Only mean
“he“ Panel dat

three 0‘“ of th.

data on trade barriers collected by UNCTAD — include: (a) subjective indexes of trade
liberalization from Papageorgiou et a1. (1991) and Thomas et al. (1991), (b) the black
market premium, (0) trade shares in GDP, (d) measures of relative domestic and
international prices, including a modiﬁed version of the Dollar (1991) index, and (d), a
measure of the indirect bias against agriculture from protection of the industrial sector
and overvaluation of the exchange rate”.

Harrison (1996) regresses GDP growth on the growth of the factors of production
— including the labor force, physical and human capital, and arable land — and the
different measures of openness (in levels or rates of change, alternatively), which are
expected to affect the change in total factor productivity (estimated as the constant in the
regression). A ﬁrst result is that in cross sectional regressions the black market premium
is the only measure of openness that presents a signiﬁcant (and negative) coefﬁcient.
When panel data with annual observations and a ﬁxed effects technique is used, however,
three out of the seven measures of openness are signiﬁcant at the 5 percent level and
another one is signiﬁcant at the 10 percent level — all with the expected sign. To deal with
short-run cyclical ﬂuctuations, the author also uses a panel of ﬁve-year averages. With
this approach, also allowing for ﬁxed effects, only three measures of openness are found
to exert a positive and signiﬁcant effect on productivity growth - two at the 5 percent

level and one at the 10 percent level. Harrison (1996) also performs a robustness analysis

 

those [like Barro (1991) and Easterly (1993)] that ﬁnd a negative relationship between
distortions in capital goods prices and economic growth” (p. 2940).

23 It is worth noting that the author does not always ﬁnd signiﬁcant rank correlations
between these alternative measures of openness, a result that is consistent with Pritchett’s
(1996) ﬁndings, and that Harrison (1996) interprets as an indication that “[the openness
measures] are not capturing the same aspects of ‘openness’” (p. 431).

25

inthe spirit of Le

 

    
  

regressions. and l

autoregressions.
between opennes
H996) study giv
higher growth; \
Noneutetess. n.
for the endoggz
to CXplajn Oup
that the Chore
hYPOQ‘tesig ‘15
Dane] data “

h°“'e\'er, (}

ShQVcn in a

in the spirit of Levine and Renelt (1992), introducing additional macro variables in the
regressions, and using only the measures of openness that had appeared as signiﬁcant in
her previous exercises. The result is that the statistical signiﬁcance of the openness
measures disappears in half of the cases. Finally, to investigate the direction of causation
between openness and growth, the author applies Granger causality tests using vector
autoregressions. As stated by Harrison (1996), “[the results] suggest that causality
between openness and grth runs in both directions” (p. 443). Overall, Harrison’s
(1996) study gives support to the hypothesis that greater openness is associated with
higher growth: whenever the former is statistically signiﬁcant, it has the appropriate sign.
Nonetheless, this result must be interpreted with caution since the author does not control
for the endogeneity problem that, as she shows, affects the openness variables when used
to explain output growth. Moreover, as stressed by the author, the results also “suggest
that the choice of the time period is critical” (p. 443): the greater support for the above
hypothesis is provided by the regressions with annual data, followed by those based on
panel data with 5-year-averages, and ﬁnally by the cross-sectional regressions. It seems,
however, that the best approach is the one based on a panel of 5-year periods. Indeed, as
shown in a study by Quah and Ranch (1990) quoted by Harrison (1996: 434), the positive
association between openness and growth when using annual data could be mostly
explained by short run cyclical ﬂuctuations. Cross-sectional regressions, on the other
hand, eliminate the large variation that has occurred over time in the developing
countries’ trade policies.

To conclude this review of the empirical evidence on the relationship between

trade openness and growth with a word of caution, it may useful to quote Helleiner’s

26

 

relationship bet“
the trade regime

unpersuaSlW sin\‘
I
Nor are compari.‘

macroeconomic ‘

changing <81le

 

are rarely availai
towel might not l
conceptual issue
different indicat-

particularly on t

5 - Methodolog-
5.l - Growth A

AS exp'
ﬁlms 0f trade

accounting CXt

uglaS Speci

Y‘: A'Kla Lt’i

“here Y is 0E

Of, r .
' especn \‘e

(1994) somewhat pessimistic appraisal of this literature: “The empirical research on the
relationship between total factor productivity (TFP) growth and output mix, imports or
the trade regime has been inconclusive. Comparisons across countries are often
unpersuasive since there are so many other inﬂuences for which it is difficult to control...
Nor are comparisons within countries over time always easy to interpret, since
macroeconomic inﬂuences upon capacity utilization typically dominate the effects of
changing output mix or incentive structure over the short- and medium-run; long run data
are rarely available for developing countries” (p. 28). On the other hand, throwing the
towel might not be the right thing to do: as stated by Rodrik (1995), “measurement and
conceptual issues aside, it is perhaps reassuring that so many studies using so many
different indicators tend to conﬁrm that countries with fewer price distortions,

particularly on the trade side, tend to grow faster” (p. 2941).

5 - Methodology and Results
5.1 - Growth Accounting

As explained in the introduction, the ﬁrst approach that we adopt to measure the
effects of trade openness on productivity growth in Latin America, is to perform a growth
accounting exercise. We do it by assuming that the production function follows a Cobb-
Douglas speciﬁcation with constant returns to scale:
Y, = A, 1e,a L3”) (1)
where Y is output, A is an index of total factor productivity, and K and L are the stocks

of, respectively, physical capital and labor. Under the assumptions of perfect competition

27

and cost minimi/
diderencing yielr
output to the rate
. , , l
lnll, / 1H): lnr

Because of the la

 

share in output ol
stocks. the labor
Working-age pm
by Nehru and D
l1Ddate it until i
‘ the Speciﬁc p
The data for Ca
memory meti
Sample of 13 1

Together, the\

and cost minimization, or is the capital share in output (0<or<1). Taking logs and ﬁrst-
differencing yields the standard grth decomposition that relates the rate of change of
output to the rates of change of TFP, capital, and the labor force:
ln(Y, / Y,__,) = ln(A, / AM) + a ln(K, /K,_,) + (1 - a) ln(L, /L,_,) (2)
Because of the lack of reliable data on factor shares, we adopt a ﬁxed average capital
share in output of 0.4 for all countries“. We calculate the rates of growth of capital
stocks, the labor force, and output using the data base on physical capital stocks,
working-age population (aged 15 to 64) and gross domestic product (GDP) constructed
by Nehru and Dareshwar (1993). This data base covers the period 1950-1990 but we
update it until 1995 using the World Bank’s data base on World Development Indicators
- the speciﬁc procedures that were used in this updating are described in the appendix.
The data for capital stocks, it is worth noting, was constructed by applying the perpetual
inventory method on the basis of the series of gross domestic ﬁxed investment. We use a
sample of 18 Latin American countries, that we choose on the basis of data availability.
Together, they account for more than 95 percent of the region’s GDP.

Tables 1.3 and 1.4 report growth decompositions based on equation (2). On
average, during the period 1950—1995, the countries considered grew at a rate of 3.6
percent, of which 48 percent is explained by capital accumulation, 44 percent by

population growth, and only 8 percent by the contribution of productivity growth. Sub-

 

2‘ According to Collins and Bosworth (1996), it is generally thought “that a plausible
range for the capital share is 0.3 to 0.4; and there is also considerable evidence that the
capital elasticity is higher in developing countries than in industrial economies” (p. 155).
A capital share of 0.4 is used by Fisher (1993) in his calculation of “Solow residuals”, by
Marfan and Bosworth (1994: 169) in their growth accounting analysis of the developing

28

period averages
period from 195
the 19905. most
however. a cone
considerable inc
growth has aver
Worthy
atstage. negativtI
l
Sartrple. 15 had n
during the 19705
difﬁcult to inten
can only lmprov
occurrence of nc
the Sectoral COn
low levels or 10
has dEClined in
also Possible 1}

Stock Series. as

[Wines of

period averages show that the contribution of TFP grth was 21.6 percent during the
period from 1950 to 1970, but fell signiﬁcantly during the 1970s and the 1980s. During
the 19903, most countries returned to their historical rates of GDP growth without,
however, a corresponding increase in their investment rates. This has been reﬂected in a
considerable increase in the average rate of TF P growth, whose contribution to GDP
growth has averaged 35 percent during the ﬁrst half of the decade.

Worthy of comment is the fact that during the 1980s the region experienced, on
average, negative growth in both GDP per worker and TFP. Out of the 18 countries in our
sample, 15 had negative rates of TFP growth during this period, while only 7 did so
during the 19705, and 4 during the 19905 (Table 1.4). Negative changes in TFP are
difficult to interpret, as it is usually thought that, at the industry level, “true” productivity
can only improve (Griliches and Lichtenberg, 1984). One possible explanation for the
occurrence of negative rates of TFP growth in the aggregate is that they reﬂect changes in
the sectoral composition of output: if output shifts towards industries characterized by
low levels or low rates of growth of productivity, it is possible to ﬁnd that productivity
has declined in the aggregate even though it has not done so at the industry level. It is
also possible that low levels of capacity utilization that are not captured in the capital

stock series, as well as high unemployment rates not reﬂected in our measure of the labor

 

countries of Latin America and Asia, and by Nehru and Dareshwar (1995: 53) in their
estimates of TFP growth.

29

force. could acct

 

periods:’.

In Table

 

 

 

is provided. This
American econo
international tra
"09611". We folio
“995), These 21
conditions are 5
exceed 40 NYC
(Cl The black .
ltl) the econor
mhjor expon
‘0 ﬁsume 11‘.
Were “Open

the Sprerr.

force, could account for the negative rates of TFP growth observed during recessionary
periods”.

In Table 1.6, preliminary evidence on the impact of openness on economic growth
is provided. This table reports growth decompositions both for the periods when the Latin
American economies were “open”, and for those in which they were “closed” to
international trade. In order to determine the years in which these economies were
“open”, we follow the criteria and timing of reforms suggested by Sachs and Warner
(1995). These authors deﬁne an economy to be “open” in a given year if all the following
conditions are simultaneously satisﬁed: (a) the coverage of nontariff barriers does not
exceed 40 percent of foreign trade, (b) the average tariff rate does not exceed 40 percent,
(0) The black market premium over the ofﬁcial exchange rate does not exceed 20 percent,
(d) the economic system is not socialist, and (e) the state does not have the monopoly on
major exports (Sachs and Warner, 1995: 22). As shown in Table 1.5, these criteria lead us
to assume that, in the period 1950-1995, the 18 Latin American countries considered
were “open”, on average, 35 percent of the time. Moreover, as highlighted in Figure 1.1,
the subperiods in which a higher fraction of the sample was considered open are the
19505 and the 19905 - almost one hundred percent of the countries were open after 1990.

Our main ﬁnding is that 14 out of 18 countries experienced faster TFP growth
during the periods of openness. On average, TFP grew at an annual rate of 1.2 percent
during the periods of Openness, and at a negative rate of 0.2 percent during those of

closedness. Among the countries with the best reactions to the opening of their

 

2’ Similarly, losses of productive capacity not captured in the calculation of capital stocks
could arise from civil wars — such as in the case of Nicaragua, for example, during the

30

economies is Ar;
During the same
of 0.2 percent. w
in 1986. Brazil a
worst performer
of openness and
Whhthecounu
and Mexico - \v
ptttious develoi
General?
reglon‘seconon
With. Indeed.
oharness. comp
ellllellt‘lrced 510.
growth during tl
mullelt mir
Growth
hith GDP Emu-
hwndmmguk
in the couhtrieg
”Spitfire”. by
in Cottnrn'CS Wit}

u\

19705.

 

economies is Argentina, whose TFP index has grown at a rate of 4.2 percent since 1991.
During the same period, Brazil (which also “opened” in 1991) has had a TFP growth rate
of 0.2 percent, while Mexico’s TFP growth rate been minus 1.9 percent, since opening-up
in 1986. Brazil and Mexico, the two biggest countries of the region, have in fact been the
worst performers in terms of the comparison of their rates of TFP growth during periods
of openness and closedness. Interestingly, in the two cases TFP growth was relatively fast
while the countries in question were closed — respectively 1.1 and 0.9 percent in Brazil
and Mexico - which may be interpreted as reﬂecting some degree of success in their
previous development strategies.

Generally speaking, there has been a correlation between the TFP growth of the
region’s economies and their overall growth performance when this is measured by GDP
growth. Indeed, the average rate of GDP growth was 4 percent during the periods of
openness, compared to 2.8 percent during those of closedness. All the four countries that
experienced slower TFP growth while they were open, also had lower rates of GDP
growth during these periods, with Brazil and Mexico experiencing the largest drops —
respectively minus 3.0 and minus 4.7 percent.

Growth in capital stocks, however, has not always been correlated with TFP nor
with GDP growth. On average, the growth rate of capital stocks has been 0.8 percent
lower during the periods of openness. Capital accumulation has been particularly slower
in the countries with the worst relative performances during the periods of openness --
respectively by 4.3 and 4.5 percent in Brazil and Mexico — but it has also been very slow

in countries with particularly good performances in terms of their rates of GDP grth

 

late 19705.

31

   

growth in the 131
0.5 percent slow
FCSponsible for r

of GDP.

 

The red
9311 be interpret
sectoral pattern
This new patte‘
has a compara
that had previ.
delflopment.
fOrmeL 11 W0

aggregate CV.

during these periods — e.g. Argentina and Guyana, whose growth in capital stocks was,
respectively, 4.0 and 3.3 percent lower after they “opened-up”. Since in most cases
growth in the labor force also has been relatively slower during the periods of openness —
0.5 percent slower, on average — the increases in the rates of TF P growth have been
responsible for more than 100 percent of the changes that have been observed in the rates
of GDP.

The reduction in the rates of capital accumulation during the episodes of openness
can be interpreted in two different ways. On one hand, it could be argued that a new
sectoral pattern of growth should develop in a context of openness to international trade.
This new pattern should beneﬁt the growth of the sectors and industries where the region
has a comparative advantage, in detriment of the import-competing sectors and industries
that had previously been promoted in the context of inward-oriented strategies of
development. Since in many cases the latter industries are more capital intensive than the
former, it would be possible to observe lower rates of capital accumulation in the
aggregate even if this is not necessarily the case at the industry level.

An alternative interpretation is that capital investment has not responded to the
new government policies because of the lack of conﬁdence of private agents in the
sustainability of the reforms. Indeed, in a context of irreversibility of investment costs it
is natural to think that the business sector should initially adopt a “wait-and-see” attitude,
until there is enough evidence on the lasting power of the reforms. This idea has been
formalized by Rodrik (1991), who proposes a simple model that links policy uncertainty
to the private investment response, and reviews the empirical work on the subject. The

author shows that “even moderate amounts of policy uncertainty can act as a hefty tax on

32

investment and

doubts as to the:

 

I
As previr

failure of our do
of capacity utilir.
working-age po
during recessior
ll? growth rat
recoveries“.
ln ord

rates ‘ 3\‘ ail al
used it to adj
refers 0my tr
unemmgym
The data or
pamdpatir

in this Tat"

Q“Slant

investment, and that otherwise sensible reforms may prove damaging if they induce
doubts as to their permanence (p. 229).”

As previously mentioned, it could be argued that our results are driven by the
failure of our data on capital stocks and the labor force to capture the changes in the rates
of capacity utilization and employment. In particular, our proxy for the labor force -— the
working-age population — could be overestimating the actual number of people employed
during recessions and underestimating them during recoveries, thus leading to predict
TFP growth rates that are too low and too high, respectively, during recessions and
recoveries“.

In order to test the robustness of our results, we collected data on unemployment
rates — available for 15 of the countries of our sample over the period 1980-199527 — and
used it to adjust the rates of grth of the labor force. In most cases the available data
refers only to “open urban unemployment”, measured as the total number of urban
unemployed as a percentage of the corresponding economically active population (BAP).
The data on the ratio of the EAP to the working age p0pulation — the so-called “rates of
participation”, available only for a few countries and periods - indicated that the changes
in this ratio have been rather small in the sampled countries. In practice, we assumed

constant rates of participation and proxied the national unemployment rates by the rates

 

2" A similar reasoning would apply to the non-consideration of changes in the rates of
capacity utilization: it would lead to overestimate the use of capital services during
recessions and to underestimate it during recoveries, thus leading to, respectively,
underestimate and overestimate the growth of TFP during recessions and recoveries.

27 The data was extracted from several issues of ECLAC’s Economic Survey of Latin
America and Statistical Yearbook of Latin America. The series were constructed for each
country by combining these two sources in a manner that led to the use of consistent
criteria for the measurement of unemployment rates over time.

33

of urban unempl

 

employment ant
obtain a proxy 0

resulting from ti-

   
  

Very small wher‘
adjusting for un

during the 1980

|
CO“11711310 eXpe‘

recovery 100k p
unemplOb'ment
recovery in the
”aims YCSul
reco‘fri' with
Period 19814

A sec
Sensitivity 1C

“mining“. f

of urban unemployment. We then calculated the annual percentage change in the rate of
employment and added the result to the rate of growth of the working age population, to
obtain a proxy of the rate of change in the employed work force. The rates of TFP growth
resulting from this adjustment are presented in Table 1.7. On average, the changes are
very small when one compares these results with those previously obtained without
adjusting for unemployment: TFP growth was 0.01 percent faster with the adjustment
during the 19803, and 0.1 percent faster during the 19905. The latter result is in fact
contrary to expectations, and shows that during the ﬁrst half of the present decade a
recovery took place in the rates of GDP growth that was not reﬂected in the rates of
unemployment - which increased in most countries. Taking this into account, the
recovery in the rates of TFP growth was even bigger than what had been suggested by our
previous results. Argentina, in particular, stands out as the most impressive example of a
recovery with increasing unemployment — the latter averaged 5.8 percent during the
period 1981-1990, and 10.4 percent during the period 1991-95.

A second test of the robustness of our results is performed by examining their
sensitivity to an adjustment made to account for changes in the rates of capacity
utilization. Since we do not have access to any direct measure of the fraction of the
capital stock that has been actually used in each stage of the business cycle, we followed
Harrigan (1996) in the calculation of a proxy for capacity output, that we then used as a
substitute for actual output in the calculation of the rates of TFP growth. Harrigan (1996)
estimates potential GDP “as the log-linear 20 year trend of actual GDP” (p.9); instead, we
estimate capacity output as the ﬁtted values of a regression of GDP on both a linear and a

quadratic trend, as well as on lagged values of GDP (included to account for serial

34

correlation). \\
application of '
the 18 countric
found to have .

The res
actual output a
of TFP grout}
rates during 1k
between the T
cOuntries cons
adjusting for (
Brazil and M.-
Iﬁrms of its T

We al
Stocks by the
showed no di
the Periods 0.
Percent, “ith
Countries 0ft

IIIVCStmem TE

.23
Seetion’ CtI

W Er

correlation). We also allowed for trend-breaks, when their presence was suggested by the
application of Vogelsang (1994) “SupF,” tests”. In practice, breaks were detected in 16 of
the 18 countries considered. In more than 80 percent of the cases, these breaks were

found to have occurred between 1978 and 1982 (Table 1.8).

The results of the calculation of TFP growth rates using capacity output instead of
actual output are presented in Table 1.9. As expected, the new results showed larger rates
of TFP growth during the 1980s — minus 1.3 instead of minus 1.5 percent -— and lower
rates during the 19908 — 1.] instead of 1.4 percent. However, on average the difference
between the TFP growth rates observed during periods of openness and closedness of the
countries considered did not change considerably: it fell from 1.57 percent without
adjusting for capacity utilization, to 1.44 percent after the adjustment. The results for
Brazil and Mexico were still the most disappointing, while Argentina still ranked ﬁrst in
terms of its TFP performance after opening-up.

We also adjusted the rates of capital accumulation by multiplying the capital
stocks by the ratio of actual to capacity output. The results, reported in Table 1.10,
showed no difference in the average change of the rates of growth of capital stocks from
the periods of closedness to those of openness: this difference was equal to minus 0.73
percent, with and without adjustments for capacity utilization. As before, the largest
countries of the region were also those with the worst performance in terms of their

investment response to more liberal trade policies

 

2' The exact procedure that was used to determine the break years is detailed in the next
section, where we apply it to the series of import- and export-output ratios, TFP indexes,

35

5.2 - Structural l

As ment

 

 

 

two types of me

outcome-based.
economy as 0pc

countries with
apply. In the pn
Openness. \n p;
order to deterr
place. Thus‘ V
of our 33mm:
peTfOrmanc e
the ”Verse ‘
the Winnie
"Du

Same 1hat
of the V0

Qfﬁte 38

5.2 - Structural Breaks

As mentioned in the review of the empirical literature on openness and growth,
two types of measures of openness have been most commonly used: policy-based and
outcome-based. In the previous section we used the ﬁrst type of criteria for deﬁning an
economy as open, and compared the corresponding performance of the Latin American
countries with the one observed during the periods in which the openness criteria did not
apply. In the present section we focus on the second type of criteria for deﬁning
openness. In particular, we examine the series of import- and export-output shares in
order to determine whether and when statistically signiﬁcant structural breaks have taken
place. Thus, we search for structural changes in the degrees of openness of the economies
of our sample that are revealed endogenously by the data. We then compare the TFP
performance of the countries in question before and aﬁer the breaks. Similarly, we follow
the reverse procedure and test for the existence of breaks in the TFP series, and compare
the countries’ degrees of openness before and alter the breaks.

The methodology that we use to determine the existence of trend breaks is the
same that Ben-David and Papell (1997) applied to a sample of 48 countries, on the basis
of the Vogelsang (1994) Sup Wald (or SupF,) tests. Their study, however, covered only 5
of the 18 countries of our sample. The data that we use for calculating trade shares was
extracted from the World Bank data bases”. The time spans for the data are as long as
1951 through 1995, and no smaller than 1965 through 1995. As before, we rely on an

updated version of the Nehru and Dareshwar (1993) data base for the TFP series.

 

as well as GDP and capital stock per worker.

36

  
  
  

'WESu
WmMRume
mmmaasofu
the tests are val
the notation of
takes place as '1‘
T3. 0 otherwise

othemise. The
K=P+Bn+

and a quadra

31:?! t B;

Re
(O'SST, ‘

The “SupFt ” tests proposed by Vogelsang (1994) are based on the estimation of a
univariate time series model under the alternative hypothesis of a one-time break in the
parameters of the trend function. The model allows for serial correlation in the errors and
the tests are valid regardless of whether the errors are stationary or have a unit root. Using
the notation of Ben-David and Papell (1997), we refer to the year in which the trend break
takes place as TB and deﬁne the following trend-break dummy variables: DUt = 1 if t >
T3, 0 otherwise; DTl = t - TB if t > T3, 0 otherwise; and DT2t = (t - TB) 2 if t > T3, 0

otherwise. The estimating equation can then be written:

I:
Rt = l~l 'l' Blt + thz + GDUt + YIDTt + 721)th + z cht-j + 5t (3)

j=1
where R, represents the variable whose series are being analyzed — e.g. the import-output
or the export-output ratios. The above equation assumes that the data contains a linear
and a quadratic trend, a speciﬁcation that we call model 1. Two other speciﬁcations are
considered: only a linear trend (model 11), and no trend at all (model 111), which
corresponds, respectively, to imposing the restrictions [32 = 72 = 0, and the restrictions
13: =71= 52:72:0-

Regression (3) is estimated for all possible breaks years TB such that: 0.15T < TB
< 0.85T, where T is the number of observations”. The number of lags included in the

regressions (“k”) is determined in the following manner. Equation (3) is estimated with

 

2" Data were available for 17 of the 18 countries of our sample: only Nicaragua had to be
excluded because of the lack of consistent data.

3° This corresponds to 15 percent trimming, which we use because of the relatively short
time spans of our data. We also performed the tests using 1 percent trimming — for which
Vogelsang (1994) also provides critical values — and found almost no changes in the
results.

37

an a priori maxi

tested using the

 

 

value. If not sig

   
  
 

lag becomes Sig
For mo-

maximum. ove

I€51ng 9 = 0. V
assumption of:
larger than the

no trend-break
selection of tht.
by Ben‘DaVid
“End‘break m
the rESults. If

only if the no.

an a priori maximum number of lags (initially 9), and the signiﬁcance of the last lag is
tested using the 10 percent value of the asymptotic normal distribution (1.6) as the critical
value. If not signiﬁcant, this lag is dropped and the model is estimated again until the last
lag becomes signiﬁcant and the ﬁnal k is determined.

For model I, the SupF, statistic proposed by Vogelsang (1994) is given by the
maximum, over all possible trend breaks, of three times the standard F -statistic for testing
0 = 'y1 = 72 = 0. Similarly, for model II, SupF, is the maximum of two times the standard
F -statistic for testing 0 = y, = 0 and, for model III, SupF, is the standard F-statistic for
testing 0 = 0. Vogelsang (1994) provides critical values for the SupF, statistic in both the
assumption of stationarity and of unit root series - the latter critical values being always
larger than the former. We adopt a conservative approach and reject the null hypothesis of
no trend-break only if the statistic exceeds the unit root critical value“. As for the
selection of the relevant model, we use the following model selection algorithm proposed
by Ben-David and Papell (1997). We ﬁrst estimate the least restrictive model I. If the no-
trend-break null hypothesis can be rejected at a level of 10 percent or higher, we report
the results. If this is not the case, we estimate model 11 and, again, we report the results
only if the no-trend-break is rejected. If model II leads to the acceptance of the null

hypothesis, we estimate model III and report the results if they indicate a trend-break.

 

3‘ This is the “conservative” approach proposed by Vogelsang (1994), who also suggests
“to not reject the null hypothesis when the statistic is smaller than the stationary critical
value” (p.11): the test would be inconclusive for values in between the stationary and the
non-stationary critical values. In our case, we only use the unit root critical values
because the application of Augmented Dickey Fuller tests to our series of trade shares
resulted in the acceptance, in the majority of the countries, of the null hypothesis of a unit
root. The same was true for the series of GDP and capital stocks per worker but not for
the TFP index. We opted, however, for the use of a uniform conservative criteria.

38

  
  
   

‘llien no model
model I.

Tables 1
import-output a
of l7 countries

export shares. C

 

We found that tl
of \4 countries
“‘35 an increas
Percent. Most
“end-break 3
The I
Shares reveg
Observed it
(list the ch;
in trade sf.
capital De
direggmn

me CaSe:

We QB“
§0\kﬂ

When no model leads to the rejection of the no-trend-break null, we report the results of
model I .

Tables 1.11 and 1.12 report the results of the SupF, tests applied to the series of
import-output and export-output ratios. Signiﬁcant trend-breaks were detected in 13 out
of 17 countries in the case of import shares, and in 14 out of 17 countries in the case of
export shares. Comparing the average trade shares in output before and after the breaks,
we found that they increased in 11 out of 13 countries for the case of imports, and in 9 out
of 14 countries for exports. The median change in import-output shares after the breaks
was an increase of 20 percent, while for the export-output shares it was an increase of 23
percent. Most trend-breaks took place during the late 19703 and early 1980s: the median
trend-break year was 1979 for imports and 1983 for exports.

The comparison of the rates of TF P growth before and after the breaks in trade
shares reveals no correlation between the changes in these growth rates and those
observed in the import- and export-output ratios. In less than 50 percent of the countries
did the changes in the rates of growth of TFP occur in the same direction of the changes
in trade shares. With regard to the changes in the rates of growth of GDP per worker and
capital per worker, we found that in 80 percent of the cases these rates changed in a
direction opposite to that of the changes in import ratios — this was true in 57 percent of
the cases for the breaks in export-output ratios.

We also estimated structural breaks in the series of TFP (Table 1.13). To this end
we constructed an index that takes the value 100 in 1950 and grows according to the TFP
growth rates calculated in our growth accounting exercise. Statistically signiﬁcant breaks

were found in 13 out of 17 countries. The median trend-break year was 1979 and the

39

   

lower than bet}-
output ratios at
and l4 percent.
We als
for 16 out of \7
these Variables
Place in the 13
Worker OCCur
QVer
r'rltios do no
determined
Emporium.
”closed-- L
trade and
Early 19%
(Mme tl
lmpjem
negatie
Eve“ t‘
Dene}.

median change in the rate of TFP growth after the break was minus 1.8 percent. In all
countries where signiﬁcant breaks were detected, the after-break rate of TFP growth was
lower than before. As for the trade shares, we typically found larger import- and export-
output ratios after the breaks: the median changes in these ratios were, respectively, 22
and 14 percent.

We also found signiﬁcant trend-breaks in GDP per worker and capital per worker
for 16 out of 17 countries (Table 1.14). In 90 percent of the cases, the growth rates of
these variables were lower after the breaks. As in the case of TFP, most trend-breaks took
place in the late 19703 and early 19803. In 11 out of 13 countries, the breaks in GDP per
worker occurred within two years of the breaks in TFP.

Overall, it is important to highlight the fact that the trend breaks in trade-output
ratios do not coincide with the dates of opening of the Latin American economics, as
determined by policy-based criteria. Indeed, the latter suggest that, except for some
temporary episodes of openness during the 19503 and 19603, most countries remained
“closed” until the late 19803 and early 19903. The dates for most trend breaks (both for
trade and for output variables), on the other hand, are concentrated in the late 19703 and
early 19803, and seem to reﬂect the effects of the external shocks that hit the region
during these periods. These terms of trade and interest rate shocks led to the
implementation of contractionary policies that resulted in the very low (and even
negative) rates of growth of the 19803 — the so-called “lost decade” for Latin America.
Even though the poor performance of this period contributed to the emergence of a new
policy stance that favored the re-orientation of the region’s trade policies, the latter only

occurred several years latter.

40

5.3 - Regressior
A prohle
liberal trade pol

trade ﬂows to (

 

 

relationship be:
variables could
it could be the t

and not the can

 

cOuntries. It is
prenous sectio
ln the 1:
eStablishjng a:
eSlimating par
represen‘umVe
considered, In
of gremh O f (
ItigTessions W
“Se Sacral C(

enema Shoe

5.3 - Regression Analysis

A problem with the comparison of growth rates before and aﬁer the adoption of
liberal trade policies, or before and after the occurrence of trend-breaks in the ratios of
trade ﬂows to GDP, is that they do not provide a basis for establishing a causality
relationship between openness and productivity growth. Changes in both types of
variables could in fact be the result of a third factor — an external shock, for example — or
it could be the case that observed increases (or declines) in openness are the consequence,
and not the cause, of improvements (or declines) in the productivity performance of
countries. It is important to keep this in mind when interpreting the results of the two
previous sections.

In the present section, we use regression analysis to address the problem of
establishing a relationship of causality between openness and growth. We do this by
estimating panel regressions of the rate of growth of GDP per worker on several variables
representative of levels and rates of changes of the degree of openness of the economies
considered. In order to isolate the effect of openness on TFP growth, we include the rate
of growth of capital stocks per worker in the above regressions. We also run separate
regressions with the growth of per worker capital stocks as the dependent variable. We
use several control variables, intended to capture scale effects, convergence processes,
external shocks, and other government policies. The basic estimating equation can thus be
written:

GRit=a+Bxir+ni+8it (4)

41

their GIL, repr
vvorlter capital :
representative r
variables; I]! is -
variables; and r

The spe
dummy Variah'.
(DREF). the sh
and imports ((1.
“Change (BLi
the degree of o
imPortance of
to have positiv I
levels and rate I
Supposed to cal
el‘llt‘tted to hal

openness,

AS prev

glow] in GDl

 

 

 

where GK, represents either the growth in per worker GDP (GYiQ or the growth in per
worker capital stock (GK-t) of country i at time t; Xit represents the set of variables
representative of the openness of country i at time t, as well as the corresponding control
variables; r]i is a country-speciﬁc effect potentially correlated with the explanatory
variables; and ait is a serially uncorrelated error.

The speciﬁc variables that we use as measures of the countries’ openness are: a
dummy variable activated when the Sachs and Warner (1995) criteria for openness apply
(DREF), the share of real total trade in real GDP (T1), the growth rates of exports (GX)
and imports (GM), and the log of one plus the premium in the black market for foreign
exchange (BLACK). We use variables representative of both levels and rates of change of
the degree of openness of the economies considered in order to determine the relative
importance of static and dynamic gains from trade”. Most of these variables are expected
to have positive coefﬁcients in the hypothesis that openness has a beneﬁcial effect on the
levels and rates of growth of productivity. The only exception is BLACK, which is
supposed to capture government restrictions on the access to foreign exchange, and is
expected to have a negative coefficient — a higher BLACK being associated with lower
openness.

As previously mentioned, in the regressions where the dependent variable is the

growth in GDP per worker, the growth in per worker capital stock (GK) is included as an

 

’2 If, for example, productivity growth is affected only by the changes and not by the
levels of openness of an economy, it can be argued that static gains from trade are more
prevalent than dynamic gains: openness affects the level of efﬁciency but not (directly)
its rates of change. This argument is made by Helliwell (1994: 265).

42

   

explanatory vat
of the countries

the second vari:

determinants 0
size of a count
other hand, ha
“tempts to ca
relativ-ety po<
e‘ 31- (1997 ).
“controlling
lTTlplerneme

CQUld OthEr

explanatory variable”. Two other control variables are included in all regressions: the log
of the countries’ total population (POP), and the log of the initial GDP per worker (1N1) —
the second variable being substituted by the log of the initial capital per worker in the
regressions where GK is the dependent variable. The variable POP is intended to capture
the positive inﬂuence of scale on the rates of growth, predicted by several models of
endogenous growth. Its inclusion is also important, in order to isolate the policy-
determinants of the share of total trade in output (T1), from the (negative) effect that the
size of a country is usually believed to exert on this variable. The variable IN], on the
other hand, has been a standard feature of the empirical estimation of growth models. It
attempts to capture the existence of a convergence or “mean reversion” process, by which
relatively poor countries would show faster rates of growth“. As emphasized by Easterly
et a1. (1997), who are also concerned with the growth eﬁ‘ects of economic reforms,
“controlling for ‘mean reversion’ is especially important... because reforms tend to be
implemented in periods of poor growth performance, and, thus, their effect on growth
could otherwise be confused with the simple dynamics of growth recovery” (p. 294).

Five other control variables are included in the regressions: the rate of change of

the terms of trade (T 07) as a measure of external shocks, the average years of secondary

 

’3 This is justiﬁed by the growth accounting equation (2). This equation implies that per
worker GDP growth can be expressed as the sum of TFP growth and the growth in per
capita capital stocks multiplied by the share of capital in output.

3‘ An extensive literature exists on the issue of convergence. Initially, this literature was
motivated by the fact that convergence is expected to occur as a result of transitional
dynamics in the context of neoclassical growth models. Some studies have also related
the existence of convergence to the international diffusion of technology, which would
lead to the faster growth of initially lagging countries. A review of the recent
controversies on the subject can be found in Durlauf (1996), as well as in the other papers
included in the corresponding issue of the Economic Journal.

43

   
  
  
   
 

schooling oi th
capital availahl
government co
variables are e.

ultile M2 is art

policies for ssh
Since in man)
POlIC)’ refom
could create
policies '6th
The
1960-1995
commute
”Emmi
basis of
Were ta]
ergem

\

35 ACC
TEgreE
“Pkg
Suggt
0L
hEYQl

hum

schooling of the population aged 15 and over (EDU) as a measure of the stock of human
capital available in each country”, the ratio of broad money to GDP (M2), the ratio of
government consumption to GDP (GOV), and the rate of inﬂation (INF). The last two
variables are expected to capture the effects of policies of macroeconomic stabilization,
while M2 is an indicator of ﬁnancial reform that measures the extent of ﬁnancial
deepening attained in each country. These three variables are included because the
policies for which they proxy have been shown to have signiﬁcant effects on growth.
Since in many cases these policies have been implemented simultaneously with trade
policy reforms, it is important to include variables that control for them: not doing it
could create doubts about whether our openness variables are capturing the effect of trade
policies alone, or that of the complete policy packages.

The data that we use is constituted by a panel of 16 countries over the period
1960-1995“. The source of the data is as follows: GY, GK, POP, and [N] were
constructed from the updated Nehru and Dareshwar (1993) database; DREF was
constructed on the basis of Sachs and Warner (1995); GX and GM were calculated on the
basis of data from the World Bank data bases; T I, TOT, BLACK, EDU, GOV, INF and M2
were taken from Easterly et a1. (1997). All variables are (mostly) ﬁve-year averages,

except for IN] and EDU which are referred to the ﬁrst year of each period, and DREF,

 

3’ According to Benhabib and Spiegel (1994), the empirical evidence from cross-country
regressions favors the idea that the stocks and not the rates of growth of human capital
“play a role in determining the growth of per capita income” (p. 166). The authors
suggest that human capital inﬂuences growth, not as another factor of production, but
through its effect on the rates of innovation and technology adoption. This is the approach
hereby adopted as a motivation for including the stock and not the rate of growth of
human capital in our growth regressions.

44

which takes tht

theyears in a g

 

 

motivated by tl
eliminate shor.

ln orde
correlation bet
of a GMM estiF
the stacking ot
ﬁlm-difference

Potentially end

 

mm? the use
E [Xltﬂ . (V8 -

11

EKX‘Hl - xvii.
E [Ohm . x“I

EquallOns (5)

exogenOus‘O .'

\
‘60me 18 C:

17 I0 be er

which takes the value 1 when the corresponding country was “open” in the majority of
the years in a given period”. The use of ﬁve-year averages instead of annual data is
motivated by the objective of concentrating on long-run effects, and is thus intended to
eliminate short-run ﬂuctuations associated with the business cycle.38

In order to deal with the problems of sirnultaneity, reverse causality, and possible
correlation between country-speciﬁc effects and the explanatory variables, we make use
of a GM estimator proposed by Blundell and Bond (1997). This estimator is based on
the stacking of equation (4) with its ﬁrst difference, and on the use of lagged levels and
ﬁrst-differences as instruments for, respectively, the ﬁrst-differences and the levels of the
potentially endogenous explanatory variables”. The speciﬁc moment conditions that

justify the use of these instruments are:

E [XKW 0 (8i, - 830.19] = 0 for s 2 2 and t 2 3 (5)
E [(xi(t-l) " Xi(t-2)) ' sit] = 0 fort ->- 3 (6)
E [(xi(t-I) " Xi(t-2)) ’ 'h] = O for t 2 3 (7)

Equations (5) and (6) are implied by the assumption that the X it variables are weakly

exogenous”, in the sense that they are potentially correlated with past and

 

3‘ Of the 18 countries considered in the previous sections, only Guyana and Nicaragua
had to be excluded because of the lack of consistent data.

37 The only additional exceptions are given by the variables taken from Easterly et a1.
(1997), for which the last period averages do not include the years 1994 and 1995.

38 This is also the procedure adopted in the cross-country studies of income convergence
that use panel data — see Loayza (1994), Islam (1995) and Caselli, et a1. (1996).

’9 A more detailed presentation of the Blundell and Bond (1997) estimator that we use is
presented in the next chapter of this dissertation (section 5)

‘° In the case of the variable INI, equation (5) can be assumed to be valid for leon the
basis of the assumption of no serial correlation in the residuals. This can be seen by
recognizing that the dependent variable can be rewritten as the lead of the ﬁrst difference
of IN], so that the model can be rearranged with [NI as the lagged dependent variable.

45

 

contemporaf
openness bei
En. - s. 1 =
Equation (7)
correlated ui
change. Thus
correlated wi

The e
ming the Gar
Arellano and
is a Sargan te
Which tests rh

InOment COnd

Serial aUtocon

 

 

contemporaneous values of the dependent variable but not with its future values - e. g.
openness being potentially “caused” by present and past growth but not by future growth:
E[Xi,oei,]=0 fort<s (8)
Equation (7) is implied by the assumption that the country-speciﬁc effects are potentially
correlated with the level of some of the explanatory variables, but not with their rate of
change. Thus, for example, we assume that country-speciﬁc characteristics are potentially
correlated with levels of openness but not with changes in those levels".

The estimation of equation (4) using the above described method is performed
using the Gauss-based program “DPD96”, whose original version is described in
Arellano and Bond (1988). This program reports two types of speciﬁcation tests. The ﬁrst
is a Sargan test for overidentifying restrictions proposed by Arellano and Bond (1991),
which tests the overall validity of the instruments by analyzing the sample analog of the
moment conditions used in the estimation process. The second type of test is a test of
serial autocorrelation (of ﬁrst and second-order) in the residuals. By construction, ﬁrst-
order serial correlation is likely to be found due to the use of ﬁrst differences, but this is
not the case for second-order serial correlation. Under both types of speciﬁcation tests,
failure to reject the null hypothesis gives support to the model.

Tables 1.15 reports estimation results of the regressions of growth in GDP per

worker on the various openness variables, including the control variables GK, IN], POP

 

" For this same reason, we assume that the variables GX and GM are not correlated with
country-speciﬁc effects. In the case of the variable INI (interpreted as a lagged dependent
variable), Blundell and Bond (1997: 12) show that the validity of an assumption such as
(7) depends on whether the difference between the initial level of [N] (at time t=0) and
its convergent level is uncorrelated with the country-speciﬁc effect itself. Since the ﬁrst

46

and TOT. The
- (5) is the on
growth have it
entirely capttn
significant ant
respectively ir
effects: faster
larger countri:-
With n

l

have the CXpet
DREF looses 1
CORSideration
{amt rates of
ﬁgm'ﬁcance u
Operiness a re:
inﬁrmity, as IT

thIOUgh Chan *

 

 

and TOT. The latter variable (T 07) does not appear to be signiﬁcant in these regressions
— (5) is the only exception — which suggests that the effects of external shocks on TFP
growth have not been signiﬁcant in the region, or, alternatively, that they have been
entirely captured by the openness variables”. The other control variables are always
signiﬁcant and have the expected signs. The coefficients on IN] and POP, in particular,
respectively indicate the existence of convergence in productivity growth and scale
effects: faster growth of countries with relatively low initial GDP per worker, and of
larger countries.

With regard to the openness variables, we ﬁnd that they are all signiﬁcant and
have the expected signs when separately included in the basic regression. However,
DREF looses its signiﬁcance when the rest of the openness variables are also taken into
consideration — regression (6) — which suggests that the latter variables account for the
faster rates of TFP growth observed during periods of openness. Similarly, T I loses its
signiﬁcance when included simultaneously with the other variables representative of
openness - regressions (4) and (6) - which could be interpreted as suggesting that trade
intensity, as measured by the ratio of trade-ﬂows to output, affects productivity mainly
through changes in the levels of efﬁciency rather than by affecting directly the rates of

change of productivity: that static gains from trade are more prevalent than dynamic ones.

 

year of our series can here be treated as random, we assume that the ﬁrst difference of IN]
is uncorrelated with 11,.

‘2 We also run a regression (not reported) excluding all openness variables and found that
TOT has a positive but not statistically signiﬁcant coefficient (with a p-value of 0.96).
The negative and signiﬁcant coefﬁcient of TOT in regression (5), on the other hand,
suggests that for a given black market premium on the exchange rate (which has a
bivariate correlation with TOT of minus 0.23) improvements in external conditions act as

47

Indeed both (

 

regardless of t

 
 
   
  

negative coet‘t
availability of
The re
eqnation ( 6) -
control \‘ariab
sample. we h:
reason We h;
Signiﬁcam 2
V'Eiriables . :
“Table 1_-
“‘31 DRE}?
Signiﬁcam
Stocks of ',
that 7‘} re,
COHStirm‘
Suggests
men ma
Control
ED U. V

a dis-1n
Open!“

Indeed, both GX and GM appear to have a positive and signiﬁcant effect on TFP growth,
regardless of the other openness variables that are included in the regressions. The
negative coefﬁcients on BLACK, on the other hand, confrrrn that lower restrictions on the
availability of foreign exchange promote faster TFP growth.

The regression results reported in Table 1.16 are based on the estimation of
equation (6) — which includes all the openness variables — but with the addition of other
control variables — namely EDU, GOV, INF, and M2. Due to the limited size of our
sample, we have been forced to include these control variables one by one. For the same
reason, we have excluded TOT, whose effect on TFP growth was shown to be not
signiﬁcant, and whose exclusion does not appear to affect the coefficients of the other
variables — as can be seen by comparing regression (6) of Table 1.15 with regression (1)
of Table 1.16. Two main ﬁndings are derived from the results in Table 1.16. The ﬁrst is
that DREF, GX, GM, and BLACK maintain their signs and signiﬁcance (or non-
signiﬁcance in the case of DREF) after the inclusion of variables that control for the
stocks of human capital as well as for other government policies. The second ﬁnding is
that T I regains its signiﬁcance when either EDU, GOV, or M2 are controlled for. This
constitutes evidence in favor of the existence of dynamics gains from trade, but it also
suggests that these are dependent on the countries’ degree of ﬁnancial development, on
their macroeconomic stability, and on their stock of human capital. None of the new
control variables, it is worth noting, has a signiﬁcant coefﬁcient, with the exception of

EDU, whose coefﬁcient is negative and has a p-value of 11 percent. This is a puzzling

 

a disincentive for TFP growth. This result is not robust, however, to the inclusion of other
openness variables, as is shown by the non-signiﬁcance of TOT in regression (6).

48

  
 

result since. at

larger human

utterpretation
same technolol
issue certainly
Tab\e
capital Stock
POSitiVe and

l'arlablEs a]

230mb of.

and Signif:

result since, after controlling for initial per worker GDP, we expected the countries with
larger human capital stocks to grow faster: as suggested by Benhabib and Spiegel (1994),
for a given technology gap, countries with more human capital should be more able “to
adapt and implement technologies developed elsewhere”(p. 1 5 6). One possible
interpretation for the negative coefﬁcient on EDU is that this variable is capturing the
same technology gap effect that causes IN] to have a negative coefﬁcient. However, the
issue certainly merits further investigation.

Tables 1.17 and 1.18 report the results of regressions with the rate of growth of
capital stocks per worker as dependent variable. They indicate that both T I and GX have a
positive and statistically signiﬁcant effect on capital accumulation, but that once these
variables are controlled for, neither GM nor BLACK have a signiﬁcant effect on the
growth of capital stocks“. The dummy variable DREF, on the other hand, has a negative
and signiﬁcant coefﬁcient when it is included together with the rest of the openness
variables. The latter result suggests that even though a larger trade intensity and faster
export growth have promoted investment in the region, the latter has responded
negatively to the implementation of more liberal trade policies. A possible interpretation
for this is that, as argued by Rodrik (1991), the private sector could be reluctant to
commit itself to the new policies until there is enough evidence that they will not be
reversed. This is an issue that should motivate great concern, as a negative investment

response to changes in economic policy can be, in itself, a factor that conspires against

 

‘3 It is important to note that regressions (1), (2) and (5) are not supported by the
speciﬁcation test of second-order serial correlation. Thus, their results should, at best, be
interpreted with great caution.

49

the sustainabi'
latter could b<

As for
investment is
improvement.
accumulation
ﬁndings on th

inclusion of a

6- sum“

l! is n(
ttade leads to
(endogenous;
the lﬂtemarjm
caPabilities 0.
“1th a dl’nanj

0n the
relationship 5

COunmeS and I

 

 

methOdoiogI-k

0f l-llege Prob
II
N

the sustainability of the new policies: a generalized belief in the eventual collapse of the
latter could become a self-fulﬁlling prophecy.

As for the control variables, we found that, as in the case of TFP growth, capital
investment is affected by both convergence and scale effects. We also found that
improvements in the terms of trade do not have a signiﬁcant effect on capital
accumulation once all openness variables are taken into consideration. Overall, the main
ﬁndings on the effects of openness on capital accumulation are not affected by the

inclusion of additional control variables“.

6- Summary and Conclusions

It is not clear, from a theoretical point of view, whether openness to international
trade leads to faster or slower productivity growth. In recent models of trade and
(endogenous) growth, this depends on the extent to which openness to trade facilitates
the international diffusion of knowledge, as well as on the initial stock of technological
capabilities of nations and the degree to which their comparative advantage lies in sectors
with a dynamic potential.

On the empirical side, a large number of studies have found that a positive
relationship seems to exist between the degree of trade openness of industries and
countries and their growth performance. Many of these studies, however, suffer ﬁom
methodological problems that lead one to question the validity of their results. Examples

of these problems are the sensitivity of the results to the speciﬁc measures of openness

 

“ As before, the variable representative of educational achievement has a negative and
signiﬁcant coefficient. So does the inﬂation rate.

50

used in eacl
econometric
causality be

Latii
of "opening
of the most 1
Latin Ameri.
found elsewl
productivity 3

In the
uDertplained r
labor and cap:
the PeriOds in
I0 intemation
Sachs and \t';
al'erage tari ff
black marker

growl of Ti

“Open”, Th] S

 

EXCEprjOnS be
I

SlnCe
be explained

capita] StOCks

used in each case, the possible endogeneity of some of the trade variables used in the
econometric studies and, in particular, the difﬁculties in establishing the direction of
causality between trade and growth.

Latin America constitutes an interesting case for the study of the dynamic effects
of “opening up”. Indeed, for decades this region has been characterized by having some
of the most restrictive trade policies in the world. However, in recent years most of the
Latin American countries have liberalized their trade regimes at a speed that is seldom
found elsewhere in the developing world. Whether this has had noticeable effects on their
productivity performance is the question that we have intended to address in this paper.

In the ﬁrst approach that we have adopted, productivity growth appears as the
unexplained residual in a growth accounting framework with two factors of production -
labor and capital. In this context, we have compared the average rate of growth of TFP in
the periods in which the Latin American economies were, respectively, open and closed
to international trade. The criteria that we have used for deﬁning openness — proposed by
Sachs and Warner (1995) — are mostly policy-based, taking into account the levels of the
average tariff rates, the average nontariff barriers to imports and the premium on the
black market exchange rate. The main ﬁnding from this exercise is that, on average, the
growth of TFP was faster during the periods in which the Latin American countries were
“open”. This result seems to apply to almost all the countries here considered, the main
exceptions being Brazil and Mexico.

Since the larger rates of TFP growth found during the periods of openness could
be explained by higher degrees of capacity utilization - not captured by our data on

capital stocks — we adjusted these rates by using, in the calculation of TFP growth, a

51

proxy of cap
country. a re
values of 0
year that or
series of G.
capacity or
we found tl
capacity uti
closedness :
adjusted for
ﬁnd as befo
the DUNS c
Womance
Also '
“Wwwt
hm0d19804
With uirh t

the Work lore

 

rtsulted in sc

 

 

proxy of capacity output instead of actual output. To this end, we estimated, for each
country, a regression of GDP on both a linear and a quadratic trend, as well as on lagged
values of GDP. In most of the countries, these regressions involved a trend-break in a
year that was determined by the application of Vogelsang (1994) “SupF,” tests to the
series of GDP. The ﬁtted values of the above regressions were assumed to proxy for
capacity output. Comparing the unadjusted rates of TF P growth with the adjusted ones,
we found that the results of the growth accounting exercise are not driven by changes in
capacity utilization. Indeed, the difference between the periods of openness and those of
closedness remains considerable, although it is slightly lower when TFP growth is
adjusted for capacity utilization — 1.44 instead of 1.57 percent, on average. Similarly, we
ﬁnd, as before, that there are only four countries in which TFP growth was slower during
the periods of openness, with Brazil and Mexico still showing the most disappointing
performances.

Also with the objective of testing the robustness of our growth accounting results,
we adjusted the rates of growth of the labor force by using data on employment for the
period 1980-95, available for 15 countries. We then compared the resulting rates of TFP
growth with those previously obtained using the working-age population as a proxy for
the work force. Although the consideration of the changes in unemployment rates
resulted in some changes in the rates of TFP growth of individual countries, there were no
signiﬁcant changes in the average rates of the 15 countries considered — 0.01 percent
during the 19808, and 0.1 percent during the ﬁrst half of the 19903. Thus, it appears that

the rates of TFP growth calculated in this paper are not sensitive to the adjustment of the

52

data on capita
capacity utiliz

A secc
the pace of ca
openness. Thi
multiplying ti
described pro.
effects of the
Which they h;
the sustainabi
ftduction in t
greater expos
econOrnies Sh
311d toward 6:
intansive that
Occur ewe“ i 1‘

As art
for deﬁning 6

mm shareg

 

endOgCDOUSI‘

 

 

data on capital stocks and the labor force for, respectively, changes in the rates of
capacity utilization and unemployment.

A second important ﬁnding is that in the majority of the Latin American countries
the pace of capital accumulation has been relatively slower during the periods of
openness. This is also true when the capital stocks are adjusted for capacity utilization, by
multiplying them by the ratio of actual to capacity output (calculated with the above-
described procedure). A possible explanation for this result could be that the beneﬁcial
effects of the trade reforms — and of the other macroeconomic and structural reforms to
which they have often been associated — are more than offset by the uncertainty regarding
the sustainability of the new government policies, thus leading to an economy-wide
reduction in the rates of investment. Alternatively, another explanation could be that the
greater exposure to foreign competition that is caused by the opening of the region’s
economies should cause a reallocation of output away from import-substituting industries
and toward export-oriented ones. Since in the region the former are oﬁen more capital-
intensive than the latter, a reduction in the overall rate of capital accumulation could
occur even if individual industries are in fact accelerating their investment plans.

As an alternative to the use of the Sachs and Warner (1995) policy-based criteria
for deﬁning an economy as open, we performed an analysis of trend breaks in trade-
output shares. Thus, we searched for structural changes in openness revealed
endogenously by the data. To this end, we applied the Vogelsang (1994) “SupF,” tests, as
they were implemented by Ben-David and Papell (1997). The results showed that most
countries experienced breaks in their series of import-output and/or export-output ratios

(76 percent in the case of imports and 82 percent for exports) and that in the majority of

53

the cases ( 7
TFP grout!
the countrie
same direct
openness ar
We
countries f0.
before the bl
Were charact
We also four
(doom-am)
the break in t
A pro
Penods of l0\
using POIicy-l

n0l pQSSible l

 

 

the cases (74 percent) the trade shares increased aﬁer the breaks. Comparing the rates of
TFP growth before and after the trend-breaks, we found that in more than 50 percent of
the countries the changes in trade-output shares were not associated with changes in the
same direction in the average rates of TFP growth: no correlation between the changes
openness and TFP growth can be extracted from this analysis.

We also estimated structural breaks in the series of TFP, and found that in all 13
countries for which these breaks were signiﬁcant, TFP growth was slower after than
before the breaks. Moreover, in the majority of the cases, the aﬁer-TFP-breaks periods
were characterized by relatively larger levels of trade openness and slower TFP growth.
We also found that in 11 of the 13 countries that experienced signiﬁcant breaks in TF P,
(downward) breaks in GDP per worker were found to have occurred within two years of
the break in the TFP index, most commonly during the 1978-1982 period.

A problem with the comparison of the rates of TF P growth observed during
periods of lower and higher openness — regardless of whether these periods are deﬁned
using policy-based criteria or on the basis of statistical tests of trend breaks — is that it is
not possible to infer the existence of a relationship of causality from openness to
productivity growth. Indeed, changes in these variables could be determined
simultaneously by a third type of variable — an external shock, for example — or it could
be the case that faster growth is propitiated by greater openness rather than the other way
around. To uncover the existence of a causality link from openness to productivity growth
in the case of Latin America, we performed a regression analysis of a panel of 16
countries over the period 1960-1995. In order to purge from the data the effect of short

run ﬂuctuations, we used ﬁve-year averages instead of annual data. The endogeneity

54

problems wen
the use of lag;
variables.
The re
mentioned fin
was observed
of the Latin A
periods of Op
370% in out
the effect oft
indicated that
market prem:
DREF Cease.
the regression
much that a
mommies,
We a.
dependent W I
13’ 8€r trademf.

 

mCiUded m i:

“as mega“ c

- l
Linn Amen

 

problems were dealt with, by using a Generalized Method of Moments estimator based on
the use of lagged levels and ﬁrst differences as instruments for some of the explanatory
variables.

The regression analysis allowed us to test (and conﬁrm) the main above-
mentioned ﬁnding from the growth accounting exercise: that a higher rate of TFP growth
was observed during the periods of openness — as deﬁned by Sachs and Warner (1995) —
of the Latin American economies. Indeed, a dummy variable that is activated during the
periods of openness (DREF) appeared positive and signiﬁcant in a regression of the
growth in output per worker on the rate of growth of capital per worker. With regard to
the effect of other measures of openness on productivity growth, our regression results
indicated that larger rates of growth in both imports and exports, as well as lower black
market premiums have led to faster TFP growth in Latin America. We also found that
DREF ceases to be signiﬁcant when the above-mentioned variables are introduced into
the regressions, which suggests that these variables account for the larger rates of TFP
gI'OWth that are observed during the periods of openness of the Latin American
ecionornies.

We also ran regressions with the rate of growth of capital per worker as a
dependent variable and found that it is positively affected by larger export growth and
lal'ger trade-output shares, but it is not signiﬁcantly affected by import growth, nor by
lOwer black market premiums. We also found that when the dummy variable DREF was
included in the regressions together with the rest of the openness variables, its coefficient
Was negative and signiﬁcant. Thus, the rates of capital accumulation observed in the

Latin American countries during the periods of openness were lower than what should be

55

expecIE“
rates of
it would
pessimis
Ir

exports ar
significan
not of the i
suggesting
CﬁlClﬂlcy r
the coeﬂicit
go‘fmment
educational .
least in the a
trade hat'e 3F
trade haVe be
maCI’OecOnOI
had a Su’PTis

Cﬂ‘eCL howe.

 

 

 

expected from the evolution of the variables that we have used to measure the levels and
rates of change of the countries’ degrees of trade openness. In the spirit of Rodrik (1991),
it would be possible to speculate that this result is associated with the harmful effects of
pessimistic expectations concerning the lasting power of the reforms.

Interestingly, the regression analysis showed that, when the growth rates of
exports and imports are controlled for, the level of the trade-output ratio is not a
signiﬁcant determinant of the growth of TFP. The signiﬁcance of the rates of change and
not of the levels of the trade ﬂows-based openness variables could be interpreted as
suggesting that these variables affect productivity mainly through changes in the levels of
efficiency rather than by affecting directly the rates of change of productivity. However,
the coefﬁcient of the trade-output share becomes signiﬁcant when the level of
government consumption to GDP, the ratio of broad money (M2) to GDP, or the level of
educational attainment of the adult population are controlled for. Thus, it appears that, at
least in the aggregate and in the context of cross-country regressions, static gains from
trade have applied equally to most Latin America countries, but dynamics gains from
trade have been dependent on the countries’ degree of ﬁnancial development, on their
macroeconomic stability, and on their stock of human capital. The latter, in particular, has
had a surprisingly negative effect on the growth of TFP and capital per worker. This
effect, however, could be interpreted as reﬂecting a convergence mechanism by which the
greater international ﬂows of knowledge that are associated with greater openness beneﬁt
more the countries that lag further behind in terms of their knowledge stock. The
existence of convergence among the Latin American countries was also suggested by the

negative and signiﬁcant coefficients of the initial levels of GDP per worker and capital

56

per worker.
worker.

We al
The motivati
less open to i
considered in
With trade p0
influence of 5
Although the
industrial COL
relevant in tl.
Latin Amem

ll lS itl
POlicy refom
such as the p
the impleme:
Gould be mg.
in fact Captu;

fﬁformy To d

 

50mg addlllll

ﬁnancial rep I

   

money to Gr

to GDP- El'e.

per worker, respectively, in the regressions of TF P growth and growth in capital per

worker.

We also included, in all regressions, the total population of the countries involved.
The motivation for this was twofold. On one hand, since larger countries are normally
less open to international trade, it is important to control for the size of the countries
considered in order to isolate the component of the trade-output shares that is associated
with trade policies. On the other hand, total population is expected to capture the positive
inﬂuence of scale on growth, predicted by several models of endogenous growth.
Although the empirical veriﬁcation of “scale effects” has been controversial in studies of
industrial countries — e. g. Jones (1995) — our regression results show that they have been
relevant in the explanation of the rates of TFP growth and capital accumulation of the
Latin American economies: ceteris paribus, larger countries have in fact grown faster.
It is important to highlight the fact that in Latin America and elsewhere trade

P01icy reforms have been implemented in conjunction with other major policy changes,
SUCh as the privatization of public enterprises, the liberalization of ﬁnancial markets, and
the inrplementation of aggressive programs of macroeconomic stabilization. Thus, it
cOuld be argued that in our regressions the variables representative of trade openness are
in fact capturing the effects on productivity growth of the complete packages of policy
I‘eforrn. To deal with this possible criticism, we have included in our main regressions
Some additional control variables that are expected to capture at least the effect of

ﬁnancial reforms and stabilization policies - respectively, through the rate of broad

money to GDP, and the inﬂation rate together with the ratio of government consumption

to GDP. Even if it is not enough to disentangle the growth consequences of the various

57

types of polic
beyond the re
of the above l
robust to the 1
Moreover. as
positive effec
government p
Also .
breaks in the
Other finding
COUntries aft,
incteasing tr:
regltssion ar.
CCOnomies - I
and half deer:
collIttries. Htl
increases in ,

negatlbe incL

Were prt

 

 

types of policies implemented simultaneously in the region - a task that certainly goes
beyond the reach of the present paper -— this exercise has allowed us to conclude that most
of the above presented ﬁndings, concerning the effects of trade policies on growth, are
robust to the inclusion of variables representative of other government policies.
Moreover, as it was previously mentioned, it appears that one particular ﬁnding — the
positive effect of the trade-output share on TFP growth — only stands when other
government policies are controlled for.

Also worthy of comment is the fact that the results from the analysis of structural
breaks in the trade and productivity series may be considered to be in conﬂict with the
Other ﬁndings of this paper. Indeed, the drop in the rates of TFP growth observed in most
countries after the trend-breaks that took place during the early 19803, in a context of
increasing trade-output ratios, can be interpreted as contradicting the result, from the
regression analysis, that changes in the degree of openness of the Latin American
e=Conomies — as measured by the growth of their trade ﬂows — have had, over the past four
and half decades, a positive effect on the productivity growth performance of these
countries. However, it can also be argued that the potentially positive effects of the
increases in the levels of openness that occurred during this period were offset by the
n'egative incentives to all types of investment, both in physical and in knowledge capital,
that were provided by the context of high macroeconomic instability in which the region

Was engulfed. This hypothesis is in fact supported by the coincidence in time of the
breaks in TFP and GDP per worker, and by the fact that in the TFP regressions the

signiﬁcance of the trade-output share increases considerably when measures of stability

are included.

58

A cc
took place I
oriented pc
import sha
opening of
few years
63an she
associatec

both a res

surpluses

A complementary argument is that the breaks in trade shares that we uncovered
took place mostly as a result of external shocks and in the context of the same inward-
oriented policies that had prevailed in the region since the 19303. Thus, the breaks in
import shares were usually associated with the oil shocks of the 19703, while the actual
opening of the Latin American economies in terms of their trade policies only happened a
few years latter, most commonly in the late 19803 and the early 19905. The breaks in
export shares also preceded the advent of outward-oriented policies and were often
associated with sharp contractions in domestic demand and highly depreciated currencies,
both a result of macroeconomic policies aimed at the generation of balance of payments
surpluses in a context of high external indebtedness of the countries in question.

During the 19903, our growth accounting exercise showed that a considerable
recovery in the rates of growth of TFP and GDP per worker has been taking place in most
countries. In this period, most Latin American economies have reencountered positive
rates of growth in GDP per worker and, as suggested by our regression results, may have
started to reap the potential beneﬁts from both their new trade policies and the changes
that occurred in their trade shares during the previous decade. In terms of capital
investment, however, the performance of most countries has been quite poor during the
periods of openness of the Latin American economies. Advancing the understanding of
this phenomena is, we believe, an issue that merits further research.

Finally, it is worth noting that, from a theoretical point of view, we should expect
different dynamic responses to the opening of an economy to international trade,
depending on its speciﬁc structural characteristics. This is especially the case if countries

do not have access to same stock of “blueprints” or, in other words, if technology

59

spillovers an
evidence or
understand
country stu

the next cl

spillovers are not instantaneous nor global in reach. It is difficult, however, to uncover
evidence on this subject in the context of a cross-country analysis. Thus, to further our
understanding of the trade-productivity link in the region, it would be useful to perform
country studies, if possible using data at the industry level. It is what we attempt to do in

the next chapter of this dissertation.

60

Agacino. R.
Industri.‘

Alan. A. an
Caribbe;
No. l 10-

Azariadis. C
uarterl‘

Balassa B, 1
M;
York, N‘

Balﬁsa. B. 1
Press.

Barro. R. J.
0&2ch

Benhabib, j
D€Ve10pi
Ben'Dav-i d.

Blundell, R_
Pane] D;

 

Bonelli, R p
tin-W?

BOWen. H
FactOr 4:“

Bufkl, S J .'
and Ihe t

 

REFERENCES

Agacino, R., Rivas, G., and Roman, E. 1993. "Apertura y Eﬁciencia Productiva: La
Industria Chilena, 1975-1988." Revista de Analisis Economico 8 (2): 93-121.

 

Alam, A. and Rajapatirana, S. 1993. “Trade Policy Reform in Latin America and the
Caribbean in the 19808.” The World Bank Policy Research Working Papers Series
No. 1104. Washington, DC: The World Bank.

Azariadis, C. and Drazen, A. 1990. “Threshold Extemalities in Economic Development.”
Quarterly Journal of Economics 105: 501-26.

 

Balassa, B. 1989. “Growth Strategies in Semi-Industrial Countries.” In Comparative
Advantage, Trade Policy, and Economic Development, edited by B. Balassa. New
York, NY: New York University Press.

 

 

Balassa, B. 1993. Policy Choices for the 19903. New York, NY: New York University
Press.

 

Barro, R. J. 1991. “Economic Grth in a Cross Section of Countries.” Quarterly Journal
of Economics 106: 407-43.

 

 

Benhabib, J. and Spiegel, M.M. 1994. “The Role of Human Capital in Economic
Development: Evidence from Aggregate Cross-Country data.” Journal of Monetary
Economics 34: 143-173.

 

Ben-David, D. and Papell, D. H. 1997. “International Trade and Structural Change.”
Journal of International Economics 43: 513-523.

 

Blundell, R. and Bond, S. 1997 . “Initial Conditions and Moment Restrictions in Dynamic
Panel Data Models.” University College London, Discussion Paper 97-07.

Bonelli, R. 1992. “Growth and Productivity in Brazilian Industries: Impacts of Trade
Orientation.” Journal of Developing Economics 39 (1): 85-109.

 

Bowen, H., Learner, E., and Sveikauskas, L. 1987. “Multicountry, Multifactor tests of the
Factor Abundance Theory.” American Economic Review 77: 791-808.

 

Burki, S. J. and Perry, G. E. 1997. The LE March: a Reform Agenda for Latin America
and the Caribbean in the Next Decade. The World Bank Latin American and
Caribbean Studies. Washington, DC: The World Bank.

 

61

Calvo. G. A.
Mimeo. l

 

Caselli, F.. F
New LOO
389.

   
 
   

Collins. S. h
versus A

Dean. J. M“
Countrie
Discussi.

Dollar. D. l
Rapidly
M

Dombusch.
30m

Durlauf. S,
as

Eﬁlerly, \

E
%

Calvo, G. A. 1996. “Why is the market so Unforgiving? Reﬂections on the Tequilazo.”
Mirneo, University of Maryland.

Caselli, F ., Esquivel, G., and F. Lefort. 1996. “Reopening the Convergence Debate: A
New Look at Cross-Country Growth Empirics.” Journal of Economic Grth 1: 363-
389.

 

Collins, S. M. and Bosworth, B. P. 1996. “Economic Growth in East Asia: Accumulation
versus Assimilation.” Brookings Papers on Economic Activity 2: 135-203.

 

De Gregorio, J. 1992. “Economic Growth in Latin America.” Journal of Developing
Economics 39 (1): 59-84.

 

Dean, J. M., Desai, S., and Riedel J. 1994. “Trade Policy Reform in Developing
Countries since 1985: A Review of the Evidence.” The World Bank Policy
Discussion Papers Series No. 267. Washington DC: The World Bank.

Dollar, D. 1991. “Outward-Oriented Developing Economies Really do Grow More
Rapidly: Evidence from 95 LDCs, 1976-85.” Economic Development and Cultural
Change 40: 523-544.

 

Dombusch, R. 1992. “The Case for Trade Liberalization in Developing Countries.”
Journal of Economic Perspectives 6: 69-86.

 

Durlauf, S. 1996. “Controversy on the Convergence and Divergence of Growth Rates.”
The Economic Journal 106: 1016-18.

 

Easterly, W. 1993. “How Much do Distortions Affect Growth?” Journal of Monetary
Economics 32: 187-212.

 

Easterly, W., Loayza, N., and Montiel, P. 1997. “Has Latin America’s Post-reform
Growth been Disappointing?” Journal of International Economics 43: 287-311.

 

ECLAC (Economic Commission for Latin America and the Caribbean). 1996. The
Economic Experience of the Last Fifteen Years: Latin America and the Caribbean,
1980-1995. Santiago, Chile: ECLAC, United Nations.

Edwards, S. 1992. “Trade Orientation, Distortions and Growth in Developing Countries.”
Journal of Developing Economics 39 (1): 31-57.

 

Edwards, S. 1993. “Openness, Trade Liberalization, and Growth in Developing
Countries.” Journal of Economic Literature 31 (3): 1358-93.

 

Edwards, S. 1995. Crisis and Reform in Latin America: From Despair to Hope. New
York, NY: Oxford University Press.

 

62

Edwards. S.
National

Feder. G. I?
ii: 595.

feenstra R
lSpecie

257.

fischet. S

Econo

Grossman
Cami

Grossma
Cam

Griliche
Proc
Gﬁllche
Lev

Crri

Haberg
lg

Harri g
Pl’t

De

Edwards, S. 1997. “Openness, Productivity and Growth: What do we Really Know?”
National Bureau of Economic Research Working Paper Series No. 5978, March.

Feder, G. 1983. “On Exports and Economic Growth.” Journal of DevelopingEconomics
12: 59-74.

 

Feenstra, R. 1996. “Trade and Uneven Growth.” Journal of Develoging Economics 49 (1)
(Special Issue: Increasing Returns, Competition and Economic Development): 229-
257.

 

Fischer, S. 1993. “The Role of Macroeconomic Factors in Growth.” Journal of Monetary
Economics 32: 485-512.

 

Grossman, G. M. Editor. 1992. Imperfect Competition and International Trade.
Cambridge, MA: The MIT Press.

 

Grossman, G. M.and Helpman, E. 1991. Innovation and Growth in the Global Economy.
Cambridge, MA: The MIT Press.

 

Griliches, Z. 1992. “Issues in Assesing the Contribution of Research and Development to
Productivity Growth.” The Bell Journal of Economics : 92-116.

 

Griliches, Z. and Lichtenberg, F. 1984. “R&D and Productivity Growth at the Industry
Level: Is There Still a Relationship.” In R&D, Patents, and Productivity, edited by
Griliches, Z. Chicago, IL: University of Chicago Press.

 

Haberger, A. 1959. "Using Resources at Hand More Efficiently." American Economic
Review 40 (1): 134-146.

 

Harrigan, J. 1996. “Econometric Estimation of Cross-Country Differences in Industry
Production Functions.” Mimeo, Federal Reserve Bank of New York.

Harrison, A. 1996. “Openness and growth: A Time-Series, Cross-Country Analysis for
Developing Countries.” Journal of Development Economics 48: 419-447.

 

Havyrlyshyn, O. 1990. “Trade Policy and Productivity Gains in Developing Countries: A
Survey of the Literature.” The World Bank Research Observer 5 (1): 1-24.

 

Helleiner, G. K. Ed. 1994. Trade Policy and Industrialization in Turbulent Times.
London: Routledge.

 

Helliwell, J. F. 1994. “Trade and Technical Progress.” In Economic Growth and the
Structure of Long Term Development, edited by Pasinetti, L. L. and R. M. Solow.
New York, NY: St. Martin’s Press.

 

63

Helliwell, J. F. and Chung, A. 1991. “Macroeconomic Convergence: International
Transmission of Growth and Technical Progress.” In International Economic
Transactions: Issues in Measurement and Empirical Research, edited by Hooper, P.
and D. J. Richardson. Chicago, IL: University of Chicago Press and NBER.

 

 

Helpman, E. 1992. “Endogenous Macroeconomic Growth Theory.” European Economic
Review 36: 237-267.

 

Islam, N. 1995. “Growth Empirics: a Panel Data Approach.” Quarterly Journal of
Economics 110: 1127-70.

 

Jones, C. I. 1995. “Time Series Tests of Endogenous Growth Models.” Quarterly Journal
of Economics 110: 494-524.

 

 

Krueger, A. O. 1997. “Trade Policy and Economic Development: How we Learn.”
American Economic Review 87 (1): 1-22.

 

Krugman, P. 1987. “The Narrow Moving Band, the Dutch Disease, and the Competitive
Consequences of Mrs. Thatcher: Notes on Trade in the Presence of Dynamic Scale
Economies.” Journal of Developing Economics 27: 41-55.

 

Learner, E. 1988. “Measures of Openness.” In Trade Policy Issues and Empirical
Analysis, edited by R. Baldwin. Chicago, IL: University of Chicago Press.

 

Lee, J. 1996. “Government Interventions and Productivity Growth.” Journal of Economic
Growth 1: 391-414.

 

Lefort, F. and Solimano, A. 1994. “Economic Growth after Market-Based Reform in
Latin America: The Cases of Chile and Mexico.” Working Paper. World Bank,
Macroeconomics and Growth Division, Whasington, DC

Levine, R. and Renelt, D. 1992. “A Sensitivity Analysis of Cross-Country Growth
Regressions.” American Economic Review 82: 942-63.

 

Loayza, N. 1994. “A Test of the International Convergence Hypothesis Using Panel
Data.” The World Bank Policy Research Working Papers Series No. 1333.
Washington, DC: The World Bank.

Lucas, R. E. Jr. 1988. “On the Mechanics of Economic Development.” Journal of
Monetary Economics 22: 3-42.

 

Lucas, R. E. Jr. 1993. “Making a Miracle.” Econometrica 61: 251-72.

 

64

Marfan, M. and Bosworth, B. 1994. “Savings, Investment, and Economic Growth.” In
The Chilean Economy: Policy Lessons and Challenge_s, edited by Bosworth, B.,
Dombusch, R., and Laban, R. Washington, DC: Brookings Institution.

 

Michaely, M. 1977. “On Exports and Economic Growth: an Empirical Investigation.”
Journal of Developing Economics 4: 49-54.

 

Minhas, B. 1962. “The Homohypallagic Production Function, Factor Intensity Reversals,
and the Heckscher-Ohlin Theorem.” Journal of Political Economy 70: 138-156.

 

Nehru, V. and Dareshwar, A. 1993. “A New Database on Physical Capital Stock:
Sources, Methodology and Results.” Revista de Analisis Economico 8 (1): 37:59.

 

Nishimizu, M. and Page Jr., J. M. 1991. “Trade Policy, Market Orientation and
Productivity Change in Industry.” In Trade Theory and Economic Reform: Essays in
Honor of Bela Balassa, edited by J. de Mello and A. Sapir. Cambridge, MA: Basil
Blackwell.

 

 

Nishinrizu, M. and Robinson, S. 1984. “Trade Policies and Productivity Changes in
Semi-Industrialized Countries.” Journal of Develomnent Economics 16: 177-206.

 

Oks, D. 1994. “Structural Reforms and Productivity Growth: Sectoral Analysis for Chile
and Mexico.” Paper presented at the Cross Fertilization Seminar Series, World Bank,
Latin America and Caribbean Region, Washington, DC

Pack, H. 1988. “Industrialization and Trade.” In Handbook of Develogment Economics,
Vol. I, edited by H.B. Chenery and TN. Srinivasan. Amsterdam: North Holland.

 

Pack, H. 1992. “Learning and Productivity Change in Developing Countries.” In Trade
Policy, Industrialization, and Development, edited by GK. Helleiner. Oxford:
Clarendon Press.

 

Papageorgiou, D., Michaely, M., and A. Choksi, eds. 1991. Liberalizing Foreign Trade:
Lessons from Experience in the Developing World. Cambridge, MA: Basil
Blackwell.

 

Pritchett, L. 1996. “Measuring Outward Orientation in LDCs: Can it be Done?” Journal
of Development Economics 49: 307-335.

Quah, D. and Rauch, J. E. 1990. “Openness and the Rate of Economic Growth.”
Manuscript, MIT.

Rivera-Batiz, L. A. and Romer, P. 1991. “International Trade with Endogenous
Technological Change.” European Economic Review 35: 971-1004.

 

65

Roberts, M. J. and Tybout, J. R. Eds. 1996. Industrial Evolution in Developing Countries:
Micro Patterns of Turnover, Productivity and Market Structure. New York, NY:
Oxford University Press.

 

Rodrik, D. 1991. “Policy Uncertainty and Private Investment in Developing Countries.”
Journal of Development Economics 36: 229-242.

 

Rodrik, D. 1992a. “The Limits of Trade Policy Reform in Developing Countries.”
Journal of Economic Perspectives 6: 87-105.

 

Rodrik, D. 1992b. “Closing the Productivity Gap: Does Trade Liberalization Really
Help?” In Trade Policy, Industrialization, and Development, edited by GK. Helleiner.
Oxford: Clarendon Press.

 

Rodrik, D. 1994. “King Kong Meets Godzilla: The World Bank and the East Asian
Miracle.” Centre for Economic Policy Research Discussion Papers Series No. 944.

Rodrik, D. 1995. “Trade and Industrial Policy Reform.” In Handbook of Development
Economics, Vol. III, edited by J. Behrman and TN. Srinivasan. Amsterdam: Elsevier
Science.

 

Sachs, J. and Warner, A. 1995. “Economic Reform and the Process of Global
Integration.” Brookings Papers of Economic Activity 1: 1-118.

 

Smith, A. 1976. An Inquiry Into the Nature and Causes of the Wealth of Nations. Oxford
: Clarendon Press.

Solimano, A. 1996. Road Maps to Prosperity: Essays on Growth and Development,
edited by A. Solimano. AnnArbor: The University of Michigan Press.

 

Solow, RM. 1956. “A Contribution to the Theory of Economic Growth.” Quarterly
Journal of Economics 70: 65-94.

 

Stokey, N. L. 1991. “Human Capital, Product Quality, and Growth.” Quarterly Journal of
Economics 106 (2): 587-616.

 

Syrquin, M. and Chenery, H. 1989. “Three Decades of Industrialization.” The World
Bank Economic Review 3 (2): 145-181.

Thomas, V., Halevi, N., and J. Stanton. 1991. “Does Policy Reform Improve
Performance?” Background Paper for World Development Report 1991. Whasington,
DC, The World Bank.

Treﬂer, D. 1995. “The Case of the Missing Trade and Other Mysteries.” American
Economic Review 85 (5): 1029-1046.

66

Tybout, J. R. 1991. “Researching the Trade-Productivity Link: New Directions.” The
World Bank Policy Research Working Papers Series No. 638. Washington DC: The
World Bank.

Tybout, J. R. 1992. “Linking Trade and Productivity: New Research Directions.” The
World Bank Research Observer 6 (2): 189-211.

 

Vogelsang, T. 1994. “Wald-Type Tests for Detecting Shiﬁs in the Trend Function of a
Dynamic Time Series.” CAE Working Paper 94-12, Cornell University. Forthcoming
Econometric Theory.

World Bank. 1987. World Development Report 1987. New York, NY: Oxford University
Press.

 

World Bank. 1993. The East Asian Miracle: Economic Growth and Public Policy. New
York, NY: Oxford University Press.

 

World Bank. 1994. Trends in Developing Economies 1994. Washington, DC: The World
Bank.

 

Young, A. 1991. “Learning by Doing and the Dynamic Effects of International Trade.”
Quarterly Journal of Economics 106 (2): 369-406.

 

67

APPENDIX A

TABLES AND FIGURES

68

_\ ...u.__...ulu..._,D vet—...; ..l_.u 4.4.-.. ....T; 14:; iw~.C,l~..Lt.Ht.~.r-H~,.W l~\...w.~\..1h~.i..i.. NH...

 

Table 1.1: Indicators of Trade Regimes Before and After Reform

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Country Unweighted Tariff Range (%) Unweighted Trade (Imports
(prc-reform year, Average of Legal Average Coverage + Exports) as
post-reform year) Tariff Rates (%) of Non-Tariff Percentage of GDP,
Barriers (%) 1980 Prices
Pre- Post- Pre- Post- Pre- Post- Pre- Post-
Reform Reform Reform Reform Reform Reform Reform Reform
(1985- (1991-
1987) 1992)
Argentina (1987, 42 (a) 15 15-1 15 5-22 31.9 8.0 38.57 54.32
1991)
Bolivia (1985, 20 (a) 8 (a) 0-20 5-10 25.0 0.0 57.51 83.97
l991-1992)
Brazil (1987, 51 21 0-105 0-65 35.3 10.0 21.17 25.27
1992)
Chile (1984, 35 11 35 11 10.1 0.0 44.96 56.34
1991)
Colombia (1984, 61 12 0-220 5-20 73.2 1.0 28.23 32.66
1992)
Costa Rica (1985, 53 (a) 15 (a) 0-1400 5-20 0.8 0.0 58.66 78.97
1992) (a)
Ecuador (1989, 37 (a) 18 0-338 2-25 (c) 59.3 - 48.73 50.84
1992) (a)
El Salvador (1987, - 10.2 - 1-30 (i) - 19.2 (d) - -
1995)
Guatemala (1985, 50 (a) 15 (a) 5-90 5-20 7.4 6.0 31.31 35.56
1992)
Guyana (1987, - 15 -- - - - - --
1995)
Honduras (1985, 41 (a) 15 (b, a) 5-90 5-20 - - 62.82 61.76
1992)
Jamaica (1981, - 20 - 0-45 - 6.6 ((1) 105.51 163.49
1991)
Mexico (1985, 24 (b) 13 (b) 0-100 0-20 12.7 20.0 22.63 34.31
1990)
Nicaragua (1985, 54 (a) 10.7 ((1) 1-100 0-10 27.8 - - ~-
1990) (6)
Paraguay (1985, 71.7 (a) 16 (a) 0-44 (0 3-86 9.9 0.0 51.01 63.14
1991-1992) (g)
Peru (1985, 1992) 64 (a) 15 (3) 0-120 5-25 53.4 0.0 30.37 41.58
(h) (g) (g)
Uruguay (1987, 32 18 10-55 12-24 14.1 0.0 38.04 45. 10
1992)
Venezuela (1989, 37 19 0-135 0-50 44.1 5.0 49.25 53.29
1991)

 

 

 

 

 

 

 

 

 

Notes: - Not available; (a) Including tariff surcharges; (b) Production-weighted average tariff; (c) Ecuador also has a

speciﬁc tariff of 40 % on automobiles; (d) 1990-1993; (e) 1986; (f) 1984; (g) 1988; (h) l991-1992; (i) 1994.
Source: Alarn and Rajapatirana (1993), Burki and Pen ( 1997), Edwards (1995), IDB (1996).

69

 

Table 1.2: Bilateral Real Exchange Rate Relative to the US. Dollar

in Selected Countries, 1980/1993

 

 

(1985:100)
Country 1980 1983 1987 1992 1993“)
Argentina 35.08 96.4 80.8 36.9 34.0
Bolivia 88.1 84.6 107.9 109.6 1 13.3
Brazil 70.7 88.7 78.0 51.7 45.7
Chile 55.3 75.3 94.8 75.1 75.2
' Colombia 79.2 78.3 1 15.9 119.9 102.6
Costa Rica 65.8 103.0 94.9 88.2 82.9
Ecuador 105.6 104.5 153.3 165.7 153.9
El Salvador 172.6 133.9 121.0 103.7 --
Guatemala 124.9 120.5 162 149.5 --
Honduras 121.6 106.2 93.2 141.5 152.3
Jamaica 60.1 54.8 80.1 94.5 70.7
Mexico 83.3 1 19.8 123.9 68.72 63.8
Paraguay 74.4 60.7 1 1 1.4 1 13.0 --
Peru 77.3 80.6 46.2 21.7 23.6
Uruguay 49.7 89.4 77.2 55.5 40.5
Venezuela 84.2 70.3 134.8 122.3 1 19.1

 

Note: Increases indicate currency depreciation against the dollar; -- not available;
(a) Preliminary.
Source: Edwards (1995).

70

Me

Nit

Par
Per

L'n

Vet

AM

Av

Av

Av

Av.

Soc

Table 1.3: Grth Decomposition: 1950/ 1995 (in percent)

 

 

Country GDP growth Capital") Labor Productivity (4)/(1)
(in percent)
(1) (2) (3) (4) (5)
Argentina 2.38 1.22 0.83 0.33 13.94
Bolivia 2.43 1.17 1.43 -0. l 7 -7.07
Brazil 5.15 2.41 1.72 1.03 19.96
Chile 3 .67 1.33 1.37 0.97 26.37
Colombia 4.55 1.77 1.82 0.96 21.06
Costa Rica 5.08 2.54 2.16 0.38 7.55
Ecuador 4.91 1.93 1.85 1.14 23.13
El Salvador 3 .57 2.08 1.45 0.04 1.03
Guatemala 3 .82 1.84 1 .76 0.23 5.96
Guyana 1.16 0.76 1.37 -0.97 -83.16
Honduras 3 .86 1.88 1.72 0.26 6.73
Jamaica 2.85 1.26 0.83 0.76 26.61
Mexico 4.69 2.62 1.77 0.31 6.57
Nicaragua 2.65 2.05 1.93 -1.33 -50.06
Paraguay 4.27 2.69 1.95 -0.37 -8.71
Peru 3.61 1.51 1.49 0.61 16.82
Uruguay 1 .93 0.42 0.43 1.08 55.68
Venezuela 3 .45 1 .44 2.22 -0.20 -5.88
Average 3.56 1.72 1.56 0.28 7.87
Av.50-70 4.50 2.01 1.52 0.97 21.61
Av.70-80 4.19 2.27 1.70 0.22 5.16
Av.80-90 0.92 0.90 1.53 -1.52 - 165.46
Av.90-95 3.88 1.03 1.49 1.36 34.96

 

Source: see text. ‘7 A capital share of 0.4 is assumed.

71

Tab

ESE
Boll
Bra;
Chil
Colt
Cos
Ecu
El 5
Que
Guy
Hor

Mei
Nit;
Pan
Pen
L’ru
Vet.
Ave

\
Sou

Table 1.4: Total Factor Productivity Growth by Sub-periods: 1950/ 1995 (in percent)

 

Country/Period 1950-1970 1970-1980 1980-1990 1990-1995 1950-1995

 

Argentina 0.77 -0.03 -2. 13 4.21 0.33
Bolivia -0.60 0.16 -0.97 2.46 -0. 17
Brazil 1.69 2.66 -1.52 0.21 1.03
Chile 0.73 0.28 0.61 4.00 0.97
Colombia 1.05 1.61 0.02 1.19 0.96
Costa Rica 1.07 -0. 19 -0.71 0.99 0.38
Ecuador 0.88 4.07 -0.93 0.41 1.14
El Salvador 0.82 - 1 .04 -1.50 2.12 0.04
Guatemala 0.51 1.41 -1.71 0.74 0.23
Guyana -1. 14 -1.31 -3.93 6.51 -0.97
Honduras 0.78 1.1 1 -1.29 -0.44 0.26
Jamaica 3.29 -3.54 0.82 0.01 0.76
Mexico 1.62 1.05 -1.85 -2.1 l 0.31
Nicaragua 1.54 -3.54 -4.61 -0.98 -1.33
Paraguay -0.60 1.83 -1.96 -0.54 -0.37
Peru 2.30 0.00 -3.21 2.85 0.61
Uruguay 1.28 1.78 -0.65 2.31 1.08
Venezuela 1.52 -2.42 -1.76 0.47 -0.20
Average 1.0 0.2 -1.5 1.4 0.28

 

Source: see text. "l A capital share of 0.4 is assumed.

72

Br

Cc

Co

Ett

Gue

Guy

Hot

Jam

Mei

Nit.

Pan

Pen

Cm

Table 1.5: Periods of Openness: 1950/1995

 

 

Country Openness Periods Openness Periods / Total Period
(percent)

Argentina Since 1991 11

Bolivia 1956-79 and 74
since 1991

Brazil Since 1991 11

Chile Since 1976 43

Colombia Since 1986 22

Costa Rica 1952-61 and 43
since 1986

Ecuador 1950-80 and 78
since 1991

El Salvador 1950-61 and 37
since 1991

Guatemala 1950-61 and 43
since 1988

Guyana Since 1988 17

Honduras 1950-61 and 37
since 1991

Jamaica 1962-73 and 41
since 1989

Mexico Since 1986 22

Nicaragua 1950-60 and 35
since 1991

Paraguay Since 1989 15

Peru 1948-67 and 50
since 1991

Uruguay Since 1990 13

Venezuela 1950-59 and 33

1989-93
Average -- 35

 

Source: see text.

73

Table 1.6: Changes in Growth Rates: Periods of Openness and Closedness of the

Economy, 1950/1995 (in percent)

 

 

 

 

Country / Total Factor Productivity Gross Domestic Product Capital Labor (3)/(6)
Variable Growth Growth Stock‘" F one") (in
percent)
Status Closed Open Open - Closed Open Open - Open - Open - Open -
Closed Closed Closed Closed Closed
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Argentina -0.2 4.2 4.4 2.1 5.0 2.9 -1.6 0.2 149
Bolivia -2.4 0.5 2.9 -O.2 3.2 3.3 0.2 0.2 88
Brazil 1.1 0.2 -0.9 5.5 2.5 -3.0 -1.7 -0.3 31
Chile -0.3 2.6 2.9 2.6 5.0 2.5 -0.2 -0.3 l 17
Colombia 0.9 1.3 0.5 4.6 4.4 -0.1 -0.2 -0.4 -343
Costa Rica 0.0 0.9 0.9 4.7 5.5 0.8 -0.2 0.1 115
Ecuador -0.7 1.5 2.3 2.0 5.5 3.6 1.2 0.1 64
E1 -0.6 0.9 1.5 2.9 4.5 1.6 0.0 0.1 95
Salvador
Guatemala 0.3 0.1 -0.2 3.9 3.7 -0.3 -0.2 0.2 87
Guyana -1.7 2.5 4.2 0.8 2.8 2.0 -1.3 -0.9 206
Honduras 0.3 0.2 -0.1 4.2 3.3 -0.9 -0.3 -0.6 9
Jamaica -0.1 1.8 1.9 1.8 4.2 2.4 0.9 -0.4 79
Mexico 0.9 -1.9 -2.8 5.7 1.0 -4.7 -1.8 -0.2 59
Nicaragua -2.2 0.4 2.6 1.8 4.4 2.6 0.1 -0.1 101
Paraguay -0.4 -0.2 0.1 4.4 3.5 -0.9 -0.8 -0.3 -15
Peru -1.3 2.6 4.0 1.7 5.6 3.8 0.5 -0.6 104
Uruguay 0.9 2.0 1.1 1.8 2.9 1.1 0.0 0.1 96
Venezuela -1.2 2.0 3 .2 2.4 5.8 3.3 0.1 0.1 94
Average -0.4 1.2 1.6 2.9 4.0 1.1 -0.3 -0.2 142

 

 

 

 

 

 

 

 

 

 

Source: see text. ‘7 Contribution to GDP growth, assuming a capital share of 0.4.

74

 

Table 1.7: Total Factor Productivity Growth?) Adjusting for Employment,

 

 

1980/1995 (in percent)
Country/ 1980-1990 1980-1990 1990-1995 1990-1995 1980-1995 1980-1995
Period Adjusted Non- Adjusted Non- Adjusted Non-
Adjusted Adjusted Adjusted
Argentina -1.82 -2. 13 5.59 4.21 0.65 -0.01
Bolivia -0.96 -0.97 1.99 2.46 0.03 0.17
Brazil -1.64 -1.52 0.25 0.21 -1.01 -0.94
Chile 0.27 0.61 4.11 4.00 1.55 1.74
Colombia 0.07 0.02 0.76 1.19 0.27 0.31
Costa Rica -0.75 -0.71 1.03 0.99 -0.16 -0. 14
Ecuador‘"’ -0.15 -0.10 0.61 0.41 0.11 0.11
Guatemala -1.44 -1.71 0.84 0.74 -0.79 -0.92
Honduras -1.36 -1.29 -0.85 -0.44 -1.19 -1.01
Mexico -1.96 -1.85 -1.66 -2.1 1 -1.86 -1.94
Nicaragua -5. 12 -4.61 0.34 -0.98 -3.86 -4.06
Paraguay -1.80 -1.96 -l.18 -0.54 -1.62 -1.64
Peru -3.13 -3.21 2.48 2.85 -1.53 -l.61
Uruguay -0.54 -0.65 2.53 2.31 0.49 0.34
Venezuela -1.47 -1.76 0.45 0.47 -0.83 -1.02
Average -1.45 -1.46 1.15 1.05 -0.65 -0.71

 

 

 

 

 

 

 

Source: see text. 0) A capital share of 0.4 is assumed. (”l Data available for 1983-1995.

75

Coun‘

‘ Arget
Boliv
Brazi
Chile
Colo
C ost;
Ecua
El 32
Goat
Guy;

HOW

Met
Nit;
Pan

Per

Uri

 

 

Table 1.8: Structural Trend Breaks in GDP, 1950/1995

 

 

 

 

Country Break in GDP

Break Year Model SupF t
Argentina 80 I 30.40"")
Bolivia 78 1 47.28")
Brazil 81 l 48.76")
Chile 83 I 14.08
Colombia 80 I 30.58"")
Costa Rica 80 I 79.38")
Ecuador 73 I 40.52")
El Salvador 79 1 1 17.06")
Guatemala 8 1 1 127.3 7")
Guyana 82 I 20.01
Honduras 78 I 55.93")
Jamaica 72 I 34.41""
Mexico 81 1 49.5 1‘"
Nicaragua 78 1 48.21")
Paraguay 82 I 53.00")
Peru 88 I 32.69""
Uruguay 82 1 37.16""
Venezuela 81 I 47.49")

 

 

 

 

 

Source: see text. "), "" and "'" stand for statistical signiﬁcance at the 1, 5, and 10 percent level,
respectively, using unit root critical values. For model 1, these are 38.35, 31.29, 27.99, respectively. For
model 11, the critical values are 30.36, 25.10, and 22.29, while for model 111 they are 22.48, 17.88, and
15.78.

76

Table 1.9: Growth Rates of Total Factor Productivity Adjusting for
Capacity Utilization, 1950/1995 (in percent)

 

 

 

 

Country/ 1950-70 1970-80 1980-90 1990-95 1950-95 Closed Open Open -
Period Closed
Argentina 0.70 0.40 -1.90 4.00 0.42 -0.02 4.00 4.02
Bolivia -0.43 -0.21 -1.00 2.60 -0. 17 -2.65 0.54 3.19
Brazil 1.99 1.58 -0.61 -0.64 1.03 1.24 -0.64 -1.88
Chile 0.60 0.06 1.29 3.34 0.94 0.27 1.78 1.51
Colombia 1.1 1 1.64 -0.22 1.11 0.93 0.87 1.16 0.29
Costa Rica 1.11 0.03 -l .00 1.1 1 0.40 0.11 0.77 0.66
Ecuador 1.26 3.26 -0.77 0.51 1.15 -0.74 1.56 2.30
El Salvador 0.87 -1.09 -1.55 2.01 0.02 -0.78 1.22 2.00
Guatemala 0.54 1.30 -1.55 0.65 0.25 0.34 0.1 1 -0.23
Guyana -1 . 16 -1.73 -2.98 4.68 -1. 17 -1.52 0.68 2.20
Honduras 0.77 1 . 19 -1.13 -1. 10 0.23 0.22 0.24 0.02
Jamaica 3.08 -2.76 0.07 2.04 0.72 -0.20 2.14 2.34
Mexico 1.59 1.00 -1.66 -1.74 0.36 0.87 -1.40 -2.27
Nicaragua 1.56 -3.89 -3.64 -3.26 -1.26 -2.36 1.30 3.66
Paraguay -0.55 1.20 -1.39 -0.70 -0.36 -0.34 -0.48 -0. 14
Peru 2.36 -0.12 -3.36 1.86 0.45 -1.48 2.56 4.04
Uruguay 1.21 1.39 -0.03 2.81 1.15 0.93 2.57 1.64
Venezuela 1.44 -2.07 -1.98 0.37 -0.22 -0.99 1.51 2.50
Average 1.00 0.07 -1.30 1.09 0.27 -0.35 1.09 1.44

 

 

 

 

 

 

 

 

 

Source: see text. ("Contribution to GDP growth, assuming a capital share of 0.4.

77

 

Table 1.10: Changes in Growth Rates of Capital Stocks, Adjusting for
Capacity Utilization: Periods of Openness and Closedness of the

Economy, 1950/1995 (in percent)

 

 

 

 

Country Growth in Adjusted Capital Stocks Growth in Non-Adjusted Capital
Stocks
Status Closed Open Open - Closed Open Open -
Closed Closed
Argentina 3.33 -0.39 -3.72 3.49 -0.60 -4.09
Bolivia 1 . 16 2.84 1.68 2.58 3.02 0.44
Brazil 6.51 3.00 -3.51 6.50 2.15 -4.35
Chile 2.95 4.51 1.56 3.52 3.08 -0.43
Colombia 4.47 4.06 -0.41 4.53 3.96 -0.57
Costa Rica 6.49 5.71 -0.78 6.59 6.04 -0.55
Ecuador 2.32 5.61 3.29 2.32 5.35 3.03
El Salvador 5.38 4.54 -0.84 5.17 5.25 0.08
Guatemala 4.77 3.54 -1.23 4.78 4.31 -0.47
Guyana 2.16 2.48 0.32 2.41 -0.76 -3.16
Honduras 5.00 3 .56 -1.44 4.93 4.29 -0.64
Jamaica 2.23 4.56 2.34 2.26 4.54 2.29
Mexico 7.66 2.69 -4.98 7.52 3.15 -4.38
Nicaragua 5.21 2.28 -2.93 5.05 5.28 0.23
Paraguay 7.60 5.10 -2.50 6.99 4.99 -2.01
Peru 3.36 3.62 0.26 3 .22 4.40 1.18
Uruguay 1.24 0.45 -0.79 1.07 1.03 -0.04
Venezuela 3.31 3.88 0.57 3.50 3.81 0.32
Average 4.18 3.45 -0.73 4.25 3.52 -0.73

 

 

 

 

 

 

 

Source: see text.

78

 

Table 1.11: Structural Trend Break Tests in Import-Output Ratios

 

 

 

 

Country Period Import-Output Ratios Changes in Growth Rates
(percent)
Break Model SupF t Change TFP GDP Capital
Year in per Stock
Import worker (p.w.)
Share (p.w.)
(%)
Argenti- 1959/95 1974 1 30.3"") -15.3 -0.8 -2.3 -3.7
na
Bolivia 1951/95 1988 I 24.5 15.9 3.0 1.2 -4.6
Brasil 1951/95 1974 1 29.6"") -7.9 -2.7 -3.4 -1.7
Chile 1951/95 1973 1 51.9") 99.3 0.8 0.6 -0.4
Colombia 1958/95 1980 I 27.9 29.9 -1.2 -1.0 0.4
Costa 1951/95 1982 1 107.6") 19.3 0.6 -0.1 -1.6
Rica
Ecuador 1951/95 1985 I 28.4"”) 31.6 -1.6 -3.0 -3.4
E1 1958/95 1973 I 35.6"" 17.8 -1.5 -2.0 -1.4
Salvador
Guatema- 1951/95 1981 1 74.6") 19.7 -1.6 -3.1 -3.9
la
Guyana 1952/93 1975 1 26.8 44.7 -0.6 -1.3 -1.7
Honduras 1958/95 1973 I 28.4"") 19.9 -1.6 -2.1 -1.4
Jamaica 1951/94 1984 1 48.4") 39.2 0.2 -0.6 -3.2
Mexico 1957/95 1980 1 43.6") 44.3 -3.3 -4.7 -3.7
Paraguay 1951/95 1987 1 34.8"" 106.4 0.5 -0.1 -1.7
Peru 1951/95 1973 1 22.5 -28.3 -3.3 -4.0 -2.4
Uruguay 1965/95 1979 1 38.0"" 48.4 -0.8 -1 .0 -0.4
Venezue- 1957/95 1976 1 50.2") 10.5 - l .5 -1.7 -0.4
In

 

 

 

 

 

Source: see text. f", "" and "'" stand for statistical signiﬁcance at the 1, 5, and 10 percent level,
respectively, using unit root critical values. For model 1, these are 38.35, 31.29, 27 .99, respectively. For
model 11, the critical values are 30.36, 25.10, and 22.29, while for model 111 they are 22.48, 17.88, and
15.78.

79

Table 1.12: Structural Trend Break Tests in Export-Output Ratios

 

 

 

 

Country Period Export-Output Ratios Changes in Growth Rates
(percent)
Break Model SupF t Change TFP GDP Capital
Year in per Stock
Export worker (p.w.)
Share (p.w.)
(%)
Argenti- 1959/95 1979 1 23. 1 -6.8 -0.2 -2. 1 -4.7
na
Bolivia 1951/95 1988 1 32.9"" -6.4 3 .0 1.2 -4.6
Brasil 1951/95 1982 I 50.3") 14.4 -1.9 -3.5 -3 .9
Chile 1951/95 1987 1 31.5"" 87.9 3.7 4.8 2.9
Colombia 1958/95 1980 1 24.6 16.3 -1.2 -1.0 0.4
Costa 1951/95 1980 1 83.8") 32.1 -1.0 -1.7 -1.9
Rica
Ecuador 1951/95 1971 11 25.2"" 47.8 0.4 0.01 -1.1
El 1958/95 197 7 I 83 .2‘" -32.9 -1.7 -2.7 -2.5
Salvador
Guatema- 1951/95 1978 1 38.3"" -0.6 -1.6 -2.8 -3.0
la
Guyana 1952/93 1986 III 22.3"" 41.6 2.4 1.7 -1.7
Honduras 1958/95 1989 I 37.4"" 2.6 -1.3 - l .9 - l .4
Jamaica 1951/94 1978 1 18.1 35.2 -1.8 -3.4 -5.5
Mexico 1957/95 1988 1 35.8"" 46.3 -1.7 -2.9 -3.0
Paraguay 1951/95 1988 1 46.1") 41.1 0.1 -0.4 -1.6
Peru 1951/95 1976 1 30.0"") -29.0 -3.5 -4.4 -3.0
Uruguay 1965/95 1982 11 30.4"") 5 1 .3 -0. 1 -0.9 -2.0
Venezue- 1957/95 1984 111 17.8"”) -7.6 1.4 0.6 -2.2
la

 

 

 

 

 

Source: see text", "" and "‘"stand for statistical signiﬁcance at the 1, 5, and 10 percent level,
respectively, using unit root critical values. For model 1, these are 38.35, 31.29, 27.99, respectively. For

model 11, the critical values are 30.36, 25.10, and 22.29, while for model 111 they are 22.48, 17.88, and
15.78.

80

Table 1.13: Structural Trend Breaks in Total Factor Productivity, 1950/1995

 

 

 

 

Country Break in Total Factor Changes in Growth Rates Changes in Trade-Output
Productivity Index (percent) Shares (percent)
Break Model SupF, TFP GDP Capital Period‘" Im- Ex-
Year per Stock ports ports
worker (p.w.)
(p.w.)

Argenti— 1988 I 20.2 1.94 -0.03 -4.9 1959/95 -2.9 -0.2
na
Bolivia 1970 11 15.9 0.6 -0.5 -2.9 1965/95 -14.6 -l7.6
Brazil 1980 1 32.7"" -2.9 -4.4 -3.6 1950/95 -18.9 13.7
Chile 1970 1 19.5 0.3 0.1 -0.6 1950/95 85.8 88.6
Colombia 1977 l 61 .9") -0.6 -0.3 0.4 1958/95 28.4 19.3
Costa 1981 1 39.6") -0.4 -1.2 -1.9 1950/95 22.1 29.2
Rica
Ecuador 1972 1 76.9") -0.3 -0.5 -1.1 1950/95 32.1 50.1
El 1979 l 73 .2") -1 .8 -3 .2 -3 .5 1958/95 -6.3 -42.4
Salvador
Guatema- 1981 1 48.0") -1.6 -3.1 -3 .9 1950/95 19.8 -9.2
la
Guyana 1963 1 26.2 2.82 3.0 0.05 1952/93 37.1 30.6
Honduras 1981 11 28.3"" -1.8 -3.1 -3.2 1958/95 -8.6 -11.7
Jamaica 1973 I 43 .3") -4.9 -6.7 -6.3 1950/94 40.7 41.7
Mexico 1981 1 40.3") -3.7 -5.4 -4.3 1957/95 43.2 116.4
Paraguay 1979 1 91.8") -1.3 -0.9 0.8 1950/95 51.6 -11.5
Peru 1987 11 22.3"'" -2.8 -4.1 -3 .0 1950/95 -25.5 -46.8
Uruguay 1979 1 50.2") -0.8 -0.7 0.] 1965/95 48.3 41.2
Venezue- 1978 III 18.5"" -2.5 -3.1 -l.6 1957/95 -2.0 -7.2
1a

 

 

 

 

 

Source: see text. (0 Period for which data on trade-output shares is available. "’, "" and "'" stand for
statistical signiﬁcance at the 1, 5, and 10 percent level, respectively, using unit root critical values. For
model 1, these are 38.35, 31.29, 27.99, respectively. For model 11, the critical values are 30.36, 25. 10, and
22.29, while for model 111 they are 22.48, 17.88, and 15.78.

81

Table 1.14: Structural Trend Breaks in GDP and Capital Stock per Worker, 1950/1995

 

 

 

 

Country Break in GDP per Worker Break in Capital Stock per Worker
Break Model Suth Change Break Model SupF; Change
Year in Year in
Growth Growth
Rate Rate
(%) (%)
Argentina 1980 I 28.8"“) -2.5 1981 II 36.5") -5.0
Bolivia 1977 I 40.9") -2.2 1979 I 38. 1"" -5.7
Brazil 1981 I 43.7") -4.0 1972 I 37.4"" -0.9
Chile 1982 II 17.5 3.0 1980 1 30.5"") 1.0
Colombia 1978 I 85.9") -0.5 1973 11 20.7 0.6

Costa Rica 1979 I 36.9"" -1 .9 1979 I 55.3") - l .7
Ecuador 1973 I 62.1") -1 .9 1972 I 44.8") - l .1
El Salvador 1979 I 55.4") -3 .2 1987 I 29.0"." -2.5

Guatemala 1980 I 90.0") -3 .2 1974 I 50.6") -1 .9

Guyana 1982 m 16.2"") 1.1 1979 I 28.3"”) -1.9
Honduras 1982 1 51.3") -2.5 1971 1 35.6"" -2.0
Jamaica 1970 1 37.8"" -6.2 1971 11 28.3"" -6.2
Mexico 1982 1 46.6") -5.3 1981 1 49.9“) -4.3
Paraguay 1980 I 53.4") -1.7 1980 I 57.9") -0.2
Peru 1988 11 24.3"”) -2.9 1971 I 31.5‘”) -2.7
Uruguay 1982 1 55.6") -o.5 1979 I 30.3"") 0.1

Venezuela 1980 III 25.1") -2.4 1975 II 24.6"") -0.3

 

 

 

 

Source: see text. ‘", "" and "'" stand for statistical signiﬁcance at the 1, 5, and 10 percent level,
respectively, using unit root critical values. For model 1, these are 38.35, 31.29, 27 .99, respectively. For
model 11, the critical values are 30.36, 25.10, and 22.29, while for model 111 they are 22.48, 17.88, and
15.78.

82

Table 1.15: GMM Estimates of Growth in GDP per Worker
(p-values in parenthesis)

 

 

 

 

(1) (2) (3) (4) (5) (6)
Regression Dif.-Lev. Dif.-Lev. Dif.-Lev. Dif.-Lev. Dif.-Lev. Dif.-Lev.
Speciﬁcation
Instruments (" Lev.-Dif. Lev.-Dif. Lev.-Dif. Lev.-Dif. Lev.-Dif. Lev.-Dif.
Growth in Capital 0.887 0.850 0.464 0.384 0.665 0.329
per Worker (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
Initial GDP per -0.019 -0.055 -0.025 -0.036 -0.026 -0.027
Worker (log) (0.01) (0.00) (0.00) (0.00) (0.00) (0.03)
Total Population 0.009 0.018 0.01 1 0.016 0.013 0.009
(log) (0.02) (0.00) (0.01) (0.01) (0.00) (0.06)
Changes in Terms -0.021 -0.024 -0.028 -0.005 -0.076 -0.032
of Trade (0.53) (0.43) (0.55) (0.86) (0.04) (0.38)
Reforms Dummy 0.017 -0.003
(0.00) (0.55)
Trade-GDP Ratio 0.035 0.002 0.015
(log) (0.01) (0.88) (0.32)
Export Growth 0.053 0.073 0.068
(0.12) (0.00) (0.05)
Import Growth 0.092 0.079 0.075
(0.00) (0.00) (0.01)
Black Market -0.044 -0.027
Premium (log of (0.00) (0.02)
one plus the
premium)
Wald Test of Joint 0.000 0.000 0.000 0.000 0.000 0.000
Signiﬁcance: p-
value
Sargan Test of 1.000 1.000 1.000 1.000 1.000 1.000
Overidentifying
Restrictions: p—
value
Test for First-Order 0.068 0.096 0.090 0.109 0.026 0.060
Serial Correlation:
p-value
Test for Second- 0.577 0.614 0.880 0.839 0.650 0.421
Order Serial
Correlation: p-value
Number of 92 92 92 92 92 92
Observations

 

Notes: (*) See text.

83

Table 1.16: GMM Estimates of Growth in GDP per Worker
Including Control Variables
(p-values in parenthesis)

 

 

 

 

 

(1) (2) (3) (4) (5)
Regression Speciﬁcation Dif.-Lev. Dif.-Lev. Dif.-Lev. Dif.-Lev. Dif.-Lev.
Instruments ") Lev.-Dif. Lev.-Dif. Lev.-Dif. Lev.-Dif. Lev.-Dif.
Growth in Capital per 0.328 0.238 0.249 0.237 0.369
Worker (0.00) (0.08) (0.17) (0.00) (0.01)
Initial GDP per Worker -0.027 -0.033 -0.034 -0.017 -0.035
(log) (0.03) (0.03) (0.09) (0.1 1) (0.05)
Total Population (log) 0.009 0.010 0.012 0.006 0.012
(0.06) (0.09) (0.13) (0.13) (0.05)
Reforms Dummy -0.003 0.0004 -0.005 -0.004 -0.006
(0.59) (0.95) (0.29) (0.52) (0.20)
Trade-GDP Ratio (log) 0.016 0.030 0.023 0.008 0.020
(0.30) (0.11) (0.01) (0.22) (0.08)
Export Growth 0.061 0.075 0.066 0.052 0.065
(0.09) (0.07) (0.18) (0.13) (0.08)
Import Growth 0.081 0.058 0.074 0.1 1 1 0.082
(0.00) (0.07) (0.02) (0.00) (0.00)
Black Market Premium -0.026 -0.022 -0.024 -0.020 -0.024
(log of one plus the (0.03) (0.08) (0.19) (0.06) (0.12)
premium)
Educational Achievement -0.015
(0.1 1)
Government Consumption -0.002
/ GDP (0.27)
Inﬂation -0.00002
(0.39)
M2 / GDP 0.0002
(0.40)
Wald Test of Joint 0.000 0.000 0.000 0.000 0.000
Signiﬁcance: p-value
Sargan Test of 1.000 1.000 1.000 1.000 1.000
Overidentifying
Restrictions: p-value
Test for First-Order Serial 0.082 0.026 0.033 0.110 0.128
Correlation:
p-value
Test for Second-Order 0.41 1 0.543 0.505 0.384 0.302
Serial Correlation: p-value
Number of Observations 92 92 92 92 92

 

Notes: (*) See text.

84

Table 1.17: GMM Estimates of Grth in Capital per Worker

(p—values in parenthesis)

 

 

 

 

(1) (2) (3) (4) (5) (6)
Regression Dif.-Lev. Dif.-Lev. Dif.-Lev. Dif.-Lev. Dif.-Lev. Dif.-Lev.
Speciﬁcation
Instruments ‘" Lev.-Dif. Lev.-Dif. Lev.-Dif. Lev.-Dif. Lev.-Dif. Lev.-Dif.
Initial Capital per -0.014 -0.034 -0.012 -0.024 -0.026 -0.027
Worker (log) (0.00) (0.00) (0.06) (0.00) (0.00) (0.03)
Total Population 0.009 0.015 0.007 0.013 0.015 0.009
(log) (0.00) (0.00) (0.05) (0.01) (0.00) (0.05)
Changes in Terms 0.083 0.083 0.075 0.047 0.059 0.024
of Trade (0.00) (0.00) (0.00) (0.22) (0.03) (0.62)
Reforms Dummy 0.001 -0.012
(0.46) (0.09)
Trade-GDP Ratio 0.019 0.001 0.026
(log) (0.03) (0.94) (0. 13)
Export Growth 0.126 0.169 0.150
(0.02) (0.00) (0.01)
Import Growth -0.019 -0.051 -0.037
(0.57) (0.06) (0.39)
Black Market -0.020 -0.002
Premium (log of (0.00) (0.91)
one plus the
premium)
Wald Test of Joint 0.000 0.000 0.000 0.000 0.000 0.000
Signiﬁcance: p-
value
Sargan Test of 0.997 0.999 1.000 1.000 0.998 1.000
Overidentifying
Restrictions: p-
value
Test for First-Order 0.767 0.368 0.01 1 0.002 0.753 0.003
Serial Correlation:
p-value
Test for Second- 0.023 0.053 0.740 0.769 0.051 0.691
Order Serial
Correlation: p-value
Number of 92 92 92 92 92 92
Observations

 

Notes: (‘) See text.

85

Table 1.18: GMM Estimates of Growth in Capital per Worker
Including Control Variables
(p-values in parenthesis)

 

 

 

 

 

(1) (2) (3) (4) (5)
Regression Speciﬁcation Dif.-Lev. Dif.-Lev. Dif.-Lev. Dif.-Lev. Dif.-Lev.
Instruments ") Lev.-Dif. Lev.-Dif. Lev.-Dif. Lev.-Dif. Lev.-Dif.
Initial Capital per Worker -0.028 -0.022 -0.020 -0.017 -0.021
(log) (0.00) (0.07) (0.09) (0.01) (0.07)
Total Population (log) 0.008 0.005 0.007 0.008 0.007
(0.09) (0.46) (0.10) (0.00) (0.16)
Reforms Dummy -0.010 -0.002 -0.014 -0.010 -0.016
(0.14) (0.80) (0.10) (0.00) (0.06)
Trade-GDP Ratio (log) 0.034 0.039 0.024 0.007 0.023
(0.02) (0.13) (0.04) (0.05) (0.04)
Export Growth 0.158 0.129 0.147 0.149 0.087
(0.02) (0.07) (0.04) (0.00) (0.08)
Import Growth -0.046 -0.046 -0.035 -0.030 0.005
(0.34) (0.37) (0.49) (0.29) (0.88)
Black Market Premium -0.001 0.008 -0.014 -0.001 -0.019
(log of one plus the (0.92) (0.71) (0.55) (0.91) (0.27)
premium)
Educational Achievement -0.020
(0.00)
Government Consumption -0.001
/ GDP (0.34)
Inﬂation -0.00005
(0.00)
M2 / GDP 0.00006
(0.90)
Wald Test of Joint 0.000 0.000 0.000 0.000 0.000
Signiﬁcance: p-value
Sargan Test of 1.000 1.000 1.000 1.000 1.000
Overidentifying
Restrictions: p-value
Test for First-Order Serial 0.005 0.036 0.001 0.000 0.019
Correlation:
p-value
Test for Second-Order 0.624 0.420 0.773 0.248 0.531
Serial Correlation: p-value
Number of Observations 92 92 92 92 92

 

Notes: (‘) See text.

86

4“"

v.

.Mu

 

Years

Figure 1.1: Open Countries as Percent of Total, 1950/1995

87

APPENDIX B

UPDATING THE NEHRU AND DARESHWAR (1993) DATA BASE

The data base constructed by Nehru and Dareshwar (1993) for the period 1950-
1990 was updated until 1995 for the 18 Latin American countries considered in this
chapter. To this end, we used information from the World Bank’s “World Development

Indicators Data Set” (WDI). A brief description of the procedures used in this updating

are described in this appendix.
1) Capital Stocks
This series was calculated using the perpetual inventory method, which is based
on the following accumulation equation:
K,= (1-d)‘K(0)+ S(1-d)i 1,i (1)
where K, is the capital stock at time t (in 1987 prices), K(0) is the initial capital stock (in
period 0), In is the Gross Domestic Fixed Investment in period t-i, and d is the
depreciation rate. Nehru and Dareshwar (1993) estimate K(0) by a modiﬁcation of a
tec=hnique proposed by Harberger(l978). The procedure is based on the assumption that
in steady state the rate of grth of output (g) is equal to the rate of growth of capital
Stock. By re-arranging (1), this rate can be written:
(Kt-Kim... =-d WK.» (2)
which, by the above assumption, implies

I<~ = I/(g+d) <3)

88

Thus, in period 0, the capital stock can be calculated as:
K(0) = I,/(g+d)
Nehru and Dareshwar (1993) calculate 11 as the ﬁtted value of a log-linear trend of It,
adjusting for trend-breaks when appropriate. The depreciation rate is assumed to be 4
percent, and g is derived from the series of real GDP at market prices. Equation (1) is
then applied to calculate the rest of the values of K,. To continue this procedure for the
post-1989 values, we used data on Gross Domestic Fixed Investment (GDFI), available
for all countries except for Argentina. For this country, only the data on Gross Domestic
Investment (GDI) was available. To solve this problem, we regressed the log of the ratio
of GDFI to GDI on a linear and a quadratic trend, using the years for which both
variables were available (from 1970 to 1990). The estimates were then used to extrapolate
the ﬁgures of GDFI from 1990 through 1995.
2) Gross Domestic Product
While comparing the WDI data for this series with the data from Nehru and
Dareshwar (1993), we found considerable discrepancies in the levels but not in grth
rates of the series. Thus, we performed the updating by multiplying the 1988 levels from
the original source by the subsequent years’ rates of growth, as derived from the WDI
data base.
3) Labor Force
Nehru and Dareshwar (1993) use the population aged 15-64 years as a proxy for
the labor force. Their data covers the period from 1960 to 1988. We updated this series
With WDI data for the period 1989-1995. For the period 1950-1959, however, we only

had access to data on total population. Thus, we regressed the log of the ratio of the

89

 

WC

011

1hr

working-age population to the total population on a linear and a quadratic trend, with data
on the period 1960-1995. We then used the ﬁtted values of this regression to extrapolate

the ﬁgures for the working-age population from 1950 through 1959.

90

Chapter 2:
OPENNESS TO TRADE AND PRODUCTIVITY GROWTH IN

LATIN AMERICAN INDUSTRIES: 1970-1994

1 - Introduction

Most of the industrial base in Latin America was developed in a context of heavy
protectionism and strong government interventionism. It was generally thought that this
was the best - and indeed the only — way of achieving rapid output and productivity
growth. This situation has changed drastically in the last decade or so. Most governments
have engaged in ambitious programs of market-oriented reforms in which trade
liberalization features prominently. Interestingly, it is now a widely accepted belief
among the region’s policy makers that the increased exposure of Latin American
industries to international trade has the potential of promoting rapid productivity growth.
This dramatic change in policy stance stems from many sources, among which can be
cited the collapse of the statist economies of Eastern Europe, the rapid export-oriented
growth of the East Asian economies, the poor economic performance of most Latin
American countries during the I980s, and the increased inﬂuence of the multilateral
ﬁnancial institutions during this period.

What have been the results of the new policies? So far, the answers to this
question have been mixed. Mexico, for example, has experienced relatively low
productivity growth in the early years after the “opening” of its economy'. The Mexican

currency crisis of December 1994, partially attributed to this poor performance, has been

91

a reminder of the difficulties and the lack of automaticity associated with the potential
grth effects of the trade reformsz. Considering the region as a whole, however, Easterly
et a1. (1997) have recently argued that its return to the historic grth rate of 2 percent
per capita in 1991-93, in the midst of a global slowdown, cannot be interpreted as a
disappointment’. Also at the aggregate level, we showed, in the ﬁrst chapter of this
dissertation, that during the period from 1950 to 1995 the Latin American economies
experienced, on average, relatively faster productivity growth during the periods in which
they were open to international trade. In a country by country growth accounting analysis,
however, we found that in a few important cases — e.g. Brazil and Mexico - this result did
not apply as total factor productivity was found to be relatively slower during periods of
openness.

From a theoretical point of view, it is not surprising that different countries have
responded in different ways to the opening of their economies to international trade. Even
the “new” models of endogenous growth, which have greatly contributed to the
understanding of the trade and growth relationship, do not provide unambiguous and
general conclusions on the growth effects of opening-up: trade “unleashes powerful
forces that speed up growth” but also “forces that are harmful to growth” (Helpman,
1992: 265). Indeed, the net grth effects from increased integration into the world

trading system can be shown to depend on the countries’ initial stocks of technological

 

‘ See Edwards (1995: 131) and the references therein.

2 See, for example, Burki and Edwards (1997).

3 The authors show that “Latin American growth has responded to changes in policy
variables as would have been predicted by the experience of other times and places, as

sumarized by a panel regression spanning a large number of countries and (mostly) 5-year
periods from 1960 to 1993” (Easterly et al. 1997: 304).

92

capabilities, and on the type of changes that trade causes in the countries’ sectoral
composition of output".

It is thus possible that the empirical studies that have focused on the effects of the
trade reforms on aggregate productivity growth have missed the actual responses to the
new policies at the sectoral level. Indeed, as stressed by the above-mentioned models of
trade and endogenous growth, it may well happen that in the high-technology sectors of
the economy the productivity gains provided by the increased access to the global stock
of blueprints and intermediate inputs that are associated with freer trade, are insufﬁcient
to compensate for the competitive disadvantages arising from the lack of a local critical
mass of knowledge. As stressed by Grossman and Helpman (1991: chapters 8 and 9) and
Feenstra (1996), this situation may arise when countries differ in size and/or prior
research experience, and there are lags in the international diffusion of technology so that,
even under free trade, countries do not have access to the same stock of knowledge. In
these circumstances, local ﬁrms in high-technology sectors might be driven out of the
market by their foreign competitors, and the net contribution of these industries to
aggregate productivity growth might suffer a reduction after the opening of the economy
to international trade.

In addition, even if one assumes that technological spillovers are instantaneous
and global in reach, trade is expected to cause an increase in the share of the sectors in
which the country has a (static) comparative advantage, and these sectors might well

have, for technological reasons, an intrinsically lower rate of productivity growth. This is

 

‘ See Grossman and Helpman (1991) and the review in the ﬁrst chapter of this
dissertation.

93

likely to occur when the countries in question are unskilled-labor abundant, as is
generally the case in Latin America. Summing-up, even if one assumes that international
trade eases the international diffusion of knowledge and thus favors faster productivity
growth in all sectors of the economy, it can also be expected to cause changes in the
composition of the economy’s output that do not necessarily favor an acceleration of
aggregate productivity growth’.

The objective of this chapter is to contribute to the research on the impact of the
“opening” of Latin America on its productivity performance, by focusing on the
developments observed at the industry level. This type of analysis has two different
motivations. The ﬁrst one is of a theoretical nature and is given by the fact that the new
models of trade and endogenous growth fail to predict an unequivocal effect of increased
openness on productivity growth at the aggregate level but do, under certain assumptions,
make testable predictions at the sectoral level. These predictions can be summarized by
saying that, in the context of global technology spillovers and relative abundance of
unskilled labor — as in Latin America - trade will lead to faster growth in productivity in
all sectors, even if it may cause the shrinking of the high-technology sectors and the

expansion of the traditional (labor-intensive) ones“.

 

’ Quoting Grossman and Helpman (1991): “the rate of innovation cannot fall in the
country that is relatively poorly endowed with human capital (. . .) Still, the labor-rich
country can experience a decline in its growth rate of real output, since its resources shiﬂ,
in accordance with comparative advantage from the production of high-technology goods
to the production of traditional goods. That is, the country specializes in the stagnant
manufacturing activity, even as it enjoys faster technological progress in its dynamic
manufacturing sector” (p. 255).

6 As stressed above, when research spillovers are geographically concentrated and the
home country is smaller in size or has an initial disadvantage in research, productivity
growth in the high-technology sectors may be subject, in addition to the general positive

94

Our second, more practical, motivation is based on the recognition that all
measures of openness are partial and arbitrary, but the aggregate ones are even more so.
In Latin America, in particular, it is a reality that the increased openness to international
trade that has been observed in the last few years has not always been uniform, and has
affected some sectors more than others. Thus, even if one believes that at the sectoral
level trade has positive effects on productivity growth, “faster productivity will be
observed in those sectors where protectionism has been reduced and will not be observed
in those still subject to trade barriers or other forms of regulation” (Edwards, 1995: 131).

More precisely, we seek, in this paper, to perform an econometric estimation of
the effect of several indicators of openness to international trade on the rates of labor
productivity and total factor productivity (TFP) grth of Latin American manufacturing
industries. To this end, we regress the rate of growth of labor productivity on indicators
of trade intensity, intra-industry trade, export growth, grth in import-output ratios, and
tariff and non-tariff barriers to trade. By controlling for the accumulation of factors of
PTOduction, we also test for the effect of these openness variables on the rate of growth of
TFP.

The regressions are run separately for each one of the countries considered. This
is justiﬁed by the fact that the relationship between openness and productivity growth is
expected to depend on the countries’ industrial structures and their historical patterns of
dCVelopment, as affected by government policies. We concentrate on the manufacturing

industries both for data availability and for theoretical reasons: most research and

\

“effects that trade exerts on the traditional sectors, to the speciﬁc negative effects of the
ll1creased competition faced by the local ﬁrms in the innovation races in which they get

95

development and international trade takes place in manufacturing, so that it appears to be
the part of the economy to which the new models of endogenous growth are best suited.
The choice of the countries on which we focus, as well as the periods and industries
covered, are dictated exclusively by data quality and availability. In the case of tariffs and
non-tariff barriers, data are available for 17 industries in three countries (Argentina,
Brazil and Mexico) during the period 1986/1993, which covers most of the recent
episodes of trade liberalization. For the rest of the openness indicators data are available
at a level of disaggregation of 28 industries, in ﬁve countries — the three already
mentioned plus Chile and Colombia — during the period 1970/1994. In order to purge
from the data the effects of the short-run ﬂuctuations in output, we construct panels in
which the data are averaged over, respectively, 3- and 5-year subperiods. The
econometric estimation is performed using the Generalized Method of Moments. This
technique allows us to control for the likely endogeneity of the indicators of openness
When analyzed as determinants of growth performance. In particular, it allows us to
Provide a possible solution to the problem of determining the direction of causation when
a relationship is found to exist between the trade and growth variables.

The paper is organized as follows. The next section provides some historical
background on the evolution of the main economic policies implemented in Latin
Arnerica from the period of import substitution to the recent trade liberalization. Section
3 Stunmarizes the economic performance of the manufacturing industries in the countries
Considered in our empirical exercise, the speciﬁcation of which is described in section 4.

\

involved. Grossman and Helpman (1991: 257).

96

Section 5 introduces the econometric procedure that we adopt. Section 6 describes the

estimation results, and section 7 offers concluding remarks.

2 - From Import Substitution to Trade Liberalization: Historical Background

During the past decade most Latin American countries have undergone a series of
market-oriented reforms, including not only the liberalization of the region’s trade
regimes, but also a heightened ﬁscal and monetary discipline, the privatization of state-
owned enterprises, and the deregulation of ﬁnancial markets. The changes in the ﬁeld of
trade policy have included a sharp reduction in the average rates of import tariffs, as well
as in their level of dispersion. In addition, the coverage of the non-tariff barriers to trade
has fallen abruptly, in many cases to the point of their complete elimination. These policy
changes have been reﬂected in signiﬁcant increases in the openness to trade of most Latin
American countries, as measured by the ratio of total trade to GDP7. In addition to these
unilateral measures, several regional integration arrangements — MERCOSUR and
NAFTA among others — have been launched or revived, further promoting outward
orientation in the region. Finally, since 1986 fourteen Latin American countries have
acceded to the GATT/ WTO multilateral trade system, a movement that reﬂects an
attﬁmpt by these countries to link the unilateral liberalization of their trade regimes to the

ne’gOtiation of reciprocal offers on the part of industrial countries”.

\
7 Indicators of trade regimes before and after the reforms can be found in the ﬁrst chapter
3f tI'lis dissertation — Table 1.1 — and Burki and Perry (1997) — chapter 11.

It IS worth noting that only eight countries from the region had participated in the
previous seven negotiation rounds since 1947 — Primo Braga et a1. (1997: 100).

97

The extent of these changes is even more surprising when one considers the
historical background against which they have occurred. Indeed, only a decade ago Latin
America’s external sector was considered to be “the most distorted in the world’”.
Furthermore, Latin American governments had been, for several decades, among the
strongest defenders of inward-oriented strategies of development. Since the 1930s and
especially after World War 11, most countries in the region had relied substantially on a
policy of import substitution as a route to industrialization. This approach, which implied
high levels of protectionism, was initially implemented as a short and medium-term
response to external shocks, but was progressively transformed into a long-term
economic philosophy based mostly on infant industry argrunents. These suggested that
the sheltering of the domestic market from foreign competition would lead to the creation
of a local industrial base and spur productivity growth through the mastering of state-of-
the-art technologies and the development of indigenous technological capabilities.

The dominance of this approach lasted, in most countries, at least until the mid
19805. However, as noted by Fishlow (1991), the exclusive reliance on the import
substitution strategy “reached its peak in the 19503 when import ratios were sharply
lowered and trade policies were consciously biased against exports” (p. 157). Starting in
the late 19603, in the midst of recurring internal and external disequilibrium, an

increasing number of countries — notably Brazil, Chile, Colombia and Mexico — adopted

_

9 Edwards (1995: 115). As this author has noted, however, the very high levels of
protection found during the mid-19805 are in part the result of measures taken as a
response to the interruption of capital inﬂows to the region that followed the 1982 “debt
crisis”.

98

policy measures aimed at reducing the anti-export bias and promoting non-traditional
exports, while maintaining the domestic market protection - Jaspersen (1997: 62).

The new, “mixed”, policy approach led to a considerable growth and
diversiﬁcation of the region’s industrial output and foreign sales during the 19708. It was
not enough, however, to deal with the effects of the adverse external shocks of the period
without resorting heavily to external borrowing. Moreover, the reluctance of most
governments to adopt adjustment and stabilization policies after the ﬁrst oil shock
reinforced the dependence of the region’s economies on the availability of funds in the
international ﬁnancial markets and contributed considerably to their external
vulnerability, especially after the new terms of trade and interest rate shocks of 1979/80.
When the Mexican government announced, in August 1982, that it could no longer meet
its international ﬁnancial obligations, the ﬂow of commercial bank loans to the region
was drastically reduced, leading most governments in the region to implement severe
contractionary policies.

Between 1981 and 1984, the Latin American Countries went from a deﬁcit of $2
billion in their trade balance to a surplus of $39 billion'°. This turnaround in the transfer
of resources to the rest of the world was achieved through sharp reductions in imports and
even greater increases in exports (minus 24.7 percent and 33.5 percent, respectively,
between 1981 and 1987 for the region as a whole)". To produce these changes, most
Countries heightened their trade restrictions and promoted considerable devaluations of

their real exchange rates. At the same time, most governments implemented drastic

\

:° Edwards (1995: 23).
’ Idem, p.25.

99

reductions in public expenditures — especially public investments and wages —— and sharp
increases in interest rates, with the natural effects on the contraction of private
investments. Mostly as a consequence of these policies, the majority of the countries in
the region experienced negative rates of growth in per capita gross domestic product
during the 1980s, which explains the usual reference to this period as the “lost decade”
for growth in Latin America”.

In the midst of the crisis, inﬂation rates soared reaching three and four digit
ﬁgures in a few cases. In some of these countries -— Argentina, Brazil and Peru - non-
orthodox stabilization programs were implemented starting in 1985/86, using price and
exchange rates freezes, at the expense of the emphasis on ﬁscal discipline, to combat the
inertial component of inﬂation. Despite short-lived periods of success, these programs
ended up causing an acceleration of the previous rates of inﬂation. Their failure, however,
exerted an important inﬂuence on the thinking of policy-makers in the region. Indeed, it
reinforced the generalized sense of frustration with “the once-dominant view based on
heavy state interventionism, inward orientation, and disregard for macroeconomic
balance”, thus contributing to the acceptance of “a new paradigm based on competition,
market orientation, and openness” — Edwards (1995: 41). The emergence of this new
policy consensus was also associated with the growing inﬂuence of the multilateral
ﬁnancial institutions - mainly the IMF and the World Bank — whose share in the ﬂow of

foreign funds to the region increased substantially during the 19808, and who routinely

 

'2 The only countries that experienced positives rates of real per capita GDP growth in the
period 1981-1990 are the Dominican Republic, Chile, Colombia, and Jamaica (ECLAC,
1996: 17). The annual GDP growth rate for the region as a whole was 1.2 percent in the
period 1980-90 - it had been 5.6 percent in the previous decade and was 3.4 percent in

100

included the implementation of market-oriented “structural reforms” as an integral part of

the conditionality of their lending programs.

3 - Growth and Structural Change in Industry: 1970-94

During the past two decades Latin America’s manufacturing sector has grown, on
average, at lower rates than total output. In the countries on which this chapter focuses,
this type of performance was most clearly seen in Argentina and Brazil, where the share
of manufacturing in GDP fell from around 35 percent in the early 19708 to levels close to
30 percent in the early 19908 (Figure 2.1).

The relatively slow pace of output growth of the Latin American manufacturing
industries in the past two and a half decades does not reﬂect, however, their overall
economic performance. Indeed, the region did experience signiﬁcant changes in the
pattern of international specialization of its economy and in the structure of its
manufacturing sector. In terms of exports, in particular, the countries here considered
underwent, with the only exception of Chile, a considerable reduction in the share of
primary products in their foreign sales, with corresponding increases in the shares of
industrial exports (Table 2.1)”. In Brazil, for example, the share of primary products in

total exports fell from 67 percent in 1970 to 20 percent in 1994.

 

the period 1990-94 (ECLAC, 1995: 65).

'3 The deﬁnition adopted for “primary products” and “industrial products” is the same
adopted by ECLAC (1992), which in turn is based upon the classiﬁcation proposed by the
United Nations Industrial Development Organization in UNIDO (1983). This concept of
industrial products is a broad one, and includes all primary products that undergo some
type of transformation process. Thus, industrial products can be divided in two groups:
semi-manufactures, or industrial products based upon natural resources (included in
Sections 0 to 4 of the SITC), and manufactures in a strict sense, or industrial products not

101

Within manufacturing, important changes also took place, both in terms of the
structure of output and especially in terms of the structure of foreign trade. To appraise
these changes, we classify manufactures into ﬁve groups, using factor intensity as the
deﬁning criteria (Table 2.A. I). We thus distinguish, ﬁrst, between the goods that are
intensive in natural resources, called “semi-manufactures”, and those that are not. Among
the latter group, we separate those that have a high input of skilled labor, which we call
“new industries”, from those that have not, and then make a further distinction within
each group based on whether the industries are labor or capital intensive".

One of the main developments, since the 19708, has been the relative expansion of
the so-called “semi-manufactures” and of the “basic input industries” (Table 2.3) —
corresponding, respectively, to industries intensive in natural ressources, and capital-
intensive industries with a low input of skilled labor. This can be associated with an
increase, mainly since the late 19708, in the relative importance in the region’s economy
of the sectors that produce industrial commodities - such as, for example, aluminium,
iron and steel, petrochemical products, pulp and paper, vegetable oils, etc. As shown by
Benavente et al. (I 996: 60), the expansion of these sectors has been promoted, since the
late 19708, through the use of ﬁscal incentives for the establishment of new plants and the
modernization of the existing ones. Thus, during the past two decades these industries

‘

(directly) based on natural resources (included 1n sections 5 to 8 of the SITC)

This classiﬁcation was proposed by ECLAC (1992,1996) and 18 mostly based on the
criteria suggested by UNIDO (1983). The industries are classiﬁed as intensive in skilled
labor on the basis of the fraction of scientists, engineers, and highly specialized workers,
in the total workforce of the corresponding industries 1n the United States. A listing of the
industries included 1n each of the ﬁve groups, by their ISIC codes, is provided 1n the
nOtes to Table 2. A. l.

102

received “a new infusion of capital and were equipped with a new generation of highly
modern production plants and up-to-date technology” (ECLAC, 1996: 78).

During the 19808, a decline in the share in total value added of the labor intensive
industries with a low input of skilled labor (“other traditional industries”), particularly of
the textile industries, was also observed. This trend has persisted in the ﬁrst half of the
l 9908, which has also witnessed an increase in the share of capital intensive “new
industries” (industries with a high input of skilled labor), associated mainly with the
recent revitalization of the automotive industry in four of the ﬁve countries considered —
the exception being Chile, where this industry is very small.

Considerable changes have also taken place in the structure of foreign trade,
although it is difficult to establish general trends at the industry level (Table 2.2). Most of
the changes, however, can be better understood by analyzing the industries’ export and
import shares in output (Table 2.3). Increases in export shares have been observed in
almost all sectors and countries. In the case of exports, they have been particularly large
in the “other traditional industries”. As previously mentioned, growth in value added was
generally slower in these sectors than in the manufacturing industry as a whole. However,
there was a very strong reorientation of output towards export demand, which led to an
increase in the sectors’ share in total exports. Also worthy of comment are: the fall in the
Share of “semi-manufactures” in total exports, which occurred in the context of higher
than average rates of output growth but relatively stable export-output ratios; and the
increasing share of the “basic inputs industries” in total exports, which was usually
associated to the conjunction of above-average output growth and increases in export-

Output ratios.

103

With respect to the aggregate export performance of the manufacturing industry,
Mexico appears as the most successful case, with an average export-output share that
went from 3.9 percent in the 19708 to 21.3 percent in the 19908. Brazil and Colombia also
showed a very good performance in terms of total industrial exports, basically doubling
their export-output ratios to levels close to 14 percent during the 19908. Argentina, on the
other hand, experienced only small increases in its manufacturing export-output ratio,
reaching 8.6 percent, on average, during the 19808, and falling back to 7.9 percent in the
early years of the present decade. Finally, Chile showed a downward trend in its
manufacturing export-output ratio, but the country still maintains the highest levels for
this index (35.6 percent during the 19908), which is explained by the very strong export
orientation of its non-ferrous metals industry - mainly cooper.

With regard to the evolution of manufacturing imports, a comparison of the
import—output ratios of the 19708 with their average levels during the 19908 shows an
upward trend in all ﬁve countries — even in Argentina and Brazil, which had reduced their
import shares during the previous decade (Table 2.3). Increases in imports were
particularly large — both relative to output and to other sectors’ imports — in the “other
traditional industries” and, with the exception of Brazil, in the “new industries” (Tables
2.2 and 2.3). Overall, Argentina and Brazil remain the most closed of our sample with
inrport-output ratios close to 10 percent in the 19908 — the corresponding indexes for
Chile, Colombia and Mexico being, respectively, 44.1 percent, 30.8 percent and 34.1
Percent. In terms of trade balances in manufacturing, Brazil is the only country to have

shown a surplus since the 19808.

104

With regard to the relative levels of labor productivity, Argentina has maintained,
at least since the early 19708, a considerable advantage over the other four countries
(Figure 2.2). Although during the 19708 all ﬁve countries experienced relatively similar
average rates of growth in value added per worker (Table 2.4), their reactions to the
external shocks of the late 19708, and especially the “debt crisis” of the 19808, were very
different. Indeed, Colombia and Mexico managed to maintain a steady trend of growth in
labor productivity — with annual rates of growth above 3 percent — while the other three
countries showed almost no growth in value added per worker in the frrst half of the
19808. After 1985, however, Brazil and especially Argentina began showing signs of
recovery and, in the early 19908, they attained annual rates of labor productivity growth
0f almost 10 percent. Chile only recovered after 1990, growing at a rate of 2.8 percent
during 1990/94.

With regard to the productivity performance of particular industries, it is clear that
Argentina’s lead in labor productivity is greatest in “semi-manufactures” and in the labor-
int8nsive “new” industries (Figures 2.3 to 2.7). Chile, on the other hand, has the lowest
levels of labor productivity in all manufacturing sectors except for the basic inputs
industries. The other three countries share similar levels of productivity in most
industries, except for the “other traditional industries” (labor intensive industries with low
input of skilled labor), in which Mexico has performed particularly well.

The evidence presented above suggests that both the composition of industrial
Value added and the rates of growth of labor productivity at the industry level have
changed considerably during the past decades. To obtain insight into the relative

importance of these two phenomena in the evolution of the rates of productivity growth

105

in manufacturing, we have performed a decomposition of these growth rates using a
methodology originally proposed by Nordhaus (1972).

This author emphasized the fact that “the aggregate growth of productivity is not
merely an average with constant weights of the sectoral productivity growth rates”".
Indeed, aggregate productivity growth can also be affected by shifts in the composition of
output that beneﬁt sectors with relatively high (or low) productivity growth rates, as well
as by shifts that beneﬁt sectors with relatively high (or low) productivity levels. More
formally, a (discrete time) version of the decomposition proposed by Nordhaus (1972)
can be described in the following terms. Let Ait denote the level of productivity (value
added per worker) of sector 1 at time t, Xit denote value added of sector i at time t, Eit
denote employment in sector i at time t, and let the subscript i = 0 refer to the aggregate
manufacturing sector. Then, aggregate productivity can be written:

A0,=X°,/Eo,=2(X,,/Ei,) (Ei,/E(,,)=21A,.,Sit
Where Sig-"(Eit / ED) is the share of sector i in aggregate employment, and the summations
are over i. The rate of change (denoted by a lower case letter) of aggregate productivity
can then be written:

a0' = (A0, - Am-1)/Am-1=(Z Ait Sit - Z Ait-l Sit-1) /A°,-1.

After some simple algebra, and denoting the initial period by a subscript t=0, the above

BXpression can be rewritten:

\

‘5 Nordhaus (1972: 519).

106

a0, = 2 ait (Xi0 / X00) + 2 ait [(Xit-l /Xo,-l) -(Xi0 / X00)]
(1) (11)
+ 2‘. (Sit - Sit-1) [(Ai0 / A00) - 1] + 2 (Sit - S,,-l) [(Ai,-l / Ao,-1)-(Aio/ A00)]
(111) (IV)
+ 2 ai,(A,-,-1 /Ao,-l) (Sit - Sit-1)
(V)
Folowing Nordhau8’(l972) terminology, we denote the ﬁrst two terms as rate terms, and
the third and fourth terms as level terms. The rate terms represent the rate of aggregate
productivity growth that would be observed if the levels of productivity were identical in
all industries. This rate effect is in turn decomposed in two parts. The ﬁrst term, denoted
the ﬁxed-weight rate term, represents the rate of productivity growth that would have
prevailed if the relative output shares had stayed constant. The second term, denoted the
Change in ﬁxed weight term, represents the effect of relaxing the previous assumption,
thus incorporating the effect of changes in output shares (still assuming identical
PTOductivity levels). A positive value for (11) indicates that output shares have shifted
toWard industries with relatively high rates of productivity growth.
The level effect is also broken into two parts. The third term, denoted the ﬁxed
Weight level term, measures the extent to which employment has tended to shiﬁ toward or
aWay from industries with initial above-average productivity levels. The fourth term,
denoted actual weight level term, measures the extent to which the shifts in employment
0Ccurred toward or away from sectors which increased their productivity levels relative to

the average. Finally, (V) captures the interaction of the rate and level terms, and

107

represents the extent to which employment has shifted toward or away from sectors with
relatively high productivity levels and growth rates”.

Table 2.5 reports the results of the above-described decomposition of the rates of
labor productivity growth for the period 1970/94, using data from the 28 sectors included
in the International Standard Industrial Classiﬁcation (ISIC) at the 3-digit level. The ﬁrst
ﬁnding suggested by these results is that the ﬁxed-weight rate term explains most of the
growth in labor productivity. In other words, the changes in the composition of output
and employment seem to have had, on average, a relatively small importance in the
explanation of productivity growth.

There are, nonetheless, two important exceptions, namely Argentina during the
early 19908, and Chile during the 19708. During these periods, the countries involved
experienced drastic changes in their degrees of openness to international trade, and their
aggregate rates of productivity growth in manufacturing were particularly high. In both
cases, however, productivity growth would have been considerably faster — respectively
55 percent and 83 percent faster — if the changes in the composition of output and
employment had not beneﬁted sectors with relatively slow productivity grth (both
cOuntries) and / or relatively low productivity levels (particularly Argentina). These
cOmpositional effects were much less visible in Brazil, Colombia, and Mexico, where
0\ltput shares also shifted toward sectors with relatively lower rates of productivity

growth, but at a much smaller rate.

\

’6 In the exercise performed by Nordhaus (1972) for the US economy during the period
1948/71, this second-order term is not reported due to its negligible value. As in the
COuntries and periods here analyzed, the changes in the variables involved are not always
Very small, we have decided to report this ﬁfth term, whose value is sometimes quite

108

Summing-up, it can be said that although the region has witnessed a decay in the
share of manufacturing in GDP, this sector’s performance in terms of exports has been
outstanding compared with that of the rest of these econorrries. Increases in export
coefficients have been observed in almost all countries and manufacturing industries.
Manufacturing imports have also expanded considerably in recent years, after having
recovered ﬁom the effects of the contractionary policies implemented in the region
during the 19808, and with the stimulus of trade liberalization policies. In terms of labor
productivity growth, Argentina and Brazil have experienced dramatic increases in their
rates of productivity growth in manufacturing, after the opening of their trade regimes,
during the early 19908. Both countries remain, however, the most “closed” of our sample
with regard to international trade, with manufacturing imports still close to 10 percent of
output in the early 19908. Mexico and Colombia, on the other hand, have shown steady
rates of productivity growth since the 19708, while Chile has only recently recovered
from a decade of stagnation in its manufacturing productivity. Although considerable
changes have taken place in terms of the composition of manufactruing value added, a
decomposition of the rates of labor productivity for the period 1970/94 shows that, with
some exceptions (Chile during the 19708 and Argentina during the early 19908), these
changes have not played an important role in the explanation of overall productivity

growth.

 

large.

109

4 - Empirical Speciﬁcation

As mentioned in the introduction, the chief objective of this paper is to study the
inﬂuence of the exposure of Latin American industries to international trade on their
productivity performance. Thus, based on the new endogenous grth literature, we
assume that the rates of growth of both labor productivity and total factor productivity are
potentially affected by variables representative of the openness of the corresponding
industries to international trade, and test empirically for the direction and signiﬁcance of
these relationships.

Our empirical speciﬁcation is based on the assumption of a neoclassical
production function with two factors of production at the industry level. More formally,
for a given country in year t, we assume that in industry i value added yit is a function of
the level of employment lit and the real capital stock k“:
ya = fr. (lira Ru) (1)
We also assume that there are Hicks-neutral technical differences over time and across
industries so that this function can be written as the product of an index of total factor
productivity ait and a function g:
fr. (lie kg.) = ar. - g (In, ka) (2)
Totally differentiating and dividing by yit yields:

d In (y,,) = d In (at) + (5,) d In (1“) + 82 d In (k,,) (3)
where the parameters 6, and 52 are the elasticities of output with respect to labor and
capital respectively. This equation can in turn be rearranged into:

d 111(er in) = d in (an) +(51+ 52 -1) ' d 111011) + 52 d 1n (kn/1r.) (4)

110

 

This expression is nothing more than the conventional growth accounting framework
introduced by Solow. It indicates that the growth in labor productivity — deﬁned as value
added per worker — can be explained by three different factors: technological change,
growth in employment, and growth in the capital-labor ratio.

This framework provides the basis for our empirical speciﬁcation. Initially,
however, we do not distinguish whether the inﬂuence of openness on labor productivity
occurs through its effect on the growth of TFP or by an alteration of the rate of growth of

the factors of production. Thus, our ﬁrst estimating equation is:

 

iii (Yir/ lit) ' 1n (Yi(r-r)/ 1104)) = a + Bo In (Yi(r-1)/ ii(r-1)) + 2i Bi OPENjit 'i' 111 'i' 3i: (5)

 

 

where or is a constant, OPEN"it are variables representative of the industry’s exposure to
international trade, 1], is an industry-speciﬁc effect that may reﬂect factors such as the
industry’s speciﬁc rate of exogenous technological change, and 8,, is a disturbance term,
which is assumed to be serially uncorrelated.

The motivation for including the initial productivity term is twofold. First, as
argued by Jong-Wha Lee (1996: 397), initial levels of state variables may affect the rate
of growth of industry labor productivity in the same way as they affect aggregate growth
in the simple Solow-type neoclassical growth model. Thus, the initial labor productivity
term can be interpreted as a measure of how far the industries are from their steady-state
levels of productivity: the farther they are, the faster they grow in their transition to
steady-state. A second motivation for the inclusion of the level variable is that it can

affect the rate of growth of TFP through the existence of technological externalities.

lll

Indeed, it can be argued that individual ﬁrms can beneﬁt from the knowledge capital
accumulated elsewhere in the economy. In one scenario, ﬁrms located in industries with
low levels of productivity would grow faster as they absorb at a low cost the results of
technology or human capital investments undertaken in high-tech sectors. It can be
argued, however, that technology spillovers are stronger within industries than across
industries: if these within-industries spillovers prevail, lower initial productivity would
lead to slower growth, thus generating divergence instead of convergence in
productivity”. Which effect dominates is an empirical question that we attempt to
answer.

Our second estimating equation is intended to measure the effect of the openness
and the initial state variables on technological change alone. A simple procedure to do so
- suggested by equation (4) — would be to include in equation (5) the rates of change of
employment and the capital-labor ratio. The problem is that data for the latter is not
available at the industry level for the countries on which we focus. However, it is possible
to circumvent this practical difﬁculty by slightly modifying the grth accounting
ﬁamework presented above and making use of data on labor shares to which we do have
access.

The additional and relatively general assumptions that are needed are: that the

function g is translog, that producers are cost-minimizers, that they are price takers in

 

‘7 The mechanism by which technology spillovers across industries could be expected to
generate convergence or divergence in productivity would be the same that operates in
the models of trade and endogenous growth described by Grossman and Helpman (1991:
chapters 8 and 9) and Feenstra (1996), where the ﬁrms’ productivity in research and
development is positively related to the stock of knowledge capital developed in the
environment in which they are located — their country or, as in the present case, their

112

input markets, and that the production function is homogeneous (not necessarily of
degree one). Under the ﬁrst assumption, the log of value added can be written as:

In yit = ln ait + 010 + alln k, + azln 1it + 013 (In ki,)2+ 014 (In 1,02 + or, Inkit - In 1,, (6)
Under the second and third assumptions, the elasticities of output with respect to labor
and capital can be shown to be equal to the shares of the corresponding factors of
production in total value added (8'it and 8““, respectively):

s'it = (d y,,/ d I") - (ln/ y“) = or2 + 2 or4 1n lit + or, In k“ (7)
8'21“; (d W d kit) ° (W ya) = 0h + 2 0&3 In 190‘ as In 111 (8)
Substituting these expressions into equation (6) and taking the ﬁrst difference of this
equation yields:

111(er y,(._.,) = 111(ar/ ai(r-1)) +‘/2 (S'r + Sli(1-1))' ln (111/ ii(r-1)) + '/2 (Ski: + Skin-1)) ° 1n (kir/ lean)
(equation 9)".

Imposing homogeneity on the production function (6) implies the following restrictions:
20:, + 015 = 0 (10)
2014 + ors = 0 (11)

Substituting these expressions into the factor shares — equations (7) and (8) — yields:

 

3'11 = a2 ' 2 0‘3 in (kit / lit) (12)
Skit = (11+ 2 0‘3 In (kn / lit) (13)
industry.

'3 It is worth noting that equation (9) can also be obtained as a discrete-time
approximation to equation (3) under the assumptions of perfect competition. Obtained in
this fashion, equation (9) is the basis of the Divisia—Tomquist index of TFP growth used
by Solow (1957). However, under the assumption of a translog production function,
equation (9) is obtained directly and is not an approximation. Since the translog
speciﬁcation is considered to be fairly general — it can interpreted as a second-order
approximation to any production function (Greene, 1997, p. 229) — it is usually thought to

113

Substituting these equations into equation (9) yields:
111 (N y,(.-.,) = 1n (air/ ai(r-l)) + (a1 + <12) - ln (1:! li(t-1)) + [(a. + as - 1)/ as] - (Slit - Sim-1))

+ (1/ 01,) [1- '/2 (8', + s',(,,,))] - (s'i, - 8204,) (14)
Similarly, an expression for the rate of growth of labor productivity can be derived by re-
arranging this equation:

in (yir/ in) ' in (Yi(r-r)/ li(t-I)) = ln (air/ 3504)) + Y 1n (iir/ li(r-1)) + ‘l’r (Slit ‘ Slur-1))
+ea-Enra-am) (H)

where y = (011 + a2 - l), (b, = y / as, 92 = 1 / 01,, and 8,, = (1/2) (s'it + s',(,,,)). It is worth
noting that since (1 + y) is the elasticity of scale, 7 can be interpreted as a measure of the
degree to which the production ftmction differs from constant returns to scale. This

provides the basis for our second estimating equation:

 

 

In (Yir/ iir) ' ln(Yi(t-1)/ [Kt—1)) = a + Bo in (Yiro/ liro) + Xi Bj OPENjit 'i’ 7 d in ii + ¢1(Slir " Slur-1))

+¢x1-iaia-amp+m+e. no

 

All the data on output, value added, employment, wages, and trade ﬂows, comes
from the data base PADI (version 2.0), developed by the United Nations’ joint

ECLAC/UNIDO Industrial and Technological Development Unit, to whom we are most

119

gratefu . The data is available, for most countries, for the period 1970-1994, and covers

 

 

provide a justiﬁcation for the use of the Divisia-Tomquist index of TFP growth.

'9 PADI stands for “Program for the Analysis of Industrial Dynamics”. It was developed
at ECLAC, in Santiago, Chile, on the basis of information from the database on industrial
statistics of the United Nations Industrial Development Organization (UNIDO). These
statistics were checked at ECLAC for consistency with those provided directly to this

114

the 28 industries included in the International Standard Industrial Classiﬁcation (ISIC) at
the 3-digit level (see Table 2.A.2).

As for OPEN 1'", we use two different types of variables representing the degree of
openness to international trade of the industries considered, namely variables based on
actual trade ﬂows, and direct measures of trade barriers based on trade policy data”.
Because the industries and periods for which data is available on the two types of
openness variables are different, we estimate two different versions of our regression
equations: in the ﬁrst, we use the set of “outcome-based” openness variables and in the
second the policy-based variables.

Within the group of openness variables based on trade ﬂows, we consider four
variables: the share of imports plus exports in the total output of the corresponding
industry (T1), the Grubbel-Lloyd index of intra-industry trade ([17), the rate of growth of
exports (GX), and the rate of growth of the import-output ratio (GMC). Each variable
measures a different dimension of the degree of openness of a given industry, so that we
use them simultaneously in our regression equations.

The ﬁrst variable that we use (T1) is usually thought to be the simplest measure of
“trade intensity” and is generally found to be positively related to output growth in
aggregate studies (Harrison, 1996: 421). When used in cross-country comparisons, this
variable has the disadvantage of being affected not only by trade policy but also by non-

policy characteristics of the countries involved, such as their size, distance to markets and

 

organization by the governments of the countries involved.
2° The two groups of variables can be classiﬁed, respectively, in what Baldwin (1989) —

quoted in Pritchett (1996: 308) — calls outcome-based and incidence-based measures of
trade barriers.

115

resource endowments“. In the present context, however, since we compare the
performance of industries within a given country, it can be argued that this criticism is of
lesser importance.

The second variable that we consider (117) is deﬁned as one minus the share of the
absolute value of the trade deﬁcit (or surplus) in total trade: it varies from zero to one,
taking the value zero when for a given industry trade ﬂows are observed in a unique
direction (either imports or exports), and the value one when the industry’s imports are
exactly equal to its exports. This measure of intra-industry trade is expected to capture the
extent to which trade in a given industry is constituted by trade in differentiated products
—Dixit and Norman (1980), Krugman (1981), and Grossman and Helpman (1991, chapter
7). Thus, it can be argued that innovation-based growth should be more prevalent in
industries with high IIT, so that these industries should be expected to grow at faster rates
for given rates of labor growth and capital accumulation. In fact, Backus, Kehoe and
Kehoe (1992) have found that, in a cross-section of countries, growth in manufacturing
labor productivity is positively correlated with the extent of intra-industry trade, a result
that they interpret as evidence in favor of the theories of endogenous growth - at least in
the manufactming sector. It is possible, however, that in some of the countries
considered, protectionist policies have led to the insulation of the industries with a bigger

potential for technology development, so that trade occurs, at best, in a unique direction

 

2‘ This problem has led several authors to adjust trade shares for the effect of non-policy
variables. Examples are given by the “structure adjusted trade intensity” measures used,
for example, by Pritchett (1996: 312): the author uses the residuals from a regression of
trade shares on the countries population, area, transports costs, and GDP per capita.
Learner (1988), on the other hand, constructs measures of openness based on the residuals
from the estimation of a Heckscher-Ohlin-Vanek model of trade, using resources supplies

116

and [IT appears to be relatively low in those industries. In any case, caution should be
used in the interpretation of this variable since, as suggested by Learner (1988: 163), high
levels of [IT could also reﬂect higher levels of aggregation in the corresponding
industries and not necessarily a higher degree of intra-industry trade.

The third and fourth openness variables that we use (GX and GMC) are
constructed as the ﬁrst-differences of the log of, respectively, total exports and the share
of imports in total output. They are measures of the rate of change of the degree of
exposure of local industries to foreign competition, either in international markets, or on
the domestic market. Using these same variables, past studies of the trade-productivity
link at the industry level have encountered a positive relationship between productivity
growth and export growth, but have also often found — with some exceptions — that
productivity growth is negatively related to the growth of the import-output ratio”. Some
authors have interpreted these relationships in the context of the so-called “Verdoon’s
Law”, that postulates a positive association between output and productivity growth,
which is “taken to reﬂect scale economies or the embodiment of new technologies during
periods of rapid investment” (Tybout, 1992: 193). As stated by Nishimizu and Page

(1991: 253), “this argument is usually cast in terms of the beneﬁts of increased demand

 

and distance to markets data to predict trade ﬂows.

’2 Nishimizu and Robinson (1984), Nishimizu and Page (1991), Tybout (1992), and
Bonelli (1992), all ﬁnd that the growth of exports is positively correlated with TFP
growth. However, only the ﬁrst paper — which focuses on Japan, Korea, Turkey and
Yugoslavia — ﬁnds a negative relationship between import substitution (deﬁned as one
minus import penetration) and TFP growth. Import substitution is found to be positively
related to TFP growth by Tybout (1991, 1992) - who describes evidence from a study of
pooled data from 3-digit industries in Chile, Colombia, Turkey and Morocco — while
Nishimizu and Page (1991), in a study of 17 countries, ﬁnd that import penetration has a
negative effect on TFP growth in the period 1973/ 85, and is not signiﬁcant when only the

117

through export expansion, but it applies equally to rapid import substitution in large
domestic markets”. A different explanation is proposed by Harrison (1996: 424), which
attributes the asymmetry between the effects of exports and imports on productivity
growth, to “estimation problems arising from sirnultaneity bias”. Indeed, exports tend to
grow faster in sectors in which countries are experiencing higher rates of productivity
growth, while imports are likely to increase in sectors with a poor productivity
performance. Thus, a proper empirical treatment of the trade and growth link should deal
explicitly with the fact that causality between the two probably runs in both directions”.

In the group of the policy-based measures of openness, we consider two variables:
the average tariff rates for each industry (TARIFF) and the corresponding coverage of
non-tariff barriers to trade (NT B). The data comes from UNCTAD’s (1994) Directory of
Import Regimes, which provides information for the product lines included in the
Standard International Trade Classiﬁcation (SITC) at the 3-digit level. In order to make
TARIFF and NT B consistent with the variables used to measure productivity, we
considered only 17 of the industries deﬁned by the SITC. These industries, as well as
their correspondence with those deﬁned by ISIC, are described in Table 2.A.3. The
countries for which we have enough data to estimate our model are Argentina, Brazil and
Mexico, over the period 1986-1993.

The reason to include TARIFF as well as NT B is that, although both variables

usually have a high correlation, their effect on productivity may be quite different. As

 

period prior to 1973 is analyzed.

2’ As stated by Harrison (1996: 441): “Does openness cause growth? Or is it the other
way around?”. In practice, the author deals with this issue ex post (after performing her
regression analysis), using vector autoregressions, and ﬁnds that “causality between

118

stressed by Nishimizu and Page (1991: 253), “the instruments by which protection is
afforded to domestic industries can affect productivity performance quite appart from the
levels of protection granted.” Indeed, non-tariff barriers to trade (N TBs) insulate domestic
producers from the effect of changes in international prices - and thus from the effects of
international productivity trends — in a way that is not possible under tariff protection.
Still, it is important to keep in mind that NT B is only an approximate measure of the
importance of non-tariff restrictions to trade, since it only indicates the fraction of the
total number of products that are subject to NTBs, without measuring how intense the
barriers in question are. As stressed by Leamer (1988: 147), “not all non-tariff barriers
can be measured, and not all barriers are equally restrictive”. On the other hand, neither
TARIFF nor NT B are free from the endogeneity problems mentioned above for the
outcome-based openness variables. As shown by Harrison (1996: 424), “trade policy
itself may be a function of other variables, including growth”. One could think, for
example, that the sectors with slower productivity grth are exactly those that spend
more efforts in lobbying for increased levels of protection, or for a slower pace of trade
liberalization. Thus, the econometric procedure adopted must deal with the potential

endogeneity of all openness variables.

5 - Econometric Methodology
The estimation procedure that we adopt takes explicitly into consideration the
possibility of simultaneity and reverse causality between the rate of productivity growth

of a given industry and the degree of openness of this industry. Furthermore, we allow for

 

openness and growth runs in both directions” (p. 442).

119

the existence of an industry-speciﬁc effect contained in the error term, and potentially
correlated with our explanatory variables. Thus, we deal explicitly with the fact that the
openness variables are likely to be endogenous in our estimating equations. To do so, we
make use of a Generalized Method of Moments (GMM) estimator of the type proposed
by Blundell and Bond (1997).

To simplify the exposition, consider the following regression equation, which
encompasses our two estimating equations,
Yi, - Yin-” = or + [30 Yard) + [3, Xit + Tl: + a“ (17)
where Yit is labor productivity in industry i at time t, Xit represents the set of explanatory
variables included in equations (5) and (16), n, is an industry-speciﬁc effect potentially
correlated with the explanatory variables, and a, is a serially uncorrelated error“. As is
well known, due to the dynamic nature of this model the conventional ﬁxed-effects
transformation does not lead to a consistent estimator. Similarly, taking the ﬁrst
difference of equation (17) eliminates the individual effect but also creates an
endogeneity problem, as the frrst-differenced lagged dependent variable is correlated with
the ﬁrst-differenced error. A solution proposed, among others, by Anderson and Hsiao
(1981) and Arellano and Bond (1991), is to estimate the ﬁrst-differenced equation by
GM, assuming that the error terms are not serially correlated and that the explanatory
variables are at least weakly exogenous. In these circumstances, it is possible to make use
of moment conditions of the following type:

B [YKW 0 (t:it - 8,049] = O for s _>_ 2 and t Z 3 (18)

 

2" Note that this model can be rewritten with Yit (the level of labor productivity) as the
dependent variable, which is the standard speciﬁcation in the literature on dynamic panel-

120

E 1X50.» a (sit - 8,049] = O for s 2 2 and t 2 3 (19)
These conditions are implied by the assumptions about the lack of serial correlation in the
error term and the weak exogeneity of the explanatory variables Xit. The meaning of the
latter assumption is that Xit is potentially correlated with both contemporaneous and past
values of the error term, but not with ﬁrture values of this variable:

E[X,-,oai,]=0 fort<s (20)
Thus, for example, rapid productivity grth — or one of its time-varying determinants —
may be the cause of greater openness in the present and future, but not in the past.

A property of the GM estimators that are based on ﬁrst-differences is that they
only make use of the time-series variation in the data - the cross-sectional variation being
lost in the ﬁrst-differencing. In the context of the present paper, this constitutes a problem
since we attempt to estimate not only the effect of greater openness on growth that arises
from the comparison of given industries during different periods of time, but also the
evidence that can be obtained from the comparison of different industries during the same
period — the cross-sectional variation.

There are also statistical problems related with the use of the ﬁrst-differences
GMM estimator. As shown by Ahn and Schmidt (1995), this estimator does not make use
of all the moment conditions that are available under fairly general assumptions, and is
not efficient under these assumptions. Furthermore, Blundell and Bond (1997) show that,
in many practical cases, lagged levels provide weak instruments for ﬁrst differences. To
deal with these problems, Blundell and Bond (1997) suggest the use of additional linear

moment conditions. Speciﬁcally, they suggest to follow Arellano and Bover (1995) in the

 

data models.

121

use of “an extended linear GMM estimator that uses lagged dijferences of Y,, as
instruments for equations in levels, in addition to lagged levels of Y,, as instruments for
equations in ﬁrst-differences” (Blundell and Bond, 1997:l). The linear moment
conditions proposed by Blundell and Bond (1997: 11) encompass non-linear moment
conditions suggested by Ahn and Schmidt (1995), and the authors show that both
methods improve upon the simple ﬁrst-difference speciﬁcation with lagged levels as
instruments.

In this paper, we utilize the system GMM estimator proposed by Blundell and
Bond (1997: 14), making use of the moment conditions implied by equations (18) and
(19) for the equations in differences”, and of the following moment conditions for the

equations in levels:

E [r], o (X,(,,) - X,(,,2))] = 0 for t 2 3 (21)
E [8,, o (X,(,,,) - X,(,,2,)] = 0 for t 2 3 (22)
E [11i 0 (Y,(,,,) - Y,,,,2))] = 0 for t 2 3 (23)
E [8,, 0 (Y,(,_,) - Y,(,_2,)] = 0 for t 2 3 (24)

Equation (21) amounts to assuming that changes in X, are uncorrelated with the industry-
speciﬁc effect. Thus, for example, even though the level of trade-intensity or the average
tariff of a given industry may be correlated with the industry-speciﬁc rate of exogenous
technological change, we assume that the rates of change of the openness variables are

not correlated with the industry-speciﬁc effect“. Equation (22) is implied by the

 

2’ In practice, because of the limited size of our sample, we limit ourselves to the use of
the second lag of the variables X, and Y,, as instruments in the equations in differences:
we actually use only the moment conditions implied by (18) and (19) in the case s=2.

2‘ Consistently, we assume that the variables GX and GMC, which measure changes in

122

assumption of weak exogeneity of the explanatory variables X,, (equation 20). The
validity of equation (23) is shown by Blundell and Bond (1997: 12) to depend “on a
restriction on the initial conditions process generating Y,,”, namely that the deviations of
the initial level of Y,, from the convergent level of this variable, which depends on the
industry-speciﬁc effect, is uncorrelated with the industry-speciﬁc effect itself. Since we
picked the ﬁrst year of our series (1970) only because it was the ﬁrst year for which
ECLAC reported industrial productivity, and since there is nothing special about the level
of productivity in the industries of our sample during this year, we might expect the
moment conditions (23) to be valid in the present case. Finally, equation (24) is implied
by the assumption of lack of serial correlation in the error term.

Based on this set of moment conditions, we construct a system GMM estimator
by stacking the equations in ﬁrst differences and the equations in levels for which
instruments are available”. As explained above, lagged levels are used as instruments in
the equations in differences — conditions (1 8) and (19) — and lagged differences (of the

explanatory and the dependent variable) are used as instruments in the equations in levels

— conditions (21) to (24)”.

 

and not levels of trade openness, are not correlated with the industry-speciﬁc effect. For
these variables, the following moment conditions are substituted for those derived from
(21) and (22):

E [11, o X,,,,,)] = 0 for t Z 2

E [8,, o X,(,,,,] = O for t 22

27 In practice, all the estimation based on GMM is performed using the Gauss-based DPD
96 program, whose original version is described in Arellano and Bond (1988).

2' The only exceptions are the variables GX and GMC, for which the instrument is the
ﬁrst lag (and not the lagged ﬁrst-difference) in the equations in levels of the levels plus
differences speciﬁcation (see note 27).

123

Since in order to control for industry-speciﬁc effects we are forced to drop the
observations of at least the ﬁrst period of our panels — which are already quite short —- we
also estimate a model in which there are no industry-speciﬁc effects. In this case, the
estimation is performed in levels, and the moment conditions are given by the
assumptions of weak exogeneity of the explanatory variables X,,, as reﬂected in equation
(20), and no serial correlation in the error term, as expressed in the following moment
condition:

E [Y,(,_,, o 8,, ] = 0 (25)

In both types of estimation procedures — levels and levels plus differences — we
report a Sargan test for overidentifying restrictions proposed by Arellano and Bond
(1991), which tests the overall validity of the instruments by analyzing the sample analog
of the moment conditions used in the estimation process. We also report tests of serial
autocorrelation of the error term 8,,. Speciﬁcally, we test whether the error term is ﬁrst- or
second-order serially correlated in the level regression and, when sufﬁcient observations
are available, we also test whether the differenced error is second-order serially correlated
in the speciﬁcation with both levels and differences (by construction, ﬁrst-order serial
correlation is likely to appear in this case). It is worth noting that both under the Sargan
and the serial correlation tests, failure to reject the null hypothesis gives support to the
model. Moreover, in the levels speciﬁcation, rejection of the hypothesis of no serial
correlation of the residuals constitutes evidence in favor of a model that allows for an

industry-speciﬁc effect.

124

6 - Estimation Results

As previously mentioned, we run two sets of regressions using, respectively, trade
ﬂows-based and trade policy-based measures of openness. In the ﬁrst case, we have data
for the period 1970-92 for Argentina, and 1970-94 for Brazil, Chile, Colombia and
Mexico. For each country, we use a panel of 28 industries (27 in the cases of Chile and
Mexico) described in Table 2.A.2, which correspond to the industries deﬁned at the 3-
digit level in the International Standard Industrial Classiﬁcation (ISIC)29.

In order to separate the long-run relationship between openness and grth — in
which we are interested — from spurious links arising from short-run ﬂuctuations, we
construct a panel where the observations are averaged over ﬁve-year periods”. The type
of short-run relationship that we attempt to purge from the data is the one that arises, for
example, in the cases where countries faced with large external shocks react with policies
that reduce the degree of openness of the economy, while the latter experiences low or
negative growth (Harrison, 1996: 434)“.

Our dependent variable is calculated as the average rate of growth of value added
per worker in each 5-year period, while initial productivity is measured as value added
per worker during the ﬁrst year of each of these periods. Similarly, the trade ﬂows-based
openness variables are calculated as 5-year-averages of the corresponding annual

variables.

 

2” In the cases of Chile and Mexico, the sector ISIC 314 (Tobacco) was excluded because
of the lack of consistent data.

3° The only exception is Argentina, for which the ﬁfth period observations correspond to
1990-1992 averages.

3‘ The author quotes Quah and Rauch (1990), “who use trade shares as a proxy for
openness, (and) ﬁnd that most of the observed positive relationship between openness

125

In the case of the trade policy-based openness variables, we have data for 17
industries in Argentina, Brazil and Mexico, during the years 1987 (1986 for Brazil),
1990, and 1993 (1992 for Mexico). We implement our model by constructing panels that
cover three 3-year periods (1986/88, 1989/91, and 1992/94), and treat the available
annual observations for tariffs and NTBs as representative of the trade policies applied
during the corresponding periods”.

Tables 2.6 to 2.10 present the results of the regressions estimated using the GMM
technique, and making use of trade ﬂows-based openness variables. Tables 2.12 to 2.14
cover the regressions that use trade policy-based measures of openness. In each Table,
regressions (1) and (2) correspond to our ﬁrst estimating equation (equation 5),
implemented successively in a levels speciﬁcation, and a differences plus levels
speciﬁcation — the second one allowing, as explained in the previous section, for the
existence of an industry-speciﬁc effect. Similarly, regressions (3) and (4) correspond to
our second estimating equation (equation 16), implemented in the two above-mentioned
speciﬁcations. As shown in section 4, regressions (3) and (4) control for the accumulation
of labor and capital so that the coefﬁcients on the openness variables should be
interpreted as representing the partial effects of changes in these variables on the rate of
TFP growth. The corresponding coefﬁcients in regressions (1) and (2), on the other hand,

measure the impact of changes in the degree of openness on labor productivity growth,

 

and grth is due to short-run cyclical ﬂuctuations” (Harrison, 1996: 434).

32 We also have data on weighted averages of tariffs and NTBs, using the imports of a
group of 120 developing countries as weights. We only report regression results using the
non-weighted TARIFF and NT B. However, similar results were obtained in (non-
reported) regressions using the weighted measures of tariffs and NTBs. Averages of
weighted and non-weighted tariffs and NTBs are presented in Table 2.A.4. Correlations

126

including both the effects that operate through changes in the rates of TFP growth, and

those that act through changes in the rates of capital accumulation and employment

growth.

6.1 - Regressions with Trade Flows-Based Openness Variables

The ﬁrst ﬁnding suggested by the regression results in Tables 2.6 to 2.10 is the
great heterogeneity existing in the relationship between trade openness and productivity
growth. Indeed, the hypothesis of a positive effect of openness - as measured by our trade
ﬂows-based variables — on the rates of growth of both labor and total factor productivity
is supported in only some of the countries considered, while a negative relationship is
found in others. Table 2.11 summarizes this ﬁnding, by listing the countries for which
statistically signiﬁcant relationships — either positive or negative - were found between
each openness variable and each measure of productivity growth”.

Overall, the result for which most supportive evidence was found is that of a
positive relationship between trade intensity and TFP growth, which appears to be present
in four of the ﬁve countries. In Colombia, however, higher trade-shares in output have
caused a decline in TFP growth. Furthermore, in the case of labor productivity growth, it
appears that only in two countries (Argentina and Brazil) does trade intensity have a

positive and signiﬁcant effect. Thus, our results show that dynamic gains from trade have

 

between the two types of measures are very high, as can be seen in Tables 2.A.5 to 2.A.7.
33 We have not included in this Table the countries for which the corresponding
regressions were not supported by the speciﬁcation tests at the 5 percent level of
signiﬁcance). When both the levels and the differences-and-levels regressions were valid,
we have reported the results of the latter, which controls for the presence of industry-
speciﬁc effects.

127

been prevalent in most countries but they have been usually limited to TFP growth — and
do not necessarily apply to capital accumulation (the other component of labor
productivity growth).

We measure the static effects of trade openness through the coefﬁcients of the
rates of change of exports and import-output ratios. The results indicate that these effects
have been positive in some countries and negative in others. Thus, in the case of export
growth, it appears that it has caused both faster TFP and labor productivity growth in two
countries (Argentina and Brazil), but has had contrary effects in two others (Colombia
and Mexico)”. The evidence for the effects of greater import penetration is even less
clear: in Brazil they are positive both for TFP and for labor productivity growth, in
Colombia they are both negative, in Argentina the effect on labor productivity is positive
but that on TFP is negative, and in Mexico only the effect on labor productivity grth is
statistically signiﬁcant (and positive).

Similarly, the productivity performance of Latin American industries does not
show an unequivocal relationship with the extent of their intra-industry trade. As with the
other openness variables, the results are mixed. It appears, however, that in most of the
cases the effects of changes in the Grubbel Lloyd index of intra-industry trade and those
of changes in the growth of import-output ratios have acted in the same direction.

Finally, evidence of convergence in productivity growth is present only in
Argentina and Brazil, and in both cases it only appears to operate through the faster TFP

growth of industries with initially lower levels of labor productivity. Returns to scale, on

 

3"A positive effect of both GX and GMC on TFP growth was also found in Chile. Nothing
can be said for labor productivity, however, since the corresponding regressions are not

128

the other hand, appear to be increasing in Argentina but decreasing in all the other

countries considered.

6.2 - Regressions with Trade Policy-Based Openness Variables

Before presenting the results of the regressions performed using trade policy-
based openness variables, it is important to note that, at least during 1986/1993, trade
liberalization policies have been implemented in very different ways in Argentina, Brazil
and Mexico. Indeed, during this period, the ﬁrst two countries made considerable cuts in
both tariffs and NTBs. Mexico, on the other hand, had already initiated its trade
liberalization process — in 1985/1986 — so that both types of trade barriers had already
undergone signiﬁcant changes.

Tables 2.A.5 to 2.A.7 report bivariate correlations among policy-based measures
of trade openness, and between these variables and the grth rates of exports and
imports. These Tables reveal at least two important differences between the trade policy
reforms of Argentina and Brazil, on one hand, and Mexico on the other. The ﬁrst is
related to the correlation between tariffs and NTBs. Indeed, in the former two countries
this correlation is positive (0.65 for Argentina and 0.34 for Brazil), while in the latter
country it is negative (-0.27). This reﬂects the fact that during the period here considered
Argentina and Brazil abruptly reduced the degree of protection of their local industries,
using cuts in both tariffs and NTBs in a complementary fashion. Mexico, which had
already made the biggest changes in its trade policies prior to 1987, made use of tariffs

and NTBs as alternative or substitute means of adjusting levels of protection from import

 

supported by speciﬁcation tests.

129

competition — e. g. making the biggest cuts of NTBs in sectors where tariffs were
experiencing the smallest reductions or were even being increased”.

A second distinguishing characteristic of the trade liberalization policies
implemented in the three countries considered is encountered in the different impacts of
these policies on the growth of foreign trade. In the case of imports, their growth appears
to be negatively correlated with both tariffs and NTBs only in Argentina (Table 2.A.5). In
Brazil, these correlations are negligible indicating that in this country low protection —
either through low tariffs or through low NTBs — is not necessarily associated with high
import growth, at least in the short run (Table 2.A.6). In the case of Mexico, the evidence
is even more puzzling, a8 NTBs show a small but negative correlation with import
growth, but this variable appears to have a considerable positive correlation with average
tariffs (Table 2.A.7). A possible explanation for this result is that in Mexico, at least
during the period on which we have data, tariffs were kept high or even increased in the
sectors that were most vulnerable to import penetration“.

With regard to exports, in Argentina their growth was negatively correlated with
the growth of imports and positively correlated with the levels of tariffs and NTBs. In

Brazil, low tariffs and NTBs were also associated with low export growth but the

 

3’ R08 (1994: 192) describes the trade reforms undertaken by the Mexican government in
1985 and 1986, which included the replacement of “direct controls” (NTBs) by tariffs.
This replacement was initiated before the July 1986 GATT membership agreement, but in
this agreement Mexico pledged to continue this type of policy, while also making a
commitment to future tariff reductions. Liberalization measures, however, were waived in
some manufacturing industries — automobiles, pharmaceuticals and electronics — where
the government had been implementing special promotion programs.

36 In fact, in January and March 1989, the Salinas Administration made an upward
adjustment in tariff rates which was motivated by “concerns about making effective
protection more uniform and, probably even more, about the surge of consumer goods

130

correlation was smaller, and there was no apparent relationship between import and
export growth. Mexico is the only country where increases in imports were associated
with increases in exports, and where at least one of the forms of protection (N TBs) was
negatively associated with exports. Low tariffs, however, were, as in the other two
countries, positively correlated with low export growth.

All in all, the above bivariate correlations indicate, ﬁrst, that trade liberalization
policies were implemented in very different ways in each country here studied and,
second, that their effects on the growth of foreign trade were also completely different. It
is therefore no surprise that the econometric estimation of the inﬂuence of trade policy on
productivity growth yields quite different results in the various countries considered.

Regression results are presented in Tables 2.12 to 2.14, and summarized in Table
2.15. The ﬁrst important ﬁnding is that in Argentina trade policies at the industry level
have had no signiﬁcant effect on either labor or total factor productivity. This is quite
surprising since, as shown in section 3 of this chapter, this country has the highest levels
of labor productivity of our sample — which should reduce its vulnerability to foreign
competition and favor the occurrence of beneﬁcial effects from trade liberalization — and
has in fact performed remarkably well, in the aggregate, after the opening of its economy
in 1990. One possible explanation is that, as shown above, the more liberal trade policies
of this country have been correlated with both higher imports and lower exports, thus
having conﬂicting consequences on productivity growth. We also know, from our
decomposition of labor productivity growth — “a la Nordhaus (1972)” —, that during the

early 19908 Argentina has witnessed considerable changes in the composition of output

 

imports during 1988” (R08, 1994: 193).

131

and employment, which have favored industries with relatively low levels and rates of
growth of labor productivity. Since the “room” for these compositional effects is
probably different within each of the 3-digit industries considered, different responses of
these (3-digit) aggregates to the same policy changes could arise.

With regard to Brazil’s and Mexico’s results, we found that in both countries tariff
reductions have had beneﬁcial effects on both labor and total factor productivity growth.
The magnitude of these effects, however, has been much larger in Mexico, where they
appear to have occurred mainly through changes in the rates of capital accumulation —
rather than through increases in the rates of TFP growth. Reductions in NTBs, on the
other hand, have been beneﬁcial to productivity growth in Mexico, but they have had the
opposite effect in Brazil.

In the case of Brazil, it is important to keep in mind that its trade policies have
had a negligible correlation with import growth at the industry level and low (negative)
correlation with export growth. Thus, Brazil’s industries have not necessarily experienced
a larger exposure to foreign competition after the implementation of more liberal trade
policies, a fact that could reﬂect the existence of other government policies that favor
local production to the detriment of imports. In this context, the beneﬁcial effect of tariff
reductions could be interpreted as reﬂecting the fact that low tariffs, even when
accompanied by other barriers to trade, allow local ﬁrms to better monitor the changes in
international prices, and thus to have a better appraisal of trends in intemational
productivity. With regard to the positive association between high NTBs and faster
productivity growth, it is worth noting that although the average coverage of NTBs has

been drastically reduced, their relative dispersion has been raised considerably (Table

132

2.A.4). The result that the sectors whose NTBs were left at higher levels have performed
relatively better can be interpreted as evidence that, in this country, the incentives arising
from government policies still prevail over market-incentives in the determination of the
industries’ decisions to make efﬁciency enhancing investments.

Finally, it is worth noting that, in the three countries considered, regression results
indicate that in recent years there was increasing divergence in either labor (Argentina) or
total factor productivity (Brazil and Mexico). Thus, the sectors with higher initial levels
of labor productivity have performed relatively better in the context of trade
liberalization, which, as previously mentioned, is consistent with the predictions of the

new theories of trade and growth.

7 - Summary and Conclusions

During the past twenty ﬁve years, manufacturing exports in several of the biggest
Latin American countries experienced unprecedented rates of growth. This can be
attributed, in part, to the effort, by some of the governments in the region, to correct the
anti-export bias of their highly “closed” trade regimes. But exports were also stimulated
by the macroeconomic policies implemented during the 19808, as a response to the “debt
crisis”. Indeed, these policies caused sharp contractions in domestic demand as well as
considerable devaluations of the local currencies, which had the effect of reorienting an
important share of industrial output towards export demand. This, together with a sharp
contraction of imports, allowed the region’s economy to accomplish an enormous
reversal of the ﬂow of resources towards the rest of the world. The cost of these policies

was not low, and a growing sense of frustration with the development strategies that had

133

been applied before the crisis began to take shape. By the end of the decade a new policy
consensus was becoming dominant among policy-makers, involving a shift towards more
market-oriented economic policies, with trade liberalization featuring prominently in the
reforms agenda. By the early 19908, most countries in the region had taken important
steps in this direction, slashing tariffs and other import barriers, while also deregulating
ﬁnancial markets, privatizing public enterprises and adopting macroeconomic policies
involving a stricter ﬁscal and monetary discipline.

Implicit in the new development strategy is the idea that there are important static
and dynamic gains to be obtained by increasing the openness to international trade of the
region’s economy. The main channel through which these gains will come, it is generally
believed, is the increase in productive efficiency associated with greater exposure to
foreign competition — either through increased imports or through a larger presence of
local ﬁrms in international markets —— and with greater local access to foreign inputs and
technologies. These trade-growth links, which had long been ignored by economic
theorists, have recently been the subject of new research, mostly associated with the new
models of endogenous growth. This new literature has formalized the idea behind the
higher-efﬁciency-through-trade “story”, but it also has highlighted potential negative
effects of trade on growth. Indeed, it has been shown that, in some circumstances, labor-
rich countries may be led, under free trade, to specialize in slow growing industries. Thus,
in the aggregate, these countries may eventually experience slower growth under free

trade, even if their sectoral growth rates are now higher”. Particularly in the context of

 

37 It is important to note, however, that slower growth is not necessarily associated with
lower welfare. An evaluation of the effects of trade on the latter must take into

134

localized research spillovers, the aggregate effect of openness on growth is theoretically
ambiguous, and depends on the resource endowments of each country — including
knowledge and technology stocks — which in turn determine the performance of
individual industries.

The present chapter has attempted to assess, empirically, the relationship between
trade openness and productivity growth, focusing on the manufacturing industries of ﬁve
Latin American countries in the period 1970-1994. To this end, we have estimated an
empirical model in which the industries’ rates of growth of labor productivity is
explained by the growth of exports, the growth of the import-output ratio, the ratio of
imports plus exports to output, and the Grubbel-Lloyd index of intra-industry trade. For
three of the ﬁve countries and during the period after 1985, we have also estimated a
model in which productivity growth is explained by trade policy-based measures of
openness, namely tariffs and non-tariff barriers to trade. In all cases, we have included in
the regressions the level of initial productivity as an explanatory variable, in order to
control for the component of productivity growth that is potentially explained by the
existence of convergence in productivity. We have also run the regressions with the
inclusion of variables that are shown to control, under fairly general assumptions, for the
accumulation of factors of production, so that the coefﬁcients on the other explanatory
variables can be interpreted as reﬂecting their effect on total factor productivity growth.
All models have been estimated using an econometric procedure, based on the

Generalized Method of Moments, that allows for the possible endogeneity of the

 

consideration the fact that under free trade consumers may have increased access to
cheaper and more varied goods.

135

explanatory variables, as well as for the existence of industry-speciﬁc effects. This
procedure allows us to interpret the results as reﬂecting causality from the explanatory to
the dependent variable, and not the other way around, which distinguishes this study from
previous attempts at uncovering the trade and growth relationship.

We have found that productivity growth at the industry level has been related to
international trade ﬂows and policies in different ways depending on the country
considered. Thus, for none of the trade ﬂows-based openness variables did we ﬁnd a
general positive relationship with productivity growth that applies to all countries. The
result that applies to most countries (four out of ﬁve) is that the level of trade openness is,
as expected, positively and signiﬁcantly associated with faster TFP growth, suggesting
the existence of dynamic gains ﬁ'om trade — the only exception is Colombia, where this
effect is negative. This result, however, does not extend to labor productivity, which is
positively affected by trade intensity in only two countries (Argentina and Brazil).

With respect to export growth, we ﬁnd that it has a positive effect on both TFP
and labor productivity growth in Argentina and Brazil, but a negative effect on these
variables in Colombia and Mexico. Similarly, growth in import-output ratios has a
positive impact on labor productivity in Argentina, Brazil, and Mexico, but a negative
one in Colombia. Moreover, only in Brazil and Chile does TF P growth signiﬁcantly
beneﬁt from larger degrees of import penetration, while there is evidence that the latter
leads to a decline in the rates of TFP growth of Argentina and Colombia. Finally, larger
intra-industry trade is positively associated with labor productivity growth in Argentina,

Brazil and Mexico, but only in Chile does it favor TF P growth. Furthermore, a negative

136

relationship between intra-industry trade and TFP growth is found in Argentina and
Colombia.

When trade policy-based openness variables were used to explain industrial
productivity grth during the recent wave of trade liberalization reforms, our ﬁndings
were even less encouraging. Indeed, regression results suggest that, contrary to
expectations, Mexico is the only country where productivity growth appears to be
negatively related to both tariffs and NTBs. In Argentina, on the contrary, the effects of
trade policies are found to be statistically non-signiﬁcant, while Brazil appears to be in an
intermediate situation, with productivity growth being affected positively by tariff cuts
but negatively by reductions in NTBs.

Overall, the empirical results obtained in this chapter indicate that the hypothesis
of a general positive relationship between openness and productivity growth at the
industry level does not encounter support in the recent experience of the largest Latin
American countries. When formulated in the context of the “new” theories of trade and
growth — as synthesized, for example, by Grossman and Helpman (1991) — this
hypothesis is based on the assumption of global and instantaneous technology spillovers.
Thus, its empirical rejection could be interpreted as suggesting that research spillovers are
in fact geographically concentrated, so that there is a role, in the determination of relative
comparative advantage, for the size and previous research experience of each country.

From a policy perspective, this implies that the dynamic gains from trade
liberalization are, to a great extent, country speciﬁc. Thus, although welfare effects do not

necessarily correlate with growth effects, it appears that the mere opening of an economy

137

to free international trade is not, at least in the short and medium term, a guaranteed road

to prosperity.

138

REFERENCES

Ahn, SC. and Schmidt, P. 1995. “”Efﬁcient Estimation of Models for Dynamic Panel
Data.” Journal of Econometrics 68: 5-27.

 

Anderson, T.W. and Hsiao, C. 1981. “Estimation of Dynamic Models with Error
Components.” Journal of the American Statistical Association 76: 598-606.

 

Arellano, M. and Bond, S. 1988. “Dynamic Panel Data Estimation Using DPD — A Guide
for Users.” London: Institute for Fiscal Studies, Working Paper 88/15.

Arellano, M. and Bond, S. 1991. “Some Tests of Speciﬁcation for Panel Data: Monte
Carlo Evidence and an Application to Employment Equations.” Review of Economic
Studies 58: 277-297.

 

Arellano, M. and Bover, O. 1995. “Another Look at the Instrumental Variable Estimation
of Error Components Models.” Journal of Econometrics 68: 29-51.

 

Backus, D., Kehoe, P., and Kehoe, T. 1992. “In Search of Scale Effects in Trade and
Growth.” Journal of Economic Theory 58: 377-409.

 

Benavente, J .M., Crespi, G., Katz, J. and Stumpo, G. 1996. “La Transformacion del
Desarollo Industrial de América Latina.” Revista de la CEPAL 60: 49—72.

 

Blundell, R. and Bond, S. 1997. “Initial Conditions and Moment Restrictions in Dynamic
Panel Data Models.” University College London, Discussion Paper 97-07.

Bonelli, R. 1992. “Growth and Productivity in Brazilian Industries: Impacts of Trade
Orientation.” Journal of Developing Economics 39 (1): 85-109.

 

Burki, SJ. and Edwards, S. 1997. “Latin America after Mexico: Quickening the Pace.” In
Annual World Bank Conference on Development in Latin America and the Caribbean
1995: the Challenges of Reform, edited by Burki, 8.1., Edwards, S. and Aiyer, S.
Washington DC: The World Bank.

Burki, S. J. and Perry, G. E. 1997. The Long March: a Reform Agenda for Latin America
and the Caribbean in the Next Decade. The World Bank Latin American and
Caribbean Studies. Washington, DC: The World Bank.

 

Dixit, A. and Norman, V. 1980. Theory of International Trade. Cambridge: Cambridge
University Press.

 

139

Easterly, W., Loayza, N. and Montiel, P. 1997. “Has Latin American Post-Reform
Growth been Disappointing?” Journal of International Economics 43: 287-311.

 

ECLAC (Economic Commission for Latin America and the Caribbean). 1992. El
Comércio de Manufacturas de América Latina: Evolucion y Estructura 1962-1989.

Estudios e Inforrnes de la CEPAL series, No. 88. Santiago, Chile: ECLAC, United
Nations.

 

ECLAC (Economic Commission for Latin America and the Caribbean). 1996. ﬁe
Economic Experience of the Last Fifteen Years: Latin America and the Caribbean,
1980-1995. Santiago, Chile: ECLAC, United Nations.

 

Edwards, S. 1995. Crisis and Reform in Latin America: From Desmir to Hope. New
York, NY: Oxford University Press.

 

F eenstra, R. 1996. “Trade and Uneven Growth.” Journal of DevelopingEconomics 49 (1)

(Special Issue: Increasing Returns, Competition and Economic Development): 229-
257.

 

Fishlow, A. 1991. “Liberalization in Latin America.” In Economic Liberalization: no
Panacea, edited by T. Banuri. Oxford: Oxford University Press.

 

Greene, W. 1997. Econometric Analysis. Upper Saddle River, New Jersey: Prentice Hall.

 

Grossman, G. M. and Helpman, E. I991. Innovation and Growth in the Global Economy.
Cambridge, MA: The MIT Press.

 

Harrison, A. 1996. “Openness and growth: A Time-Series, Cross-Country Analysis for
Developing Countries.” Journal of Development Economics 48: 419-447.

 

Helpman, E. 1992. “Endogenous Macroeconomic Growth Theory.” European Economic
Review 36: 237-267.

 

Jaspersen, F. 1997. “Growth of the Latin American and East Asian Economies.” In
Pathways to Growth: Comparing East Asia and Latin America, edited by N. Birdsdall
and F. Jaspersen. Washington DC: Inter-American Development Bank.

Krugman, P. 1981. “Intraindustry Specialization and the Gains from Trade.” Journal of
Political Economy 89: 959-973.

 

Learner, E. 1988. “Measures of Openness.” In Trade Policy Issues and Empirical
Analysis, edited by R. Baldwin. Chicago, IL: University of Chicago Press.

 

Lee, J. 1996. “Government Interventions and Productivity Growth.” Journal of Economic
Growth 1: 391-414.

 

I40

Nishimizu, M. and Page Jr., J. M. 1991. “Trade Policy, Market Orientation and
Productivity Change in Industry.” In Trade Theory and Economic Reform: Essays in
Honor of Bela Balassa, edited by J. de Mello and A. Sapir. Cambridge, MA: Basil
Blackwell.

 

 

Nishimizu, M. and Robinson, S. 1984. “Trade Policies and Productivity Changes in
Semi-Industrialized Countries.” Journal of Development Economics 16: 177-206.

 

Nordhaus, W.D. 1972. “The Recent Productivity Slowdown.” Brooking Papers on
Economic Activity 3: 493-536.

 

 

Meller, P. 1994. “The Chilean Trade Liberalization and Export Expansion Process 1974-
90.” In Trade Policy and Industrialization in Turbulent Times, edited by GK.
Helleiner. London: Routledge.

 

Ocampo, J .A. 1994. “Trade Policy and Industrialization in Colombia.” In Trade Policy
and Industrialization in Turbulent Times, edited by GK. Helleiner. London:
Routledge.

 

 

Primo Braga, C. A., Nogués, J ., and Rajapatirana, S. 1997. “Latin America and the
Caribbean in the World Economy.” In Annual World Bank Conference on
Develomnent in Latin America and the Caribbean 1995: the Challenges of Reform,
edited by Burki, S.J., Edwards, S. and Aiyer, S. Washington DC: The World Bank.

 

Pritchett, L. 1996. “Measuring Outward Orientation in LDCs: Can it be Done?” Journal
of Development Economics 49: 307-335.

 

Quah, D. and Rauch, J. E. 1990. “Openness and the Rate of Economic Growth.”
Manuscript, MIT.

Ros, J. 1994. “Mexico’s Trade and Industrialization Experience since 1960.” In Trade
Policy and Industrialization in Turbulent Times, edited by GK. Helleiner. London:
Routledge.

Sachs, J. and Warner, A. 1995. “Economic Reform and the Process of Global
Integration.” Brookings Papers of Economic Activity 1: 1-118.

Solow, R. 1957. “Technical Change and the Aggregate Production Function.” Review of
Econonrics and Statistics 39: 312-320.

Tybout, J. R. 1991. “Researching the Trade-Productivity Link: New Directions.” The
World Bank Policy Research Working Papers Series No. 638. Washington DC: The
World Bank.

141

Tybout, J. R. 1992. “Linking Trade and Productivity: New Research Directions.” 11E
World Bank Research Observer 6 (2): 189-211.

 

UNCTAD. 1994. Directory of Import Regimes. New York: United Nations.

 

UNIDO. 1983. La Industria Mundial en 1980. New York: United Nations.

 

142

APPENDIX A

TABLES AND FIGURES

143

Table 2.1: Composition of Merchandise Exports, 1970/1994 (percent)

 

 

Country/ Sector 1970 1980 1994?“)
Argentina

1. Primary Products 61.3 48.9 36.3
2. Semi-Manufactures") 25.6 30.6 33.4
3. Manufactures 13.1 20.5 30.3
4. Other 0.1 0.0 0.0
Total 100.0 100.0 100.0
Brazil

1. Primary Products 67.0 30.2 19.9
2. Semi-Manufactures") 21 .2 36.0 30.3
3. Manufactures l 1.5 33.6 48.0
4. Other 0.3 0.2 1.8
Total 100.0 100.0 100.0
Chile

1. Primary Products 12.9 20.2 30.4
2. Semi-Manufactures") 84.9 74.4 56.3
3. Manufactures 2.2 5.3 8.9
4. Other 0.0 0.1 4.4
Total 100.0 100.0 100.0
Colombia

1. Primary Products 85.1 70.3 51.5
2. Semi-Manufactures") 7.7 1 1.3 10.6
3. Manufactures 6.6 16.1 31.6
4. Other 0.6 2.3 6.3
Total 100.0 100.0 100.0
Mexico

1. Primary Products 45.3 80.7 38.2
2. Semi-Manufactures") 26.7 9.7 13.5
3. Manufactures 27.7 9.5 47.6
4. Other 0.3 0.1 0.6
Total 100.0 100.0 100.0

 

Source: ECLAC (1996), p. 113.

0) Food products, beverages, tobacco, wood products, pulp and paper products, industrial chemicals,
petroleum reﬁneries, and petroleum and coal products (ISIC groups 311, 313, 314, 331, 341, 351, 353,
354). "" Data available for 1993 in the case of Colombia, and 1992 for Mexico.

144

Table 2.2: Composition of Industrial Value Added,
Exports and Imports, 1970/1994 (percent)

 

 

 

 

Country/Sector Value Added Exports Imports
1970/80 1981/90 1991/94 1970/80 1981/90 1991/94 1970/80 1981/90 1991/94
Argentina")

1. Traditional 59.3 61.1 57.1 77.7 76.2 73.1 34.4 37.1 32.9
Industries

1.1. Semi- 40.2 47.1 45.4 64.8 63.6 59.9 30.4 33.0 23.0
Manufactures '

1.2. Other 5 19.1 13.9 11.6 12.8 12.6 13.2 4.0 4.1 9.9
2. Basic Inputs ° 8.7 9.6 10.3 4.4 10.7 8.3 19.4 10.7 6.7
3. New Industries 32.0 29.3 32.6 18.0 13.1 18.6 46.1 52.2 60.4

3.1. Labor— 13.3 10.3 1 1.1 9.3 6.4 7.4 28.2 32.5 34.9
Intensive ‘

3.2. Capital- 18.8 19.0 21.5 8.7 6.7 11.3 17.9 19.7 25.5
Intensive ‘

Brazil
1. Traditional 46.6 48.0 47.9 79.6 58.6 49.3 32.0 38.9 42.1
Industries

1.1. Semi- 29.9 32.2 33.6 69.0 46.1 35.7 29.3 34.6 36.3
Manufactures '

1.2. Other " 16.7 15.8 14.4 10.6 12.5 13.6 2.7 4.3 5.8

2. Basic Inputs ° 14.3 13.6 14.6 4.4 15.4 20.7 14.7 7.7 5.0

3. New Industries 39.1 38.4 37.5 16.0 25.9 30.0 53.2 53.4 52.9

 

3.1. Labor- 20.2 20.1 18.4 7.3 11.2 14.3 34.6 33.5 32.4
Intensive “

3.2. Capital- 18.9 18.2 19.1 8.6 14.7 15.7 18.6 19.9 20.6
Intensive °
Chile

1. Traditional 49.4 53.1 56.5 19.6 33.5 41.3 36.8 39.1 33.8
Industries

1.1. Semi- 35.3 43.9 47.7 19.0 32.2 37.1 29.3 27.8 21.5
Manufactures '

1.2. Other " 14.1 9.1 8.8 0.5 1.2 4.2 7.4 11.3 12.3

2. Basic Inputs ° 26.6 29.3 22.6 78.3 63.6 52.5 6.2 6.1 7.0

3. New Industries 24.0 17.7 21.0 2.2 2.9 6.2 57.0 54.8 59.2
3.1. Labor- 9.5 6.3 7.2 0.8 1.0 2.7 32.3 31.2 33.6
Intensive "

3.2. Capital- 14.5 11.4 13.8 1.3 1.9 3.5 24.7 23.6 25.6

Intensive °

 

145

Table 2.2 (cont’d)

 

 

 

 

Country/Sector Value Added Exports Imports

1970/80 1981/90 1991/94 1970/80 1981/90 1991/94 1970/80 1981/90 1991/94
Colombia
1. Traditional 67.6 65.5 61.3 77.0 76.7 72.5 34.1 38.3 37.6
Industries

1.1. Semi- 43.6 46.4 42.3 44.6 48.4 32.2 30.6 35.0 31.3
Manufactures '

1.2. Other “ 24.1 19.1 18.9 32.4 28.3 40.3 3.5 3.3 6.3
2. Basic Inputs ° 8.7 10.9 12.3 5.5 9.1 8.5 11.2 11.0 9.5
3. New Industries 23.7 23.6 26.4 17.5 14.2 19.1 54.7 50.7 52.9

3.1. Labor- 8.4 8.6 8.5 7.7 7.4 9.6 29.2 29.8 28.4
Intensive ‘

3.2. Capital- 15.3 15.0 17.9 9.8 6.7 9.5 25.5 21.0 24.6
Intensive °
Mexico
1. Traditional 61.5 59.3 55.9 57.3 41.7 23.8 29.5 32.8 29.3
Industries

1.1. Semi- 41.9 39.9 39.4 44.7 33.7 13.7 25.6 28.0 20.0
Manufactures '

1.2. Other " 19.6 19.4 16.6 12.6 8.0 10.2 3.8 4.8 9.2
2. Basic Inputs c 10.4 11.5 11.1 13.7 16.0 8.1 9.2 9.1 8.8
3. New Industries 28.2 29.2 33.0 29.0 42.3 68.1 61.4 58.1 61.9
3.1. Labor- 12.] 11.6 12.1 10.6 13.3 36.2 33.6 34.8 35.7

Intensive "
3.2. Capital- 16.1 17.6 20.9 18.4 29.0 31.9 27.7 23.3 26.2
Intensive °

 

§Ourcez see text.

( ) Data available for 1970-93. ‘ Food products, beverages, tobacco, wood products, pulp and paper
PFOducts, industrial chemicals, petroleum reﬁneries, and petroleum and coal products (ISIC groups 311,
3 1 3, 314, 331, 341, 351, 353, 354). " Textiles, wearing apparel, leather products, footwear, rubber
Products, pottery, china and earthenware, glass and glass products, and other miscellaneous manufactures
(ISIC groups 321, 322, 323, 324, 355, 361, 362, 390). ° Plastic products, other non-metallic mineral
p"(N.‘iucts, iron and steel, and non-ferrous metals (ISIC groups 356, 369, 371, 372). " Furniture, printing
2nd publishing, non-electrical and electrical machinery (ISIC groups 332, 342, 382, 383).
Pharmaceutical and other chemical products, fabricated metal products, transport equipment, and

professional and scientiﬁc equipment (ISIC groups 352, 381, 384, 385).

146

Table 2.3: Export and Import Coefficients in Manufacturing
Output, 1970/1994 (percent)

 

 

 

 

 

Country/Sector Export Shares Import Shares
1970/80 1981/90 1991/94 1970/80 1981/90 1991/94
Argentina 0)
1. Traditional Industries 9.0 10.2 9.7 4.8 4.1 6.1
1.1. Semi-Manufactures ' 10.9 11.0 10.3 6.1 4.7 5.2
1.2. Other " 4.7 7.5 7.4 1.9 2.0 9.7
2. Basic Inputs Industries ‘ 3.5 11.0 8.1 21.8 8.3 8.7
3. New Industries 3.7 4.1 4.2 12.6 12.8 20.3
3.1. Labor-Intensive ‘ 4.9 5.9 5.5 20.2 23.5 35.5
3.2. Capital-Intensive ‘ 2.9 3.1 3.5 7.8 7.4 12.6
Total Manufactures 6.8 8.6 7.9 8.5 6.9 10.6
Brazil
1. Traditional Industries 1 1.4 12.6 13.8 5.5 3.8 7.9
1.1. Semi-Manufactures ' 14.2 13.4 14.0 7.2 4.7 9.5
1.2. Other " 5.1 10.4 13.5 1.6 1.6 3.9
2. Basic Inputs Industries c 2.3 11.9 19.3 9.5 2.7 3.1
3. New Industries 3.6 9.3 12.3 14.2 8.9 14.5
3.1. Labor-Intensive ‘ 3.4 8.6 12.5 19.9 11.8 18.9
3.2. Capital-Intensive ° 3.7 10.0 12.1 9.2 6.3 10.7
Total Manufactures 7.5 l 1.4 14.1 9.0 5.3 9.4
Chile
1 - Traditional Industries 16.9 24.4 26.4 24.8 26.1 26.7
1 - l - Semi-Manufactures ‘ 21.2 27.5 27.8 27.0 21.6 19.8
1 -2- Other I 2.7 6.2 18.5 22.6 53.0 67.7
2- Basic Inputs Industries ° 150.7 89.8 73.8 9.5 8.0 12.3
3 - New Industries 5.1 7.3 11.5 103.7 133.0 136.9
3 ~ 1 - Labor-Intensive " 4.9 7.8 14.8 149.8 223.9 232.9
-2 - Capital-Intensive ° 5.2 7.2 9.8 74.2 87.4 88.9
mManufactures 46.6 40.6 35.6 36.8 37.7 44.1

 

147

Table 2.3 (cont’d)

 

 

 

 

Country/Sector Export Shares Import Shares
1970/80 1981/90 1991/94 1970/80 1981/90 1991/94

Colombia

1. Traditional Industries 7.1 8.4 15.8 9.5 12.3 18.0
1.1. Semi-Manufactures ' 6.0 6.8 9.6 12.5 14.9 20.2
1.2. Other" 9.7 13.1 32.9 3.1 4.3 12.0

2. Basic Inputs Industries ° 4.3 6.8 10.4 27.7 23.9 25.5

3. New Industries 4.8 4.3 10.1 46.8 47.3 63.5
3.1. Labor-Intensive " 6.8 7.1 16.6 76.1 83.9 108.6
3.2. Capital-Intensive ° 4.0 3.0 7.2 32.4 29.3 43.3
Total Manufactures 6.4 7.3 13.7 19.4 21.5 30.8

Mexico

1. Traditional Industries 3.6 5.4 8.2 5.3 7.6 16.7
1.1. Semi-Manufactures ‘ 3.8 5.8 5.8 6.3 8.7 14.7
1.2. Other " 3.0 4.6 16.4 2.5 4.6 23.4
2. Basic Inputs Industries ° 5.0 11.3 14.2 10.0 11.1 27.0
3. New Industries 4.5 15.0 47.5 26.8 30.3 67.5
3.1. Labor-Intensive ‘ 4.2 13.9 83.9 37.9 52.1 122.4
3.2. Capital-Intensive ° 4.6 15.5 30.0 19.8 18.6 41.5
Total Manufactures 3.9 8.9 21.3 1 1.3 14.6 34.1

 

Source: see text.

0’ Data available for 1970-93. ‘ Food products, beverages, tobacco, wood products, pulp and paper
PFOducts, industrial chemicals, petroleum reﬁneries, and petroleum and coal products (ISIC groups 311,
3 13, 314, 331, 341, 351, 353, 354). " Textiles, wearing apparel, leather products, footwear, rubber
PFOducts, pottery, china and earthenware, glass and glass products, and other miscellaneous manufactures
(I SIC groups 321, 322, 323, 324, 355, 361, 362, 390). ° Plastic products, other non-metallic mineral
Products, iron and steel, and non-ferrous metals (ISIC groups 356, 369, 371, 372). Furniture, printing
and publishing, non-electrical and electrical machinery (ISIC groups 332, 342,382,383).
Pharmaceutical and other chemical products, fabricated metal products, transport equipment, and

professional and scientiﬁc equipment (ISIC groups 352,381, 384, 385).

148

Table 2.4: Rates of Growth of Industrial Labor Productivity"),

 

 

 

 

 

1970/1994 (percent)
Country/Sector 1970/1980 1980/1990 1990/1994 1970/1994
Argentina”1
1. Traditional Industries 2.6 2. 1 3 .8 2.6
1.1. Semi-Manufactures ' 1.9 2.5 2.3 2.2
1.2. Other 5 2.9 0.2 6.3 2.1
2. Basic Inputs Industries ° 2.0 1.3 14.7 3.3
3. New Industries 4.2 2.1 21.4 5.8
3.1. Labor-Intensive " 5 .3 1.0 25.5 5.8
3.2. Capital-Intensive ° 3.4 2.7 19.3 5.0
Total Manufactures 3.1 2.2 10.3 3.6
Coefﬁcient of Variation ‘ 1.4 1.1 9.4 1.7
Brazil
1. Traditional Industries 3.2 0.4 8.7 2.9
1.1. Semi-Manufactures ' 3.3 0.7 8.7 3.1
1.2. Other " 2.7 -0.1 7.5 2.3
2. Basic Inputs Industries ° 1.1 1.1 12.3 2.9
3. New Industries 1.7 0.3 10.3 2.5
3.1. Labor-Intensive “ 2.4 -0.1 9.7 2.5
3.2. Capital-Intensive ° 1.3 0.8 10.0 2.5
Total Manufactures 2.4 0.4 9.7 2.7
goefﬁcient of Variation ‘ 0.9 0.5 1.8 0.3
Chile
1 - Traditional Industries 5.0 -1.4 4.4 2.2
1 - l . Semi-Manufactures ' 5.2 -1.6 3.0 1.9
1 -2. Other 5 1.8 -2.4 4.1 0.4
2- Basic Inputs Industries ° -2.2 0.2 -1.4 -1.1
3 - New Industries 4.6 -1.1 5.5 2.3
3 ~ 1 . Labor-Intensive ‘ 3.6 -1.8 6.8 1.8
3 2. Capital-Intensive = 5.3 -0.9 4.9 2.6
Total Manufactures 2.8 -1.1 2.8 1.2
wcient of Variation ‘ 3.1 1.0 3.1 1.5
149

Table 2.4 (cont’d)

 

 

 

Country/Sector 1970/1980 1980/1990 1990/1994 1970/1994
Colombia
1. Traditional Industries 2.5 2.5 2.1 2.4
1.1. Semi-Manufactures ‘ 3.5 1.5 1.4 2.3
1.2. Other” 0.3 3.8 1.6 2
2. Basic Inputs Industries ° 1.5 6.3 2.0 3.6
3. New Industries -0.2 4.5 7 .4 3.0
3.1. Labor-Intensive ‘ 0.3 2.4 5.1 1.9
3.2. Capital-Intensive ° -0.5 6.0 8.5 3.6
Total Manufactures l .8 3 .3 3 .5 2.7
Coefficient of Variation ‘ 1.6 2.1 3.1 0.9
Mexico
1. Traditional Industries 2.6 3.5 1.9 2.8
1.1. Semi-Manufactures ‘ 1.8 3.9 2.4 2.8
1.2. Other " 3.9 2.3 0.6 2.7
2. Basic Inputs Industries ° 2.7 4.6 1.6 3.3
3. New Industries 3.2 2.4 3.3 2.9
3.1. Labor-Intensive ‘ 3.5 1.1 3.4 2.5
3.2. Capital-Intensive ° 3.0 3.1 3.3 3.1
Total Manufactures 2.7 3.2 2.3 2.9
Coefﬁcient of Variation ‘ 0.8 1.4 1.2 0.3

 

Source: see text.

0’ Geometric means of annual rates. "2 Data available for 1970-93. ' Food products, beverages, tobacco,
Wood products, pulp and paper products, industrial chemicals, petroleum reﬁneries, and petroleum and
Coal products (ISIC groups 311, 313, 314, 331, 341, 351, 353, 354). " Textiles, wearing apparel, leather
PI‘Oducts, footwear, rubber products, pottery, china and earthenware, glass and glass products, and other
miscellaneous manufactures (ISIC groups 321, 322, 323, 324, 355, 361, 362, 390). ° Plastic products,
Other non-metallic mineral products, iron and steel, and non-ferrous metals (ISIC groups 356, 369, 371,
3 72). " Furniture, printing and publishing, non-electrical and electrical machinery (ISIC groups 332, 342,
3 82 , 383). ° Pharmaceutical and other chemical products, fabricated metal products, transport equipment,
and professional and scientiﬁc equipment (ISIC groups 352, 381, 384, 385). ‘Calculated on the basis of

the sectors 1.1, 1.2, 2, 3.1, and 3.2.

150

Table 2.5: Decomposition of the Rates of Labor Productivity Growth
in 28 Industrial Sectors(*), 1970/1994

 

 

 

 

Country / Period 1970/1980 1980/ 1990 1990/ 1994 1970/ 1994
Argentina
Rate Terms
1. Fixed Weight 3.13 1.90 16.13 4.19
2. Change in Fixed Weight -0.19 -0.26 -0.62 -0.18
Level Terms
3. Fixed Weight 0.59 0.92 -3.50 0.16
4. Change in Fixed Weight -0.08 0.05 -0.24 0.00
Interaction of Rate and Level -0.21 -O.14 -1.35 -0.33
Terms
Aggregate Productivity Growth 3.23 2.47 10.42 3.84
Brasil
Rate Terms
1. Fixed Weight 3.43 0.69 10.03 3.45
2. Change in Fixed Weight -0.39 -0.16 -0.08 -0.30
Level Terms
3. Fixed Weight -0.06 0.22 0.08 0.09
4. Change in Fixed Weight 0.10 0.13 0.03 0.09
Interaction of Rate and Level -0. 18 -0.31 -0.23 -0.24
Terms
Aggregate Productivity Growth 2.89 0.56 9.83 3.08
Chile
Rate Terms
1. Fixed Weight 6.23 1.26 3.39 3.67
2. Change in Fixed Weight -3.46 -0.41 -0.07 -l .61
Level Terms
3. Fixed Weight 1.66 -0.90 -0.49 -0.26
4. Change in Fixed Weight -0.88 -0.66 0.12 -0.13
Interaction of Rate and Level -0. 13 -0.29 -0.09 -0.19
Terms
Aggregate Productivity Growth 3.41 -1.00 2.85 1.48

 

151

Table 2.5 (cont’d)

 

 

 

Country / Period 1970/1 980 1980/ 1990 1990/1994 1970/1994
Colombia
Rate Terms
1. Fixed Weight 1.98 4.54 3.37 3.45
2. Change in Fixed Weight -0.40 -O.93 -0.08 -0.74
Level Terms
3. Fixed Weight 0.29 -0.17 0.19 0.03
4. Change in Fixed Weight 0.03 0.00 0.04 0.08
Interaction of Rate and Level -0.10 -0.07 -0.03 -0.08
Terms
Aggregate Productivity Growth 1.81 3.37 3.49 2.74
Mexico
Rate Terms
1. Fixed Weight 2.85 3.31 2.32 2.92
2. Change in Fixed Weight -0.28 -0. 12 0.04 -0.13
Level Terms
3. Fixed Weight 0.42 0.16 0.00 0.27
4. Change in Fixed Weight -0.11 0.00 0.02 -0.08
Interaction of Rate and Level -0.09 -0.10 -0.06 -0.09
Terms
Aggregate Productivity Growth 2.78 3.25 2.33 2.90

 

Source: see text.

Note: (*) Sectors as deﬁned in Table 2.A.1.

152

Table 2.6: GMM Estimates of Industry Labor Productivity Growth
on Trade Flows-based Openness Variables: Argentina, 1970/1992
(p-values in parenthesis)

 

 

 

 

(1) (2) (3) (4)
Regression Speciﬁcation Levels Diﬁ-Lev. Levels Dif.-Lev.
Instruments ('1 Levels Lev.-Dif. Levels Lev.-Dif.
Initial Labor Productivity 0.03 -0.0002 0.004 -0.002
(0.34) (0.77) (0.02) (0.05)
Trade Intensity‘I 0.004 0.012 -0.010 0.009
(0.37) (0.00) (0.00) (0.02)
Index of Intra-Industry 0.032 0.010 0.006 -0.005
Tradeb (0.00) (0.00) (0.00) (0.07)
Growth of Exports 0.131 0.136 0.067 0.101
(0.00) (0.00) (0.00) (0.00)
Growth in Import -Output 0.090 0.084 -0.029 -0.028
Ratio Coefficientc (0.00) (0.00) (0.00) (0.00)
Constant 0.01 1 -0.002
(0.81) (092)
Growth in Employment 0.292 0.365
(0.00) (0.00)
Change in the Labor Share in 4.143 4.690
Value Added (0.00) (0.00)
Change in the Labor Share in -9.970 -10.561
Value Added times one (0.00) (0.00)
minus the Labor Share“
Wald Test of Joint 0.00 0.00 0.00 0.00
Signiﬁcance: p-value
Sargan Test of 0.211 0.832 0.245 0.998
Overidentifying Restrictions:
P-value
Test for First-Order Serial 0.264 0.006 0.035 0.001
Correlation:
P‘Value
TeSt for Second-Order Serial 0.376 0.886 0.762 0.539
OITeIation: p-value
I”h-lh'rber of Observations 140 l 12 140 1 12

 

N\
otes: (*) See text. (a) Log of the ratio of total trade to total output. (b) Log of one minus the ratio of the
abSO lute value of the trade deﬁcit to total trade. (c) Growth of imports minus growth of total output. ((1)
e average of the current and the lagged labor share is used.

153

Table 2.7: GMM Estimates of Industry Labor Productivity Grth
on Trade F lows-based Openness Variables: Brazil, 1970/1994

(p-values in parenthesis)

 

 

 

 

(1) (2) (3) (4)
Regression Speciﬁcation Levels Dif.-Lev. Levels Dif-Lev.
Instruments ('1 Levels Lev.-Dif. Levels Lev.-Dif.
Initial Labor Productivity 0.015 -0.002 0.010 -0.002
(0.00) (0.22) (0.00) (0.02)
Trade Intensity‘ 0.010 0.018 0.008 0.012
(0.00) (0.00) (0.00) (0.00)
Index of Inna-Industry 0.007 0.010 0.001 -0.002
Tradeh (0.03) (0.00) (0.65) (0.44)
Growth of Exports 0.066 0.067 0.122 0.051
(0.00) (0.00) (0.00) (0.00)
Growth in Import-Output 0.129 0.089 -0.002 0.032
Ratio Coefﬁcientc (0.00) (0.00) (0.85) (0.11)
Constant -0.148 -0.1 13
(0.00) (0.00)
Growth in Employment -0.550 -0.463
(0.00) (0.00)
Change in the Labor Share in ~15.411 ~18.907
Value Added (0.00) (0.00)
Change in the Labor Share in 20.392 25.313
Value Added times one (0.00) (0.00)
minus the Labor Share‘I
Wald Test of Joint 0.00 0.00 0.00 0.00
Signiﬁcance: p-value
Sargan Test of 0.068 0.655 0.346 0.993
Overidentifying Restrictions:
P-Value
Test for First-Order Serial 0.012 0.000 0.228 0.001
Correlation:
P‘Value
Tesr for Second-Order Serial 0.499 0.972 0.057 0.709
ol‘l’elation: p-value
Number of Observations 140 112 140 112

 

N\‘
otes: (*) See text. (a) Log of the ratio of total trade to total output. (b) Log of one minus the ratio of the
bSOlute value of the trade deﬁcit to total trade. (c) Growth of imports minus growth of total output. (d)
e average of the current and the lagged labor share is used.

154

Table 2.8: GMM Estimates of Industry Labor Productivity Growth
on Trade Flows-based Openness Variables: Chile, 1970/1994

(p-values in parenthesis)

 

 

 

 

(1) (2) (3) (4)
Regression Speciﬁcation Levels Dif.-Lev. Levels Dif.-Lev.
Instruments ('5 Levels Lev.-Dif. Levels Lev.-Dif.
Initial Labor Productivity -0.006 -0.005 0.006 -0.002
(0.14) (0.08) (0.25) (0.33)
Trade Intensity‘ 0.022 0.043 0.010 0.026
(0.00) (0.00) (0.00) (0.00)
Index of Intra-Industry 0.006 0.064 0.015 0.039
Trade” (0.29) (0.00) (0.01) (0.00)
Growth of Exports 0.054 0.031 0.098 0.081
(0.00) (0.01) (0.00) (0.00)
Growth in Import-Output 0.016 -0.014 0.030 0.056
Ratio Coefficient‘ (0.78) (0.25) (0.55) (0.05)
Constant -0.01 3 -0.060
(0.77) (021)
Growth in Employment -0.596 -0.579
(0.00) (0.00)
Change in the Labor Share in -8.197 -4.127
Value Added (0.01) (0.10)
Change in the Labor Share in 12.234 6.616
Value Added times one (0.00) (0.06)
minus the Labor Share‘i
Wald Test of Joint 0.00 0.00 0.00 0.00
Signiﬁcance: p-value
Sargan Test of 0.061 0.850 0.628 0.997
Overidentifying Restrictions:
p—value
Test for First-Order Serial 0.004 0.007 0.012 0.007
Correlation:
p-value
Test for Second-Order Serial 0.739 0.020 0.968 0.460
Correlation: p-value
Number of Observations 135 108 I35 108

 

Notes: (*) See text. (a) Log of the ratio of total trade to total output. (b) Log of one minus the ratio of the
absolute value of the trade deﬁcit to total trade. (c) Growth of imports minus growth of total output. (d)
The average of the current and the lagged labor share is used.

155

Table 2.9: GMM Estimates of Industry Labor Productivity Growth
on Trade F lows-based Openness Variables: Colombia, 1970/1994
(p-values in parenthesis)

 

 

 

 

(1) (2) (3) (4)
Regression Speciﬁcation Levels Dif.-Lev. Levels Dif.-Lev.
Instruments (‘7 Levels Lev.-Dif. Levels Lev.-Dif.
Initial Labor Productivity -0.004 0.004 -0.023 0.0001
(0.27) (0.00) (0.00) (0.88)
Trade Intensity' 0.012 -0.001 -0.002 -0.0005
(0.00) (0.67) (0.00) (0.80)
Index of Inna-Industry -0.012 -0.004 -0.002 -0.014
Trade” (0.00) (0.01) (0.00) (0.00)
Growth of Exports -0.041 -0.064 -0.003 -0.014
(0.01) (0.00) (0.72) (0.10)
Growth in Import-Output -0.052 -0.029 -0.054 -0.025
Ratio Coefﬁcient‘ (0.00) (0.00) (0.00) (0.00)
Constant 0.021 0.233
(0.58) (0.00)
Growth in Employment -0.296 -0.403
(0.00) (0.00)
Change in the Labor Share in 32.717 20.142
Value Added (0.00) (0.04)
Change in the Labor Share in -46.l86 -30.501
Value Added times one (0.00) (0.00)
minus the Labor Share‘
Wald Test of Joint 0.00 0.00 0.00 0.00
Signiﬁcance: p-value
Sargan Test of 0.287 0.659 0.285 0.997
Overidentifying Restrictions:
p-value
Test for First-Order Serial 0.567 0.266 0.443 0.144
Correlation:
p-value
Test for Second-Order Serial 0.996 0.572 0.543 0.590
Correlation: p-value
Number of Observations 140 112 140 1 12

 

Notes: (‘) See text. (a) Log of the ratio of total trade to total output. (b) Log of one minus the ratio of the
absolute value of the trade deﬁcit to total trade. (c) Growth of imports minus growth of total output. ((1)
The average of the current and the lagged labor share is used.

156

Table 2.10: GMM Estimates of Industry Labor Productivity Growth
on Trade Flows-based Openness Variables: Mexico, 1970/1994
(p-values in parenthesis)

 

 

 

 

(1) (2) (3) (4)
Regression Speciﬁcation Levels Dif.-Lev. Levels Dif.-Lev.
Instruments 1') Levels Lev.-Dif. Levels Lev.-Dif.
Initial Labor Productivity -0.028 0.005 -0.026 0.002
(0.00) (0.00) (0.00) (0.00)
Trade Intensity' 0.013 0.0004 0.009 0.008
(0.00) (0.78) (0.00) (0.00)
Index of Intra-Industry 0.004 0.018 -0.0002 -0.001
T rade" (0.56) (0.00) (0.97) (0.82)
Growth of Exports -0.132 -0.091 -0.108 -0.059
(0.00) (0.00) (0.00) (0.00)
Growth in Import-Output 0.048 0.021 0.008 0.001
Ratio Coefﬁcientc (0.01) (0.00) (0.49) (0.92)
Constant 0.290 0.277
(0.00) (0.00)
Growth in Employment -0.256 -0.191
(0.02) (0.00)
Change in the Labor Share in 0.561 -2.857
Value Added (0.31) (0.00)
Change in the Labor Share in -0.246 4.819
Value Added times one (0.76) (0.00)
minus the Labor Shared
Wald Test of Joint 0.00 0.00 0.00 0.00
Signiﬁcance: p-value
Sargan Test of 0.319 0.909 0.433 1.000
Overidentifying Restrictions:
p-value
Test for First-Order Serial 0.071 0.010 0.260 0.013
Correlation:
p-value
Test for Second-Order Serial 0.639 0.079 0.905 0.068
Correlation: p-value
Number of Observations 135 108 135 108

 

Notes: (*) See text. (a) Log of the ratio of total trade to total output. (b) Log of one minus the ratio of the
absolute value of the trade deﬁcit to total trade. (c) Growth of imports minus growth of total output. (d)
The average of the current and the lagged labor share is used.

157

Table 2.11: Summary of Regression Results from Tables 2.6 to 2.10

 

 

 

 

 

 

 

Variables/ Countries where a signiﬁcant effect on Countries where a signiﬁcant effect on
Countries labor productivity growth was found total factor productivity growth was
found
Increases Reductions Increases Reductions
Trade Intensity Argentina -- Argentina Colombia
Brazil Brazil
Chile
Mexico
Export Growth Argentina Colombia Argentina Colombia
Brazil Mexico Brazil Mexico
Chile
Growth in Argentina Colombia Brazil Argentina
Import-Output Brazil Chile Colombia
Ratios Mexico
Index of Intra- Argentina Colombia Chile Argentina
Industry Trade Brazil Colombia
Mexico

 

 

 

 

 

Source: see text.

158

 

Table 2.12: GMM Estimates of Industry Labor Productivity Growth
on Trade Policy-based Openness Variables: Argentina, 1986/1993
(p-values in parenthesis)

 

 

 

 

(1) (2) (3) (4)
Regression Speciﬁcation Levels Dif.-Lev. Levels Dif-Lev.
Instruments (‘1 Levels Lev.-Dif. Levels Lev.-Dif.
Initial Labor Productivity 0.025 0.008 0.008 -0.003
(0.05) (0.10) (0.72) (0.50)
Average Tariffs 0.002 -0.0002 -0.0004 0.003
(0.45) (0.95) (0.86) (0.15)
Coverage of Non-Tariff -0.001 -0.0002 -0.00002 0.0004
Barriers (0.08) (0.83) (0.99) (0.53)
Constant -0.213 -0.028
(0.22) (0.88)
Growth in Employment -0.659 -0.101
(0.50) (0.68)
Change in the Labor Share in 0.200 11.009
Value Added (0.99) (0.00)
Change in the Labor Share in 0.090 -16.036
Value Added times one (0.96) (0.62)
minus the Labor Shared
Wald Test of Joint 0.074 0.000 0.001 0.000
Signiﬁcance: p-value
Sargan Test of 0.866 0.452 0.914 0.643
Overidentifying Restrictions:
p-value
Test for First-Order Serial 0.258 0.574 0.495 0.854
Correlation:
p-value
Test for Second-Order Serial 0.848 -- 0.062 --
Correlation: p—value
Number of Observations 51 34 51 34

 

Notes: (‘) See text. (a) The average of the current and the lagged labor share is used.

159

Table 2.13: GMM Estimates of Industry Labor Productivity Growth
on Trade Policy-based Openness Variables: Brazil, 1986/1994

(p-values in parenthesis)

 

 

 

 

(1) (2) (3) (4)

Regression Speciﬁcation Levels Dif.-Lev. Levels Dif.-Lev.
Instruments ‘2 Levels Lev.-Dif. Levels Lev.-Dif.
Initial Labor Productivity ~0.050 0.013 0.030 0.008

(0.09) (0.00) (0.05) (0.00)
Average Tariffs -0.004 -0.004 0.0002 -0.003

(0.00) (0.00) (0.71) (0.00)
Coverage of Non-Tariff 0.002 0.005 -0.002 0.005
Barriers (0.01) (0.00) (0. 10) (0.00)
Constant 0.626 -0. 184

(0.04) (0.19)
Growth in Employment 0.861 -0.780

(0.01) (0.00)

Change in the Labor Share in -3.128 -13.379
Value Added (0.90) (0.44)
Change in the Labor Share in 7.308 16.524
Value Added times one (0.81) (0.43)
minus the Labor Shared
Wald Test of Joint 0.00 0.00 0.00 0.00
Signiﬁcance: p-value
Sargan Test of 0.212 0.187 0.355 0.585
Overidentifying Restrictions:
p-value
Test for First-Order Serial 0.303 0.670 0.111 0.961
Correlation:
p-value
Test for Second-Order Serial 0.998 -- 0.685 --
Correlation: p-value
Number of Observations 51 34 51 34

 

Notes: (') See text. (a) The average of the current and the lagged labor share is used.

160

(p-values in parenthesis)

Table 2.14: GMM Estimates of Industry Labor Productivity Growth
on Trade Policy-based Openness Variables: Mexico, 1986/1994

 

 

 

 

(1) (2) (3) (4)
Regression Speciﬁcation Levels Dif.-Lev. Levels Dif.-Lev.
Instruments (‘1 Levels Lev.-Dif. Levels Lev.-Dif.
Initial Labor Productivity -0.061 0.022 -0.014 0.020
(0.08) (0.00) (0.45) (0.00)
Average Tariffs -0.022 -0.016 -0.005 -0.014
(0.01) (0.00) (0.02) (0.00)
Coverage of Non-Tariff -0.007 -0.003 -0.001 -0.001
Barriers (0.02) (0.08) (0.26) (0.06)
Constant 0.949 0.226
(0.02) (0.28)
Growth in Employment -0.619 -0.615
(0.01) (0.00)
Change in the Labor Share in 7.807 2.154
Value Added (0.00) (0.10)
Change in the Labor Share in -10.012 -2.521
Value Added times one (0.00) (0.16)
minus the Labor Share‘
Wald Test of Joint 0.046 0.001 0.000 0.000
Signiﬁcance: p-value
Sargan Test of 0.210 0.235 0.067 0.163
Overidentifying Restrictions:
p-value
Test for First-Order Serial 0.094 0.054 0.182 0.022
Correlation:
p-value
Test for Second-Order Serial 0.145 -- 0.479 --
Correlation: p-value
Number of Observations 51 34 51 34

 

Notes: (*) See text. (a) The average of the current and the lagged labor share is used.

161

Table 2.15: Summary of Regression Results from Tables 2.12 to 2.14

 

 

 

 

 

Variables/ Countries where a signiﬁcant effect on Countries where a signiﬁcant effect on
Countries labor productivity growth was found total factor productivity growth was
found

Increases Reductions Increases Reductions
Tariffs Brazil -- Brazil --
(reductions in) Mexico Mexico
Non-Tariffs Mexico Brazil Mexico Brazil
Barriers

(reductions in)

 

 

 

 

 

Source: see text.

162

 

(WM)

0

70717273747576777879808182838485868788899091929394
Souroozsootoxt.

Figure 2.1: Manufacturing Share in GDP, 1970/1994

163

.1 Argentina

+ Brasil
+ Chile
-I— Colombia

 

 

—l— Mexico

(1985 USS Dalian)

40000

 

 

35000
30000
25000
. _ ﬂ
+ Argentina
21] ii [1 D + Brasil
+ Chile
15000 + Colombia
. —-0— Mexico
10000
5000

 

0
70717273747576717879800182638485868708899091929394
Sourcezeeetexi

Figure 22: Labor Productivity in Manufacturing, 1970/1994

3

i

(1905 USS 00110")

S

 

o
70717273747576777679808182838485868788899091929394
Sourcezseetext

Figure 2.3:. Labor Productivity in Semi-Manufactures, 1970/1994

165

i

E

(1985 USS Dollars)

.a

 

A A_-_ar_-
v nluwlnuuu

+ Brazil
+ Chile
+ Colombia
:7 M09

 

 

 

 

70717273747578777879808182838435868708899091929394

Source: see text.
Figure 2.4: Labor Productivity in Other Traditional Industries

(excluding Semi-Manufactures), 1 970/ 1 994

166

§§§§§

(1985 US: Dollar!)

+ Argentina
+ Brazil
15000
+ Chile
+ Colombia
10000
:1— Mexlgo

 

§

 

O

70717273747576777879808182838485868788899091929394

Source: see text
Figure 2.5: Labor Productivity in Basic-Inputs Industries,

1970/1994

167

$3

320000

(1985

 

0
70717273747576777879808182838485868788899091929394
Source: see text.
Figure 2.6: Labor Productivity in New Labor-Intensive Industries,

1 970/ l 994

168

(1985 U3 Dollars)

.§§§§§§§§§

 

70717273747576777879808182839495368788990091929394
Sourcezaeetext.

Figure 2.7: Labor Productivity in New Capital-Intensive
Industries, 1970/1994

169

APPENDIX B

ADDITIONAL TABLES

170

Table 2.A.1: Classiﬁcation of Manufacturing Industries According to Factor Intensity

 

Industry Denomination Description

 

1. Traditional Industries

1.1. Semi-Manufactures ' Industrial products based upon natural resources

1.2. Other b Labor-intensive industrial products, not based upon natural resources
and with a low input of skilled labor

2. Basic Inputs Industries ° Capital-intensive industrial products, not based upon natural resources
and with a low input of skilled labor

3. New Industries

3.1. Labor-Intensive “ Labor-intensive industrial products not based upon natural resources
and with a high input of skilled labor
3.2. Capital-Intensive ° Capital-intensive industrial products not based upon natural resources

and with a high input of skilled labor

 

Source: see text.

‘ Food products, beverages, tobacco, wood products, pulp and paper products, industrial chemicals,
petroleum reﬁneries, and petroleum and coal products (ISIC groups 311, 313, 314, 331, 341, 351, 353,
354). " Textiles, wearing apparel, leather products, footwear, rubber products, pottery, china and
earthenware, glass and glass products, and other miscellaneous manufactures (ISIC groups 321, 322, 323,
324, 355, 361, 362, 390). ‘ Plastic products, other non-metallic mineral products, iron and steel, and non-
ferrous metals (ISIC groups 356, 369, 371, 372). " Furniture, printing and publishing, non-electrical and
electrical machinery (ISIC groups 332, 342, 382, 383). ° Pharmaceutical and other chemical products,
fabricated metal products, transport equipment, and professional and scientiﬁc equipment (ISIC groups
352, 381, 384, 385).

171

Table 2.A.2: Codes and Descriptions of Sectors at the 3-digit Level
of the International Standard Industrial Classiﬁcation

 

 

Code ISIC Description
31 1 Food products
313 Beverages
314 Tobacco
321 Textiles
322 Wearing Apparel
323 Leather Products (Except Footwear)
324 Footwear
331 Wood Products
332 Fumiture
341 Pulp and Paper Products
342 Printing and Publishing
351 Industrial Chemicals
352 Pharmaceutical and other Chemicals
353 Petroleum Reﬁneries
354 Miscellaneous Petroleum and Coal Products
355 Rubber Products
356 Plastic Products
361 Pottery, China, Earthenware
362 Glass Products
369 Other Non-Metallic Mineral Products
371 Iron and Steel
372 Non-Ferrous Metals
381 Fabricated Metal Products, Except Machinery and
Equipment
382 Machinery, Except Electrical
383 Electrical Machinery
384 Transport Equipment
385 Professional and Scientiﬁc Equipment
390 Other Manufacturing Industries

 

Source: see text.

172

Table 2.A.3: Codes and Descriptions of Sectors used in the Regressions
of Industry Labor Productivity Growth on Trade Policy-based Openness Variables
(Tables 11, 12 and 13)

 

 

Sector Number Code SITC Code ISIC Description

1 110 311 +313 +314 Food products, Beverages and
Tobacco

2 140 353 + 354 Mineral Fuels

3 150 372 Non Ferrous Metal Industries

4 210 351 + 352 Chemical Products

5 220 371 Iron and Steel

6 231 382 Non-Electrical Machinery

7 232 383 Electrical Machinery

8 233 384 Transport Equipment

9 241 323 Leather Products (except Footwear)

10 242 355 Rubber Manufactures

11 243 331 Wood Products

12 244 341 Pulp and Paper Products

13 245 321 + 322 Textiles and Clothing

14 246 361 + 362 + 369 Non-Metallic Mineral Products

15 247 332 Furniture

16 248 324 Leather Footwear

17 249 385 Professional, Scientiﬁc and

Controlling Equipment

 

Source: see text.

173

Table 2.A.4: Tariff and Non Tariff Barriers to Trade: Averages and Coefficients of
Variation across 17 Sectors("‘) in Argentina, Brasil and Mexico, 1986/1993

 

 

 

Country, Tariffs Non Tariff Barriers Weighted Weighted Non
Period Tariffs(") Tariffs Barriers(")
Average Coef. of Average Coef. of Average Coef. of Average Coef. of
Varia- Varia- Varia- Varia-
tion tion tion tion
Argentina
1987 29.5 18.3 38.6 76.2 27.7 21.0 36.4 84.0
1990 21.2 8.8 12.5 228.1 21.0 9.2 13.9 208.4
1993 11.7 39.4 0.4 406.3 11.7 41.2 2.1 318.6
Brazil
1986 59.4 37.2 37.8 79.4 56.6 38.5 40.2 76.6
1990 31.6 40.0 4.7 233.6 31.1 47 .1 10.3 222.4
1993 13.7 40.9 3.0 374.9 14.2 36.1 5.6 402.7
Mexico
1987 12.3 34.6 9.0 133.0 11.6 40.7 16.4 137.2
1990 11.6 36.5 6.2 224.5 1 1.4 41.9 12.5 182.2
1992 13.8 20.8 3.7 155.4 13.6 22.9 10.1 184.3

 

 

 

 

 

Source: see text.

Notes: (‘) See Table 2.A.2 for a description of the sectors; (") Weights are based on the 1985 imports of
a group of 120 developing countries.

174

Table 2.A.5: Bivariate Correlations between Tariffs, Non-Tariff Barriers to Trade, and
the Growth of Exports and Imports: Argentina, 1986/1993 (*)

 

 

 

 

 

 

 

Tariffs Non-Tariff Weighted Weighted Growth in
Barriers Tariffs NTBs Imports
(NTBS)
Non Tariff 0.65
Barriers
(NTBs)
Weighted 0.98 0.63
Tariffs
Weighted 0.60 0.97 0.60
NTBs
Growth in -0.23 -0.34 -0.22 -0.33
Imports
Growth in 0.33 0.53 0.36 0.55 -0.20
Exports

 

 

 

 

 

 

Source: see text.
(*) 17 sectors, 3-year periods.

175

 

Table 2.A.6: Bivariate Correlations between Tariffs, Non-Tariff Barriers to Trade, and

the Growth of Exports and Imports: Brazil, 1986/1993 (*)

 

 

 

 

 

 

 

Tariffs Non-Tariff Weighted Weighted Growth in

Barriers Tariffs NTBs Imports
(NTBS)

Non Tariff 0.34

Barriers

(NTBS)

Weighted 0.98 0.33

Tariffs

Weighted 0.13 0.89 0.14

NTBs

Growth in -0.02 0.02 -0.02 -0.02

Imports

Growth in 0.10 0.14 0.09 0.01 0.06

Exports

 

 

 

 

 

 

Source: see text.
(*) l7 sectors, 3-year periods.

176

 

Table 2.A.7: Bivariate Correlations between Tariffs, Non-Tariff Barriers to Trade, and

the Grth of Exports and Imports: Mexico, 1986/1993 (*)

 

 

 

 

 

 

 

Tariffs Non- Tariff Weighted Weighted Growth in

Barriers Tariffs NTBs Imports
(NTBS)

Non Tariff -0.27

Barriers

(NTBS)

Weighted 0.93 -0.28

Tariffs

Weighted -0.47 0.77 -0.52

NTBS

Growth in 0.50 -0.12 0.46 -0.21

Imports

Growth in 0.41 -0.24 0.40 -0.39 0.20

Exports

 

 

 

 

 

 

Source: see text.

(‘) 17 sectors, 3-year periods.

177

 

Chapter 3:

WHAT CAUSES VIOLENT CRIME?

l - Introduction

A growing concern across the world is the heightened incidence of criminal and
violent behavior. Rampant criminal behavior is a major concern in a variety of countries,
ranging from the United States to the so-called transition economies of Eastern Europe
and the developing countries in Sub-Saharan Africa and Latin America and the
Caribbean.1 A recent paper on the topic states that, “Crime and violence have emerged in
recent years as major obstacles to the realization of development objectives in the
countries of Latin America and the Caribbean” (World Bank 1997, abstract). In fact,
crime rates for the world as whole have been rising Since the mid-19708, as illustrated in
Figure 3.2. The grong public awareness is justiﬁed because rampant crime and
violence may have pernicious effects on economic activity and, more generally, because
they directly reduce the quality of life of all citizens who must cope with the reduced
sense of personal and proprietary security. Despite the fact that violent crime is emerging
as a priority in national policy agendas worldwide, we actually do not know what are the
economic, social, institutional, and cultural factors that make some countries have higher

crime rates than others over time.

 

' See, for example, the attention given to the “rising crime wave” in Latin America and
the Caribbean in “Law and Order,” Latin Trade, June 1997, “Mexico City Crime Alarms
Multinationals,” The Wall Street Journal, October 29, 1996, p. A18, and “Reform
Backlash in Latin America,” The Economist, November 30-December 6, 1996.

 

 

178

At least since the publication of Becker (1968), the economics profession has
analyzed the determinants of criminal behavior from theoretical and empirical points of
view. Most empirical studies have addressed the issues associated with criminal behavior
within cities and across regions within countries, especially the United States; yet very
few empirical studies have addressed the question of why crime rates vary across
countries and over time. This paper is an attempt to ﬁll this vacuum in the economics

literature.

We assembled a new data set of crime rates for a large sample of countries for the
period 1970-1994, based on information from the United Nations World Crime Surveys.
Then, we propose a simple model of the incentives faced by individuals to commit
crimes, and explicitly consider possible causes of the persistence of crime over time
(criminal inertia). The empirical implementation of the model estimates several
econometric models attempting to capture the determinants of crime rates across
countries and over time. The empirical models are ﬁrst run for cross-sections and then
applied to panel data. Working with panel data (that is, pooled cross-country and time-
series data) allows us to consider both the effect of the business cycle (i.e., GDP growth
rate) on the crime rate and the presence of criminal inertia (accounted for by the inclusion
of the lagged crime rate as an explanatory variable). Furthermore, the use of panel data
techniques will allow us to account for unobserved country-speciﬁc effects, for the likely
joint endogeneity of some of the explanatory variables, and for the existence of some

types of measurement errors afflicting the data of reported crimes.

179

Some of the interesting results are the following: Greater inequality is associated
with higher intentional homicide and robbery rates, but the level of income per capita is
not a signiﬁcant determinant of national crime rates. “Deterrence” effects are also shown
to be signiﬁcant. Contrary to our expectations, national enrollment rates in secondary
education and the average number of years of schooling of the population appear to be
positively (but weakly) associated with higher homicide rates. We address this puzzle
(also found in other empirical studies) when the regression results are presented. Drug
production and drug possession are both Signiﬁcantly associated with higher crime rates.
Regarding dynamic effects, we ﬁnd that the homicide rate rises during periods of low
economic activity. Also, we ﬁnd that crime tends to persist over time (criminal inertia),
even after controlling for other determinants of criminal behavior. All these results are
robust to models that take into account the likely joint endogeneity of the explanatory
variables, the presence of country-speciﬁc effects, and certain types of measurement

errors in reported crime rates.

The rest of the paper is organized as follows: Section 11 provides a selective
review of theoretical and empirical contributions to the economics literature dealing with
criminal behavior. Section III presents a simple economic model of criminal behavior that
begins with an individual-level analysis of the incentives to commit crimes, and then
considers time effects. Under a couple of assumptions, the model provides a framework
to analyze the empirical determinants of national crime rates. Section IV presents the data
sets used in the econometric estimation, describing the sources of the data as well as its

basic statistical characteristics. Section V presents the econometric models used for

180

estimating the impact of selected variables on national crime rates, and interprets the
results of each econometric exercise. Section VI presents the conclusions of the paper and

suggests future directions for research.

2 - Literature Review

In his Nobel lecture, Becker (1993, 390) emphasized that the economic way of
looking at human behavior “implie[s] that some individuals become criminals because of
the ﬁnancial and other rewards from crime compared to legal work, taking account of the
likelihood of apprehension and conviction, and the severity of punishment.” More recent
literature has emphasized the role of time effects and criminal inertia that may result from
social interactions, or delayed responses to surges in criminal activity on the part of

police and judicial systems.

The theoretical and empirical literature has considered the role of three types of
economic conditions in determining the incidence of criminal activity, namely the
average income of the communities involved, the pattern of income distribution, and the
level of education. Fleisher (1966) was a pioneer in studying the role of income on the
decision to commit criminal acts by individuals, and stated that the “principal theoretical
reason for believing that low income increases the tendency to commit crime is that the
probable cost of getting caught is relatively low... because [low-income individuals] view
their legitimate lifetime earning prospects dismally they may expect to lose relatively
little earning potential by acquiring criminal records; furthermore, if legitimate earnings

are low, the opportunity cost of time actually spent in delinquent activity, or in jail, is

181

also low” (F leisher 1966, 120). However, the level of legal income expected by an

individual is not the only relevant “income” factor; the income level of potential victims
also matters. The higher the level of income of potential victims, the higher the incentive
to commit crimes, especially crimes against property. Thus, according to Fleisher (1966,
121), “[average] income has two conceptual inﬂuences on delinquency which operate in

opposite directions, although they are not necessarily equal in strength.”

Fleisher’s (1966, 128-129) econometric results showed that higher average family
incomes across 101 US. cities in 1960 were actually associated with lower court
appearances by young males, and with lower numbers of arrests of young males for the
crimes of robbery, burglary, larceny, and auto theft. 2 The author also found that the
difference between the average income of the second lowest quartile and the highest
quartile of households tended to increase city arrest and court-appearance rates, but the
coefﬁcient was often small in magnitude, and became statistically insigniﬁcant when the

regressions were run for high-income communities alone.

The effects of income levels and distribution on crime were further analyzed by
Ehrlich (1973, 53 8-540). He argued that payoffs to crime, especially property crime,
depend primarily on the “opportunities provided by potential victims of crime,” as
measured by the median income of the families in a given community. The author

assumed that, “the mean legitimate opportunities available to potential offenders,” may

 

2 The crime-reducing effect of income appeared robust to various regression
speciﬁcations that controlled for preferences or tastes of different communities. For
example, average family incomes tended to reduce crime even when taking into account
the shares of the local young populations that were composed of African-Americans,
divorced or single mothers, and irnrrrigrants.

182

be approximated by, “the mean income level of those below the state’s median [income]”
(p. 539). For a given median income, income inequality can be an indication of the
differential between the payoffs of legal and illegal activities. In his econometric analysis
of the determinants of state crime rates in the US. in 1960, Ehrlich (1973, 546-551)
found that higher median family incomes were associated with higher rates of murder,
rape, and assault, and with higher rates of property crimes, such as burglary. In addition, a
measure of income inequality — the percentage of families below one-half of the median
income — was also associated with higher crime rates. The former ﬁnding contradicts
Fleisher (1966), but the latter ﬁnding on the role of income inequality supports F leisher’s
ﬁndings that inequality is associated with higher crime rates. Both Fleisher (1966, 136)
and Ehrlich (1973, 555) considered the effect of unemployment on crime rates, viewing
the unemployment rate in a community as a complementary indicator of income
opportunities available in the legal labor market. 3 In their empirical studies, however,
both authors ﬁnd that unemployment rates were less important determinants of crime

rates than income levels and distribution.

Another important factor related to the effect of economic conditions on crime is
the level of education of the population, which can determine the expected rewards from
both legal and criminal activities. In addition, Usher (1993) has argued that education
may also have a “civilization” effect, tending to reduce the incidence of criminal activity.

However, after controlling for income inequality and median income, Ehrlich (1975a,

 

3 In the words of F leisher (1966, 121), “in attempting to estimate the effect of income on
delinquency, it is important to consider the effects of both normal family incomes and
deviations from normal due to unemployment.”

183

333) found a positive and Signiﬁcant relationship between the average number of school
years completed by the adult population (over 25 years) and particularly property crimes
committed across the US. in 1960. Four possible explanations of this puzzling empirical
ﬁnding were provided by the author. First, it is possible that education may raise the
marginal product of labor in the crime industry to a greater extent than for legitimate
economic pursuits (Ehrlich 1975a, 319). Second, higher average levels of education may
be associated with less under-reporting of crimes (Ehrlich 1975a, 333). Third, it is
possible that education indicators act as a “surrogate for the average permanent income in
the population, thus reﬂecting potential gains to be had from crime, especially property
crimes” (Ehrlich 1975a, 333). Finally, combined with the observation that income
inequality raises crime rates, it is possible to infer that certain crime rates are “directly

related to inequalities in schooling and on-the-job training” (Ehrlich 1975a, 335).

Together with the relationship between economic conditions and crime, one of the
main issues in the pioneering studies of Becker (1968) and Ehrlich (1973, 1975b, 1981)
was the assessment of the effects of police presence, convictions, and the severity of
punishments on the level of criminal activity. Individuals who are considering whether to
commit crimes are assumed to evaluate both the risk of being caught and the associated
punishment. The empirical evidence ﬁ'om the United States conﬁrmed that both factors

have a negative effect on crime rates — see Ehrlich (1973, 545, and 1996, 55).

Analysts often make a subtle distinction between the “deterrent” effects of
policing and convictions and the “incapacitation” effects of locking-up (or killing, in the

case of capital punishment) criminals who may have a tendency to rejoin the crime

184

industry once they are released. As stated by Ehrlich (1981, 311), “deterrence essentially
aims at modifying the ‘price of crime’ for all offenders,” while incapacitation — and for
that matter, rehabilitation — acts through the removal of, “a subset of convicted offenders
from the market for offenses either by relocating them in legitimate labor markets, or by
excluding them from the social scene for prescribed periods of time.” The author showed
that, in theory, the effectiveness of rehabilitation and incapacitation, vis-a-vis the purely
deterrent approach to crime control, depends on the rate of recidivism of offenders, and
on their responsiveness to economic incentives — i.e., changes in the “price of crime.”4
For example, the relatively higher rates of recidivism observed for property crimes — in
comparison to violent crimes (Leung 1995, 66) — may imply that incapacitation and/or
rehabilitation are more appropriate means for controlling these types of crime than
deterrence policies. However, if property offenders respond readily to economic

incentives, the argument would be the opposite.

Since most forms of punishment that incapacitate offenders also involve deterrent
effects -— e. g. imprisonment — it is often difficult to evaluate empirically the importance of
each type of action. Using estimates based on regression results for the US. states in
1960, Ehrlich (1981) concluded that, “in practice the overwhehning portion of the total

preventive effect of imprisonment is attributable to its pure deterrent effect.” Moreover,

 

‘ Ehrlich (1981, 311-313) showed that the reduction in crime that follows the
rehabilitation and/or incapacitation of past offenders is partially compensated by the entry
or reentry of new offenders into the market, attracted by the temporary increase in the
returns from crime that follows the departure of individual offenders. The author
demonstrates that while the aggregate response of the supply of offenses to the removal
of past offenders - through incapacitation or rehabilitation — decreases with the elasticity

185

Ehrlich (1975b) found that capital punishment provisions across the US. tended to
reduce crime rates primarily through their deterrent effect, rather than through their
incapacitation effect. Levitt (1995) addressed these issues jointly with one of the most
recurrent problems in the aforementioned literature; namely, the author attempts to assess
whether the seemingly negative relationship between crime rates and arrest rates were the
product of deterrence effects, incapacitation, or measurement errors associated with the
fact that crime tends to go unreported.’ The author ﬁnds that most of this negative
relationship in the US. is due to deterrence effects, and not measurement error or

incapacitation, for most types of crime.

Another important consideration for assessing the effectiveness of deterrence is
the individual’s attitude towards risk, because an individual’s expected utility ﬁ'om illegal
income will be affected by his/her tastes for the risk involved. Becker (1968, 178) and
Ehrlich (1973, 528), for example, established that a risk-neutral offender will tend to
spend more time in criminal activity than a risk-averse individual. Another implication of
assuming risk-aversion is that raising the probability of conviction may have a greater

deterrent effect than raising the severity of punishment (Becker 1968, 178).

 

of this function (with respect to the return from offenses), the efﬁcacy of general
deterrence increases with this elasticity.

’ The crime rate is the number of crimes over population, while the arrest (or conviction)
rate is the number of arrests (convictions) over the number of reported crimes. Therefore,
it is possible that a negative relationship may exist between these two variables simply
because under-reporting would produce a downward bias in the crime rate, while raising
the arrest (conviction) rate. However, the relationship between these variables may be
more complex, because the number of arrests (convictions) also depends on the number
of reported crimes.

186

Some recent contributions to the theoretical literature consider the possible
endogeneity of the perceived probability of punishment of offenders, and emphasize that
the timing of the rewards and punishments from crime will affect the incidence of
criminal activity over time. Davis (1988), for example, considers a model where the
probability of a criminal being caught at any point in time is positively related to both the
intensity of the individual’s criminal activity, and to the rate of enforcement at that point
in time. The author stresses that this probability is a component of the rate used by
potential offenders to discount future streams of income from illegal activities, and
derives optimal crime rates for given levels of punishment and rates of enforcement.
Leung (1995) extends this type of model by considering an inﬁnite time horizon, and by
introducing recidivism into the analysis. The author allows the individual’s number of
previous convictions to affect the probability of a new conviction when a past offender
commits new crimes, as well as the severity of the corresponding punishment. In Leung’s
(1995) model, past criminal records also reduce the returns from engaging in legal
activities, both through stigma and human capital effects. The latter are associated with
the depreciation of past skills and the foregoing of new investments in education during

the period spent on illegal activities or in jail.

Sah (1991) studied a different relationship between the intensity of crime rates
over time and the probability of apprehension. The author argued that individuals living
in areas with high crime-participation rates can perceive a lower probability of
apprehension than those living in areas with low crime-participation rates, because the

resources spent in apprehending each criminal tend to be low in high crime areas. An

187

important implication of this analysis is that “past crime breeds future crime” (Sah 1991,
1282). In a similar analysis, Posada (1994) presented a simple model where a random
increase in crime rates can result in a permanent increase in the crime rate, when the
increase in crime is not compensated by a proportional increase in the resources spent in
the detection and punishment of crimes, which results in a lower perceived rate of

apprehension.

Glaeser, Sacerdote, and Scheinkman (1996) emphasized the role of local social
interactions in determining crime rates in US. cities. In contrast to Sah (1991) and
Posada (1994), who emphasized the effect of what we call “systemic” interactions (i.e.,
an individual’s perceived probability of apprehension depends on society’s crime crate),
Glaeser et a1. (1996) argued that “local” interactions among individuals act through the
transfer of information between agents regarding, “criminal techniques and the returns to
crime, or interactions result from the inputs of family members and peers that determine
the costs of crime or the taste for crime (i.e., family values), and monitoring by close
neighbors” (Glaeser, et al. 1996, 512). A notable implication of the local interactions
approach is that crime rates across communities need not converge. For the purposes of
this paper, the implication of systemic and local interactions is that countries may

experience criminal inertia over time.

In sum, the economics literature on crime has transited from an emphasis on
economic conditions (including education) and deterrence effects to more recent

considerations of factors that may explain how crime is propagated over time and within

188

communities. In the following section we attempt to organize some of the ideas addressed

in the literature in a simple framework.

3- A Simple, Reduced-Form Model of Criminal Behavior

We now present a simple model of criminal behavior that may help us organize
ideas and motivate the variables postulated as determinants of crime rates in the empirical
section of the paper.6 We ﬁrst model criminal behavior from the perspective of the
individual and then aggregate to the national level to obtain a reduced-form equation of
the causes of national crime rates.

The basic assumptions are that potential criminals act rationally, basing their
decision to commit a crime on an analysis of the costs and beneﬁts associated with a
particular criminal act. Furthermore, we assume that individuals are risk neutral, and
respond to changes in the probability of apprehension and the severity of punishment.
Thus, individuals will commit a crime whenever its expected net beneﬁts are large
enough. Equation (1) below says that, for a particular individual, the expected net beneﬁt
(nb) of committing a crime is equal to its expected payoff (that is, the probability of not
being apprehended (1 -pr) times the loot 1), minus the total costs associated with planning
and executing the crime (6'), minus the foregone wages from legitimate activities (w),
minus the expected punishment for the committed crime (pr *pu):7

nb = (1-pr)*l - c - w - pr‘pu (l)

 

6 For a comprehensive survey of models of criminal behavior, see Schmidt and Witte
(1984, 165-182).

189

Assuming that individuals have some “moral values,” the expected net beneﬁts of
a crime would have to exceed a certain threshold before she/he commits a crime. The
individual’s threshold would be determined by her/his moral stance (m), to which we can
assign a pecuniary value to make it comparable to the other variables in the model.
Equation (2) establishes this relationship between the decision to commit a crime and the
net beneﬁts of such behavior:

d = 1 when nb 2 m

d = 0 when nb <m (2)
where d stands for the decision to commit the crime (d = l) or not to commit the crime
(d = 0).

In the empirical section of the paper, we estimate a model in which the dependent
variable is the national crime rate and the explanatory variables are a number of national
economic and social characteristics. We ﬁrst link those characteristics with the elements
entering the individual decision to commit a crime. Then, we aggregate over individuals
in a nation to obtain a reduced-form expression for the country’s crime rate in terms of
the underlying socio-economic variables. (Figure 3.1 summarizes the discussion below.)

The ﬁrst underlying variable is individual education (e), which may impact on the
decision to commit a crime through several channels. Higher levels of educational
attainment may be associated with higher expected legal earnings, thus raising w. Also,

education, through its civic component, may increase the individual’s moral stance, m.

 

7 Lower-case letters represent the variables related to a particular individual (not
necessarily a representative individual in society). Upper-case letters represent society’s
averages for the respective variables.

190

On the other hand, education may reduce the costs of committing crimes (i.e., reducing c)
or may raise the crime’s loot, I, because education may open opportunities for an
individual to enter higher-paying crime industries. Hence the net effect of education on
the individual’s decision to commit a crime is, a priori, ambiguous. We can conjecture,
however, that if legal economic activities are more skill- or education-intensive than
illegal activities, then it is more likely that education will induce individuals not to
commit crimes. In addition, following Tauchen and Witte (1994), it is possible that
school enrollment alone (independently of the level of educational attainment) will
reduce the time available for participating in the crime industry. Hence, the empirical
section will assess the effect of both secondary enrollment rates and educational
attainment on crime rates.

The individual’s past experience in criminal activities (d1. 1) is another important
underlying variable that affects in several ways the decision to commit a crime. First,
convicts tend to be stigmatized in the legal labor market, thus having diminished
employment opportunities and expected income (lower w). Second, criminals can learn
by doing, which means that the costs of carrying out criminal acts, c, may decline over
time. Third, people tend to have a reduced moral threshold, m, after having joined the
crime industry. The past incidence of crime in society (Dr. 1), through the local social
interactions noted in our literature survey, also affect the individual’s decision by both
reducing the costs of carrying out criminal activities (lower c) and impairing civic moral
values (lower m). These arguments strongly suggest the possibility of criminal inertia,

that is, present crime incidence explained to some extent by its past incidence.

191

The level and growth of economic activity (EA) in society create attractive
opportunities for employment in the legal sector (higher w) but, since they also improve
the wealth of other members of society, the size of the potential loot from crime, I, also
rises. Therefore, the effect of heightened economic activity on the individual’s decision to
commit a crime is, in principle, ambiguous. The effect of income inequality in society
(INEQ) will depend on the individual’s relative income position. It is likely that in the
case of the rich, an increase in inequality will not induce them to commit more crimes.
However, in the case of the poor, an increase in inequality may be crime inducing,
because such an increase implies a larger gap between the poor’s wages and those of the
rich, thus reﬂecting a larger difference between the income from criminal and legal
activities (higher l-w). A rise in inequality may also have a crime-inducing effect by
reducing the individual’s moral threshold (lower m) through what we could call an “envy
effect”. Therefore, a rise inequality will have a positive impact on (at least some)
individuals’ propensity to commit a crime.

The existence of proﬁtable criminal activities (DRUGS) in some countries means
that the expected loot from crime is larger in those countries than in others. The most
important example of proﬁtable criminal activities is the illicit drug trade (other two are
gambling and prostitution). Countries where the raw materials for illicit drugs are easily
obtained (such as Colombia, Bolivia, and Peru in the case of cocaine) or countries that are
located close to high drug consumption centers (such as Mexico in relation to the United
States) have frequent and highly proﬁtable opportunities for criminal activities. These
activities not only consist of drug production and trade themselves, but also involve the

element of violence and ofﬁcial corruption required for them to carry on.

192

The strength of the police and the judicial system (JUST) increases the probability
of apprehension (pr) and the punishment (pa) for criminal actions, thus reducing the
incentive for an individual to commit a crime. This is the crime deterrence effect. It
should also be noted that the past incidence of crime in society (D1. 1) may determine an
individual’s perceived probability of apprehension (pr) via systemic interactions, as
discussed above.

Finally, there are other factors that may affect an individual’s propensity to
commit crimes (other) such as cultural characteristics (religion and colonial heritage, for
example), age and sex (young males are said to be more violent prone than the rest of the
population), the availability of ﬁre arms in the country, and the population density where
the individual lives (urban centers would facilitate the social interaction through which
crime skills are transmitted). These other factors can affect the individual’s decision to
commit a crime mainly through the cost of planning and executing the crime (c) and
through his/her moral threshold (m).

Considering the effects summarized in Figure 3.1, and substituting them into
equations (1) and (2), we have that a given individual will commit a crime (d=1) if the

following inequality (3) holds:

I(e, EA, INEQ, DRUGS, JUST) - co, d.-., banana»)

- w (e, (7.4, EA) - pr (JUST) * pu - m (e, 61,4, DH, INEQ,0t/ier)2 0

Rewriting this condition as a function f of the underlying individual and social variables,

we obtain the following reduced-form expression:

d=1 if f(e, al.-.. D..., E’A, INEQ, DRUGS, JUST, otlier) 2 0

193

¢>d=1iff(‘1’)20 (4)
where ‘1’ is a vector of the underlying determinants of crime. Assuming both a linear
probability model for the decision to commit a crime and a linear functional form for f:
we obtain the following individual regression equation:

d = [3' ‘1’ + p (5)
The assumption of linearity in both the functional form of f and the probability model are,
of course, arbitrary. They are chosen because they allow the aggregation of equation (5).
Given that our data is not individual but national, our regression equation must be
speciﬁed in terms of national rates, which is obtained by averaging equation (5) over all
individuals in a country and over a given time period,

Dt = B ‘i’r + Vt (6)
That is,

Crime Rate“ = ,80 + ,6] EDUC i, t + ﬂz Lagged crime rate ,3, + ,63 EA 5,; +

.64 DRUGS 2;: + .65 JUST i,t + .86 OTHER it + m + 51",! (7)
where the subscripts i and t represent country and time period, respectively; and 0 is an

unobserved country-speciﬁc effect.

4 - The Data

A full description of the variables (and their sources) used in this paper is
presented in the Appendix. Curious readers are urged to examine the descriptions and
tables included therein. This section brieﬂy describes the data used to calculate the

national crime rates and the set of explanatory variables.

194

4.1 - National Crime Rates

The empirical implementation of the theoretical model proposed above will rely on
crime rates, which were based on the number of crimes reported by national justice
ministries to the United Nations World Crime Surveys. The econometric analysis will
focus on the determinants of “intentional homicide,” and robbery rates between 1970 and
1994.8 All crime rates are expressed as the number of reported crimes in each category
per 100,000 inhabitants. AS shown in Table 3.1, there is a considerable variation in the
crime-related variables. However, it is worth noting that most countries did not report

data for the entire period nor for every type of crime.

Figure 3.2 shows the evolution of the population-weighted average rate of
intentional homicides in the group of 34 countries for which there was data available in
each 5-year sub-period. As mentioned in the introduction, the world’s intentional
homicide rate has been increasing steadily, at least since the early 19803, with a notable
acceleration during recent years. Figures 3.3 and 3.4 show the evolution of the median
intentional homicide rate in each ﬁve-year period for the whole sample of countries,
while separating groups of countries by income levels and regions. We use the median
rate to describe the evolution of homicide rates because this measure is less sensitive to
the inﬂuence of outliers than the mean rate. Figure 3.3 shows that much of the increase

was due to increases in the median homicide rates of middle-low and low-income

 

8 Drug possession crime rates and the lagged values of the intentional homicide and
robbery rates were also used as explanatory variables. “Total” homicide statistics were
collected for this project, but were not used in the econometric analysis because we feared
that this broader deﬁnition of “homicide” was subject to more deﬁnition differences
across countries than “intentional” homicide.

195

countries (where the former had a GNP per capita ranging from $766 US dollars in 1995
to $3,035, and the latter had an income per capita of $765 or less). Figure 3.5 shows that
the highest homicide rates are found in Latin America and the Caribbean, followed by
Sub-Saharan Aﬁica. In these regions, and in the developing countries of Europe and
Central Asia, considerable increases in intentional homicide rates have been observed in
the early nineties. However, it should be noted that the sample of Sub-Saharan African
countries is quite small and varies across sub-periods, thus the evolution of the median
rate for this group may reﬂect the inclusion of outliers in the latter two periods (e.g.,
Swaziland and Sao Tome & Principe have high crime rates, but we only have data for the

last two periods).

Figures 3.5 and 3.6 show the evolution of intentional homicide rates in South America
and Mexico, and in Central America and the Caribbean, respectively.’ Regarding Figure
3.5, it is interesting to note that only Argentina and Chile experienced a decline in their
homicide rates since the early 19703, when both countries faced severe economic and
political crises. Colombia experienced the most noticeable increase in the homicide rate,
jumping from an average of approximately 16 intentional homicides per 100,000
inhabitants during 1970-1974 to over 80 in 1990-1994, possibly reﬂecting the rise of the
drug trafﬁcking industry in that country. Figure 3.6 shows that several small economies,

such as Bahamas, Jamaica, Nicaragua, and El Salvador, have had higher intentional

 

9 The homicide rates for Argentina, Brazil, Colombia, Mexico, and Venezuela were
constructed from data provided by the Health Situation Analysis Program of the Division
of Health and Human Development, Pan-American Health Organization, from the PAHO
Technical Information System. This source provided us with data on the annual number
of deaths attributed to homicides, which come ﬁ'om national vital statistics systems.

196

homicide rates than most large Latin American countries. All of these countries have
experienced rates in excess of 20 intentional homicides per 100,000 population.
Furthermore, Bahamas, Barbados, Jamaica, and Trinidad and Tobago have experienced
considerable increases in their crime rates since the early I970s. Of the small countries,
only Costa Rica has experienced a steady decline of its intentional homicide rate. Thus,
the rise in the overall homicide rate in Latin America and the Caribbean can be attributed
to an upward trend in criminal activity in most countries of the region (with a few
exceptions such as Argentina, Chile, and Costa Rica), with a few outliers that have

experienced dramatic increases in criminal activity (Bahamas, Jamaica, and Colombia).

4.2 - Explanatory Variables

Following the simple model presented in the previous section, we have selected a set
of explanatory variables that proxy for the main economic determinants of crime rates, as
well as for some of the non-pecuniary factors that may affect the decision to perform

illegal activities.

As a proxy of the average income of the countries involved in our econometric
study, we use the Gross National Product (GNP) per capita, in prices of 1987. The ﬁgures
were converted to US. dollars on the basis of the methodology proposed by Loayza et a1.

(1998), which is based on an average of real exchange rates.lo In the regressions that are

 

'° Most of the data was provided by Loayza et al. (1998). For some countries not covered
by these authors, however, the conversion factors were constructed on the basis of
information from World Bank databases.

197

based on both cross-sectional and time-series data, we also used the rate of grth of

GDP, calculated on the basis of ﬁgures expressed in 1987 prices (in local currency).

The degree of income inequality was measured by the Gini index and by the
percentage of the national income received by the lowest quintile of a country’s income.
Both variables were constructed on the basis of the data set provided by Deininger and
Squire (1996); we used what these authors have termed “high quality” data for the
countries and years for which it was available, and otherwise calculated an average of
alternative ﬁgures (also provided by Deininger and Squire, 1996). The Gini coefﬁcients
which were originally based on expenditure information were adjusted to ensure their

comparability with the coefﬁcients based on income data. ”

Two educational variables were used, as measures of the stock and the ﬂow of
investment in human capital in a given country. These are, respectively, the average years
of schooling of the population over 15 years of age, as calculated by Barro and Lee
(1996), and the secondary enrollment rate, which was taken from World Bank databases,
and is deﬁned as the number of people (of all ages) enrolled in secondary schools,

expressed as a percentage of the total population of secondary school age.12

Another type of economic incentive to commit crime that we considered was the
existence of proﬁtable criminal “industries”. In particular, we focused on the existence, in

a given country, of considerable production and/or distribution of illegal drugs. The

 

” We followed, in this respect, the suggestion of Deininger and Squire (1996, 582) of
adding to the indices based on expenditure the average difference of 6.6 between
expenditure-based and income-based coefﬁcients.

198

choice of this particular crime industry was motivated not only by the fact that the drug
trade is known to be highly proﬁtable but also because, at least in some countries — e.g.
the US. — it is also known to use a very “violence-intensive” technology. The latter
aspect of this industry, and the intellectual and moral decay associated with the
consumption of the substances in question, can be expected to generate externalities for
the proliferation of other violent crimes. We used two speciﬁc variables as measures of
the size of the illegal drug industry. The ﬁrst was the number of drug possession offenses
per 100,000 population, which we calculated on the basis of data from the United
Nations’ Crime Surveys. It is worth noting that this variable does not measure the extent
of actual drug consumption in a given country, but only the ﬁaction of that ﬁgure that is
considered illegal in the country’s legislation, and that has been detected by the law
enforcement agencies. Thus, the variable in question reﬂects not only the size of the drug-
consurning population, but also the degree of tolerance for drug consumption in the
corresponding society. The second measure that we used is a “dummy” variable that takes
the value one when a country is listed as a signiﬁcant producer of any illegal drug in any
of the issues of the US. Department of State ’8 International Narcotics Control Strategy

Report - which has been published on an annual basis since 1986.

Regarding the negative incentives to commit crime, we used several variables to
proxy for the probability of being caught and convicted when performing an illegal
activity, and for the corresponding severity of the punishments. To capture the ﬁrst

component of the crime deterrence efforts of a given society, we used both the number of

 

'2 “Net” enrollment rates (the fraction of people of secondary-school age who are enrolled

199

police personnel per 100,000 inhabitants, and the conviction rate of the corresponding
crime, deﬁned as the ratio of the number of convictions to the number of reported
occurrences of each type of crime. Table 3.1 shows summary statistics for both variables,
which were constructed on the basis of data provided by the United Nations, in its World
Crime Surveys. ‘3 We also collected information provided by Amnesty International
about the existence of the death penalty in countries across the globe, which we use as an

indicator of the severity of punishments.

Other determinants of the intensity of criminal activity highlighted by the
theoretical model presented above include factors that reduce both the pecuniary and the
non-pecuniary cost of engaging in illegal activities. These factors may act by facilitating
the development of social interactions between criminals and would-be criminals.
Assuming that these interactions are more prevalent in urban agglomerations than in rural
areas, we use the rate of urbanization as a possible factor in explaining crime rates across
nations. We also include in our empirical exercise the proportion of the total population
encompassed by males belonging to the 15-29 age group, which is -— at least in the US. —

the demographic group to which most criminals belong.

The taste or preference for criminal activity may also be inﬂuenced by cultural
characteristics of the countries involved. As countries with common cultural traits may

also share similar economic characteristics, it is important to control for the former in

 

in secondary school) are not available for a large number of developing countries.

'3 The conviction rates reported in Table 3.1 are ﬁve-year averages, rather than annual
observations. The averages provide better descriptions of the convictions rates because
reported convictions are often associated with crimes committed in previous years, but

200

order to obtain an accurate appraisal of the effect of the latter on the determination of
national crime rates. With this end in mind, we employed religion and regional
“dummies” in our cross-sectional regressions. The ﬁrst set of variables — related to
Buddhist, Christian, Hindu, and Muslim countries — was constructed on the basis of
information from the CIA F actbook, and each variable takes the value one for the
countries in which the corresponding religion is the one with the largest number of
followers. Regional dummies were constructed for the developing countries of Sub-
Saharan Africa, Asia, Europe and Central Asia, Latin America and the Caribbean, Middle
East and Northern Africa, all based on the regional deﬁnitions employed by the World
Bank and the International Monetary Fund. Finally, we used a variable from Easterly and
Levine (1997) that measures the likelihood that two randomly selected people from a
given country will not belong to the same ethno-linguistic group. This index is only
available for 1960, and hence it should be interpreted with caution. The objective is to
capture not only cultural effects on crime that may be derived from a common set of

values, but also any potential effects from cultural polarization.

5- Empirical Implementation

A version of the regression equation derived ﬁ'om our model is ﬁrst run for simple
cross-sections and then applied to panel data. On the one hand, cross-sectional
regressions are illustrative because they emphasize cross-country variation of the data,

allowing us to analyze the effects of variables that do not change much over time. On the

 

the annual rates are constructed with contemporary observations of the number of
reported crimes.

201

other hand, working with panel data (that is, pooled cross-country and time-series data)
allows us to consider both the effect of the business cycle (i.e., GDP growth rate) on the
crime rate and the presence of criminal inertia (accounted for by the inclusion of lagged
crime rate as an explanatory variable). Furthermore, the use of panel data will allow us to
account for unobserved country-speciﬁc effects, for the likely joint endogeneity of some
of the explanatory variables, and for some types of measurement errors in the reported
crime rates.

As dependent variables, we consider the incidence of two types of crime, namely,
intentional homicide and robbery. Under-reporting is a major problem related to the
available measures of crime. It is well known that mis-measurement of the dependent
variable does not lead to estimation biases when the measurement error is uncorrelated
with the regressors. This condition, however, is very likely to be violated in the case of
crime under-reporting given that the degree of mis-measurement is surely related, for
instance, to the average income of the population, its level of education, and the degree of
income inequality, which are considered as explanatory variables in our empirical model
of crime. Of all types of crime, intentional homicide is the one that suffers the least from
under-reporting because corpses are more difﬁcult to ignore than losses of property or
assaults. Therefore, most of the analysis will concentrate on the regressions that have the
intentional homicide rate as the dependent variable. To the extent that intentional
homicide is a good proxy for overall crime, the conclusions we reach apply also to
criminal behavior broadly understood. However, if intentional homicide proxies mostly
for violent crime, then our results apply more narrowly. Hence we also focus on the

determinants of robbery rates. Robberies are crimes against property that include a

202

violent component; they are deﬁned as the taking away of property from a person,
overcoming resistance by force or threat of force. We believe that victims of robberies
may have stronger incentives to report them than victims of only theft or assault.

For ease of exposition, we ﬁrst present the cross-sectional regression results and

then the panel regression results.

5.1 - Cross-Sectional Regressions

Tables 2 and 3 report the results from cross-sectional regressions for intentional
homicides and robbery rates, respectively. These regressions use country averages of the
relevant dependent variables for the period 1970-94, but the averages were calculated

using only the annual observations for which the homicide data was available.

Table 3.2 shows that the Gini index of income distribution has a positive
coefﬁcient, which is signiﬁcant in all the regressions, revealing that countries with more
unequal distributions of income tend to have higher crime rates than those with more
egalitarian patterns of income distribution. In addition, regression (2) includes an
alternative measure of the distribution of income; namely, the share of national income
received by the poorest 20 percent of the population. The negative and signiﬁcant
coefﬁcient of this variable tells us that crime tends to decline as the poorest quintile
receives higher shares of national income. Income (i.e., GNP) per capita seems to be
negatively associated with the incidence of intentional homicides, as reﬂected in its
negative coefﬁcient, but this result is signiﬁcant at conventional levels in only one of the

sixteen regressions presented in Table 3.2. The combination of an insigniﬁcant effect of

203

the income per capita with a signiﬁcant effect of the distribution of income may indicate
that changes in income distribution, rather than changes in the absolute levels of poverty,

are associated with changes in violent crime rates.

Regarding education, the results in Table 3.2 show that the average years of
schooling, or the level of educational attainment of the population, has a negative
coefﬁcient in 12 out of the 15 regressions that include this variable, but the coefficient is
not signiﬁcant in any speciﬁcation. In equation (3) we use the secondary enrollment rate
(or the ﬂow of human capital) instead of the attainment variable. Contrary to our
expectations, the coefﬁcient of the enrollment rate is positive, but also insigniﬁcant. As
elaborated in our theoretical model, the relationship between educational variables and
crime rates can be ambiguous. However, from an empirical point of view, these results
may be explained by an implicit relationship between the extent of crime under-reporting
and the level of education of the population; that is, an increase in education may induce
people to report more crimes, thus producing a rise in reported crime rates. Also, the two
education variables are in fact negatively correlated with the homicide rate and at the
same time highly correlated with both per capita GNP (correlation about 0.5) and the Gini
index (correlation about -0.55). Therefore, it is quite possible that the expected crime-
reducing effects of education are captured by the measures of both national income per
capita and income distribution, also present in the homicide rate regression equation. We
will reconsider the effect of the educational variables when we discuss the panel data

results.

204

Regressions (4) to (6) in Table 3.2 examine the relationship between deterrence
and incapacitation effects and intentional homicide rates. The presence of police seems to
reduce crime, but the negative coefﬁcient is not signiﬁcant. The coefﬁcients
corresponding to the conviction rate are statistically different from zero, even after
including the variable that controls for the existence of the death penalty, which may
indicate that high convictions rates tend to deter criminal activity independently of the
incapacitation effect of the death penalty. However, as for most results of these OLS
cross-sectional regressions, this result must be regarded as preliminary given that the
negative relationship between homicide and conviction rates may be due to measurement
error in the number of homicides, which is both the numerator of the homicide rate and
the denominator of the conviction rate (see Levitt 1995)." We reexamine this issue in the
context of panel data analysis, in which correction for measurement error is possible to
some extent. In regressions not reported in Table 3.2, we included subjective indices of
the quality of the state apparatus instead of the police and conviction rates. Neither the
index of rule of law nor the index of absence of corruption turned out to be signiﬁcant.
The lack of signiﬁcance of the estimated coefﬁcients on these subjective indices of the
rule of law and absence of corruption in the bureaucracy may be due to the fact that they
are highly correlated with other important explanatory variables in the regression,

namely, per capita GNP, the Gini index, and the measures of educational stand.

 

” An indication that the negative relationship between homicide and conviction rates may
be partially spurious is given by the suspicious jumps in the ﬁt of the regression when the
conviction rate is included as an explanatory variable.

205

Table 3.2 also shows that the incidence of intentional homicides is statistically
larger in countries that produce drugs. The drug possession crime rate, which proxies for
the effects of both illegal drug consumption and for the violence emanating from the
distribution of illegal drugs, is also positively associated with the intentional homicide
rate, but it is signiﬁcant in only two of the 16 speciﬁcations. These results give credence
to the popular view that violent crimes increase with drug trafficking and consumption. It
remains to be studied, however, whether the incidence of homicides in drug producing
and/or consuming countries is directly affected by drug-related activities or is also the
result of crime externalities of these activities. The latter would be the case if, for
example, criminal organizations established to deal with drugs are also used to manage

other forms of criminal endeavors.

In the cross-sectional regressions considered in Table 3.2, the urbanization rate
appears not to be signiﬁcantly associated with the homicide rate. This result may be due
to the high correlation between the urbanization rate and other economic variables, such
as income per capita, the Gini index, and, especially, the education variables. Still, we
expected that the urbanization rate could provide information on the strength of social
interactions in the formation of criminal behavior; this information would not be
necessarily captured by the other indicators of economic development. We will
reconsider this issue when discussing the robbery regressions and the panel data

regressions for the homicide rate.

We examine the importance of other variables that in principle may be related to

the incidence of intentional homicides. We do it by including them one by one in a core

206

regression that considers per capita GNP, the Gini index, the average years of schooling,
the urbanization rate, the drug producers dummy, and the drug possession crime rate as
explanatory variables.” In these additional regressions (also presented in Table 3.2), we
ﬁnd the rather surprising result that the index of ethno-linguistic fractionalization, which
has been used as a proxy for social polarization and conﬂict (see Easterly and Levine
1997), is negatively associated with the rate of intentional homicides, though this
association is only marginally signiﬁcant. Regarding the religion dummies, Christian
countries seem to have signiﬁcantly higher homicide rates, while Hindu and Muslim
countries seem to have lower homicide rates than the average, even after controlling for
other possible determinants of crime rates. Of the regional dummies, South and East
Asian countries seem to have signiﬁcantly lower homicide rates than the average, while

Latin America seems to have higher rates than the average.l6

Table 3.3 reports the cross-sectional regression results for the incidence of
robberies. As mentioned, these results should be interpreted with caution given that the
robbery rates may suffer from under-reporting more severely than the intentional

homicide rates.

 

‘5 We do not include the homicide conviction rate in the core regression for two reasons;
ﬁrst, the variables to be examined are likely to also proxy for the strength of the police
and judicial system; and second, the inclusion of the conviction rate reduces the sample
size of the estimated regression by about 25%.

'6 We also ran regressions that included an index of the coverage of ﬁrearm regulations
and the share of national population encompassed by males of 15-29 years of age as
explanatory variables — see Table 3A] for a description of these variables. However, the
results showed that these variables were not signiﬁcant determinants of intentional
homicide rates. In addition, we collected information regarding the incidence of ﬁrearms
in a group of countries, but this data was only available for a small group of countries, the

207

The results of the robbery regressions are in several respects similar to those for
the homicide rate. The level of per capita income is not a signiﬁcant determinant of
robbery rates, but a worsening of income inequality is statistically related to higher
robbery rates. However, the drug producers dummy appears to be less important in the
robbery regressions than in the homicide regressions. The coefﬁcient of the secondary
enrollment rate is also positive in regression (3), and is actually more signiﬁcant than in
the corresponding homicide regression. However, the deterrence and incapacitation
variables appear with noticeably different coefﬁcients in the robbery regressions. First,
the presence of police personnel variable turns out to have a positive and signiﬁcant
coefﬁcient, which may reﬂect that police presence is endogenous. The conviction and
death penalty variables introduced in regression (5) and (6) appear with the expected

negative signs, but neither is statistically signiﬁcant.

An interesting result, that contrasts with those of the homicide regressions, is that
the urbanization rate seems to have a positive and signiﬁcant association with the robbery
rate; the coefﬁcient is signiﬁcant in 14 of the 16 speciﬁcations. This result may indicate
that this type of crime may be related to population density and the social interactions that
arise from it. As in the homicide regression, the index of ethno-linguistic fractionalization
is also not a signiﬁcant determinant of robbery rates. Regarding the religion and regional
dummy variables, the results reported in Table 3.3 are consistent with the results in Table
3.2, but with the additional ﬁnding that Sub-Saharan Aﬁican countries also tend to have a

signiﬁcantly higher robbery rate than the average.

 

regressions contained only 18 countries, and the coefﬁcient of this variable was also

208

5.2 - Panel Regressions

The cross-sectional results emphasize the cross-country variation of crime rates
and their determinants. However, further analysis is possible given that the available data
on crime rates and their determinants allow the use of an unbalanced panel with ﬁve-year
periods. The time-series dimension of the data can add important information and permit
a richer model speciﬁcation. First, we would like to test whether the crime rate varies
along the business cycle by including the ﬁve-year average GDP growth rate in the
regression model; this test could not be done using cross-sectional data averaged over a
long period of time (1970-94). Second, we would like to test whether there is inertia in
crime rates, by including the lagged crime rate in the model. Third, we would like to
control for the likely joint endogeneity of some of the explanatory variables and the bias
due to under-reporting. And, fourth, we would like to control for the presence of
unobserved country-speciﬁc effects.

Our preferred panel estimation strategy follows the Generalized Method of
Moments (GMM) estimator proposed by Chamberlain (1984), Holtz-Eakin, Newey and
Rosen (1988), Arellano and Bond (1991), and Arellano and Bover (1995), which has
been applied to cross-country studies by Caselli, Esquivel and Lefort (1996) and Easterly,
Loayza and Montiel (1997). The following is a brief presentation of the GM estimator

to be used.‘7

 

statistically insigniﬁcant.
'7 For a concise presentation of the GMM estimator addressed to a general audience, see
the appendix of Easterly, Loayza, and Montiel (1997) and chapter 8 of Baltagi (1995).

209

We will work under two econometric models. In the ﬁrst one, we assume that
there are no unobserved country-speciﬁc effects. In the second one, we allow and control
for them. Why do we also work with the constrained model of no country-speciﬁc
effects? The data requirements to handle appropriately the presence of country-speciﬁc
effects (namely, a minimum of three consecutive observations per country in the sample)
produce the loss of a large amount of observations in our panel, which is of rather limited
coverage to start with. Considering the model without country-speciﬁc effects increases

the number of observations at the cost of estimating a more restricted model.

5.2.1 - Assuming no unobserved country-speciﬁc effects

Consider the following regression equation,
yrt = ay 131-1 + ﬂXu + 5i,t (8)
where y represents a crime rate, X represents the set of explanatory variables other than
the lagged crime rate, a is the error term, and the subscripts i and t represent country and
time period, respectively.

We would like to relax the assumption that all the explanatory variables are
strictly exogenous (that is, that they are uncorrelated with the error term at all leads and
lags). Relaxing this assumption allows for the possibility of simultaneity and reverse
causality, which are very likely present in crime regressions. We adopt the assumption of
weak exogeneity of at least some of the explanatory variables, in the sense that they are
assumed to be uncorrelated with future realizations of the error term. For example, in the

case of reverse causality this weaker assumption means that current explanatory variables

210

may be affected by past and current crime rates but not by future crime rates. In practice
we assume that all variables are weakly exogenous except for the drug producers dummy

and the GDP growth rate.

Furthermore, we would like to allow and control for the possibility that errors in

the measurement of the crime rate (which are imbedded in the error term a) be correlated
with some of the explanatory variables. This would be the case if, for instance, the degree
of crime under-reporting decreases with the population’s level of education. As explained
below, our method of estimation corrects this type of nus-measurement bias, as long as
the error in measurement is not serially correlated.

Under the assumption that the error term, 6‘, is not serially correlated, the
assumption of weak exogeneity of the explanatory variables implies the following
moment conditions,

E[Xi, {-5 - a," t] = 0 for s 2 l (9)
These moment conditions mean that the observations of X lagged one or more periods are
valid instruments for the corresponding contemporaneous observations.

Given that the lagged crime rate is also measured with error, it must also be
replaced by an instrument. Again, under the assumption that , is not serially correlated,
observations of the crime rate lagged two or more periods are valid instruments for the
lagged crime rate, yt. 1. That is, the following moment conditions apply,

Elyi, (-5 maj, t] = 0 forsZZ (10)

211

5.2.2 - Allowing and controlling for unobserved country-speciﬁc effects
Consider the following regression equation,
J’i,t = ay i,t-1 + ﬂXm + n: + £13: (11)
Equation (11) differs from (8) in that it includes 77;, an unobserved country-speciﬁc
effect. The usual method to deal with the Speciﬁc effect in the context of panel data has
been to ﬁrst-difference the regression equation (Anderson and Hsiao, 1981). In this way
the speciﬁc-effect is directly eliminated from the estimation process. First-differencing
equation (1 1), we obtain
yr: -yi,r-1 = 610 i,t-1 -yi,t-2) + ﬂ(Xi,t -Xi,t-1) + (82;: - Sit-1) (12)
The use of instruments is again required to deal with several problems: ﬁrst, the
likely joint endogeneity of the explanatory variables, X; second, the fact that mis-
measurement in the contemporaneous crime rate may be correlated with the explanatory
variables; third, the fact that the lagged crime rate is likely to be measured with error; and

fourth, the fact that by differencing, we introduce by construction a correlation between

the new error term, r: ,3, - e i, t-1,and the differenced lagged dependent variable,

yi, (-1 - ”1.2. Under the assumption that the error term, a, is not serially correlated, the

following moment conditions apply in relation to, respectively, the lagged dependent

variable and the set of explanatory variables,

ElVi, t-s '(6i,t-€i,t-1)]=0 f0r823 (13)

Ele, t-s -(ar,r- tel-5-1)] =0 forszz (14)
Arellano and Bond (1991) develop a consistent GMM estimator based on moment

conditions similar to those in equations (13) and (14). However, for reasons explained

212

below, we will use an estimator that complements these moment conditions (applied to
the regression in differences) with appropriate moment conditions applied to the
regression in levels. Before explaining the statistical advantages of the estimator that
combines differences and levels regressions over the simple difference estimator, a
conceptual justiﬁcation for our approach is the following. This paper studies not only the
time-series determinants of crime rates but also their cross-country variation, which is
eliminated in the case of the simple difference estimator.

Alonso-Borrego and Arellano (1996) and Blundell and Bond (1997) show that
when the lagged dependent and the explanatory variables are persistent over time, lagged
levels of these variables are weak instruments for the regression equation in differences.
The instruments’ weakness has repercussions on both the asymptotic and small-sample
performance of the difference estimator. AS the variables’ persistence increases, the
asymptotic variance of the coefﬁcients obtained with the difference estimator rises (that
is, the asymptotic precision of this estimator deteriorates). Furthermore, Monte Carlo
experiments show that the weakness of the instruments produces biased coefﬁcients in
small samples; this bias is exacerbated with the variables’ over time persistence, the
importance of the speciﬁc-effect, and the smallness of the time-series dimension. An
additional problem with the simple difference estimator relates to measurement error:
Differencing may exacerbate the bias due to errors in variables by decreasing the signal-
to-noise ratio (see Griliches and Hausman, 1986).

On the basis of both asymptotic and small-sample properties, Blundell and Bond
(1997) suggest the use of the Arellano and Bover (1995) estimator in place of the usual

difference estimator. Arellano and Bover (1995) present an estimator that combines, in a

213

system, the regression in differences with the regression in levels. The instruments for the
regression in differences are the lagged levels of the corresponding variables; therefore,
the moment conditions in equations (1 3) and (14) apply to this ﬁrst part of the system.
The instruments for the regression in levels are the lagged diﬂ'erences of the
corresponding variables. These are appropriate instruments under the following two
assumptions: First, the error term r: is not serially correlated. And second, although there
may be correlation between the levels of the right-hand side variables and the country-
speciﬁc effect, there is no correlation between the differences of these variables and the
speciﬁc effect. The second assumption results from the following stationarity property,
ED’i, t+p ' m] = Eli’i, t+q ' 771']
and ElXi, t+p - ml= Eer, t+q - ml for aIIp and q (15)
Therefore, the moment conditions for the second part of the system (the

regression in levels) are given by:
Elm, t—s -yi, t-s-I) - (771+ 6w] = 0 for S=2 (16)

E[(Xi, t-s -Xi, t-s-I) '(771‘+ 8139]: 0 for 5:1 (17)

5.2.3 - Summary of the Methodology.

The estimation strategy proposed in this paper can deal with unobserved ﬁxed effects
in a dynamic (lagged-dependent variable) model, joint endogeneity of the explanatory
variables, and serially-uncorrelated crime rate mis-measurement. The moment conditions
presented above can be used in the context of the Generalized Method of Moments

(GMM) to generate consistent and efﬁcient estimates of the parameters of interest

214

(Arellano and Bond, 1991; and Arellano and Bover, 1995). Speciﬁcally, in the model that
ignores unobserved country-speciﬁc effects, the moment conditions in equations (9) and
(10) are used; and in the model that allows and controls for unobserved speciﬁc effects,
the moment conditions in equations (13), (14), (16) and (17) are used.”

The consistency of the GMM estimator depends on whether lagged values of the
crime rate and the other explanatory variables are valid instruments in the crime
regression. To address this issue we present two speciﬁcation test, suggested by Arellano
and Bond (1991). The ﬁrst is a Sargan test of over-identifying restrictions, which tests the
overall validity of the instruments by analyzing the sample analog of the moment

conditions used in the estimation process. The second test examines the hypothesis that

the error term £1"; is not serially correlated. In the levels regression we test whether the
error term is ﬁrst- or second-order serially correlated, and in the system difference-level
regression we test whether the differenced error term is second-order serially correlated
(by construction, it is likely that this differenced error term be ﬁrst-order serially
correlated even if the original error term is not). Under both tests, failure to reject the null

hypothesis gives support to the model.

5.2.4 - Results
Table 3.4 reports the GM estimates from the panel regressions for the intentional
homicide rate, both ignoring and controlling for unobserved country-speciﬁc effects. It

must be noted that, given that we are controlling for possible problems of simultaneity

 

'8 We are grateful to Stephen Bond for providing us with a program to apply his and
Arellano’s estimator to an unbalanced panel data set.

215

and reverse causality, we can interpret the estimated coefficients not simply as partial
associations but as effects of the explanatory variables on homicide rates. As in the cross-
sectional regressions, we consider a “core” set of explanatory variables consisting of the
GDP growth rate, the (log) of GNP per capita, the Gini index, the average years of
schooling of the population older than 15 years of age, the urbanization rate, a dummy for
whether the country produces illegal drugs, the drug possession crimes rate, and (except
for the ﬁrst regression) the lagged homicide rate. To this core set, we add in turn the
secondary enrollment rate, the ratio of policemen per inhabitant in the country, and the
homicide conviction rate.

The ﬁrst regression in Table 3.4 considers a static speciﬁcation (that is, one
excluding the lagged crime rate as explanatory variable). This speciﬁcation is rejected by
the error serial-correlation tests; therefore, its estimated coefﬁcients cannot offer valid
conclusions. The correlation of the error term in this regression signals that relevant
variables with high over-time persistence were omitted; these variables can be the lagged
homicide rate (which makes the model dynamic) and/or the country-speciﬁc effect. When
the lagged homicide rate is included in subsequent regressions, both the hypothesis of
lack of residual serial correlation and the hypothesis of no correlation between the error
term and the instruments (Sargan test) cannot be rejected, and, thus, the dynamic model is
supported by the speciﬁcation tests. The dynamic model with country-speciﬁc effects
(regressions (7) and (8)) is also supported by the Sargan and second-order serial
correlation tests.

From the regressions ignoring country-speciﬁc effects (regressions (2) to (6)) and

those accounting for them (regressions (7) and (8)), the most robust and signiﬁcant results

216

in relation to the core variables are the following: First, the business cycle effect,
measured by the coefﬁcient on GDP growth rate holding constant average per capita
income, is statistically signiﬁcant and shows that, as expected, crime is counter-cyclical;
stagnant economic activity induces heightened homicide rates. Second, higher income
inequality, measured by the Gini index, increases the incidence of homicide rates; this
result survives the inclusion of lagged homicide rates and is strengthened when
unobserved country-speciﬁc effects are taken into account. The only regression where the
Gini coefﬁcient loses its statistical signiﬁcance is the one that allows for time-speciﬁc
effects. In addition, the combination of signiﬁcant effects of the business cycle and
income distribution tells us that the rate of poverty reduction may be associated with
declines in crime rates.'9 Third, higher drug related activity, represented by both drug
production and drug possession, induces a higher incidence of intentional homicide. It
must be noted that the drug producers dummy loses some of its signiﬁcance when time
effects are allowed, and the drug possession crimes rate is not robustly signiﬁcant when
country-speciﬁc effects are accounted for. Fourth, the lagged homicide rate has a positive
and signiﬁcant impact on current rates, which is evidence of criminal inertia, as predicted
by recent crime theoretical models. The size of the coefﬁcient on the lagged homicide
rate decreases but remains Signiﬁcant when country-speciﬁc effects are controlled for,

which indicates that country-speciﬁc factors explain only a portion of criminal inertia.

 

'9 The abolute level of poverty (usually measured as the percentage of people below a
certain level of income) is determined by the national income and its pattern of
distribution. Hence, when GDP grows, while holding the Gini index constant, the abolute
level of poverty declines.

217

As in the cross-sectional regressions, the level of income per capita does not have
an independent, signiﬁcant effect on the homicide rate. The results concerning the
urbanization rate are not robust to the issue of country-speciﬁc effects. In the model
without country-speciﬁc effects, the urbanization rate does not affect signiﬁcantly the
homicide rate. However, when country-speciﬁc effects are controlled for, the
urbanization rate is associated with larger homicide rates.20

The puzzle concerning the lack of a signiﬁcantly negative association between a
country’s educational stand and its homicide rate is somewhat clariﬁed in the panel
regressions that account for country-speciﬁc effects. When a country’s educational stand
is proxied by the secondary enrollment rate, its effect on homicide rates is signiﬁcantly
positive.” However, when the average years of schooling in the adult population is used
to proxy for the country’s educational position, it has a signiﬁcant crime-reducing impact.
The contrast between the results obtained using secondary enrollment rates and average
years of schooling may indicate that the efforts to educate the young may not reduce
crime immediately but eventually lead to a reduction of crime, especially of the violent
sort.

In regressions (4) and (5) we examine the effect of the strength of the police and
judicial system in deterring crime. The proxies we use are, in turn, the rate of policemen

per inhabitant in the country and the homicide conviction rate. Both variables are subject

 

2° It must be noted that the differences between the results found in the levels and
differences speciﬁcations are not solely the result of controlling for country-speciﬁc
effects, for in the latter case the sample size is much smaller than in the former.

2' The fact that the coefficient on secondary enrollment remains positive even after
accounting for criminal inertia and country-speciﬁc effects makes it unlikely that this

218

to joint endogeneity in crime regressions, and the conviction rate may be spuriously
negatively correlated with the homicide rate given the mis-measurement in the number of
homicides. Because of these reasons, the panel GMM estimator is clearly superior to the
cross-sectional results. Since we are instrumenting for both the policemen rate and the
conviction rate (and the speciﬁcation tests support the model), we conclude that the
negative and signiﬁcant coefﬁcient on both proxies means that a stronger police and
judicial system does lead to a lower incidence of homicides.

In regression (6) we examine the importance of time-speciﬁc effects. We ﬁnd that
in the period 1990-94, the world has experienced a statistically signiﬁcant increase in
homicide rates relative to those in the late 19708 and early 1980s; this rise in homicide
rates cannot be fully explained by the evolution of the crime determinants in the core
model.

Table 3.5 shows the GMM estimates for the panel regressions for the robbery rate.
The model speciﬁcation without a lagged dependent variable or a country-speciﬁc effect
is strongly rejected by the residual serial correlation tests. In contrast to the homicide
regressions, the dynamic speciﬁcation of the crime regression that ignores country-
speciﬁc effects is also rejected by the residual serial correlation test. Therefore, we must
base our conclusions on the dynamic speciﬁcation that accounts for speciﬁc effects. This
prevents us from analyzing the role of the proxies for the strength of the police and
judicial system given that the inclusion of these variables limits dramatically the sample

size available for estimation of the speciﬁc-effect model.

 

controversial coefﬁcient Sign is due to the omission of some relevant variable in the
homicide rate regression.

219

The results of the dynamic model that controls for country-speciﬁc effects for the
robbery rate are virtually the same as the corresponding ones for the homicide rate”:
Stagnant economic activity (low GDP growth) promotes heightened robbery rates; the
counter cyclical behavior of the robbery rate appears to be larger than that in the case of
the homicide rate. Larger income inequality (high Gini index) induces an increase in the
incidence of robberies, but not to the same extent as in the case of homicide rate. The
robbery rate exhibits a signiﬁcant degree of inertia, which is somewhat larger than that of
the homicide rate. The urbanization rate has a signiﬁcant positive impact on the incidence
of robberies; this impact appears to be larger than in the case of homicides. Although the
secondary enrollment rate has a puzzling positive effect on robbery rates, the level of
educational attainment of the adult population has a robbery-reducing impact. The drug
possession crimes rate is positively associated with the robbery rate. Finally, as in the
homicide regressions, the level of per capita income does not appear to be robustly

correlated with the robbery rate.

6 - Conclusions

The conclusions that can be derived from the theoretical model and the empirical
ﬁndings regarding potentially ﬁ'uitful directions for future research and possible policy

implications fall under two headings: the good news and the bad news.

 

22 The remarkable similarity between the homicide and robbery regression results gives

credence to our interpretation of the homicide rate as a relatively broad proxy for criminal
behavior.

220

The bad news ﬁrst. Some bad news are related to the results of the dynamic panel
estimation methods (GMM). The results show that economic downturns and other non-
economic shocks, such as a rise in drug trafficking, as in Colombia in the 19705, can raise
the national crime rate. The econometric results also suggest that the rise in the crime rate
may be felt long after the initial shock — countries can be engulfed in a crime wave. The
policy implication of this ﬁnding is that policy-makers should act to counter the crime
wave, if not, a country may get stuck at an excessively high crime rate.

Although we do not know the precise channels through which a crime shock tends
to be perpetuated over time, the existing literature proposes three possible channels:
systemic interactions, local interactions, and recidivism. Future research should attempt
to clarify which one of these is at work, but this research would probably need to rely on
individual-level analysis, because local interactions and recidivism are forces that are
determined by an individual’s location with respect to her local community and her past

criminal record, respectively.

The good news. Two important determinants of crime rates — inequality and
deterrence — are, we believe, “policy-sensitive” variables. Policy-makers facing a crime
wave should then consider a combination of counter-cyclical re-distributive policies (e.g.,
targeted safety nets) and increases in the resources devoted to apprehending and
convicting criminals — a “carrots-and-stic ” policy response would seem to be
appropriate, especially during economic recessions. Regarding the crime-inducing effect
of inequality, our empirical ﬁndings suggest that there is, “a social incentive for

equalizing training and earning opportunities across persons, which is independent of

221

ethical considerations or any social welfare function” (Ehrlich 1973, 561). In addition,
our empirical ﬁndings regarding criminal inertia imply that current crime rates respond to
current policy variables with a lag. Sah (1991, 1292) observed that, “This apparent lack
of response is a source of frustration for politicians as well as for law enforcement
ofﬁcials... Such reactions, though understandable, may be inappropriate if they are caused

by an inadequate understanding of the dynamics of crime.”

Future research in this area should attempt to solve the crime-education puzzle
present in our empirical ﬁndings. We have provided a result which may prove to be one
of the clues to solve the puzzle: there is a delayed effect of educational effort on crime
alleviation, that is, the crime-reducing effect of education does not materialize when the
young are being educated but mostly when they become adults. Another clue to the

puzzle may be obtained by considering the indirect effects of education on inequality.

This paper was motivated by the impression that crime has pernicious effects on
economic activity, and may also reduce welfare by reducing individuals sense of personal
and proprietary security. Indeed, a fertile area for future research is to attempt to measure
the effects of criminal behavior on economic growth and welfare. We suspect that there
are many ways of measuring the economic costs of crime, ranging from the costs of
maintaining an effective police and judicial system, to estimates of the forgone output.

However, the overall effects on welfare may be more difﬁcult to assess.

222

REFERENCES

Alonso-Borrego, C., and Arellano, M. 1996. “Symmetrically Normalized Instrumental
Variable Estimation Using Panel Data.” CEMFI Working Paper No. 9612,
September.

Anderson, T.W., and Hsiao, C. 1981. “Estimation of Dynamic Models with Error
Components.” Journal of the American Statistical Association 76: 598-606.

 

Arellano, M., and Bover, O. 1995. “Another Look at the Instrumental-Variable
Estimation of Error-Components Models.” Journal of Econometrics 68: 29-52.

 

Arellano, M., and Bond, S. 1991. “Estimation of Dynamic Models with Error
Components.” Review of Economic Studies 58: 277-297.

 

Baltagi, B. H. 1995. Econometric Analysis of Panel Data. New York: John Wiley &
Sons.

 

Barro, R., and Lee, Jong-Wha. 1996. “New Measures of Educational Attainment.”
Mimeographed. Department of Economics, Harvard University.

Becker, G. S. 1993. “Nobel Lecture: The Economic Way of Looking at Behavior.”
Journal of Political Economy 101: 385-409.

 

Becker, G. S. 1968. “Crime and Punishment: An Economic Approach.” Journal of
Political Economy 76: 169-217. Reprinted in Chicago Studies in Political Economy,
edited by G.J. Stigler. Chicago and London: The University of Chicago Press, 1988.

 

 

Blundell, R., and Stephen B. 1997. “Initial Conditions and Moment Restrictions in
Dynamic Panel Data Models.” Discussion Papers in Economics 97-07, Department of
Economics, University College London.

Caselli, F., Esquivel, G., and Lefort, F. 1996. “Reopening the Convergence Debate: A
New Look at Cross-Country Growth Empirics”. Journal of Economic Growth 1: 363-
389.

 

Chamberlain, G. 1984. “Panel Data.” In Handbook of Econometrics Vol.2. Z. Griliches
and M. D. Intriligator, eds.

 

Davis, M. L. 1988. “Time and Punishment: An Intertemporal Model of Crime.” Journal
of Political Economy 96: 383-390.

 

223

Deninger, K., and Squire, L. 1996. “A New Data Set Measuring Income Inequality.” IE
World Bank Economic Review 10 (3): 565-592.

 

Easterly, W., and Levine, R. 1997. “Africa’s Growth Tragedy: Policies and Ethnic
Divisions.” The Quarterly Journal of Economics [November].

 

Easterly, W., Loayza, N., and Montiel, P. 1997. “Has Latin America’s Post-Reform
Growth Being Disappointing?” Journal of International Economics 43: 287-311.

 

Ehrlich, I. 1973. “Participation in Illegitimate Activities: A Theoretical and Empirical
Investigation.” Journal of Political Economy 81: 521-565.

 

Ehrlich, I. 1975a. “On the Relation between Education and Crime.” In Education, Income
and Human Behavior, edited by F .T. Juster. New York: McGraw-Hill.

 

 

Ehrlich, I. 1975b. “The Deterrent Effect of Capital Punishment: A Question of Life and
Death.” American Economic Review [December]: 397-417.

 

Ehrlich, I. 1981. “On the Usefulness of Controlling Individuals: An Economic Analysis
of Rehabilitation, Incapacitation and Deterrence.” American Economic Review 71(3):
307-322.

 

Ehrlich, I. 1996. “Crime, Punishment, and the Market for Offenses.” Journal of Economic
Perspectives 10: 43-68.

 

 

F leisher, B. M. 1966. “The Effect of Income on Delinquency.” American Economic
Review 56: 118-137.

 

Glaeser, E. L., Sacerdote, B., and. Schneinkman, J. A. 1996. “Crime and Social
Interactions.” Quarterly Journal of Economics 111: 507-548.

 

Griliches, Z., and Hausman, J. 1986. “Errors in Variables in Panel Data.” Journal of
Econometrics 31: 93-118.

 

Holtz-Eakin, D., Newey, N., and Rosen, H. S. 1988. “Estimating Vector Autoregressions
with Panel Data.” Econometrica 56: 1371-1395.

 

Leung, S. F. 1995. “Dynamic Deterrence Theory.” Economica 62: 65-87.

 

Levitt, S. D. 1995. “Why Do Increased Arrest Rates Appear to Reduce Crime:
Deterrence, Incapacitation, or Measurement Error?” National Bureau of Economic
Research Working Paper 5268, September, Cambridge, Massachusetts.

224

Loayza, N., Lopez, H., Schmidt-Hebbel, K. and Serven, L. 1998. “A World Savings
Database.” Mimeographed, Policy Research Department, The World Bank,
Washington, DC.

Newman, G., and DiCristina, B. 1992. “Data Set of the lst and 2nd United Nations World
Crime Surveys.” Mimeographed. School of Criminal Justice, State University of
New York at Albany.

Posada, C. E. 1994. “Modelos economicos de la criminalidad y la posibilidad de una
dinamica prolongada.” Planeacion y Desarrollo 25: 217-225.

 

Sah, R. 1991. “Social Osmosis and Patterns of Crime.” Journal of Political Economy 99:
1272-1295.

 

Schmidt, P., and Witte, A. D. 1984. An Economic Analysis of Crime and Justice: Theory,
Methods, and Applications. New York: Academic Press, Inc.

 

 

Tauchen, H., and Witte, A. D. 1994. “Work and Crime: An Exploration Using Panel
Data.” National Bureau of Economic Research Working Paper Series No. 4794, July.

Usher, D. 1993. “Education as a Deterrent to Crime.” Queen’s University, Institute for
Economic Research, Discussion Paper No. 870, May.

World Bank. 1997. “Crime and Violence as Development Issues in Latin America and

the Caribbean.” Mimeographed, Office of the Chief Economist, Latin America and
the Caribbean, The World Bank, Washington, DC.

225

APPENDIX A

DATA DESCRIPTION AND SOURCES

This appendix presents the data used in this paper, with special attention to the
variables related with crime rates, conviction rates and police personnel. Table 3.A.1
provides the description and sources of all the variables that were used. References are
provided for details on the variables that have been previously used in other academic
papers. In the case of the crime-related data, even though the information that was used is
publicly available, additional work was required in order to assemble the variables
actually used in the econometric estimations.

These variables were constructed with information provided by the United Nations,
through its Crime Prevention and Criminal Justice Division. The United Nations has
conducted, since 1978, ﬁve Surveys of Crime Trends and Operations of Criminal Justice
Systems. Each survey has covered periods of 5 to 6 years, requesting crime data from
government officials covering the period from 1970 to 1994. The statistics included in
these surveys represent the official statistics of member countries of the United Nations.
They have been compiled by the United Nations on the basis of questionnaires distributed
to member countries, as well as yearbooks, annual reports, and statistical abstracts of
these countries. The United Nations Surveys are available on the intemet at
http://www.ifs.univie.ac.at/~uncjin/wcs.html#wcsl23 (March 10, 1998).

In order to construct series covering the period 1970/1994 for the largest number of

countries, the ﬁve U.N. Surveys were used. When these surveys overlap (in 1975, 1980,

226

1986 and 1990), the information from the latest survey was used. It is worth noting that
most of the countries did not respond to all surveys, so that missing values are a common
occurrence in these series. The deﬁnitions of the various crimes are stable across the
surveys and are detailed in Table 3.A.1. However, as stated by Newman and DiCristina
(1992), who constructed a data set with the information of the ﬁrst and second surveys,
the deﬁnitions “were applied as far as possible”. Moreover, they add, “it will be
recognized that, owing to the immense variation in criminal justice systems around the
world, these categories are of necessity rough” (Newman and DiCristina 1992, 6).

In addition to assembling the series for the yearly number of crimes and convictions
in each country, we conducted a “cleaning” of the data. This process, inherently based on
arbitrary judgments, was nonetheless guided by the following criteria. We analyzed the
evolution of the variables over time, searching for large and discontinuous changes. More
speciﬁcally, we looked for situations where a change in the order of magnitude of the
variables (e.g., ten-fold or hundred-fold increases) occurred ﬁom one survey to the other.
In the cases where it was apparent that, in each new survey, the level of a speciﬁc
variable experienced this type of abrupt and permanent change, all the observations for
the corresponding country and variable were dropped for the period in question. This
decision was based on the assumption that these changes could only be explained by
changes in the deﬁnitions or criteria used in the collection of the data by the respondents
of the corresponding questionnaires. In addition, when these deﬁnition changes were
apparent in only one small subperiod (e. g., corresponding to only one survey or subperiod

thereof) this subperiod was dropped for the corresponding country and variable.

227

APPENDIX B

TABLES AND FIGURES

228

Table 3.1: Summary Statistics for Crime, Convictions and Police Rates

 

 

 

 

Variables No. of Obs. Mean Standard Min. Max. No. of Countries
Deviation

Crime Rates:"'

Intentional Homicides 1579 6.834 1 1.251 0 142.014 128

Robbery 1251 55.902 95.973 0 676.840 120

Drug Possession 1037 69.990 128.301 0 1358.524 99

Conviction Rates?”

Intentional Homicides 183 69.730 204.260 0 2694.643 80

Robbery 23 1 42.266 64.1 10 0 675.604 72

Police Personnel" 486 329.262 310.264 1.598 2701.31 104

 

‘Per 100,000 inhabitants, annual data.

I”Percent of number of crimes, 5-year-averages.

229

Table 3.2: OLS Cross-Sectional Regressions of Intentional

Homicide Rate, 1970/1994 (p-values in parenthesis)

 

 

 

(1) (2) (3) (4) (5) (6) (7)
GNP Per Capita -.004 -.096 -.278 -.090 .014 -.078 -.032
(.981) (.577) (.125) (.649) (.935) (.628) (.885)
Gini Index .035 .035 .038 .043 .052 .041
(.019) (.034) (.025) (.014) (.002) (.025)
Average Years of -.027 -.017 .011 .013 .079 -.052
Schooling (.744) (.814) (.901) (.885) (.384) (.598)
Urbanization Rate .000 .002 .004 .005 .001 .001 -.001
(.989) (.791) (.625) (.593) (.920) (.919) (.886)
Drug Producers .670 .912 .390 .711 1.305 1.311 .667
Dummy (.074) (.012) (.272) (.069) (.002) (.001) (.093)
Drug Possession .002 .001 .003 .002 .001 .001 .004
Crimes Rate (.329) (.694) (.090) (.359) (.616) (.758) (.127)
Income Share of the -20.405
Poorest Quintile (.001)
Secondary Enrollment .009
Rate (.314)
Police —.001
(.214)
Conviction Rate -.001 -.002
(.001) (.000)
Death Penalty -.659
(.011)
Index of Ethno- -.665
Linguistic (.200)
Fractionalization
Constant -.066 3.190 1.109 .213 -.755 -.322 .270
(.963) (.003) (.396) (.885) (.619) (.821) (.887)
R2 .285 .386 .213 .303 .502 .599 .31 1
Adjusted R2 .200 .306 .127 .192 .405 .501 .198
Number of 58 53 62 52 44 42 51
Observations

 

230

Table 3.2 (cont’d)

 

 

 

(8) (9) (10) (11) (12) (13) (14) (15) (16)

GNP Per Capita -.006 -.077 -.038 -.069 -.030 -.132 .012 .046 -.024
(.974) (.674) (.831) (.708) (.870) (.489) (.948) (.801) (.898)

Gini Index .035 .031 .030 .028 .029 .029 .038 .024 .034
(.021) (.041) (.048) (.073) (.076) (.059) (.015) (.143) (.027)

Average Years of -.028 -.060 -.049 -.073 -.025 -.009 -.046 -.042 -.038

Schooling (.735) (.474) (.559) (.409) (.768) (.911) (.595) (.611) (.660)

Urbanization Rate .000 .001 -.001 .002 .002 -.004 .001 -.004 .001
(.977) (.931) (.944) (.813) (.808) (.666) (.915) (.612) (.882)

Drug Producers .653 .624 .706 .582 .760 .751 .690 .558 .633

Dummy (.087) (.090) (.057) (.121) (.049) (.043) (.067) (.135) (.097)

Drug Possession .002 .003 .002 .003 .002 .002 .002 .002 .002

Crimes Rate (.328) (.196) (.232) (.214) (.359) (.273) (.247) (.255) (.312)

Buddhist Dummy .140

(most common (.737)

religion)

Christian Dummy .437

(most common (.087)

religion)

Hindu Dummy -.816

(most common (.1 1 1)

religion)

Muslim Dummy -.541

(most common (.158)

religion)

Sub-Saharan .457

Africa Dummy (.307)

South and East -.663

Asia Dummy (.073)

Eastern Europe and .321

Central Asia (.4 1 2)

Dummy

Latin America .488

Dummy (.110)

Middle East -.378

Dummy (.530)

Constant -.052 .545 .605 .967 .232 l .420 -.275 .226 . 159
(.971) (.702) (.675) (.538) (.872) (.376) (.848) (.871) (.913)

R7 .286 .326 .320 .3 13 .299 .330 .294 .320 .290

Adjusted R2 .186 .231 .225 .217 .201 .236 .195 .225 .191

Number of 58 58 58 58 58 58 58 58 58

Observations

 

231

Table 3.3: OLS Cross-Sectional Regressions of Intentional
Robbery Rate, 1970/1994 (p-values in parenthesis)

 

 

 

(l) (2) (3) (4) (5) (6) (7)

GNP Per Capita .061 -.169 -.127 -.129 -.101 -.161 .280

(.821) (.556) (.616) (.653) (.741) (.619) (.430)
Gini Index .091 .089 .085 .052 .060 .108

(.000) (.000) (.001) (.098) (.082) (.000)
Average Years of .113 -.021 .133 -.033 -.028 .061
Schooling (.360) (.861) (.290) (.825) (.856) (.673)
Urbanization Rate .020 .030 .022 .023 .025 .025 .020

(.108) (.023) (.040) (.070) (.062) (.078) (.121)
Drug Producers .139 .378 .206 .154 .699 .673 .276
Dummy (.795) (.517) (.682) (.774) (.336) (.370) (.637)
Drug Possession .004 .005 .004 .004 .005 .005 .004
Crimes Rate (.223) (.155) (.097) (.131) (.1 1 l) (.1 15) (.229)
Income Share of the -27.715
Poorest Quintile (.006)
Secondary Enrollment .021
Rate (.11 1)
Police .002

(.132)
Conviction Rate -.003 -.001
(.697 (.885)
Death Penalty -.567
(289)

Index of Ethno- .349
Linguistic (.663)
Fractionalization
Constant -2.85 1 4.447 -2.055 -2.023 .694 l . 146 -5.229

(.179) (.012) (.246) (.346) (.807) (.699) (.100)
R2 .452 .374 .469 .495 .404 .4 10 .460
Adjusted R2 .375 .283 .402 .406 .264 .235 .355
Number of 50 48 54 48 38 36 44
Observations

 

232

Table 3.3 (cont’d)

 

 

 

(8) (9) (10) (ll) (12) (13) (14) (15) (16)
GNP Per Capita .065 -.061 .073 -.075 .007 -.213 .026 .135 .061
(.809) (.812) (.790) (.774) (.979) (.465) (.922) (.596) (.821)
Gini Index .092 .088 .093 .076 .072 .077 .088 .067 .091
(.000) (.000) (.000) (.001) (.006) (.001) (.000) (.005) (.000)
Average Years of .114 .018 .121 .009 .125 .130 .142 .082 .113
Schooling (.354) (.881) (.340) (.942) (.302) (.274) (.259) (.480) (.360)
Urbanization .019 .021 .020 .023 .023 .016 .019 .012 .020
Rate (.11 1) (.063) (.110) (.050) (.057) (.181) (.115) (.322) (.108)
Drug Producers .201 -.020 .135 -.044 .429 .177 .087 -.295 .139
Dummy (.708) (.968) (.803) (.932) (.439) (.731) (.870) (.575) (.795)
Drug Possession .003 .005 .003 .005 .003 .004 .003 .004 .004
Crimes Rate (.231) (.089) (.251) (.090) (.232) (.147) (.395) (.153) (.223)
Buddhist Dummy -.639
(most common (.271)
religion)
Christian Dummy 1.054
(most common (.007)
religion)
Hindu Dummy .249
(most common (.735)
religion)
Muslim Dummy -1.420
(most common (.019)
religion)
Sub-Saharan 1.083
Africa Dummy (.105)
South and East -1.127
Asia Dummy (.042)
Eastern Europe -.648
and Central Asia (.268)
Dummy
Latin America 1.292
Dummy (.010)
Middle East
Dummy dropped
Constant -2.852 -1.993 -3.083 -.569 -2.069 .222 -2.467 -2.007 -2.851
(.178) (.315) (.172) (.796) (.330) (.929) (.249) (.315) (.179)
R2 .468 .540 .453 .520 .485 .504 .468 .533 .452
Adjusted R2 .379 .463 .362 .440 .400 .421 .379 .455 .375
Number of 50 50 50 50 50 50 50 50 50
Observations

 

233

Table 3.4: GMM Estimates: Panel Regressions of Intentional Homicide Rate

(p-values in parenthesis)

 

 

 

 

 

(1) (2) (3) (4) (5)
Regression Speciﬁcation Levels
Instruments (‘) Levels
GDP Growth Rate -0.101 -0.064 -0.056 -0.047 -0.034
(0.000) (0.000) (0.000) (0.000) (0.01 1)
GNP per Capita -0.305 0.026 0.017 -0.049 -0.021
(0.161) (0.588) (0.740) (0.039) (0.748)
Gini Index 0.034 0.021 0.016 0.016 0.012
(0.060) (0.000) (0.000) (0.001) (0.1 17)
Average Years of 0.007 0.015 -0.073 0.011
Schooling (0.923) (0.591) (0.000) (0.848)
Urbanization Rate -0.000 -0.002 -0.002 0.003 -0.003
(0.971) (0.216) (0.143) (0.095) (0.378)
Drug Producers Dummy 0.196 0.338 0.238 0.311 0.648
(0.564) (0.006) (0.000) (0.000) (0.000)
Drug Possession Crimes 0.004 0.001 0.001 0.002 0.001
Rate (0.000) (0.058) (0.074) (0.000) (0.259)
Lagged Homicide Rate 0.737 0.761 0.723 0.570
(0.000) (0.000) (0.000) (0.000)
Secondary Enrollment 0.000
Rate (0.912)
Police -0.000
(0.019)
Conviction Rate -0.006
(0.009)
Constant 2.354 -0.478 -0.1 17 0.581 0.631
(0.175) (0.154) (0.668) (0.042) (0.349)
Sargan Test of 0.545 0.397 0.51 1 0.365 0.369
Overidentifying
Restrictions: p-value
Test for F irst-Order 0.000 0.530 0.879 0.647 0.888
Serial Correlation: p -
value
Test for Second-Order 0.006 0.91 1 0.202 0.284 0.550
Serial Correlation: p -
value
Number of Observations 153 (68) 85 (45) 76 (42) 49 (27) 31 (21)

(Countries)

 

234

Table 3.4 (cont’d)

 

 

 

 

 

(6) (7) (8)
Regression Speciﬁcation Levels Dif.-Lev.
Instruments (*) Levels Lev. Dif.
GDP Growth Rate -0.052 -0.051 -0.036
(0.000) (0.000) (0.001)
GNP Per Capita -0.046 -0.014 -0.207
(.343) (0.289) (0.000)
Gini Index 0.008 0.021 0.036
(0.335) (0.000) (0.000)
Average Years of Schooling 0.023 -0.040
(0.257) (0.001)
Urbanization Rate -0.002 0.004 0.004
(0.340) (0.130) (0.063)
Drug Producers Dummy 0.246
(0.135)
Drug Possession Crimes Rate 0.001 0.000 0.001
(0.083) (0.299) (0.047)
Lagged Homicide Rate 0.893 0.664 0.640
(0.000) (0.000) (0.000)
Secondary Enrollment Rate 0.009
(0.000)
1980-84 Period Dummy -0.036
(0.530)
1985-89 Period Dummy 0.071
(0.299)
1990-94 Period Dummy 0.141
(0.051)
Constant 0.322
(.468)
Sargan Test of Overidentifying 0.397 0.589 0.839
Restrictions: p-value
Test for F irst-Order Serial 0.357 0.278 0.278
Correlation: p-value
Test for Second-Order Serial 0.767 0.280 0.319
Correlation: p-value
Number of Observations (Countries) 86 (46) 60 (22) 54 (20)

 

 

(*) In the levels speciﬁcation, all variables are assumed to be only weakly exogenous, except
for the GDP growth rate and the Drug Producers Dummy which are assumed to be strictly
exogenous. The second lag is used as an instrument for the lagged crime rate. As for the
other variables, the instrument used is the ﬁrst lag. The only exception to the previous rule is
regression (4), where the Gini index and the urbanization rate are assumed to be strictly
exogenous due to limitations in the sample size. In the speciﬁcation that includes both
differences and levels, the lagged ﬁrst differences are used as instruments in the equations in
levels, with the exception of the lagged crime rate for which we use the second lag of the
ﬁrst difference, and the GDP growth rate which is assumed to be strictly exogenous. In the
equations in differences, all ﬁrst differences are assumed to be strictly exogenous, except for
the lagged ﬁrst difference of the crime rate, which is instrumented with the third lag of the

crime rate (in level).

235

Table 3.5: GMM Estimates: Panel Regressions of Robbery Rates

(p-values in parenthesis)

 

 

 

 

 

 

(I) (2) (3) (4)
Regression Speciﬁcation Levels Dif.-Lev.
Instruments (*) Levels Lev.-Dif.
GDP Growth Rate -0.069 -0.096 -0.091 -0.072
(0.009) (0.000) (0.000) (0.000)
GNP per Capita 0.533 0.162 0.038 -0.045
(0.076) (0.017) (0.219) (0.035)
Gini Index 0.137 0.038 0.006 0.011
(0.000) (0.000) (0.003) (0.009)
Average Years of Schooling -0.010 0.031 -0.025
(0.866) (0.045) (0.093)
Urbanization Rate -0.000 -0.005 0.008 0.011
(0.980) (0.038) (0.000) (0.000)
Drug Producers Dummy 0.625 -0.478
(0.053) (0.000)
Drug Possession Crimes Rate 0.007 0.000 0.001 0.001
(0.000) (0.879) (0.012) (0.019)
Lagged Robbery Rate 0.891 0.833 0.839
(0.000) (0.000) (0.000)
Secondary Enrollment Rate 0.002
(0.191)
Constant -6.683 -1 .791
(0.013) (0.008)
Sargan Test of Overidentifying Restrictions: p-value 0.156 0.339 1 0.611 0.628
Test for First-Order Serial Correlations: p-value 0.004 0.091 I 0.057 0.053
Test for Second-Order Serial Correlation: p -value 0.045 0.313 | 0.760 0.539
Number of Observations (Countries) 133 (56) 77 (39) 1 58 (20) 50 (17)

 

(*) In the levels speciﬁcation, all variables are assumed to be only weakly exogenous, except
for the GDP growth rate and the Drug Producers Dummy which are assumed to be strictly
exogenous. The second lag is used as an instrument for the lagged crime rate. As for the other
variables, the instrument used is the ﬁrst lag. The only exception to the previous rule are
regressions (4) and (5), where the GNP per capita, the Gini index, the average years of
schooling and the urbanization rate are assumed to be strictly exogenous due to limitations in
the sample size. In the speciﬁcation that includes both differences and levels, the lagged ﬁrst
differences are used as instruments in the equations in levels, with the exception of the lagged
crime rate for which we use the second lag of the ﬁrst difference, and the GDP growth rate
which is assumed to be stricly exogenous. In the equations in differences, all ﬁrst differences
are assumed to be strictly exogenous, except for the lagged ﬁrst difference of the crime rate,
which is instrumented with the third lag of the crime rate (in level). In regression (8), the Gini
index is also assumed to be strictly exogenous due to limitations in the sample size.

236

 

 

Individual education (e): le ::> 77, lc, Tw, Tm

Individual criminal experience (d,-1): 1d,-) :> lc, 1w, im

Past incidence of crime in society (D14): lDtJ :> lo, M

Level and grth of economic activity (EA): TEA => ll, lw

Income inequality (INEQ): TINEQ :> Tﬂ-w), lm

Existence of proﬁtable criminal activities (DRUGS): TDRUGS :> T 1

Strength of police and justice system (JUST): TJUST: Tpr, Tpu

Other factors that affect the propensity to commit a crime (other):
lather :> lc, 1m

 

Figure 3.1: Underlying Determinants of Criminal Activities

237

 

 

 

 

 

 

 

c 7..
.9
s a.
Q
8
s 5‘
°.
8 44 ........
I
a 34 ...............................
E’
30: 2- ...................................
E 1]
0 t t :
1 970—74 1 975-79 1 980-84 1985-89 1 990-94

Period
NOTE: Weighted average calculated using the following sample of 34 countries: Argentina, Australia, Austria, The Bahamas,
Bahrain, Barbados, Bulgaria, Canada, Colombia, Costa Rica Cyprus, Denmark, Egypt, Germany, Greece, India, Indonesia,
Italy, Japan, South Korea, Kuwait, Malaysia, Mexico, Norway, Poland, Qatar, Singapore, Spain, Sweden, Syria, Thailand,
Trinidad and Tobago, United States, and Venezuela.

Figure 3.2: The World: Intentional Homicide Rate
(population-weighted average)’

238

int. Horn. per 100,000 Pop.

10

0

1 970—74

 

Middle-Low

 

Low

Middle-High \‘

High

NOTE: Income groups are deﬁned in terms of per capita income. Low-income = $765
or less, middle-low = 3766-51035, middle-high = $3,036—S9,385, and high = $9,386
or more, based on GNP per capita as of 1995.

 

 

 

1975-79 1980-84 1985-89 1990-94
Period
Figure 3.3: Median Intentional Homicide Rates by Income
Groups, 1970/1994

239

Int. Horn. per 100,000 Pop.

 

14

 

 

 

 

 

 

 

8 i
Africa l
5 .......................................................................................
4 ...................................................................... .-EUTlQJDiI
Asia 3 Central
2 ................................ T ...................... 1" rm...”
Middle East 8. North Africa
0 e . i
1970-74 1975-79 1980-84 1985-89 1990-94
Period
Figure 3.4: Median Intentional Homicide Rates by Regions,
1970-1994

240

 

 

 

 

 

 

    

 

  
 

 

     

 

 

 

 

 

 

 

E . E E E E
E E E
E lE El - e
3 E E E E E
E
E. E i E E
E E E E E
H b E e
E E E ElE ......
E
E .4 tr 41311.4 '1 I
. E E
E E _E E 1 I _
l+-:.._.l r.rIEr
_ _ E E
wilflllﬂlj 1 I ocEEmeq
_ E E E E. E
E4 Alﬁtmj a $303:

   

 

 

E. E E. / ‘I sees

   

   

 

 

E
E , E, .
E E_, ...
lEl-_,_.._l l, . +1.. I
E . _ I sea
I ... ... r. .
iEallliiErl Ill» 14 l -i E
E. E . / I ﬁes:
E .. _. .... / .r/
E ,r ,r __ . 3 E E
Elli E E E _ E E m 0 0
m w m w m w w 2 1

cascaded 8°69 .8. 82256...

Figure 3.5: Intentional Homicide Rates in South America and

Mexico, 1970/1994

241

“ 1990-94
1980-84

1 970-74

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

w m w m m m
cascaded 83x: .8 822on

l E E E E E
, E E E E E
E E E E E
b E E E E
a E E E
E E r
. E E E E
i E E E1 E H _
E E E E, E
E E E1 E a
- E E4 En E e t
- E E ElE E E
rlEllE E E A
- E E E
E E E
.E alE. P
Te
Ell .E ...E .E E 5 0E

dong-um _m
3:0
e530

«Enema

05.00
-. 8.558
mama—.2
859.0:
063.56
mes—Eon
89E»
82 830

3398
3088.2

SENSE.
quacam

Figure 3.6: Intentional Homicide Rates in Central America

and the Caribbean, 1970/1994

242

APPENDIX C

ADDITIONAL TABLES

243

Table 3.A.]: Description and Source of the Variables

 

 

Variable Description Source
Intentional Death purposely inﬂicted by another Constructed from the United Nations
Homicide Rate person, per 100,000 population. World Crime Surveys of Crime Trends and
Operations of Criminal Justice Systems,
various issues, except for Argentina, Brazil,
Colombia, Mexico, and Venezuela. The
data is available on the intemet at
http://www.ifs.univie.ac.at/~uncjin/wcs.htm
l#wcs123.
The data on population was taken from the
World Bank’s International Economic
Department database.
For the ﬁve Latin American countries listed
above, the source for the number of
homicides was the Health Situation
Analysis Program of the Division of Health
and Human Development, Pan-American
Health Organization, from the PAHO
Technical Information System. This source
provided us with data on the annual
number of deaths attributed to homicides,
which come from national vital statistics
systems.
Robbery Rate Total number of Robberies recorded Same as above.
by the police, per 100,000
population. Robbery refers to the
taking away of property from a
person, overcoming resistance by
force or threat of force.
Conviction The number of persons found guilty Same as above.
Rates (of of a speciﬁc crime (Intentional
Intentional Homicides, Theft, Robbery, or
Homicides, Assault) by any legal body duly
Theft, Robbery, authorized to do so under national
and Assault) law, divided by the total number of
the corresponding crime (in
percentage).
Police Number of police personnel per Same as above.
100,000 population.
Drug Possession Number of drug possession offenses Same as above.
Crime rate per 100,000 population.

 

244

Table 3.A.] (cont’d)

 

 

Variable Description Source
Drug Producers Dummy that takes the value one for International Narcotics Control Strategy
Dummy the countries which are considered Report, US. Department of State, Bureau
signiﬁcant producers of illicit drugs. for International Narcotics and Law
Enforcement Affairs, various issues.
Gini Index Gini Coefﬁcient, aﬁer adding 6.6 to Constructed from Deininger and Squire

Average years
of Schooling

Secondary
Enrollment

GNP per capita

Growth of GDP

Urbanization
Rate

Political

Assassinations
Rate

the expenditure-based data to make
it comparable to the income-based
data.

Average years of Schooling of the
Population over 15.

Ratio of Total Enrollment,
regardless of age, to the population
of the age group that oﬁicially
corresponds to the secondary level
of education.

Gross National Product expressed in
constant 1987 US. dollars prices.

Growth in the Gross Domestic
Product expressed in constant 1987
local currency prices.

Percentage of the total population
living in urban agglomerations.

Number of political assassinations
per 100,000 population.

(1996). The dataset is available on the
intemet from the World Bank’s Server, at
http://www.worldbank.org/htmI/prdmg/grth
web/datasets.htm.

Barro and Lee (1996). The dataset is
available on the intemet from the World
Bank’s Server, at
http://www.worldbank.org/html/prdmg/grth
web/datasets.htm.

World Bank, International Economic
Department data base.

Same as above.

Same as above.

Same as above.

Easterly and Levine (1997). The dataset is
available on the intemet from the World
Bank’s Server, at
http://www.worldbank.org/hthprdmg/grth
web/datasets.htrn.

 

245

Table 3.A.1 (cont’d)

 

 

Variable Description Source
Dummy for War Dummy for war on national territory Same as above.
on National during the decade of 1970 or 1980.
Territory
Absence of ICRG index of corruption in International Country Risk Guide.
Corruption government, ranging from 1 to 6,
Index with higher ratings indicating few
ethical problems in conducting
business.
Rule of Law ICRG measure of Law and Order Same as above.
Index Tradition, ranging ﬁ'om 1 to 6, with
lower ratings indicating a tradition
of depending on physical force or
illegal means to settle claims, as
opposed to a reliance on established
institutions and laws.
Index of Measure that two randomly selected Easterly and Levine (1997). The data-set is
ethnolinguistic people from a given country will not available on the intemet from the World
fractionalization belong to the same ethnolinguistic Bank’s Server, at
group (1960). http://www.worldbank.org/html/prdmg/grth
web/datasets.htm.
Buddhism Dummy for countries where CIA Factbook. The data is available on the
Dummy Buddhism is the religion with the intemet at
largest number of followers. http://www.odci.gov/cia/publications/pubs.
html.
Christian Dummy for countries where Same as above.
Dummy Christian religions are the ones with

Hindu Dummy

Muslim Dummy

Africa Dummy

the largest number of followers.

Dummy for countries where
Hinduism is the religion with the
largest number of followers.

Dummy for countries where Islam is
the religion with the largest number
of followers.

Dummy for Developing Countries
of Sub-Saharan Africa.

Same as above.

Same as above.

Classiﬁcation used in the Data Bases of the
World Bank International Economic
Department.

 

246

Table 3.A.l (cont’d)

 

 

Variable Description Source

Asia Dummy Dummy for Developing Countries Same as above.
of Asia.

Europe and Dummy for Developing Counties Same as above.

Cental Asia of Europe and Cental Asia.

Dummy

Latin America
Dummy

Middle East
Dummy

Africa and Latin
America
Dummy

Index of Firearm
Regulations

Alcohol
Consumption

Death Penalty

Dummy for Developing Counties
of Latin America.

Dummy for Developing Counties
of the Middle East and Northern
Africa.

Dummy for Developing Countries
of Africa and Latin America.

Measure of restrictions affecting
ownership, importing and mobility
of hand guns and long guns in the
early 1990s. Weights of .5, .25 and
.25 were given to the resulting
measures (2 given to county if it
prohibits or resticts all ﬁrearms; 1
given to county if it prohibits or
resticts some ﬁrearms; 0 given to
county if it does not have either
prohibitions or restictions on
ﬁrearms) regarding ownership,
imports and movement,
respectively.

Annual alcohol consumption per
capita in lites, covering the period
1982-1991.

Dummy for counties whose laws
do (1) or do not (0) provide for the
death penalty. Some counties
experienced changes, either
abolishing or imposing the death
penalty during 1970-94. Hence
period averages range between 0
and 1.

Same as above.

Same as above.

Same as above.

United Nations International Study on
Firearm Regulation at http://www.ifs.
univie.ac.at/~uncjin/ﬁrearms/

Alcoholism and Drug Addiction Research
Foundation (Toronto, Ontario, Canada) in
collaboration with the Programme on
Substance Abuse of the World Health
Organization. International Proﬁle: Alcohol
& Other Drugs, 1994.

 

 

Amnesty International. List of Abolitionist
and Retentionist Counties at
http://www.amnesty.org/ailib/intcam/dp/
abrelist.htm#7

 

247

Table 3.A.1 (cont’d)

 

 

Variable Description Source

Ratio of Males Ratio of number of males aged 15 to Pre-forrnatted projection tables in the
Aged 15 to 29 29 (34) to total population. World Development Indicators database of
(34) to Total the World Bank.

Population

 

248

Table 3.A.2: Summary Statistics of Intentional Homicide Rates by Country

 

 

(Annual Data)
County No. of Mean Standard Min. Max. First Year Last Year
Obs. Deviation
lndustalized and High-
Income Developing
Counties

Austalia 22 2.432 0.728 1.586 3.789 1970 1994
Austia 25 2.332 0.339 1.804 3.191 1970 1994
The Bahamas 22 27.950 20.587 6.322 83.088 1970 1994
Belgium 3 3.010 0.323 2.648 3.268 1983 1994
Bermuda 15 4.403 5.182 0.000 17.036 1980 1994
Canada 22 2.355 0.247 0.633 2.732 1970 1994
Cyprus 25 3.551 4.407 0.633 15.902 1970 1994
Denmark 25 3.706 1.939 0.507 6.013 1970 1994
Finland 20 5.445 2.568 2.192 10.061 1975 1994
France 17 3 .348 1 .963 0.400 4.937 1970 1994
Germany 21 3.432 0.231 3.045 3.886 1970 1990
Hong Kong 12 1.794 0.339 1.285 2.454 1980 1994
Israel 15 4.841 1.089 2.210 6.286 1975 1994
Italy 25 4.151 1.353 2.293 7.284 1970 1994
Japan 25 1.489 0.361 0.980 2.106 1970 1994
Kuwait 23 5.447 3.163 0.879 1 1.814 1970 1994
Luxembourg 2 7.586 0.01 1 7.578 7.594 1986 1990
Netherlands 16 1 1.1 15 2.476 7.303 15.994 1975 1990
New Zealand 17 1.201 0.502 0.586 2.411 1970 1986
Norway 21 0.959 0.618 0.205 2.546 1970 1990
Portugal 14 4.165 0.628 2.559 4.873 1977 1990
Qatar 25 2.100 0.767 1.103 3.674 1970 1994
Singapore 25 2.400 0.596 1.526 3.828 1970 1994
Spain 23 1.956 1.463 0.083 5.010 1970 1994
Sweden 25 4.06 2.885 1.243 9.532 1970 1994
Switzerland 18 1.883 0.899 0.395 3.188 1970 1994
United Arab Emirates 6 3.589 1.013 2.325 5.149 1975 1980
United Kingdom 15 1.920 0.394 1.481 2.566 1970 1986
United States 22 8.386 1.096 6.436 10.105 . 1970 1994

Latin America and the

Caribbean
Antigua & Barbuda 2 7.238 1.168 6.412 8.065 1985 1986
Argentina 18 5.159 1.347 3 .489 9.079 1970 1993
Barbados 16 5.909 2.3 17 2.893 1 1 .664 1970 1990
Belize 6 21.506 5.567 12.623 25.647 1975 1980
Brazil 16 14.497 4.270 7.699 21.614 1977 1992
Chile 16 5.662 3 .049 2.206 14.127 1970 1994
Colombia 19 44.962 26.634 13.895 86.044 1970 1994
Costa Rica 18 9.218 5.120 3.779 19.122 1970 1994

 

249

Table 3.A.2 (cont’d)

 

 

County No. of Mean Standard Min. Max. First Year Last Year
Obs. Deviation

Cuba 7 4.248 1.585 3.176 7.718 1970 1977
Dominica 7 0.080 0.0132 0.032 0.120 1980 1986
Ecuador 9 7.156 6.559 0.325 17.930 1970 1994
El Salvador 4 25.304 7.083 15.024 30.213 1970 1973
Guyana 7 7.873 1 .257 6.939 10.426 1970 1976
Honduras 12 7.1 10 3.087 3.327 13.326 1975 1986
Jamaica 20 19.536 7.949 7.596 41.678 1970 1994
Mexico 25 18.037 2.019 12.723 22.419 1970 1994
Nicaragua 5 21.297 3 .853 15.520 25.376 1990 1994
Panama 6 10.932 2.899 7.590 14.692 1975 1980
Peru 13 2.172 1.212 .035 4.777 1970 1986
St. Kitts & Nevis 9 6.592 3.468 2.347 11.450 1980 1990
St. Lucia 1 3.232 n.a. 3.232 3.232 1980 1980
St. Vincent & the Gre. 9 14.441 4.505 9.116 20.896 1980 1991
Suriname 9 7.605 9.908 1.089 30.757 1975 1986
Trinidad & Tobago 18 6.786 1.493 4.991 10.357 1970 1990
Uruguay 12 5.376 1.160 3 .680 7.367 1980 1994
Venezuela 23 10.017 2.643 7.280 15.833 1970 1994

Eastern Europe & Cental

Asia
Armenia 5 3.002 1.641 1.718 5.726 1986 1990
Azerbaijan 5 7.877 1.194 6.733 9.602 1990 1994
Belarus 5 7.142 1.666 5.316 9.193 1990 1994
Bulgaria 21 5.161 2.413 3.255 10.800 1970 1994
Croatia 5 10.584 3.635 7.283 14.925 1990 1994
Czech Republic 16 1.190 0.306 0.716 2.046 1975 1990
Estonia 5 15.774 7.21 1 8.685 24.350 1990 1994
Georgia 3 2.610 2.601 0.959 7.216 1990 1994
Gibraltar 8 2.389 3.524 0 2.532 1975 1986
Greece 25 1 .466 0.670 0.301 8 1970 1994
Hungary 15 3.853 0.470 2.981 4.517 1980 1994
Kazakstan 9 10.478 3.160 7.020 15.244 1986 1994
Kyrgyz Republic 5 10.912 2.547 8.191 13.720 1990 1994
Latvia 9 8.377 4.738 3.945 16.589 1986 1994
Lithuania 9 7.052 3.966 3.457 14.055 1986 1994
Macedonia 5 0.825 0.293 0.592 1.324 1990 1994
Malta 15 2.047 1.350 0.288 4.428 1980 1994
Moldovia 9 7.035 2.242 4.564 1 1.465 1986 1994
Poland 21 1.660 0.316 1.001 2.327 1970 1990
Romania 9 4.500 1.866 2.194 6.516 1986 1994
Russian Federation 9 11.928 5.738 6.301 21.815 1986 1994
San Marino 6 2.729 4.615 0 11.111 1970 1975
Slovak Republic 5 2.271 0.315 1.760 2.554 1990 1994

 

250

Table 3.A.2 (cont’d)

 

 

County No. of Mean Standard Min. Max. First Year Last Year
Obs. Deviation
Slovenia 9 4.1 17 0.609 3.181 5.209 1986 1994
Tajikistan 4 2.541 0.462 2.055 3.168 1987 1990
Turkey 6 16.556 1.960 14.090 19.931 1970 1975
Ukraine 15 5.196 1.538 3.439 8.804 1980 1994
Yugoslavia,FR(Serbia) 7 13.007 3.237 10.939 19.934 1975 1990
Middle East & North Africa
Algeria 6 0.924 0.336 0.524 1.468 1970 1975
Bahrain 15 1.388 1.172 0.382 5.042 1970 1990
Egypt,Arab Rep. 23 2.337 1.023 1.392 4.172 1970 1994
Iraq 9 10.517 2.212 8.076 13.482 1970 1978
Jordan 20 3 .352 1.756 1.822 7.038 1975 1994
Lebanon 9 15.495 12.478 4.479 42.898 1970 1988
Morocco 14 1 .071 0.386 0.689 2.157 1970 1994
Oman 6 0.824 0.893 0 2.461 1970 1975
Saudi Arabia 10 0.767 0.168 0.519 1.062 1970 1979
Syrian Arab Republic 22 4.083 1.431 1.964 6.263 1970 1994
Sub-Saharan Aﬁica
Botswana 10 10.179 1 .742 6.652 13.03 1 1980 1990
Burundi 7 1.088 0.164 0.758 1.284 1980 1986
Cape Verde 1 5.242 5.242 5.242 1979 1979
Ethiopia 5 10.185 2.678 5.682 12.298 1986 1990
Liberia 5 2.615 1.635 0.530 5.028 1982 1986
Madagascar 15 6.597 13.327 0.468 53.432 1975 1994
Malawi 7 2.762 0.483 2.059 3.399 1980 1986
Mauritius 15 2.652 0.399 2.081 3.448 1970 1994
SaoTomeandPrincipe 5 118.429 21.184 90.749 142.014 1990 1994
Senegal 6 2.186 0.277 1.914 2.598 1975 1980
Seychelles 6 4.467 2.234 1.642 8.335 1975 1980
South Africa 6 22.874 4.358 18.249 29.853 1975 1980
Sudan 15 5.406 1.301 3.262 7.045 1970 1994
Swaziland 5 68.048 8.495 58.813 81.738 1986 1990
Zambia 6 8.605 1 .293 6.973 10.160 1975 1980
Zimbabwe 10 9.544 4.909 4.336 18.344 1975 1994
South and East Asia
Bangladesh 12 2.541 0.392 1.984 3.340 1975 1986
China 5 0.965 0.078 0.867 1.076 1981 1986
Fiji 15 2.635 1.117 0.329 4.670 1970 1986
India 17 4.814 2.200 2.655 8.085 1970 1994
Indonesia 19 0.895 0.212 0.108 1.127 1970 1994
Korea,Rep. 19 1 .444 0. 168 1.235 1 .834 1970 1994
Malaysia 20 1.883 0.367 1.050 2.397 1970 1994
Maldives 5 1 .900 0.983 0.463 3.060 1986 1990
Myanmar 5 0.703 0.091 0.563 0.818 1986 1990

 

251

Table 3.A.2 (cont’d)

 

 

County No. of Mean Standard Min. Max. First Year Last Year
Obs. Deviation
Nepal 13 1.584 0.571 0.387 1.994 1970 1986
Pakistan 1 1 6.069 0.728 4.661 7.034 1970 1980
Papua New Guinea 2 2.080 0.143 1.979 2.181 1975 1976
Philippines 8 9.509 8.378 2.598 29.355 1970 1980
Sri Lanka 17 12.174 10.007 6.295 48.358 1971 1989
Thailand 12 21.506 1 1.354 7.556 41.776 1970 1990
Tonga 1 1 7.519 5.163 1.074 14.286 1975 1990
Vanuatu 4 0.881 0.343 0.678 1.395 1987 1994
Western Samoa 5 1.976 1.266 0.613 3.125 1990 1994

 

252

"‘llll'lliillliiill’5