INEQUALITY AND ECONOMIC GROWTH:
EVIDENCE FROM ARGENTINA’S PROVINCES USING SPATIAL
ECONOMETRICS
DISSERTATION
Presented in Partial Fulfillment of the Requirements for
The Degree Doctor of Philosophy in the Graduate
School of The Ohio State University
By
Alejandro A. Cañadas, M.B.A.
* * * * *
The Ohio State University
2008
Dissertation Committee: Approved by
Professor Claudio Gonzalez-Vega, Adviser
Professor Mark Partridge ______
Professor Joseph Kaboski Adviser Graduate Program in Agricultural, Environmental and Development Economics
Copyright by Alejandro Cañadas 2008
iii ABSTRACT
This dissertation analyzes whether or not the spatial distribution of inequality in the provinces of Argentina affects real per capita economic growth. The primary objective is to decouple the effect of inequality into within inequality, which is the own province i level of inequality, and the spillover of inequality from the closest provinces to province i. Furthermore, another important objective is to decouple the influence of inequality into long-run and short-run effect
To accomplish this, I based the analysis on a framework used by Partridge (2005), which starts considering a very simple model, called a “parsimonious” model with a few key variables. Building on that simple model I started adding a set of important control variables in order to get a more fully specified model, called “base” model. The main idea of using this methodology is that the “parsimonious” models, with only a few variables (income distribution and a few other control variables), not only reduces multicollinearity but also it is a test for robustness in the relationship between inequality and growth (Perotti, 1996; Panizza, 2002; Partridge, 2005).
In addition, following Partridge (2005), I considered that income distribution might have an entirely separate effect at the middle versus the tails of the distribution.
Therefore, I decided to include the Gini that controls for the overall distribution, and the third Quantile share (Q3) that controls for middle-class consensus and the role of the
ii median voter. The purpose of having two variables of income distribution is that when
the Q3 is included in the model, the Gini controls for the overall distribution, especially
at the tails, while Q3 controls for middle-class consensus and the role of the median
voter.
Additionally, a key difference from Partridge (2005) framework, apart from the decoupled effect of inequality into within inequality, which is the own province i level of inequality, and the spillover of inequality from the closest provinces to province i, is the explicit consideration of possible spatial autocorrelation in the models. To achieve this, I used two of the simplest spatial specifications: the spatial lag and spatial error models.
In the dissertation I have found very robust evidence that the own province i inequality, and the inequality of the neighboring provinces of province i, affects negatively the economics growth of the provinces of Argentina in the period 1991-2002.
Morerover, I have also found robust evidence that the third Quantile (Q3) affects negatively the economics growth, which is not consistent to the vibrancy of the middle class. The overall pattern of my results are not consistent with a long-run classical/incentive interpretation but to a political economy interpretation, in which the distortionary redistribution policies and social or political conflict are generated by the difference in inequality among provinces.
.
iii
Dedicated to my beloved family, my lovely wife, Cynthia, my son Santiago, and my daughter María Camila
iv ACKNOWLEDGMENTS
This dissertation is the end product of a five-year journey that began when
I started working toward my Ph.D. at The Ohio State University. Many people have walked (and stumbled) with me throughout these years. First and foremost, I would like to thank my advisor Dr. Claudio Gonzalez-Vega. His encouragement and guidance have been invaluable to go through some turbulent moments of the Ph.D. program, particularly the first year. I also want to thank Don Claudio for giving me the opportunity to work as his assistant since 2003. I learned a great deal from him and I will always remember him as a smart thinker, generous person, and enthusiastic teacher.
I also want to thank Dr. Mark Partridge and Dr. Joe Kaboski, who played a fundamental role in helping me develop this research. They were always ready to read my draft, give me precious advice, and offer suggestions that help me to be ready for the job market. Moreover, I am very grateful to Dr. Dave Kraybill and Dr. Ian Sheldon for teaching the best classes I have ever had and inspiring the topics for this dissertation. I am also very thankful to Stan Thompson, Fred Hitzhusen, Mario Miranda, and specially my advisor from the PFF Program (Preparing Future Faculty) Dr. Robert Ebert, from
Baldwin Wallace College, for all his support.
I am very grateful to Ricardo Martinez ([email protected] ) from the CEPAL office in Buenos Aires, who provided me with Argentina's provincial per
v capita GDP and to Dr. Leonardo Gasparini from CEDLAS, Universidad Nacional de La
Plata, Argentina ([email protected] ), who offered me useful comments in the manipulation of the survey from the EPH.
Working as a staff member at AEDE, I have had the pleasure to work with Joan
Weber and Susan Miller, who always have been very kind to me.
During these years, I shared wonderful moments with fantastic people that I want
to mention: Franz Gomez-Soto, Francisco Monge-Ariño, Erik Davidson, Mauricio
Ramirez, Maria Jose Roa, Carlos Alpizar, Jose Pablo Barquero, Malena Svarch, Paula
Cordero-Salas, Carolina Castilla, Emilio Hernandez, Scott Pearson, Carolina Prado, and
Marcelo Villafani.
I extend my love to my family, my dad, mom, Angeles and Marita, as well as my
friends, Hernan Bourbotte, Diego Sica, Octavio Groppa, Mariano Massano, Juan Pablo
Tiepolt, Jill Gerschutz, Ana Maria Gilmore, and William Hamant, and I thank them for
believing in me and for supporting my dreams from a distance.
Nothing would have been possible without my wife’s unconditional support, care
and love. She gives me the strength and courage to do things I would have never
imagined I could. I thank God for her and for our precious little son, Santiago, and our
daughter, María Camila, and for all God’s strength through all these years.
vi VITA
March 13, 1972………Born – Jujuy, Argentina.
1995 – 1996…………..Economist, Arthur Andersen
1997…………………..B.S. (Licenciatura) Economics, Universidad Católica Argentina
1996 – 2000…………..Marketing Researcher, Telefónica de Argentina
2000 – 2003…………..Masters of Business Administration,
University of Dayton, Ohio
2004– 2008…………...Graduate Research Associate, Rural Finance Program,
Agricultural, Environmental and Development Economics,
The Ohio State University
vii
FIELDS OF STUDY
Major Field: Agricultural, Environmental and Development Economics
Minor Fields: Development Economics
viii TABLE OF CONTENTS
Page Abstract...... ii
Dedication...... iv
Acknowledgements...... v
Vita...... vii
List of Tables ...... xiii
List of Figures...... xvi
Chapters
1. Introduction...... 1
1.1 Motivation...... 1 1.2 Growth, Distribution and Poverty...... 3 1.3 Spatial Dependence and Convergence...... 6 1.4 Research Questions and Objectives...... 6 1.5 Research Strategy...... 8 1.6 Hypotheses...... 9 1.7 Contents ...... 11
1.8 Significance and Relation to the Present State of Knowledge...... 11 1.9 The Influence of Inequality on Growth ...... 13
ix 2. Argentina...... 18
2.1 Argentina, a Beautiful Country...... 18 2.2 Initial Conditions ...... 19 2.3 Argentina as a Puzzling Country ...... 23 2.4 What Went Wrong, and When?...... 26 2.5 Volatility of Growth...... 27 2.6 The Argentinean Economy ...... 30 2.7 The Argentinean Economy ...... 32 2.8 A Caudillo Country...... 38
3. Economic Growth in Argentina...... 41
3.1 Per Capita Income in Latin America: A Long-Run Comparative Perspectives...... 43 3.1.1 Historical Per Capita GDP Estimates for Latin America...... 43 3.1.2 Income Convergence in Latin America ...... 46 3.2 Comparative Perspective ...... 47 3.3 The Data...... 49 3.3.1 Changes in geographical coverage...... 50 3.3.2 The New EPH Continua...... 51 3.3.3 Limitations ...... 53 3.4 Convergence Concepts and Spatial Effects ...... 54 3.4.1 Spatial Effects in the Analysis of Regional Income Convergence56 3.4.2 Exploratory Spatial Data Analysis of Argentina’s Income Convergence ...... 57 3.5 Spatial Autocorrelation ...... 58 3.5.1 Local spatial autocorrelation...... 62 3.6 Income Comparison among the Provinces of Argentina ...... 66
x 4. Inequality in Argentina ...... 68
4.1 Importance of the Study of Inequality ...... 69 4.2 Relationships among Poverty-Growth- Inequality…………………………………………………………………..69 4.3 Relationship between Inequality and Growth...... 70 4.4 Inequality in Argentina ...... 73 4.5 Inequality in Latin America...... 76 4.6 Regional Inequality in Argentina...... 81 4.7 Spatial Autocorrelation of Income Inequality...... 82 4.7.1 Local Spatial Autocorrelation for Income Inequality ...... 84
5. Empirical Implementation ...... 88
5.1 Spatial Econometrics ...... 88 5.1.1 The Problem of Spatial Autocorrelation ...... 89 5.1.2 Spatial Lag Operator ...... 93 5.1.3 Spatial Autocorrelation in a Regression Model...... 94 5.2 Inequality-Economic Growth Models...... 98 5.3 Regression Specification...... 100 5.3.1 Spatial Econometric Model Specification ...... 101 5.3.2 Spatial Lag Model...... 102 5.3.3 Spatial Error Model...... 103 5.4 Empirical Results...... 104 5.4.1 Parsimonious Long Run Model ...... 104 5.4.2 Base Long Run Model ...... 109 5.4.3 Gini Spillover Effect Model...... 116 5.4.4 Pooled-OLS Models...... 117 5.4.5 Instrumental Variables Model...... 134
xi 5.4.5.1 Durbin-Wu-Hausman Tests for Endogeneity ...... 134 5.4.6 Spatial Pooled-OLS Models ...... 137 5.4.7 Panel Data Models ...... 138 5.4.7.1 Fixed Effect Model ...... 142 5.4.7.2 Random Effect Model...... 144
6. Conclusions...... 149 6.1 Summary...... 149 6.2 The Main Results ...... 150 6.3 Contributions...... 154 6.4 Policy Implications ...... 157 6.5 Limitations and Future Research ...... 157
Bibliography ...... 168
Appendices...... 185 Appendix A: Tables and Figures for Chapter 1...... 185 Appendix B: Maps from Argentina ...... 192 Appendix C: Geary-Khamis method of aggregation ...... 205 Appendix D: Tables and Figures for Chapter 3...... 214 Appendix E: Tables and Figures for Chapter 4 ...... 238
xii LIST OF TABLES
Table Page
2.1 Comparison of per capita GDP among developed countries, 1820-1990…25
2.2 Summary statistics for the annual rate of growth of per capita GDP for
Argentina, 1810-2004 (percentage) ...... ……………28
2.3 Comparison of GDP (PPP) and per capita GDP (PPP) for the seven largest Latin
American economies, in 2007...... 30
2.4 Comparison of labor force, unemployment rate, poverty and inequality for the
seven largest Latin American economies, in 2007 ...... 31
A.1 Population living below the US$ 1 poverty line, 1990 and 2001 ...... 188
A.2 Population living below the US$ 2 poverty line, 1900 and 2001...... 189
A.3 Indicators of inequality for selected Latin American countries, the United
States, and Italy, late 1990s...... 190
D.1 Average per capita income growth rates for the seven major Latin American
economies, 1810-2004 ...... 215
D.2 Average annual rates of growth of per capita GDP for regions of the world,
1820-2004 (percentage)...... 221
D.3 Economic Regions and Provinces in Argentina. EPH...... 223
D.4 Urban Conglomerates by Provinces in Argentina...... 224
xiii D.5 Moran’s I statistics for the provincial per capita GDP of Argentina, 1980-
2002………………...... 228
D.6 Summary of local Moran statistic as a measure of spatial association: real per
capita GDP by quadrants, 1980-2002 ...... 232
E.1 Distribution of household per capita income in Argentina (deciles shares and
income ratios), 1992-2005 ...... 240
E.2 Inequality Indices from household surveys in major provincial cities in
Argentina, 1992-2005 ...... 241
E.3 Inequality in Latin America between 1950 and 2000. Measured by Gini
coefficients...... 243
E.4 Changes in inequality measured by percentage points of Gini Coefficient using
household surveys in each country ...... 244
E.5 Bonferroni and the Tukey’s tests to determine means differ in Gini coefficient
among regions in Argentina, 1991-2002 ...... 247
E.6 Changes in Gini coefficient, third quantile (Q3), top 10 percent and bottom 20
percent shares in income of the population by region, between 1991 and 2002
(percentage)...... 248
E.7 Estimates of the Moran’s I statistic for the provincial Gini coefficients of
Argentina, 1991-2002 ...... 250
5.1 Results from parsimonious long-run models ...... 108
5.2 Base Long-Run Models ...... 115
xiv 5.3 Growth Model with Gini Spillover Effects...... 117
5.4 Pooled-OLS Models...... 126
5.5 Robust, Clusters, FGLS, IV, and GMM Models ...... 133
5.6 Spatial autocorrelation test statistic for the spatial lag and error models . 140
5.7 Descriptive Statistics of the Panel Data Variables...... 139
5.8 Panel Data Models of Real Per Capita Growth ...... 147
5.9 Fixed Effect and Random Effect Models of Real Per Capita Growth...... 148
xv LIST OF FIGURES
Figure Page
2.1 Map of the Republic of Argentina, its main cities, and neighboring countries
...... …… 21
2.2 Comparison of per capita GDP relative to the US for Argentina, Brazil and
Mexico, 1820-1990...... …… 24
2.3 Annual rate of growth of per capita GDP for Argentina, 1810-2004 (percentage)
...... …… 27
2.4 Coefficient of variation of the log of inter-annual per capita GDP for each
decade in Argentina, 1810-2004...... …… 29
A.1 Gini coefficient for the income distribution in Latin America, 1950-2000 186
A.2 Poverty rates in Latin America, 1950-2000...... …… 187
A.3 Trends in inequality in major Latin American countries from the early 1980s to
mid-2000s (Gini coefficients)...... 191
B.1 Satelital map of Argentina using Google Earth...... 193
B.2 Satelital map of Argentina in South America using Google Earth...... 194
B.3 Satelital map of the provinces of Argentina using Google Earth ...... 195
B.4 Map of the provinces of Argentina...... 196
B.5 Map of the regions of Argentina...... 197
B.6 Per capita income growth by province (1992-2002) using STATA...... 198
xvi B.7 Gini Coefficient by province in 1991 using STATA...... 199
B.8 Gini Coefficient by province in 2002 using STATA...... 200
B.9 Area of provinces in Argentina using STATA ...... 201
B.10 Density of population in the provinces of Argentina using STATA...... 202
B.11 Two main clusters for the real per capita income in the provinces of Argentina
using the Moran I, 1980-2002...... 203
B.12 Two main clusters for the Gini in the provinces of Argentina using the Moran I,
1991-2002 ...... 204
D.1 Per Capita GDP for seven major Latin America economies, 1900-2004 216
D.2 Per Capita GDP for Argentina, 1810-2004...... 217
D.3 Average annual GDP growth rates experienced by the seven largest Latin
American economies between 1900 and 2004, with their corresponding (logged)
initial per capita income level in 1900...... 218
D.4 Cross-country standard deviation of per capita GDP for the seven largest Latin
America economies, 1900-2004 ...... 219
D.5 Coefficient of variation of log of per capita GDP for the seven largest Latin
America economies, 1900-2004 ...... 220
D.6 Per capita GDP for the United States, United Kingdom, Germany, Norway,
Australia, Japan and Argentina, 1820-2004...... 222
xvii D.7 Average annual GDP growth rates experienced by 23 provinces of Argentina,
1980-2002, with their corresponding (logged) initial per capita income level in
1980...... 225
D.8 Coefficient of variation of the log of provincial real per capita GDP for 23
provinces and the capital city (Ciudad Autónoma de Buenos Aires) in
Argentina, 1980-2002 ...... 226
D.9 Moran's I statistic for the provincial real per capita GDP of Argentina, 1980-2002
...... 227
D.10 Provincial coefficient of variation of the log of provincial real per capita GDP
for 23 provinces and the capital city of variation and the Moran's I statistic,
Argentina, 1980-2002 ...... 229
D.11 Local Moran’s I statistic for the provincial real per capita GDP in 1980. 230
D.12 Local Moran’s I statistic for the provincial real per capita GDP in 2002. 231
D.13 GDP participation of the five richest provinces in Argentina, 1980, 1991 and
2002...... 234
D.14 Population participation among the five largest regions...... 235
D.15 Population shares of the five largest provinces in Argentina, 1980, 1991 and
2002...... 236
D.16 Comparison of provincial per capita income relative to the country’s average,
for 1980, 1991 and 2002 ...... 237
xviii E.1 Decomposition of a change in distribution and poverty into growth and
distributional effects...... 239
E.2 Gini Coefficient for Argentina, from the distribution of per capita household
income, 1992-2005 ...... 242
E.3 Provincial Gini coefficients for Argentina. Averages for 1991-2002...... 245
E.4 Regional Inequality in Argentina, as shown by Gini coefficients. Averages for
1991-2002 ...... 246
E.5 Moran’s I statistic for the provincial Gini coefficients of Argentina, 1980-2002
...... 249
E.6 Local Moran’s I statistic for the Gini coefficients provincial in 1991...... 251
E.7 Local Moran’s I statistic for the Gini coefficients in 2002...... 252
E.8 Local Moran’s I statistic for the Gini coefficients in 2001...... 253
xix CHAPTER 1
INTRODUCTION
1.1 Motivation
This dissertation is about the effects of the inequality in the income distribution on per capita GDP growth in the provinces of Argentina.
A recent report of the World Bank claims that the levels of inequality in Latin
America are well above those of developed countries. The Gini coefficient for the region is about 0.55, compared to 0.37 for developed countries (de Ferranti et al., 2004). The report also indicates that, together with Sub-Saharan Africa, Latin America has long been known as the region with the highest inequality in the world, with a Gini coefficient above 0.50 since the 1960s, as shown in Figure 1 in Appendix A. While inequality in
Argentina has been below the region’s average, it has been increasing over time.
The fact that the first of the Millennium Development Goals is to “eradicate extreme hunger and poverty by 2015” shows that poverty is perceived as one of the most important problems in the world. Both poverty and persistent inequality are major challenges in Latin America.
Tables 1 and 2 in Appendix A show that, by 2001, some 128 million people in
Latin America lived below the US$2-a-day poverty line. Of those, 10 million lived below
1 the US$1-a-day poverty line. Nevertheless, if we look at the incidence of poverty in Latin
America and the Caribbean, as a share of the total population, the region’s poverty levels are among the lowest, well below those for East Asia, South Asia, and Sub-Saharan
Africa. In contrast to those regions, poverty in Latin America is less a rural and more an urban phenomenon.
In the long term, the region has experienced important gains in alleviating poverty. Figure 2, in Appendix A, shows that, while measured using a poverty line of
US$ 2 per person a day, in Latin America the incidence of poverty declined rapidly between 1950 and 1980, but this process slowed down and the poverty rate was still above 20 percent by 2000.
Despite this progress, the problem continues to be acute. According to estimations of the Economic Commission for Latin America and the Caribbean for 2006, 71 million persons (13.4 percent of the total population of Latin America) were extremely poor, while the number of poor people (including those 71 million) was estimated at 194 million, equivalent to 36.5 percent of the region’s population (ECLAC, 2007).
According to some authors, in Latin America the problems of poverty and inequality have been related to weak economic growth (Perry et al., 2006). Inequality, in turn, may have also influenced economic growth itself. Thus, inequality matters per se and because of its association with other critical outcomes, such as poverty and a country’s growth performance.
Both within the region and in comparison to other regions, inequality varies a lot.
In most Latin American countries, the richest 10 percent of the individuals earned between 40 and 47 percent of total income, while the poorest 20 percent received only
2 between 2 to 4 percent (Table 3 in Appendix A). In contrast, the richest 10 percent in the
United States received 31 percent of total income while, in Italy, individuals in this decile earned 27 percent. Even the most egalitarian countries in Latin America (Costa Rica and
Uruguay) show comparatively high levels of income inequality. This concentration has been substantially higher than in OECD countries, Eastern Europe, and most of Asia.
Only some countries in Africa and the successor states of the former Soviet Union show comparable inequality (de Ferranti et al., 2004). Moreover, when inequality is measured as the shares of the richest and the poorest quintiles, Latin America is the least equitable region in the world.
In Argentina, inequality has dramatically increased since 1950. The Gini coefficient for the household per capita income distribution in the urban areas increased from 0.421 in 1992 to 0.535 in 2003 (CEDLAS, 2003). Even if observations for the recent years of economic crisis are ignored, the trend toward increased inequality is evident. This trend has steadily reduced the difference between Argentina and other major countries with more concentrated distributions in Latin America (Figure 3 in
Appendix A). No other Latin American country has experienced such deep distributional changes as Argentina (Gasparini and Sosa Escudero, 2001). This dissertation addresses the consequences of this inequality in distribution on regional economic growth.
1.2 Growth, Distribution and Poverty
The performance of an economy is usually described in terms of mean variables, such as per capita GDP or disposable income. If per capita GDP increases, namely, if
“the economy grows,” performance is considered to be positive. While important, this
3 evaluation is incomplete, because it does not consider the distribution of income.
Actually, an increase in mean income can reflect different combinations of poverty and distributional changes. However, the linkages among growth, poverty, and income distribution are complex (Ferreira, 1998). Some authors (Ravallion, 2001; Besley and
Burgess, 2003) illustrate the relationships between poverty, income growth, and inequality by providing examples of how both growth and changes in inequality influence poverty.
A widespread concern for pro-poor growth has resulted in part from evidence showing that, in some countries, the fruits of economic growth have not been equally shared by the population and from evidence that during some growth events the well- being of the poor actually decreased (Perry et al., 2006). The relationship is even more relevant if an economy experiences negative growth, when poverty deepens further. In
Latin America, growth and poverty patterns have differed substantially across countries and within countries over time. There are cases of sustained growth and poverty reduction, like Chile, along with unfavorable experiences in terms of poverty, such as
Argentina and Venezuela.
This mix of experiences makes the analysis of pro-poor growth particularly rich.
Moreover, the issue has been at the core of political debates. Recent episodes of
economic growth combined with unchanged or even increasing poverty have made some
people question the proposition that growth is strongly linked to poverty reduction.
Others have been concerned about the ways in which initial poverty influence growth and
have explored the persistence of poverty traps (Azariadis, 1996a, 1996b). This
4 dissertation is concerned with how inequality may, in turn, influence the process of growth itself.
A multiplicity of factors induce changes in incomes. These changes usually modify several dimensions of the income distribution, like the mean income, its degree of dispersion, and the mass below certain cut-off income level. In this sense, growth (linked to shifts in the mean of the distribution), changes in inequality (linked to variations in the dispersion of income), and changes in poverty (linked to movements toward the lower tail of the distribution) are all particular manifestations of changes in the whole distribution.
Thus, in a given period, one should not think of changes in poverty as being caused by growth and changes in inequality but, rather, all three are outcomes of changes in the distribution caused by other determinants. Current inequality may have consequences, however, on future growth and other outcomes. This dynamic relationship constitutes the concern of this dissertation.
Researchers have found it useful, nevertheless, to decompose changes in the distribution of income in a given period into two dimensions: changes in its central position (growth) and changes in its dispersion (inequality). Each of these dimensions implies, in turn, changes in the concentration in the lower tail of the distribution
(poverty). From this static perspective, changes in poverty are presented as “the result” of growth and changes in inequality (Bourguignon, 2004).
This decomposition does not predict, however, how different degrees of inequality may influence the rate of economic growth over time or the reverse question of how different patterns of growth may result in diverse paths of inequality. This dissertation focuses on the first issue: the relationship between income inequality and
5 income growth over time; that is, it examines the dynamic relationship between previous levels of inequality and current growth, in a regional context and taking into account the spillover effects of inequality among neighboring provinces in Argentina .
1.3 Spatial Dependence and Convergence
In the regional economics literature, the role of spatial relationships has only in recent times caught attention, while the earlier literature on regional inequality was practically silent on the difficulties that spatial data create and on the insights to be gained from them. These spatial dimensions are not ignored here.
Over the last decade, there have been an increasing number of empirical studies on regional convergence. As Rey and Janikas (2005) emphasize, there are promising opportunities for integration of the inequality and convergence literatures as well as for a reassessment of the relationship between growth and inequality (Benabou, 1996). The possibility of integration is currently considered modestly at the international level, while such articulation at the sub-national and regional levels remains essentially unexplored. I would like to fill this gap by analyzing the relationships among inequality, per capita
GDP growth, and spatial dependence, applied to the regions of Argentina.
1.4 Research Questions and Objectives
The basic research questions that I address in the dissertation are:
1) Does inequality in income distribution affect economic growth in the provinces of
Argentina?
6 2) Is the relationship between inequality and growth influenced by the spatial
distribution of inequality in the provinces of Argentina?
To answer these research questions, I approach them in several steps:
1. Does spatial clustering help in explaining differences in the growth of per capita
income across regions in Argentina?
2. Does spatial clustering help in explaining differences in the inequality of the
distribution of per capita income across regions in Argentina?
3.a Are the relationships between the own province i inequality and economic
growth, on the one hand, and the inequality of the neighboring provinces and
economic growth in province i , on the other, negative or positive for
Argentina’s provinces?
3.b. Is this relationship between distribution and growth reinforced when controlling
for spatial dependence, by using spatial econometrics procedures?
3.c. What are likely explanations of the relationship found?
The general and specific objectives of the dissertation are closely interrelated. These objectives are:
1. To compare the path of economic growth in Argentina with economic growth in
selected countries in Latin America and regions of the world and gain an
understanding of the country’s overall growth experience.
2. To determine whether there has been convergence in per capita income across
regions in Argentina.
7 3. To assess the importance of spatial clustering on the processes of growth
convergence and income inequality.
4. To identify the relationship between inequality and economic growth in
Argentina. In particular, the purpose is to isolate the influence of provincial
inequality on the current differential rates of growth across regions. Moreover, I
want to determine whether or not inequality has a short-run or a long-run effect
on growth.
5. To consider some potential explanations of the results as a gateway towards
future research.
These are the main purposes of the dissertation. The results are expected to assist in identifying pro-poor growth policies for Argentina.
1.5 Research Strategy
Most of the empirical income distribution and growth literature relates to comparisons across countries, while the literature that investigates the inequality-growth relationship across regions within countries has mostly focused on developed economies.
This dissertation applies these approaches to a middle-income country that has experienced substantial volatility in GDP growth rates, Argentina.
The strategy for the empirical analysis relies on a framework used by Partridge
(2005), which starts by considering a very simple model of regional growth, called a
“parsimonious” model, with only a few key variables. Building on this simple model, I add a set of important control variables in order to get a more fully specified model,
8 called the “base” model. In addition, I apply some spatial econometrics techniques in order to test for the presence of spatial autocorrelation among the units of analysis and, if it is appropriate, to control for it.
To gain further understanding about the relationship between distribution and
growth, this dissertation attempts to test some hypotheses. Since the research question is
broken into several parts, the hypotheses follow a four-step scheme. If any one of the null
hypotheses is rejected, alternative hypotheses that reflect my expectations are proposed.
1.6 Hypotheses
First step :
Null hypothesis
H0: If one takes into account spatial clustering in the provinces of Argentina, this will not help to explain better the growth in per capita income across regions.
Alternative hypothesis
H 1: If one takes into account spatial clustering in the provinces of Argentina, this will help to explain better the growth in per capita income across regions (a more significant and more powerful influence).
Second step :
Null hypothesis
H0: If one takes into account spatial clustering in the provinces of Argentina, this will not help to explain better the inequality in the distribution of per capita income in each region.
9 Alternative hypothesis
H 1: If one takes into account spatial clustering in the provinces of Argentina, this will help to explain better the inequality in the distribution of per capita income in each region (a more significant and more powerful influence).
Third step:
Null hypothesis
H0: Higher levels of inequality of the per capita income distribution in each different province and its neighboring provinces will not be associated with differences in the provincial rates of growth of per capita income.
Alternative hypothesis
H 1: Higher levels of inequality of the per capita income distribution in each different province and its neighboring provinces will be negatively associated with the provincial rates of growth of per capita income.
Fourth step:
Null hypothesis
H0: Controlling for spatial dependence, the negative relationship between inequality and growth will remain unchanged.
Alternative hypothesis
H 1: Controlling for spatial dependence, the negative relationship between inequality and growth will be stronger (more significant and more powerful).
10 1.7 Contents
The dissertation is divided into six chapters. Chapter 2 focuses on the country,
Argentina, analyzing its historical development in comparison to other countries as well
as considering its recent political context. Chapter 3 assesses the economic growth of
Argentina, analyzing its evolution and the effects of the spatial distribution of growth
among its provinces. This includes an analysis of the convergence hypothesis at the
regional level. Chapter 4 reviews the evolution of inequality in Argentina and its
relationship with economic growth. In this chapter, I also explore the spatial effects of the
provincial distribution of income. Chapter 5 determines the extent to which income
inequality affects real per capita income growth in the long-run and the short-run, while
controlling for substantive and nuisance spatial dependence, as practiced in spatial
econometrics. Also, this chapter attempts to find whether spillovers of inequality exist
among provinces. Chapter 6 concludes with the main findings of this dissertation.
1.8 Significance and Relation to the Present State of Knowledge
Today, there is much interest in the spatial patterns of inequality and in the
dynamics of geographic income disparities. Since Krugman (1999), there have been
concerns with levels of spatial income inequality, their persistence, and the fundamental
processes that give rise to them. These issues have been investigated across the global
economy down to the level of the neighborhood. Surveys of the literature on convergence
at the international and regional level can be found in: de la Fuente (1997), Durlauf and
Quah (1999), Temple (1999), Florax and Folmer (2002), Fingleton (2003), Islam (2003),
and Magrini (2004).
11 Despite the reappearance of interest in regional economic growth and inequality, the geographical dimensions of the data underlying empirical analyses have received much less consideration. This has been the role of the literature on spatial econometrics
(Anselin and Florax, 1995; Anselin, 2001; Anselin, 2002) and spatial statistics (Getis et al., 2004).
Evidence suggesting that physical location and geographical spillovers matter more than traditional macroeconomic factors are grounded in Quah (1996) and Moreno and Trehan (1997). Modern applications of formal spatial econometrics methods to the question of regional convergence have produced new insights about the nature of spatial economic change (Rey and Montouri, 1999; Fingleton and López-Bazo, 2006).
These trends can also be found in the literature on convergence and inequality in
Latin America. For example, in the past decade there has been a growing literature on the empirical analysis of regional growth and inequality in many countries of the world, including some from Latin America, such as Dobson and Ramlogan (2002) and Serra et al. (2006) for Latin America; Gerber (2002) and García-Verdú (2002) for Mexico; Aroca and Bosch (2000) for Chile; Ferreira (1998) and Azzoni (2001); for Brazil; and Utrera and Korosch (1998) and Marina (2002) for Argentina. Despite this rich empirical literature, relatively few studies have compared the rates of growth of per capita income and inequality within national systems. Moreover, comparative studies have tended to concentrate on the more advanced economies, such as the European countries, the US,
Canada and Australia, among others.
The underlying geographic dimensions of regional growth processes have not received, however, attention in these investigations. In the case of Argentina, none of the
12 studies has used any spatial econometrics methods to analyze the geographical dimensions of the data.
I expect that, in Argentina, not only inequality within the province but also the spillovers of inequality from neighboring provinces will affect economic growth.
Furthermore, there are short-run and long-run effects of inequality on economic growth. I attempt to disentangle these effects in this dissertation. While, chapter 4 focuses on the provinces own inequality in a long-run framework, chapter 5 specifies econometric models to capture the effects of inequality in neighboring provinces as well as their short- run and long-run effects on economic growth.
In summary, due to the fact that there is a lack of spatial considerations in the recent literature on inequality and growth in Latin America and, specifically, in
Argentina, this dissertation attempts to close the gap. The goal is to reconsider the relationships between inequality and growth, taking into account the impact of spatial location. Thus, a contribution of this dissertation is the association of physical location and geographical spillovers to the regional levels of inequality and growth.
1.9 The Influence of Inequality on Growth
The key question for this dissertation is how inequality influences growth. This is a complex relationship, as there are multiple channels through which inequality affects growth with a positive influence and multiple channels through which inequality affects growth with a negative influence. The net outcome depends on which influences dominate. Thus, if the relationship between inequality and growth is ambiguous at the
13 theoretical level, the actual impact of inequality on economic growth is an empirical question.
The literature review identifies several of these channels. On the one hand, there are several positive influences of inequality on growth, usually associated with a
“classical” economic approach. This approach stresses how inequality enhances incentives that increase efficiency, growth and capital accumulation. For example:
1) Inequality in income distribution may increase the rate of savings
(through different propensities to save across income classes) and
capital accumulation and, thereby, accelerate growth (Kuznets, 1955;
Kaldor, 1956). 1
2) Inequality may signal opportunities to improve one’s income (or one’s
position in the income distribution), through greater work effort and
diligence, as well as Schumpeterian factors such as entrepreneurship,
risk-taking, and innovation, which are also sources of growth (Siebert,
1998; Bell and Freeman, 2001).
On the other hand, inequality may have negative effects on growth. There are at least five conceptual reasons why this might be the case.
1) If credit or insurance markets are imperfect, economic agents may
depend entirely on their initial wealth to undertake important
investment projects. Given these imperfections, the poor would be
1 This view suggests that inequality may have a positive effect on income growth if it is conducive to greater savings. Research by the Rural Finance Program at The Ohio State University has accumulated evidence that the poor save a higher proportion of their income than the rich, contrary to Kaldor (Adams, 1983). The individual amounts involved are small, though, and incomplete financial markets reduce the extent to which these savings are intermediate for swifter growth.
14 unable to invest in socially efficient (that is, profitable) projects (Galor
and Zeira, 1993; Banerjee and Newman, 1993; Aghion et al., 1999;
Levine, 2004).
2) The second conceptual reason why inequality may lead to lower
growth rates includes several political economy dimensions. In
societies with high degrees of concentration of power and wealth, the
elites may have more success in selecting economic strategies that
benefit them rather than middle and lower income groups. Their rent-
seeking behavior usually distorts resource allocation in the economy,
affecting factor productivity and hurting economic growth (Perotti,
1993; Bertola 1993; Alesina and Rodrik, 1994: Persson and Tabellini,
1994; Benabou, 1996; Barro, 2000; Leon 2007).
3) The third conceptual reason why inequality may lead to lower rates of
economic growth emerges when income inequality encourages social
conflict and results in more crime and illegal activities, which
discourage investment and weaken property rights (Hibbs, 1973;
Venieris and Gupta, 1986; Gupta, 1990, Alesina and Perotti, 1996;
Benhabib and Rustichini, 1996; Banerjee and Duflo, 2000). Inequality
of wealth and income motivates the poor to engage in crime, riots, and
other disruptive activities. The stability of political institutions may
even be threatened by revolution, so that laws and other rules have
shorter expected duration and greater uncertainty. The participation of
the poor in crime and other antisocial actions represents a direct waste
15 of resources because the time and energy of the criminals and of the
law enforcers are not devoted to productive efforts. Moreover, the
threats to property rights deter investment. Through these various
dimensions of socio-political unrest, more inequality tends to reduce
the rate of growth of an economy.
4) Inequality adversely affects social capital, which is typically defined as
the level of trust, civic norms, and social networks in a society that
facilitate contracts and transactions and maintain social stability
(Knack and Keefer, 1997; Kawachi et al., 1997; Nan, 2000; Caramuta,
2005).
5) Inequality adversely affects the set of opportunities specifically related
to education, distribution of assets and land, political influence, public
infrastructure and health (Sen, 1992; Roemer, 1998; Ferreira, 2001;
Bourguignon, Ferreira, and Menéndez, 2003). Thus, more inequality
implies a more narrow set of opportunities in society and less
productivity of resources, which tends to reduce the rate of growth of
an economy.
6) Inequality may have an entirely separate effect at the middle versus the
tails of the distribution. For example, Easterly (2001) argues that a
middle-class consensus promotes growth by encouraging stability,
mass education, better public services, and property rights. This
“consensus” appears closely related to the social capital literature that
stresses cohesiveness and trust (Bowles and Gitis, 2002; Henry, 2002;
16 Glaeser at al., 2002), and to the existence of income mobility across
society (Partridge, 2005).
The influence of inequality in the income distribution on economic growth may
have different short-run and long-run effects (Forbes, 2000; Partridge, 2005), and
different effects in the rural versus the urban areas. 2 For example, Fallah and Partridge
(2006) have found for the U.S. that in the urban areas there is a positive relationship between inequality and growth, while in the rural areas there is a negative relationship between inequality and growth. The authors reinvestigated the inequality-growth relationship using U.S. county-level data and they found that the inequality-growth relationship could completely vary even within states. In urban areas, because factors such as agglomeration economies and specialization of labor play a primary role in generating economic growth, greater income inequality intensifies the market rewards for the most able, attracting more skilled and specialized workers. Conversely, in smaller rural communities, more intimate personal relationships and lack of anonymity mean that greater income inequality takes on a personal nature that weakens social cohesion and, in turn, economic growth
My expectation is that, for environments like Argentina in recent times, the net effect of inequality of growthis likely to be negative. This dissertation attempts to establish if this is the case.
2 Following Kucera (2002), suppose a ‘‘middle-class consensus’’ leads to greater taxes used to fund an improved education system. Short-run growth may decline through higher taxes, but the long-run effects are positive, as the workforce becomes more productive. Similarly, because increased inequality is associated with liquidity/ credit constraints, it can produce greater cyclical volatility, which can depress short-run investment and growth (Aghion, Caroli, and García-Peñalosa, 1999).
17 CHAPTER 2
ARGENTINA
2.1 Argentina, a Beautiful Country
Argentina is a beautiful, wealthy country with abundant natural resources, a large territory with low density of population, and a fascinating economic history. Argentina’s experience of economic growth has been perplexing. Scholars like della Paolera and
Taylor (2003) have examined “the Argentina puzzle,” wondering why the country was very rich until at least around 1913 and today it is relatively poor. This deterioration may help to explain why psychoanalysis and the nostalgia of the tango music are so popular in
Argentina. Its affluent past is revealed by the “Belle Époque splendor” of Buenos Aires, which was developed as the economic, cultural, and political center of the country with a
French accent . In contrast, the number of scavengers that comb its streets reveals its current poverty.
Argentina’s economic history has been marked both by the impressive performance of its per capita income growth from 1880 to 1913 and by its ups and downs afterwards. This volatility has reflected the fragility of its political economy. The
Executive Branch of Government has repeatedly violated the rule of law, and democratic
18 institutions have broken down many times during the 20th century. For this reason, the state has been able to expropriate private savings often, through hyperinflation or devaluation. Cavallo, Minister of Finance in the early 1990s, felt the need to set up a currency board in order to gain the people’s confidence in the economic system.
This dissertation considers how one particular initial condition, income inequality, affects the process of economic growth within and among economic regions, while controlling for regional spillovers. High levels of poverty and an unequal distribution of income may create poverty traps or vicious circles that negatively affect growth (Perry et al., 2006).
This is not, however, a dissertation about the economic history of Argentina.
Instead, the goal of this chapter is to describe some facts of Argentina’s economic history directly related to its economic growth and income distribution. Further, Argentina’s growth record will be compared to the record of other countries, to highlight its unique evolution.
2.2 Initial Conditions
The sources of economic growth can be traced to a variety of factors: by and large, investment that increases the quantity and improves the quality of existing physical, human, and natural resources and that raises the productivity of resources through institutional change and technological progress. These sources of growth are influenced by differences in a country’s initial conditions:
a. Endowments of physical, human, and natural resources.
b. Levels of per capita income, poverty, and the distribution of wealth.
19 c. Climate and other sources of risk.
d. Size of the population and the country’s stage in the demographic
transition.
e. Domestic and international migration.
f. Opportunities for international trade.
g. Basic scientific and technological research and development capabilities.
h. Stability and flexibility of political and social institutions.
Some of these key initial conditions will be described next. Argentina has an area of almost 3.8 million square kilometers. Of these, 2.8 million are on the mainland and the rest are in Antarctica. Its borders with Uruguay, Brazil, Paraguay, Bolivia and Chile form a perimeter of 9,376 Km. and the territory bordered by the Atlantic Ocean is 4,725 Km. long (see Figure 2.1 and Maps 1 to 4 in Appendix B).
20
Source: http://www.argentina.gov.ar/argentina/portal/paginas.dhtml?pagina=1486
Figure 2.1: Map of the Republic of Argentina, its main cities, and neighboring countries.
21 Argentina is the ninth largest country in the world and the second largest in South
America, after Brazil. Its main geographic feature is the enormous contrast between the rich plains of the Pampas in the northern half, the sterile and stony plateaus of Patagonia, and the impressive rocky Andes.
The population is about 40 million (July 2007 est.). Population density is a low 13
inhabitants per square kilometer (see Maps 9 and 10 in Appendix B). However, almost
half of its population lives in the city and the province of Buenos Aires. The demographic
transition is essentially over and population grows 0.9 percent per year (2007 est.). The
population is mostly of Spanish and Italian descent (97 percent), a result of massive
European immigration. Between 1896 and 1914, the growth in the immigrant population
directly contributed 35 percent of the increase in the total population (Taylor, 1994). The
rate of growth of border immigrants was particularly high between 1947 and 1960 and
between 1970 and 1980 (Marshal and Orlansky, 1983). Domestic migration was at its
highest between 1947 and 1960, from the north of the country to the growing
manufacturing district of Buenos Aires.
Argentina consists of 23 provinces and one autonomous city ( Ciudad Autónoma
de Buenos Aires or Capital Federal ). To rule its administration, each province has
adopted its own Constitution, in accordance with the National Constitution.
The instability of Argentina’s political system has always been a serious
weakness. Between 1854 and 1996, the president should have been elected for 6-year
terms so, in 142 years, Argentina should have had 24 presidents. The country actually
had 44 presidents; and each president on average was in power for a little more than 3
years. Before 1930 more stable political institutions characterized the country, while after
22 1930 the instability of its institutions was the norm. 3 Between 1930 and 1966, only two
presidents, Agustin P. Justo (1932-1938) and Juan Domingo Peron (1946-1955), could
complete their presidencies. After 1966, again only two presidents, Raul R. Alfonsin
(1983-1989) and Carlos S. Menem (1989-1999), could complete their presidencies. The
44 presidents that governed the country from 1854 to 1996 had 114 ministers of the
economy; that is on average each one was leading the economy for one year and three
months. 4
2.3 Argentina as a Puzzling Countr y
Although one of the world's wealthiest countries 100 years ago, during most of the 20th century Argentina suffered from recurring economic crises, persistent fiscal and current account deficits, high inflation, escalating external debt, and capital flight.
Many scholars believe that Argentina has not been a typical “developing country”.
Instead, uniquely, it achieved development and then it lost it again.
Around 1800, Argentina’s per capita income was well above that of its neighbors and was similar to levels observed in Europe or the United States at that time. Coatsworth
(1998) estimates Argentina’s per capita income at 102 percent of the US level in 1800, compared to 66 percent for the South American region. Maddison (1995) places US per
3 For example, in 76 years, from 1854 to 1930, it should have had 13 presidents, and Argentina actually had 18. Only one president, Miguel Juarez Celman (1886-1890), could not finish his period due to a revolutionary coup in 1890. In 66 years, from 1930 to 1966, it should have had 11 presidents, and Argentina had 26 instead. During this period, only two presidents, Agustin P. Justo (1932-1938) and Juan Domingo Peron (1946-1955), could complete their mandates. The shortest period in power for a president in Argentina was Arturo Rawson with 3 days, Hector J. Campora for 49 days, and Eduardo A. Lonardi for 54 days (de Pablo, 2005). 4 However, regarding ministers of economy, there is no discontinuity before and after 1930, as it was the case for presidents. For example, in 76 years, from 1854 to 1930, Argentina had 55 ministers of economy (on average 1 year and 5 months in power); while in 66 years, from 1930 to 1966, Argentina had 62 ministers of economy (on average slightly over 1 year in power) (de Pablo, 2005).
23 capita income at $1,287 in 1820 (in 1992 international dollars) compared to $1,228 in
Western Europe and $1,236 in the “Western offshoots” (Australia, Canada, New Zealand, and the United States). Using the Ferreres (2005) database, Figure 2.2, places Argentina’s per capita income at 87 percent of the US level in 1820. At that time, Argentina’s per capita income was 1.63 times the level for Brazil and 1.25 times the level for Mexico.
Figure 2.2 also shows that, after some improvement at the end of the 19 th century,
Argentina’s per capita income relative to the US levels has been declining, particularly
since 1930, to converge to values similar to those of Brazil and Mexico in 1990.
1.00
0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
0.10
0.00 1820 1850 1870 1900 1930 1960 1990
Argent ina/ USA Brazil/ USA Mexico/ USA
Note: Per capita GDP is measured in constant Geary Khamis US$ Million of 1990.
Source: Constructed by the author using the Ferreres (2005) dataset.
Figure 2.2: Comparison of per capita GDP relative to the US for Argentina, Brazil and
Mexico, 1820-1990.
24 From the 1880s to 1913, Argentina experienced an average rate of growth of per capita GDP of 3 percent per year, which was a spectacular performance. Argentina’s per capita income had risen from about 67 percent of developed-country levels in 1870 to 90 percent in 1900 and to 100 percent in 1913. Table 2.1 compares the performance of a sample of countries from 1820 to 1990. By 1913, it must have seemed that the process of convergence was almost complete and that Argentina had become an “advance” economy. Argentina’s per capita GDP level in 1913 (measured in constant Geary Khamis
US$ of 1990) even exceeded the levels of middle-income European countries, such as
France and Germany. Only the United Kingdom and the four Western offshoots surpassed it, and it was well above the levels in Southern European countries such as
Italy and Spain. Further, Argentina’s 1913-1930 income level was clearly among the world’s top ten. From that point in history, a dramatic reversal started for Argentina.
Argentina USA UK France Germany Australia Italy Spain 1820 1,120 1,285 1,771 1,230 1,008 1,586 1,117 1,063 1850 1,297 1,817 2,383 1,685 1,338 3,187 NA 1,147 1870 1,562 2,454 3,292 1,876 1,734 3,946 1,500 1,376 1900 2,918 4,091 4,633 2,876 2,841 4,463 1,785 2,040 1913 4,038 5,301 5,076 3,485 3,475 5,715 2,563 2,255 1930 4,303 6,212 5,240 4,531 3,671 4,975 2,918 2,802 1960 5,884 11,328 8,645 7,543 7,685 8,865 5,916 3,437 1990 6,600 23,214 16,411 18,093 15,932 17,043 16,320 12,210
Note: Per capita GDP is measured in constant Geary Khamis US$ Million of 1990.
Source: Constructed by the author using the Ferreres (2005) dataset.
Table 2.1: Comparison of per capita GDP among developed countries, 1820-1990.
25 2.4 What Went Wrong, and When?
Argentina thus qualifies as a unique country in the modern era because of its fall from riches to poverty. This fact alone has produced much scholarly work, in an attempt to explain how it came about? And when?
According to one school of thought, the decline began in 1913, as the Pampas became fully established: growth slowed down because the country proved unable to industrialize and diversify effectively (Di Tella, 1986). This year represents the moment of closest convergence between Argentina and OECD per capita income levels.
One traditional view places the beginning of Argentina’s retardation after 1929, with serious divergence beginning with the postwar autarkic policies associated with the
Peronist government (Taylor, 1992). Díaz Alejandro (1974) also dates the end of the
Belle Époque as late as the beginning of the Great Depression, in 1929.
Liberals, for their part, have traditionally held the administrations of Péron (1946-
55) responsible, with their quasi-fascist pursuit of autarky and a state-run economy.
Leftists offer a more precise date: 24 March 1976, when the cruelest dictatorship of
South America's recent history took power. There is some truth in all of these views. Yet, the most powerful factor in Argentina's decline has been its ever-unstable politics since
1930, when a military junta took power, ending seven decades of civilian constitutional
government.
Can all of Argentina’s problems be attributed to poor policies or domestic
conditions? External conditions also played a part. It was not just the Great Depression,
but also the devastating recession during World War I that encouraged Argentina to
doubt the merits of an export-oriented and capital-importing strategy. Instead, it
26 eventually subscribed to inward-looking economic policies and import substitution (Díaz
Alejandro, 1984).
2.5 Volatility of Growth
0.2500
0.2000
0.1500
0.1000
0.0500
0.0000
-0.0500 1811 1819 1827 1835 1843 1851 1859 1867 1875 1883 1891 1899 1907 1915 1923 1931 1939 1947 1955 1963 1971 1979 1987 1995 2003
-0.1000
-0.1500
-0.2000
-0.2500
Note: Per capita GDP is measured in constant Geary Khamis US$ Million of 1990.
Source: Constructed by the author using the Ferreres (2005) dataset.
Figure 2.3: Annual rate of growth of per capita GDP for Argentina, 1810-2004
(percentages).
27 Figure 2.3 shows a high volatility of per capita GDP growth. Table 2.2 identifies five periods: (1) from 1810 to 1840, growth was slow and more stable; (2) from 1841 to
1881, oscillations from expansion to contraction were wider; (3) from 1882 to 1918, growth was faster but exceptionally volatile, and the annual rate of growth showed a standard deviation of 9.6 percentage points; (4) from 1919 to 1945, volatility declined but it was still significant; (5) from 1946 onwards, volatility was higher than the previous period. Negative rates of growth were particularly notable during World War I, the Great
Depression, and during the second part of the 20th century.
Compound Simple Period Growth Growth Std. Dev. Min Max Obs 1810-1840 0.25 0.24 0.014 -1.67 5.48 30 1841-1881 0.63 0.49 0.037 -7.87 10.21 41 1882-1918 2.64 1.69 0.096 -21.24 22.14 37 1919-1945 1.28 1.16 0.044 -8.65 9.56 27 1946-2004 1.18 0.96 0.051 -11.83 9.18 59
Note: Per capita GDP is measured in constant Geary Khamis US$ Million of 1990. Column 1 is the compound growth for each period. Column 2 is the simple average growth for each period. Column 3 is the standard deviation for column 1.
Source: Constructed by the author using the Ferreres (2005) dataset.
Table 2.2: Summary statistics for the annual rate of growth of per capita GDP for
Argentina, 1810-2004 (percentages).
28 Figure 2.4 shows the coefficient of variation of logged inter-annual per capita income for each decade for the period 1810-2000. The figure shows that, in Argentina, the variation of per capita GDP started to increase in 1830. Then, a tremendous episode of increasing volatility took place in Argentina during the 1870-1880 period. Since then, there was a trend toward lower variability until 1970, followed by another episode of increasing volatility, interrupted at the beginning of the 1990s.
0.03
0.025
0.02
0.015
0.01
0.005
0 1810 1820 1830 1840 1850 1860 1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000
Note: Per capita GDP is measured in constant Geary Khamis US$ Million of 1990.
Source: Constructed by the author using the Ferreres (2005) dataset.
Figure 2.4: Coefficient of variation of the log of inter-annual per capita GDP for each
decade in Argentina, 1810-2004.
29 2.6 The Argentinean Economy
Today, the GDP (PPP) is $524 billion (2007 est.), which makes Argentina the
25th economy in the world and the third economy in Latin America, after Brazil and
Mexico. Per capita GDP is $13,000 (2007 est.), which represents the 82th “richest” economy in the world and the second richest economy in Latin America, after Chile.
(Table 2.3 compares the seven largest economies in Latin America). Valued added to the
GDP by sector is 6 percent from agriculture, 29 percent from industry, and 65 percent from services (2007 est.).
GDP (purchasing Latin GDP - Latin power parity) America per capita World America Country Billion World Rank Rank (PPP) Rank Rank Brazil $1,840 11 1 $9,700 98 6 Mexico $1,400 14 2 $12,500 85 4 Argentina $523.7 25 3 $13,000 82 2 Venezuela $335 32 4 $12,800 83 3 Colombia $320.4 36 5 $7,200 116 8 Chile $234.4 46 6 $14,400 77 1 Peru $217.5 50 7 $7,600 113 7
Source: Constructed by the author using The World Factbook in 2007.
Table 2.3: Comparison of GDP (PPP) and per capita GDP (PPP) for the seven largest
Latin American economies, in 2007.
30 Table 2.4 compares the labor force, unemployment rate, poverty, and inequality measures for the seven largest Latin American economies, by 2007. About 23 percent of
Argentina’s population is below the poverty line (January-June 2007). In the first quarter of 2007, the richest 10 percent of the population earned 35 percent of the total income, while the poorest 10 percent of the population received only 1 percent. The Gini coefficient was 0.483 (June 2006), which is the lowest inequality level among the seven largest Latin American economies.
Population below Share Share Labor force Unemployment poverty Highest Lowest Gini Country (millions) rate (%) line (%) 10% 10% coefficient Brazil 99.47 9.8 31 © 31.3 0.7 56.7 Mexico 45.38 3.7 14 (d) 35.6 1.6 54.6 Argentina 16.1 (a) 8.9 23 (b) 35.0 1.0 48.3
Colombia 20.65 10.6 49 © 34.3 8.0 53.8 Chile 6.97 7 18 © 47.0 1.2 53.8 Peru 9.4 7.4 45 (e) 37.2 0.8 49.8 Venezuela 12.5 9.1 38 *d) 36.5 0.8 49.1
Notes: (a) urban areas only; (b) January-June 2007; (c) for 2005; (d) food-based poverty. Asset based poverty amounted to more than 40% (2006) ; (e) for 2006.
Source: the first three columns are constructed by the author using The World Factbook in 2007. The last three columns are constructed by the author using data from Gasparini, Gutierrez, and Tornarolli (2007).
Table 2.4: Comparison of labor force, unemployment rate, poverty and inequality for the seven largest Latin American economies, in 2007.
31 2.7 Currency Board
During the 1980s, Argentina experienced a very difficult political and economic transition. The country had been governed by a dictatorship since 1976, and it had been living a period of terror, while a “dirty war” took the lives of 30,000 people . 5 The
country was affected by the debt crisis, which had severe financial consequences for all
Latin American countries. The external debt had increased at an annual growth rate of
19.9 percent between 1976-1978, 42 percent between 1979-1981, and 8 percent between
1985-1986, which resulted in the outflow of capital (de Pablo, 2005).In 1982, the military
government started the “Falkland Islands War” against the United Kingdom . Defeat
meant the end of the dictatorship and the beginning of democracy . 6 In December 1983,
Alfonsín became the president of the first democratic government in this period, characterized by a severe depreciation of the currency and high inflation rates. During the hyperinflation episode, the monthly rate of inflation reached over 200 percent in July of
1989 (de Pablo, 2005). The difficult and complex economic situation ended Alfonsín’s presidency five months before the completion of his mandate, and Menem became president on July 8, 1989.
Menem was president for two complete periods, from 1989 to 1999. With the help of Domingo Cavallo, his economy minister from 1991 to 1996, he put an end to hyperinflation through the “Convertibility” Plan, a currency-board scheme that fixed the
5 In a coup on March 24, 1976, a military junta seized power in Argentina and went on a campaign to wipe out left-wing terrorism with terror far worse than the one they were combating. Between 1976 and 1983 - under military rule - thousands of people, most of them dissidents and innocent civilians unconnected with terrorism, were arrested and then vanished without a trace. See http://www.desaparecidos.org/arg/ 6 The Falkland Islands (Islas Malvinas) are a group of islands in the south Atlantic. The two main islands, East Falkland and West Falkland, lie 480 km east of the Argentina coast. About 200 smaller islands form a total land area of approximately 12,200 square km. The capital and only town is (Port) Stanley. Argentina has claimed the islands since 1820. Britain had occupied and administered the islands since 1833 and had consistently rejected Argentina's claims. http://www.yendor.com/vanished/falklands-war.html . The Falkland war took the lives of 890 Argentineans and 700 Britons, plus 3,345 people injured.
32 peso by law and limited the money supply to the stock of foreign currency reserves in the
Central Bank. The main consequence of the currency-board was a successful period of low inflation and strong economic growth. The annual inflation rate fell to 25 percent in the twelve months immediately after the adoption of the currency board. By the end of
1993, inflation was close to 10 percent per year, and it remained at very low levels thereafter.
Economic activity rebounded in 1991, after a decade of stagnation, with real GDP growing at an average of 8 percent per year between 1991 and 1994, supported by monetary stability and structural reforms. The Menem and Cavallo team privatized almost all the enterprises once owned by the state and welcomed foreign investors.
Capital inflows were considerable, leading to rapid credit growth and a consumption and investment boom. Contagion from the 1995 Mexican crisis interrupted this strong macroeconomic performance, but growth rebounded in the two years that followed.
An interesting question is why the Convertibility Plan fell apart. Mostly, Argentina's fiscal policy was incompatible with the currency board. The country has always had persistent fiscal deficits. Even as the economy grew, the public debt expanded from 29 percent of GDP in 1993 to 41 percent in 1998. On the one hand, Menem spent a lot in order to get his second reelection and in his unsuccessful attempt to get a third term. On the other hand, regardless of important macroeconomic events, structural weaknesses continued and growing vulnerabilities remained largely unaddressed up to 1998.
According to Daseking et al. (2004):
• Fiscal deficits continued to increase after 1991. Although the currency board
regime put an end to bank financing of the fiscal deficits, it provided no safeguard
33 against excessive capital market borrowing, thereby increasing the country’s
vulnerability to shifts in market conditions.
• Exports, although growing at a robust pace, did not keep up with sharply rising
imports, leading to wider trade deficits and further external debt accumulation.
The limited contribution of the export sector was a key weakness, in part
reflecting the sharp appreciation of the real exchange rate in the early stages of
stabilization. Argentina’s export base remained largely concentrated, both in
terms of goods and destination, thus increasing its exposure to undiversified price
shocks and to macroeconomic conditions in key trading partners, notably Brazil.
• Even though some reforms were put into practice, the labor market remained
largely rigid. These rigidities were a factor in increased unemployment and lower
job security. A flexible labor market was all the more critical, given Argentina’s
currency board, requiring domestic prices and wages to adjust in order to mitigate
the impact of external shocks on output and employment.
• Even with significant growth throughout the 1990s, the financial system remained
undeveloped, encouraging heavy dependence on foreign borrowing and extensive
informal dollarization. In addition, the banks’ exposure to the public sector grew
steadily.
• The currency board provided the necessary monetary stability, but it limited the
government’s ability to implement monetary policy and it did not prevent
currency mismatches.
34 • Finally, although a pension reform involving a switch to individual accounts was
an important accomplishment, which should have saved the government money in
the long run, at first it produced an extra annual bill equal to 1.5 percent of GDP.
By the late 1990s, balance-sheet weaknesses had dramatically increased, with a large share of the banks’ and public sector liabilities denominated in foreign currency, raising the risk of an eventual exit from the currency board. After nearly a decade of good macroeconomic performance, Argentina’s economy fell into recession by the end of
1998. Argentina did not suffer during the Asian crisis in 1997, but growth started to decelerate during the second half of 1998.
According to Daseking et al. (2004), the immediate trigger for the downturn was the 1998 Russian default, which provoked a stampede of capital from emerging markets.
This financial contagion caused a reversal of capital flows and a sharp rise in Argentina’s country risk premia. Several other factors contributed as well, including a cyclical correction (with output growing above potential for several years up to 1998), the deterioration in the international terms of trade, further losses of competitiveness after
Brazil, its main export market, devalued its currency in 1999, and the generalized “flight- to-quality” in all assets after the decline in NASDAQ in 2000. In addition, a continuous worsening of social conditions since the mid-1990s contributed to widespread dissatisfaction and further weakening of political support for needed policies, including fiscal adjustment.
The end of Menem’s mandate was pretty severe, with an external debt equivalent to more than 400 percent of annual exports. The economy slowed down and ultimately fell into a complete depression, with investors' fears growing in the wake of Russia's
35 default, Brazil's devaluation, and the political dissension caused by Menem’s unpopular efforts to run for a constitutionally prohibited third term.
In 1999, Fernando de la Rúa became president, through a weak coalition with leftist parties, but the economic situation was already destined to fail. Russia's default was the last straw, while devaluation in Brazil threw Argentina into recession. Unable to devalue, Argentina had to hope that deflation would eventually improve its competitiveness. But deflation is painful for democratic governments, and de la Rúa's was weak and divided. His last move was to bring Cavallo back, whose costly efforts to avoid devaluation precipitated financial collapse.
The government tried several measures to cut the fiscal deficit and promote confidence, while it received large IMF credit facilities, but nothing worked to revive the economy. Depositors began withdrawing money from the banks in late 2001, and the government responded with strict limits on withdrawals. That year, Argentina suffered its worst economic collapse in more than a century. In December, after $20 billion had fled the country and bank deposits were frozen, a bizarre combination of unemployed rioters and “pot-banging middle-class protesters” showed their unhappiness with the government. Finally, when street protests turned deadly, de la Rúa, the Radical president, was forced to resign in December of 2001.
Output fell by 10 percent in 2002, inflation rose back to the levels of the early
1990s, the government defaulted on its debt, the banking system was almost paralyzed, and the Argentine peso reached a low of 4 per dollar in mid-2002. Social indicators deteriorated dramatically, with nearly 50 percent of the population under the poverty line by May 2002, according to the Instituto Nacional de Estadísticas y Censos (National
36 Institute of Statistics and Census, INDEC)7, unemployment reached an alarming 21.5
percent of the labor force, and income inequality, as measured by the Gini coefficient,
rose to nearly 60 percent.
The crisis had both real and financial dimensions. On the financial side, Argentina
faced continuous capital flight, particularly towards the end of 2001. On the real side,
Argentina was increasingly unable to export, reflecting the loss in competitiveness
relative to Brazil, Asia, and other emerging countries. In addition, the lack of exchange
rate flexibility under the currency board required domestic prices and wages to adjust in
order to absorb shocks. A rigid labor market, however, contributed to increased
unemployment, with significant bearings on household vulnerability.
After de la Rúa, three stand-in presidents came and went in a week. One of them,
Rodriguez Saa, declared the largest sovereign default in history on Argentina's foreign
public debt of $80 billion, but he stepped down only a few days later when he failed to
gain political support from the provincial governors.
Duhalde became President in January of 2002, and he took to pieces the currency-
board system that had pegged the peso to the US dollar at par for a decade. He also
decreed that dollar deposits and loans be converted to pesos at different exchange rates
for firms and households in various sectors. When the peso depreciated and inflation
rose, Duhalde’s government switched utility fees to pesos and froze them, reduced
creditors' rights, and levied high taxes on exports. It was a mistaken effort to help
debtors, industrialists, and the middle classes at the expense of banks, privatized firms,
and exporters.
7 See http://www.indec.gov.ar/
37 The economic slump of 2001-02, which followed three years of recession, has left many social problems. In 2002, per capita income was 22 percent below its 1998 level.
At its peak, unemployment reached 18 percent (or 21 percent if those on an emergency welfare program are included), but it had fallen to around 9 percent in 2007. More than half of all Argentines dropped below the national poverty line. Much of the growth of the
1990s was wiped out.
The economy recovered strongly from the crisis, inflation started to fall, and
Duhalde called for special elections. Kirchner was elected President, taking office in May of 2003, and he continued the restrictions imposed by Duhalde. With the reemergence of double-digit inflation in 2005, the Kirchner administration pressured businesses into a series of agreements to hold prices down. The government also restructured its debt in
2005 and paid off its IMF obligations in early 2006, reducing Argentina's external debt burden. Real GDP growth averaged 9 percent during the 2003-06 period, expanding government revenues and keeping the budget in surplus. By 2008, when Kirchner’s wife,
Cristina Fernandez, became president, this approach was already showing signs of fatigue, and the future is uncertain.
2.8 A Caudillo Country
For the first half of the 19th century, Buenos Aires, whose leaders wanted a unitary republic under their guidance, fought the federalist caudillos of the interior over the shape of the new nation. In the end, the two sides settled for a tie, which materialized in the Constitution of 1860. The outlines of that deal survive today. It gives disproportionate influence to small, backward provinces. Given their population, they are
38 over-represented in the Senate and the lower house of Congress. Even allowing for their poverty, they get more than their fair share of federal revenues, under a transfer system.
These funds pay for a multitude of public employees who make up a political clientele.
Provincial politics and the caudillo system have had a lasting influence. This is one of the sources of Argentina's persistent populism and its history of political movements. In Argentina, populism is as close to fascist corporatism as it is to socialism, involving strong leaders who distort the distinction between government and state. Unlike
Chile or Uruguay, Argentina failed to develop a stable two- or three-party system of conservatives and liberals or, later, social democrats. Instead, Argentina got two populist movements, Radicalism and Peronism. Both were ambivalent towards capitalism, and both raised public employment.
Perón built his movement as an unstable coalition between corporatist trade unions and the conservative provincial caudillos. The result was “a state-dominated society”. Workers, businesses and other interest groups entered into pacts with the state rather than seeking political change. The system left little space for democratic opposition, and it thus triggered military intervention and political violence (de Pablo,
2005). Political differences were fought out within parties (often violently) rather than between them. The orderly and democratic handover from Menem to de la Rúa in 1999 was much celebrated, because it was the only such transition from one party to another since before 1930. Significantly, no Radical president since 1928 has finished his term.
Under the dictatorship, military followers promoted monetary policies that shaped the
1980s as a decade characterized by debt, inflation, and recession.
39 Indeed, Argentina is a beautiful, wealthy country with abundant natural resources, a large territory, and low density of population. However, Argentina is a puzzling country, due to its fascinating economic history. Although it suffered its worse economic collapse in 2001, Argentina’ s incidence of poverty was the second lowest, after Chile, and its inequality (measured by the Gini coefficient) was the lowest among the seven largest economies in Latin America in 2007.
40 CHAPTER 3
ECONOMIC GROWTH IN ARGENTINA
This chapter explores the second research question: whether consideration of spatial clustering helps in explaining differences in growth in per capita income across regions in Argentina. In addition, it addresses two objectives of this dissertation:
1) To compare the path of economic growth in Argentina and in selected countries in Latin America and other regions of the world and to gain an understanding of the country’s overall growth experience. This comparison sets the stage for a more detailed exploration of the recent growth experience and the extent to which it may have been influenced by the initial income distribution and the historical experience of the country.
2) To determine whether there is convergence in real per capita income across regions of Argentina, a determinant of the country’s distribution of income across regions. In the following chapter, the dissertation will explore the extent to which the distribution of income within each region influences its own growth performance and that of its neighbors.
41 Today, per capita income in Latin America is about 30 percent of per capita income in the developed world, on the basis of population-weighted averages, and about
25 percent of US levels. By 1930, however, Argentina had been the seventh largest economy in the world. At that time, Argentina’s per capita GDP was 70 percent of the US per capita GDP.
Current differences in development between Latin America and the developed world did not appear overnight. They have been likely the result of historical processes that, in some cases, go back to the colonial period. For example, de Ferranti et al. (2004) claim that, to understand the region’s contemporary situation, one needs to be aware of the role played by the colonial inheritance (characterized by the extremely high inequality that emerged soon after the Europeans began to colonize) and the institutional framework put in place at the time (which allowed a small group of elites to protect the large rents they were enjoying and excluded most of the population from access to land, education, and political power). Both the initial inequality and the institutions that emerged were shaped mostly by factor endowments, which privileged the establishment of large cultivated areas and of extractive activities that relied on forced labor, rather than by the nature of the colonial powers.
This type of argument is put forward by, among others, Engerman and Sokoloff
(2006), who believe that the impact of the colonial legacy can be observed not only in the current high levels of income inequality but also in the persistent poverty. This is so because the institutional arrangements that place the economic opportunities created in the development process beyond the reach of broad segments of society are likely to result in lower growth rates, as modern economies require broad participation in
42 entrepreneurship and innovation. Thus, Engerman and Sokoloff believe that the gap in per capita incomes between Latin America and the richer countries began to emerge in the 18th and 19th centuries.
3.1 Per Capita Income in Latin America: A Long-Run Comparative Perspective
There are two main steps in considering the evolution of Latin America’s income
levels over time. The first is gathering historical time-series data on which to base the
debate. The second is recognizing that the exercise of evaluating the evolution of the region is comparative in nature and, therefore, that it requires deciding which country or region to use as the benchmark.
3.1.1 Historical Per Capita GDP Estimates for Latin America
It is not easy to find historical data for any Latin American country. I use Prados de la Escosura (2005) data, which show GDP estimates for nine Latin American economies. Also, I use a database compiled by Ferreres, from the Foundation “Norte y
Sur,” in collaboration with the Catholic University of Argentina. This source has GDP estimates for the seven largest economies in Latin America since 1900. The data are in
1990 international Geary-Khamis dollars, which is a sophisticated aggregation method of calculating purchasing power parity (PPP). See Appendix C for an explanation of the
Geary-Khamis method. An important characteristic of the Ferreres database is that it has
GDP estimates for Argentina from 1810. This is the only database with this kind of spell for Argentina that I am aware of.
43 Table 1 in Appendix D compares the rates of growth of per capita income for the seven major Latin American countries, with a combined population that represented almost 90 percent of the whole region’s population in 2003. These growth rates are presented at roughly decadal benchmarks, for the period 1890–2000 and for longer periods since 1810 for Argentina and since 1850 for the rest of the countries. Decades are arbitrary partitions and they do not always correspond to actual episodes of growth and recession, but they facilitate the comparisons.
Over the 1870–2000 period, Chile experienced the highest rate of per capita income growth, followed by Venezuela and Mexico. Of the seven countries, Colombia and Peru experienced the lowest rate. Brazil and Argentina were intermediate cases, with an estimated growth rate of 1.54 and 1.34 percent per year, respectively. At the 1.34 rate, per capita GDP doubles roughly every 51 years, so today’s per capita GDP for Argentina would be about eight times the level observed in the late 1870s.
Table 1 in Appendix D suggests that, for most of the countries, the 1938–1980
period was the most productive. This was especially true for Brazil and Mexico. The
exceptions were Chile, where per capita income growth accelerated later, and Argentina,
where growth had been swifter earlier.
Except for Chile, however, the last two decades of the 20th century were not very
positive (Peru and Venezuela actually experienced negative per capita income growth),
due to two adverse episodes. The first one is “the lost decade” of the 1980s, following the
Latin American debt crisis, when Argentina experienced negative growth. The second
one is the period following the Asian financial crisis of 1997 and the Russian financial
crisis in 1998. Had it not been for the positive performance of the region during 1990–
44 1997, when all seven countries enjoyed positive growth (and when two of them,
Argentina and Chile, enjoyed growth rates that more than doubled their historical trends), the last part of the 20th century would have been more dramatic than it actually was.
Argentina experienced outstanding per capita GDP growth rates during 1870-
1890, at 2.5 percent, 1960-1970, at 2.8 percent, and 1990-1997, at 4.9 percent. The last financial crisis affected this country during 2000-2002, with a negative rate of growth
(minus 5.9 percent).
Figure 1 in Appendix D plots the per capita GDP trends for seven Latin American
countries. Although the figure shows some dispersion in GDP levels (especially between
1948 and 1986), Argentina’s per capita GDP was almost always higher than the per
capita GDP for all other countries except during 1948-1996, when Venezuela’s was the
highest, and since 1999, when Chile’s per capita GDP became the highest in the region.
Chile’s per capita GDP started its recent steady upward growth in 1986, following
its market-oriented reforms, with a small drop in 2002. Mexico’s per capita GDP had
been growing close to Peru’s figures until 1966, when Mexico started a period of steady
upward growth, interrupted between 1984 and 1998. Almost all the countries, except for
Colombia, reached their peak in the early 1980s. The first one to experience the decline
was Venezuela, in 1980. Argentina reached a first peak in 1983, like Brazil, while
Mexico, Chile, and Peru reached their peak in 1984. Mexico took 18 years to get back to
this previous highest level, Argentina took 14 years and reached a new peak in 2001,
Brazil took 10 years, and Chile took 8 years, while Venezuela and Peru never reached
those levels again.
45
3.1.2 Income Convergence in Latin America One interesting question is whether the evidence that emerges from the long-run trends supports the hypothesis of income level convergence among Latin American countries (Serra et el., 2006). That is, over the past century or so, have the countries that were initially poorer grown faster than those that were initially richer? To explore the empirical evidence on this issue, Figure 3 in Appendix D compares the average annual growth rates experienced by the different countries between 1900 and 2004, with their corresponding (logged) initial per capita GDP level in 1900. The figure shows a negative correlation between these two variables. The estimated slope of the regression line is
−0.54, and it has an associated standard error of 0.21, which are statistically significant at
5 percent. Although one has to be careful in extrapolating results based on only seven countries, the evidence would indicate that the poorer countries in the early 1900s grew faster over the ensuing 104 years than the initially richer countries. This would lend some support to the hypothesis of convergence of incomes across the Latin American countries during this period.
Figures 4 and Figure 5 in Appendix D show the cross-country standard deviation of logged per capita GDP and the coefficient of variation of logged per capita GDP.
These measures of income dispersion allow an alternative way to explore the possibility of convergence. The figures suggest that the dispersion of cross-country per capita income increased during the first ten years of the 20th century (1900–1910) and decreased after World War I. Dispersion increased again during the Great Depression of the1930s and after World War II, before falling throughout the years until 1980, when it reached its historical low. No major changes have been observed since.
46 Thus, although convergence in per capita GDP levels over the 1900–2004 period has been interrupted by periodic increases in cross-country inequality, the evidence reveals a convergence trend.
3.2. Comparative Perspective
How does Latin America’s per capita GDP perform in comparison to other countries and regions of the world? What has been Argentina’s experience in this context? Usually, historical comparisons have taken the United States as reference, as the leading performer during this period. Here, I take a broader view and consider the performance of several groups. These include: (a) the group of North America, formed by the US and Canada; (b) European developed countries that belong to the OECD (France,
Germany, Italy, Austria, Switzerland, the Netherlands and Belgium); (c) Spain, a country with which Latin America shares some institutional background; (d) Northern Europe
(Norway, Finland, Denmark, Sweden); (e) Oceania (Australia and New Zealand); (f) East
Asia (Hong Kong, China; the Republic of Korea; Singapore; and Taiwan, China), to take account of the “Asian miracle,” and (g) Latin America (Argentina, Brazil, Chile,
Colombia, Mexico, Peru and Venezuela). Table 2 in Appendix D reports the corresponding rates of growth of per capita GDP since 1820.
During the second half of the 19th century, Australia and New Zealand were the fastest-growing economies. Latin America had an explosive starting phase during 1870-
1900 and 1870-1929, which was led by Argentina’s performance. The main difference is that Latin America got behind during 1929-1938 and during and after World War II.
Except for Latin America, most groups outperformed their earlier growth pace. Thus,
47 early the Latin American countries lagged in their per capita growth rates. Starting in the
1950s, nevertheless, East Asia outperformed Latin America, North America and Oceania, and in the 1970s, 1980s and 1990s, it became the fastest-growing group.
Figure 6 in Appendix D illustrates the evolution of per capita GDP for Argentina, the USA, UK, Germany, Norway, Australia, and Japan. Several messages emerge from
Figure 6. First, during 1820-1905, Australia and the UK outperformed the US. Second, from 1820 to 1883, Argentina and Germany shared the same level of per capita income, although below the US levels. From 1883 to 1935, Argentina enjoyed a higher per capita income than Germany. Afterwards, they were very close again, until World War II.
Argentina started to diverge from the general upward trend in four moments in time: (a) after 1913, when the Pampas became fully established and growth slowed down because the country proved unable to industrialize and diversify effectively (Di Tella, 1986); (b) after 1929, at the beginning of the Great Depression; (c) after 1949, with the postwar autarkic policies associated with two Peronist governments (Taylor, 1992); and (d) after
1974, coinciding with the third government of Péron (1973-1974), his death and his wife’s Presidency (1974-1976), which was interrupted by a military dictatorship. Third, the US outperformed Australia and the UK in the earlier 1920s; from that time on, the US had an impressive performance. Finally, Japan’s per capita GDP started to grow after
World War II and more extraordinarily since the 1960s.
These examples show that some regions or countries that started with lower relative incomes broke with their historical path and did not continue in this position forever, as East Asia has shown and, particularly, as the case of Japan reveals. In contrast,
48 the puzzle is Argentina, which started with income levels similar to countries like
Germany, Italy, Spain, or Sweden and now is quite poor compared to these countries.
3.3 The Data
I have data on real per capita GDP for 23 provinces and the capital city ( Ciudad
Autónoma de Buenos Aires ) in Argentina, available for 1980-2002 from the ECLAC office in Buenos Aires.
Data about household income and characteristics come from the EPH (Encuesta
Permanente de Hogares). The EPH is the main household survey in Argentina. It has
been carried out since the early 1970s by the INDEC ( Instituto Nacional de Estadística y
Censos ). It covers 31 large urban areas, which are home to 71 percent of the urban
population. Since, in Argentina, the share of the population in urban areas is 87 percent
(one of the highest in the world), the sample of the EPH represents 62 percent of the total
population of the country. The EPH is a rotating panel, which implies that a household
remains in the survey for one year and a half. The EPH panels have been available to the
public only since 1998.
The microdata of the EPH have been available for the Greater Buenos Aires
(GBA) since October of 1974. Since then, two major changes have been implemented by
the INDEC. First, the survey has been extended to cover all large urban areas in the
country, with at least one observation in each province. The second major change has
been the recent implementation of the Encuesta Permanente de Hogares Continua : the
survey is now carried out over the whole year and not in just two rounds (May and
October) as before.
49 3.3.1 Changes in geographic coverage
As mentioned above, the EPH was extended from the GBA to 31 large urban
areas in the country. This geographical extension has taken place gradually over time.
Some conglomerates were included in the 1980s, while others were added in the 1990s.
The last conglomerates were added in 2002.
This process of progressively adding urban areas calls for a careful treatment of
the data. Comparisons of statistics should take into account that a different number of
urban areas are available at each point in time. Given this problem, several studies have
taken an extreme solution: they have restricted the analysis to the Greater Buenos Aires.
In fact, this is the only possibility if one wants to construct a series starting in the mid-
1970s.
I have tried to include the majority of the conglomerates and, as a result, I have
been able to use the EPH only since the early 1990s. I have selected a set of 16 urban
areas with consistent EPH microdata since 1992, including: Capital Federal, Conurbano
Bonaerense, Comodoro Rivadavia, Córdoba, Jujuy, La Plata, Neuquén, Paraná, Río
Gallegos, Salta, San Luis, San Juan, Santa Rosa, Santa Fe, Santiago del Estero, and
Tierra del Fuego. It is important to clarify that Capital Federal (the city of Buenos Aires) and Conurbano Bonaerense (the suburbs) are a single conglomerate (the Greater Buenos
Aires). However, INDEC differentiates the two areas with different codes.
From 1998 on, I could extend the analysis to 29 urban areas, including data on
Bahía Blanca, Catamarca, Concordia, Corrientes, Formosa, La Rioja, Mar del Plata,
Mendoza, Posadas, Resistencia, Río Cuarto, Rosario, and Tucumán.
50 The EPH uses the word “conglomerate” to describe an urban area with over
100,000 inhabitants. In each one of the 23 provinces there is at least one conglomerate, corresponding to the capital city.
The INDEC has divided the country into six geographical regions, as reported in
Table 3 in Appendix D. From table 4 in Appendix D, out of 24 provinces (including
Capital Federal), there is only one conglomerate in 17 of the provinces. There are three provinces with two conglomerates (Sante Fe, Córdoba, and Chubut) and three provinces with three conglomerates (Entre Ríos, Corrientes, and Gran Buenos Aires). Finally, there is only one province, Buenos Aires, with four conglomerates. The last column reports the region to which each province corresponds.
3.3.2 The New EPH Continua
The EPH was carried out twice a year, in May and October. During 2003, a major methodological change was implemented by INDEC, including changes in the questionnaires and in the timing of the survey visits. The EPH is now conducted over the whole year. INDEC publishes statistics by quarters and semesters, and it is supposed to share the microdata with the same frequency. However, so far only a reduced version of the dataset of the new EPH-Continua (EPHC) for the fourth quarter of 2003 has been released to the public.
The questionnaire of the new EPHC is intended to improve the report of labor variables and income, in particular those related to informal jobs and public programs.
The EPHC includes some additional questions beyond the original EPH to capture income from vouchers, tips, and other items not included in regular wages. It is also
51 richer in the questions on incomes from self-employment and income from public programs, charity, and child labor.
The EPHC includes some questions on the household’s strategy to finance expenditures (dis-saving, borrowing, selling assets, and others) and on non-monetary items of household income. Unfortunately, these are only binary “yes-or-no” questions, and thus they are not useful for a traditional poverty and inequality analysis.
So far, only the microdata for the IV quarter of 2003 are available to the public.
Moreover, only a reduced version of the dataset is in the INDEC web page, which significantly constraints the possibility of studying the new survey. For instance, with the data available, researchers have to rely on the INDEC’s estimates of total household income, since it is impossible to construct this variable because data on secondary jobs and non-labor income were not included in the published dataset. This situation affects my research by limiting the information only until May 2003, since it is impossible to construct some variables such as education and migration with the limited information of the EPHC available so far.
The EPHC differs from the EPH in another relevant dimension: non-response and inconsistency in the income answers seems to be a more important issue in the new survey.
Due to the fact that I have data on real per capita GDP for 23 provinces and the capital city available for the 1980-2002 period, and the significant change in the methodology of the household surveys, I have decided to use the household data from the
EPH from 1991 to 2002. Finally, because the conglomerate “Viedma-Carmen de
Patagones- 93,” corresponding to the province of Rio Negro, was not included in the EPH
52 and it was not included in the EPHC until late 2002, I am not considering this province in my analysis. For this reason, during my period of analysis, 1991-2002, I can work with
22 provinces and the capital city ( Ciudad Autónoma de Buenos Aires ).
3.3.3 Limitations
Household surveys in Argentina do not cover rural areas. This limitation,
although certainly relevant, is not as important as in other countries, since the share of the
rural population in Argentina is small (less than 15 percent).
The World Bank’s Encuesta de Impacto Social de la Crisis en Argentina (ISCA)
included some small towns in rural areas. From the information of that survey, the
income distribution in rural areas turns out to be not substantially different from the
income distribution in urban areas. The Gini coefficient for the distribution of household
per capita income is 0.474 in urban areas, 0.482 in rural areas, and 0.475 for the whole
country. This result suggests that the urban inequality statistics can be taken as a good
approximation for the national figures.
The Encuesta Permanente de Hogares has an additional limitation: it only covers
large conglomerates (more than 100,000 inhabitants). However, given the high
concentration of Argentina’s population in a few large cities (especially in the Greater
Buenos Aires), the coverage of the EPH turns out to be high (71 percentage of the urban
population). The Encuesta de Condiciones de Vida includes observations from small
cities. In fact, 33 percent of the observations are collected in cities of less than 100,000
inhabitants.
53 3.4 Convergence Concepts and Spatial Effects
A first concept of convergence pertains to the decline in the cross-sectional
dispersion of per capita incomes. Several different measures have been employed to
examine this form of convergence, including the (unweighted) standard deviation
(Carlino and Mills, 1993) and the coefficient of variation (Bernard and Jones, 1996) of
the log of per capita income. I use these measures in Figures 4 and 5 in Appendix D for
Latin American countries. This form of convergence has been referred to as sigma-
convergence, and it has attracted much attention in the economic geography literature
(Kuznets, 1955; Easterlin, 1960a, 1960b; Williamson, 1965; Amos, 1988, 1989; Coughlin
and Mandelbaum, 1988; Fanan and Casetti, 1994).
A second form of convergence occurs when poor regions grow faster than rich
regions, resulting in the former eventually catching up to the latter in per capita income
levels. To test for this form of convergence, numerous studies have employed a cross-
sectional specification as follows:
Yi, t+ k ln =α + β ln ()Yit, + ε it (1) Yi, t
where Yi, t is the per capita income in province i and year t; α and β are
parameters to be estimated; and εit is a stochastic error term. Following Baumol (1986),
the convention has been to interpret a negative estimate for β as support for the
convergence hypothesis, since such an estimate would suggest that the growth rates in per
capita income over a k-year period were negatively correlated with starting incomes.
Thus, this second form of convergence has been called β -convergence. It is important to
54 recognize that a finding of a negative convergence parameter does not necessarily imply declining cross-sectional variance in income levels ( σ -convergence).
Many authors have studied the β -convergence approach for Latin American countries. Most recently, Serra, Pazmino, Lindow, Sutton and Ramirez (2006) studied the convergence between 1970 and 2002 of per capita output for regions within six Latin
American countries: Argentina, Brazil, Chile, Colombia, Mexico and Peru, following the
β -convergence approach (using Barro and Sala-i-Martin, 1991a, 1991b, and 1992).
These authors found that poor and rich regions within each country converged at very low rates over the past three decades. Specifically, for Argentina they found that, from
1970 to 2002, there is no evidence of convergence among its provinces. Their estimates for the speed of β -convergence suggest that poor provinces did not catch up with the rich ones. For the 31-year period and for sub-periods of 10 years, the estimated speeds of absolute convergence yielded insignificant coefficients. In addition, these authors found that there is no evidence of conditional convergence either within subnational regions or if accounting for structural shocks. Even when they accounted for shocks in manufacturing, there is no evidence of convergence. 8
These results for Argentina are similar to those from previous studies by Garrido et al. (2000), Marina (2000), and Figueras et al. (2003). However, another study by
Madariaga et al. (2005) analyses the process of per capita income convergence among 23 provinces in Argentina for the 1983-2002 period. Accounting for spatial autocorrelation
8 In the empirical observations on convergence, if one allows for heterogeneity across economies and, in particular, if one drops the assumption that all economies have the same parameters and, therefore, the same steady-state positions, a concept of conditional convergence emerges. The main idea is that an economy grows faster, the further it is from its own steady-state value. In practice, β -convergence applies if the growth rate of per capita GDP is negatively related to the starting level of per capita GDP after holding other variables constant, such as initial levels of human capital, measures of government policies, and so on. (Barro, R., and X. Sala-i-Martin, 2004).
55 among provinces, they found that there is conditional convergence among Argentinean provinces and, a positive and significant impact of agglomeration variables on the growth rate. Hence, their results show that ignoring spatial autocorrelation due to geographic proximity leads to misleading estimations and that it underestimates the speed of convergence, particularly for provinces that are distant from Buenos Aires. Therefore, in the next section, I test the hypothesis of the presence of spatial autocorrelation in my income data.
3.4.1 Spatial Effects in the Analysis of Regional Income Convergence
In general, the unit of analysis has been an individual region observed within a cross section or in a time series. Implicitly, each region has been viewed as an independent entity and the potential for observational interactions across space has gone largely ignored. While technology spillovers (Krugman, 1987; Jones, 1997) have been identified as key mechanisms that may lead to convergence, the geographic dimensions of these spillovers have most of the time been ignored. Alternatively, Bernard and Jones
(1996) suggest that comparative advantage leading to regional specialization in tradable goods sectors may result in lack of convergence at the aggregate level.
While theoretically intriguing, such arguments have not been formally incorporated in the empirical models used to examine the convergence hypothesis at the regional scale. However, by recognizing such forms of interaction as cases of substantive spatial dependence (Anselin and Rey, 1991), a rich body of spatial process models becomes available for the study of regional income convergence.
56 In addition to the substantive form of spatial dependence, the geographic organization of the observations in regional convergence studies may give rise to a second type of spatial dependence. This can result from a mismatch between the spatial boundaries of the market processes under study and the administrative boundaries used to organize the data. Spatial dependence due to this form of boundary mismatch has been referred to as nuisance dependence, since it is reflected in a spatially autocorrelation error term (Anselin and Bera, 1998).
For Argentina during the 1980-2002 period, I consider both σ - and β - convergence and I investigate the spatial dimension of Argentina’s income dynamics.
Attention is first directed toward β -convergence, followed by σ -convergence and the related spatial patterns.
3.4.2 Exploratory Spatial Data Analysis of Argentina’s Income Convergence
Figure 7 in Appendix D compares the average annual growth rates experienced by
the different provinces between 1980 and 2002, with their corresponding (logged) initial
per capita GDP level in 1980. The figure shows a negative correlation between these two
variables. The estimated slope of the regression line is −0.87, and it has an associated
standard error of 0.5, which are statistically significant at 10 percent. This evidence
would indicate that the poorer provinces in the early 1980s grew faster over the ensuing
22 years than the initially richer provinces. This would lend some support to the
hypothesis of convergence of incomes across the provinces of Argentina during this
period.
57 Rey and Montouri (1999) argue that σ -convergence is only concerned with the provincial dispersion of income distribution and that it does not take into account any geographic pattern. Instead, it seems important to analyze the dynamic behavior of income dispersion across provinces in Argentina as well as to explore its geographic dimensions.
Figure 8 in Appendix D shows the pattern of per capita GDP dispersion in
Argentina, measured by the coefficient of variation for the natural log of real provincial per capita incomes. Although there is a general negative trend, the graph shows two clear tendencies: declining per capita income dispersion until 1990 and increasing per capita income dispersion after 1990, which coincided with rapid growth during the Menem period.
3.5 Spatial Autocorrelation
Spatial autocorrelation occurs when the spatial distribution of the variable of interest exhibits a systematic pattern (Cliff and Ord, 1981). I compute three measures of global spatial autocorrelation: Moran’s I (Moran 1948), Geary’s c (Geary 1954), and
Getis and Ord’s G (Getis and Ord, 1992).
Moran’s I is given by:
N N
∑ ∑ wij Z i Z j I = i=1 j = 1 (2) S0 m 2
where wij denotes the elements of the spatial weights matrix W corresponding to the
location pair [( i,j ); j], Zi= Y i − Y , Yi denotes the value taken on by the variable Y of
58 interest al location i; Y denotes the mean of variable Y , S= w , m= Z2 / N , 0 ∑i ∑ j ij 2 ∑ i i and N is the number of provinces.
Figure 9 in Appendix D displays the Moran’s I statistic. Under the null hypothesis of no global spatial autocorrelation, the expected value of I is given by:
−1 E( I ) = (3) (N − 1)
If I is larger than its expected value, then the overall distribution of variable Y is characterized by positive spatial autocorrelation. This means that the value taken on by Y at each location i tends to be similar to the values taken on by Y at spatially contiguous locations. If I is smaller than its expected value, then the overall distribution of variable Y is characterized by negative spatial autocorrelation. This means that the value taken on by
Y at each location I tends to be different from the values taken on by Y at spatially
contiguous locations. Inference is based on z-values, computed by subtracting E(I) from I and dividing the result by the standard deviation of I:
I− E( I ) z = (4) I sd( I )
zI follows a normal distribution (asymptotically), so that its significance can be evaluated by means of a standard normal table (Anselin, 1992).
Table 5 in Appendix D presents the estimates for the Moran’s I statistic for the provincial per capita incomes of Argentina, for the 1980-2002 period. The table shows that there is strong evidence of spatial dependence (positive spatial autocorrelation), as the statistic is highly significant for all the years at p = 0.01. This implies that the value taken by per capita income at each province i tends to be similar to the values taken by
59 per capita income at spatially contiguous locations. In other words, in any given year the provincial per capita income displays a high degree of spatial autocorrelation.
Figure 10 in Appendix D displays both the provincial per capita income dispersion in Argentina, as measured by the coefficient of variation of the natural log of provincial real per capita GDP, and the spatial autocorrelation (Moran's I statistic) for the provincial income over the same period. The simple correlation between the Moran’s I statistic and the coefficient of variation is 0.4 over the 22-year period. Interestingly, the measure of spatial autocorrelation also tends to co-move with the measure of income dispersion, until 1995.
Looking at the Moran’s I statistic, it can be claimed that, even though spatial autocorrelation is statistically significant throughout the period, its magnitude seems to weaken (specifically around 1990-1991 and in 1998-2000).
After comparing the Moran’s I statistic with the coefficient of variation, four clear patterns can be identified.
First, for 1980-1983 the coefficient of variation and the Moran’s I statistic were increasing slightly. This period coincided with the debt crisis in Latin America and in
Argentina, a period of high inflation (reaching its peak of 27 percent per month in July of
1982), the end of the dictatorship after the “Falkland Islands War,” and the beginning of the democracy. 9
Second, 1984-1990 is a period of income convergence, as shown by the decline in the coefficient of variation, which is also accompanied by a decline in Moran’s I statistic.
This period saw a severe inflation and several attempts by the government to control it.
9 In Argentina the dictatorship ended with its defeat in the Falkland Islands War on June 14, 1982. That was the transition to democracy, in which R. Alfonsín from the Unión Cívica Radical party (UCR) won the election on October 30, 1983 and became president on December 10, 1983.
60 The monthly inflation rate reached 43 percent in June, 1985. In June 14, 1985, the Plan
Austral was an ambitious set of economic policies adopted to combat inflation, which dropped under the 10 percent range until October of 1987, when it reached 30 percent. 10
The period finished with a phase of hyperinflation between January and July of 1989.
The monthly rate of inflation reached 209.1 percent in July of 1989, which was equivalent to 3.8 percent daily (de Pablo, 2005). 11
Third, during the 1991-1995 period, the coefficient of variation as well as the
Moran’s I statistic increased somewhat. The later reached its peak in 1995. Thus, the process of convergence came to an end. This period coincided with the currency board regime and the market oriented policies implemented by Menem.
Finally, during the 1996-2002 period, the evolution of the two measures moves in opposite directions. This period corresponded to the interruption of the country’s strong macroeconomic performance because of the contagion from the 1995 Mexican crisis and the financial debt crisis in December of 2001.
I would like to highlight three conclusions. First, despite the process of per capita income convergence, in 1990 there is a clear break with this trend, according to the coefficient of variation. Second, during this difficult period in Argentina, even though there is strong evidence of positive spatial autocorrelation, which is statistically significant throughout the whole period, the absolute value of the Moran I. actually declined. This statistic reflects the degree of provincial clustering. This result suggests that there was an evolution in the clustering process of the provincial per capita income,
10 The Plan Primavera was launched on August 2, 1988 to reduce the rate of inflation. 11 The difficult and complex economic situation ended Alfonsín’s presidency five months before the completion of his mandate. Menem became president on July 8, 1989 and he completed two presidential periods that ended on December 10, 1999.
61 which was a consequence of the political and economic forces that prevailed during this period.
Third, the relatively high (low) income provinces tend to be located nearby other high (low) income provinces more often than would be expected due to random chance.
If this is the case, then each province should not be viewed as an independent observation, as has been implicitly assumed in previous studies of regional income convergence (Rey and Janikas, 2005).
3.5.1 Local spatial autocorrelation
Appendix D offers a more disaggregated view of the nature of spatial autocorrelation for the initial (Figure 11) and final (Figure 12) years. Each figure contains a Moran scatterplot, recommended by Anselin (1995), which plots the standardized income of a province against its spatial lag (also standardized). A province’s spatial lag is a weighted average of the incomes of its neighboring provinces, with the weights being obtained from the simple contiguity matrix.
The Moran scatterplot is a plot of Wz versus z, where W denotes a row-
standardized spatial weights matrix and z is equal to:
z=( Y − Y) / sdY ( ) . The oblique line represents the linear regression line obtained by
regressing Wz on z, and its slope equals Moran’s I (in this case I = 0.624 for 1980, and I =
0.329 for 2002).
The Moran scatterplot is divided into four quadrants, each of which represents a
different kind of local spatial association between a province and its neighbors:
62 • The upper right quadrant represents spatial clustering of a high-income
province with high-income neighbors (HH-quadrant I). In general, these
locations are associated with positive values of the local Moran Ii .
• The upper left quadrant represents spatial clustering of a low-income province
surrounded by high-income neighbors (LH -quadrant II). In general, these
locations are associated with negative values of the local Moran Ii .
• The lower left quadrant represents spatial clustering of a low-income province
surrounded by low-income neighbors (LL-quadrant III). In general, these
locations are associated with positive values of the local Moran Ii ; and
• The lower right quadrant represents spatial clustering of a high-income
province with low-income neighbors (HL-quadrant IV). In general, these
locations are associated with negative values of the local Moran Ii .
Quadrant III and I relate to positive forms of spatial dependence while the
remaining two correspond to negative spatial dependence.
Used in combination with the global measures of spatial dependence (underlying
Figure 9), the scatterplot offers a visual intuition on the overall stability of the global
pattern of spatial dependence as well as the capacity to discover local regimes of spatial
dependence that may depart from the overall pattern. The Moran for province i takes the following form:
n n xi 2 Ii, t= ∑ w ij x j , t where m0= ∑ x i , t (5) m0 j=1 i
In (5), wij is an element of a spatial weights matrix W such that wij = number of
hours that it takes to drive from location i to location j using the actual routes available in
63 each province; x j, t is the natural log of real per capita income in province j in year t
(measured as a deviation from the mean value for that year); and n is the number of provinces considered in the analysis.
Viewing Figures 11 and 12 together suggests that there was no modification of the overall structure of spatial dependence between 1980 and 2002. More specifically, except for province 1 (Buenos Aires), which is in quadrant II all the time, only two provinces seem to be near the border between quadrant I and IV (13-Corrientes) and quadrant III and IV (5-Mendoza). Three provinces changed quadrant in 1991: La Pampa, from quadrant IV to I, San Juan, from quadrant III to II, and San Luis, from quadrant III to IV. All other provinces started in a specific quadrant in 1980 and continued in the same quadrant at the end of the period.
The results from the application of the local Moran statistic to the income values in each of the years in the sample are summarized in table 6 in Appendix D, which shows the dynamics of spatial dependence among the provinces of Argentina. For each year, the table shows the number of the quadrant that represents each province in the Moran’s scatterplot. The local pattern of spatial association tends to reflect the global trend of positive spatial association reported earlier. More specifically, over 95 percent of the local indicators that are significant fall in either quadrant I or III, reflecting HH and LL clustering, respectively.
Second, two main regional clusters persist throughout the whole period. The first
cluster is the high-income province with high-income neighbors represented by the
provinces of Patagonia (Chubut, Neuquén, Santa Cruz, and Tierra del Fuego), plus La
Pampa, each of which appears in quadrant I. All the other provinces, except for Buenos
64 Aires and Corrientes, constitute a second cluster of a low-income province surrounded by low-income neighbors, each of which falls in quadrant III the vast majority of the years
(see map 11 in Appendix B). These results provide fairly strong evidence that the clustering in these two regions is not due to chance alone. It appears, therefore, that the positive correlation between the global Moran’s I statistic and the measure of income dispersion, depicted in Figures 10 in Appendix D, is due to a strengthening of the regional clusters during periods of income divergence (at least until 1995), rather than the emergence of newly formed clusters.
Third, with only few exceptions, just five provinces, Corrientes, San Juan, San
Luis, La Pampa and Mendoza, experienced a permanent change in their spatial
dependence dynamics among all other provinces of Argentina. These results provide
reasonably strong evidence that, even in the case of Argentina, which during this period
experienced institutional change, hyperinflation, major economic reforms, a brutal
recession, and several international shocks, the dynamics of the positive spatial
dependence among its provinces has being stable throughout the period.
Finally, as a measure of robustness of these results, I have estimated the global
and local measures of spatial autocorrelation with a different W matrix. For the earlier
results, I used a spatial weights matrix W such that wij = number of hours that it takes to drive from the capital city of each province i to the capital city of province j, using seven hours as the cutoff point and the actual routes available in each province. Alternatively, I
calculated a spatial weights matrix W such that wij = distance in kilometers between each provincial capital city (location i) and each of the other provincial capital cities (location
65 j) within 800 km, using the actual routes available in Argentina, and I obtained the same
significant results.
3.6 Income Comparison among the Provinces of Argentina
The contribution of various regions to the Argentinean GDP in 1980, 1991 and
2002 is illustrated in Figure 13 in Appendix D. The figure reveals the predominant role of
Buenos Aires province along with the Capital City (Ciudad Autónoma de Buenos Aires ).
Both together add up to approximately 60 percent of the total output of the country.
Three provinces follow them: Córdoba, Santa Fe, and Mendoza, which together account for 20 percent of the total GDP. The remaining 20 percent is divided among 20 provinces. See the per capita income growth by province during the 1992-2002 period in
Map 6 of Appendix B.
Figure 13 also shows that, even though the leading role of Buenos Aires province diminished from 35.5 percent of the GDP in 1980 to 33 percent in 2002, while the Capital
City increased its participation from 25 percent in 1980 to 25.4 percent in 2002, in general the shares of Córdoba, Santa Fe, and Mendoza in the total GDP have been constant.
The same comparison emerges if we consider the population of these richest provinces relative to the total population of the country (Figure 14 in Appendix D).
Thus, Buenos Aires province and the Capital City accounted for almost 50 percent of the population in 1980 and 46.5 percent in 2002, while Cordoba, Santa Fe and Mendoza accounted for 22 percent of the total population in 1980 and 20 percent in 2002.
66 Figure 15 in Appendix D compares the participation in the total GDP of the main five regions of Argentina. The region with the largest participation in the total GDP is
Pampeana, because it concentrates Buenos Aires, Córdoba and Santa Fe.
Figure 16 in Appendix D compares each province’s per capita income relative to the country’s average for 1980, 1991 and 2002. The horizontal line equal to one gives the country’s average. Consistently, over the whole period, six provinces and the capital city have enjoyed per capita incomes higher than the national average: Tierra del Fuego,
Santa Cruz, Chubut, La Pampa, Rio Negro and Neuquén, of which all (except for La
Pampa) are provinces in the Patagonia region. These provinces belong to quadrant I, in
Figures 11 and 12 (see Appendix D). Also the Capital City, Tierra del Fuego, and San
Luis have experienced some volatility in their relative incomes.
67 CHAPTER 4
INEQUALITY IN ARGENTINA
This chapter addresses the second research question: whether spatial clustering helps in explaining differences in the inequality of the distribution of per capita income across regions in Argentina.
The chapter has three main purposes. First, it shows the importance of studying
inequality, particularly in its relationship with poverty and growth. Second, it assesses the
income distribution of Argentina and it compares it to that of other Latin American
countries. Third, it analyzes Argentina’s regional differences in income distribution
inequality using spatial econometrics tools.
Most economic analysis is concerned with inequality in the distribution of some
measure of individual well-being. Inequality generally refers to a measure of dispersion
in the distribution. Most measures used are consistent with certain desirable attributes,
known as axioms of inequality measurement (Atkinson, 1970; Cowell and Jenkins, 1995;
Cowell, 1998).
68 4.1 Importance of the Study of Inequality
As de Ferranti et al. (2004) conclude, inequality is pervasive. It characterizes every aspect of life, including access to education, health, and public services. It prevents access to land and other assets, and it affects the functioning of credit and formal labor markets. It excludes people from the attainment of political voice and influence
(inequality of agency). Inequality in Latin America has been rooted in exclusionary institutions that have been perpetuated ever since colonial times. It has reduced the impact of economic growth on poverty reduction and as this dissertation assesses, inequality has been bad for aggregate economic growth, especially when it has seen associated with unequal access to credit and education. Besides, inequality is associated with a greater prevalence of social conflict and violence and it may impair an economy’s ability to respond effectively to macroeconomics shocks (de Ferranti et al., 2004).
4.2 Relationships among Poverty-Growth-Inequality
Bourguignon (2004) describes changes in poverty in a given period as reflecting growth in mean income and changes in the distribution of relative income. The decomposition illustrated in Figure 1 in Appendix E corresponds to an identity described as the “Poverty-Growth-Inequality Triangle.”
A change in the distribution of income can actually be decomposed into two effects. First, there is the effect of a proportional change in all incomes that leaves the distribution of relative income unchanged ( growth effect). Second, there is the effect of a change in the distribution of relative incomes (which, by definition, is independent of the mean), known as a distributional effect.
69 In Figure 1 in Appendix E, the poverty headcount is the area under the density curve to the left of the poverty line (here set at US$1 a day). 12 The movement from the
initial to a new distribution goes through an intermediate step, namely the horizontal shift
of the initial density curve to curve (I). Because of the logarithmic scale, this change
corresponds to the same proportional increase of all incomes and thus stands for the pure
“growth effect.” Then, the movement from curve (I) to the new distribution occurs at
constant mean income and it corresponds to the “distribution” effect. Both growth and
inequality changes thus play a role in generating changes in poverty. The shaded areas to
the left of the poverty line show these changes.
4.3 Relationship between Inequality and Growth
Beyond the Bourguignon (2004) identity, former levels of inequality may have an influence on future growth. The literature on inequality and economic growth is indeed rich. However, the empirical evidence about the relationship is mixed. On the one hand, the literature that uses OLS regressions over a cross-section of nations generally finds that initial inequality is negatively related with future growth, when considering over 30 years (Alesina and Perotti, 1994). On the other hand, the literature using panel data over shorter periods generally finds a positive inequality-growth relationship (Li and Zou,
1998; Forbes, 2000).
In particular, Forbes (2000) suggests that one reason for the conflicting
inequality-growth results in the literature is that the relationship may differ for short and
for long periods (for example 5-10 versus 25-30 years). Forbes also notes that panel
12 This figure shows the density of the distribution of income, which is the share of the population at each level of income, where income is represented on a logarithmic scale on the horizontal axis.
70 techniques, such as fixed-effect estimators, capture how time-series changes in inequality within a country (or state) affect changes in its growth rate over a short period. In contrast, Barro (2000) argues that OLS models capture how persistent cross-sectional differences in inequality affect long-run growth rates, which is more relevant to understanding growth disparities. Therefore, the two methods may reflect different responses.
Recently, the influences of the New Economic Geography and spatial econometrics have shed light on the relationship between regional inequality and economic growth. There are a few studies of US regional growth, such as Partridge
(1997). Using pooled OLS models, this author finds that inequality is positively related to
growth. Panizza (2002), using panel data with fixed-effect models, finds that small
specification changes can turn around these results. Thus, cross-state results can suffer
from the same sign changes that characterize cross-country studies, when switching from
OLS to panel approaches. Again, these results may suggest different short-term and long-
term influences. This calls for careful specifications of the relationship.
An important advantage of these types of studies is that regions or states can be
used as good laboratories to examine inequality-growth issues. For example, Partridge
(1997, 2005) and Panizza (2002) both note that many of the hypotheses about these
relationships should apply to states, because they are essentially small open economies
with distinct histories and institutions. These authors also claim that, among states, there
appears to be sufficient variation in income distribution to produce differential outcomes,
due to large factor flows across states. In contrast, greater legal and informational barriers
would limit the flow of resources among countries, especially for low-income economies.
71 This, in turn, would reduce the factor flows that produce larger growth rate differentials.
Consequently, any income distribution-growth relationship should be much easier to detect using data for states rather than countries (Siebert, 1998).
Partridge (2005) studies these issues using data for 48 US states, over the 1960-
2000 period, and he finds that inequality is positively related to long-run growth. In his paper, Partridge makes four important contributions. First, regarding the ambiguous findings in the literature when moving from cross-sectional to panel data methods,
Partridge suggests that, instead of considering them as conflicting, these results should be considered as complements in the analysis. In general, conflicting results from various methodologies may not be a signal of lack of robustness, if there are separate long-run and short-run linkages. The results support Temple (1999), who argued that a variety of cross-sectional and panel approaches are necessary to fully understand the determinants of growth. Partridge’s conclusion is that “by examining separate short-run and long-run models, researchers can gain a more complete picture of transitory and dynamic responses and a better understanding of how policy affects economic processes” (p.389).
Second, after allowing for short-run and long-run effects, Partridge (2005) controls for different effects for the tails and the middle of the distribution. In this, he follows Easterly (2001), who argues that a middle-class consensus promotes growth by encouraging stability, mass education, and better public services. In effect, Partridge finds that a more vibrant middle class, measured by the middle-quintile income share
(Q3), is positively related to growth.
Third, following Kaldor (1956), Partridge (2005) argues that income inequality generates incentives for resources to be channeled into more efficient uses and is
72 conductive to saving and capital accumulation. This may explain the positive inequality- growth relationships found. However, these hypotheses assume that there is sufficient factor mobility in a given society, which may not be true for some developing countries.
In fact, Partridge (2005) makes it clear that these results are derived from the experience of advanced economies. It is thus interesting to test these hypotheses for developing countries, where high inequality and slow growth have been present.
4.4 Inequality in Argentina
In order to assess the dynamics of income inequality in Argentina, I will use data
from the SEDLAC, which is a database of socio-economic statistics constructed from
microdata coming from the Latin American and Caribbean (LAC) household surveys and
developed by CEDLAS (Universidad Nacional de La Plata) and The World Bank’s LAC
poverty group (LCSPP). All estimates are computed from the Encuesta Permanente de
Hogares (EPH). This survey has been carried out by the Instituto Nacional de Estadística
y Censos (INDEC) since the early 1970s in the Greater Buenos Aires area and since the
1980s in most large cities (with over 100,000 inhabitants), in two rounds: May and
October.
During 2003, a major methodological change was implemented by INDEC, including changes in the questionnaires and in the timing of the survey visits. The new survey (known as EPH Continua or EPH-C) is now conducted over the whole year.
INDEC also started to provide population weights that take the income non-response problem into account. To assess the impact of these methodological changes, I present three sets of statistics for 2003 in most tables: one computed from the EPH carried out in
73 May, and two computed from the EPH-C of the second semester of 2003. One of them is generated with the old weights (ignoring income non-response) and the other two use the new weights.
The EPH-C covers 28 conglomerates or large urban areas, which are home to around 70 percent of the Argentine urban population. Since the share of urban areas in
Argentina is 87 percent of the total population, the sample of the EPH represents around
60 percent of the total population of the country. Household surveys in Argentina cover only urban areas (the same problem is found in Uruguay). However, both Argentina and
Uruguay are two of the most urban countries in the world, with over 85 percent of the population living in cities.
In Argentina, like in many Latin American countries, household surveys have experienced significant improvements. In particular, major changes have been implemented since the early 1990s. Although these changes are very welcome, they pose significant problems for comparison purposes within countries over time. This is one reason why I decided to present data for Argentina since 1992. Other reasons are the incorporation of major cities in 1998 and the change in methodology.
Each decile in Table 1 in Appendix E includes an equal number of individuals
(not households). The income ratio 10/1 is the mean income of decil ten divided by the mean income of decil one. The ratio 90/10 is the mean income of percentile 90 divided by the mean income of percentile ten. Finally, the ratio 95/80 is the mean income of percentile 95 divided by the mean income of percentile 80.
The richest 10 percent of the population earned up to 40 percent of the total income, a peak reached in the first semester of 2002. By the second semester of 2005,
74 this share had declined lo levels similar to those for a decade before. However, the post- crisis shares are higher than the pre-crisis shares of the Menem’s period. In contrast, the poorest 10 percent of the population earned as little as 1 percent of total income (2001 to
2003). These extremes coincided with the recent financial crisis. Afterwards, however, the share of the poorest 10 percent never recovered to its 1992 level. This share was higher in the pre-crisis period. Thus, the rapid growth and stability of the pre-crisis era seem to have been associated with less inequality.
Table 2 in Appendix E shows several inequality indices related to the distribution
of per capita household income: the Gini coefficient, the Theil index, the Coefficient of
Variation (CV), the Atkinson index with parameters 0.5, 1 and 2, and the generalized
entropy index, with parameters 0 and 2 (the Theil index is the entropy index with
parameter 1). A simple correlation analysis among these indices shows that the Gini
coefficient is highly correlated with the Theil index (95 percent), the Atkinson index with
parameters 0.5, 1 and 2 (99 percent, 99 percent, and 96 percent, respectively), and with
the generalized entropy index with parameter 0 (99 percent). Therefore, I can analyze the
inequality in the distribution of income by just looking at the Gini coefficient. All the
inequality indices were calculated from the various editions of the household surveys
(Encuesta Permanente de Hogares, EPH ). Table 2 compares all the indices taking care of
all the modifications in the EPH between 1992 and 2005.
Figure 2 in Appendix E shows the Gini coefficient for Argentina. A major
increase in inequality took place in the country since 1992 (when the Gini was 0.45
percent). After the Argentinean crisis in December of 2001, the Gini jumped to 0.533,
75 then it reached its maximum level during the first half of 2003, at 0.537. Since then, the level of inequality has declined slightly.
4.5 Inequality in Latin America
What explains the high level of inequality observed in Latin America? To a large extent, most interpretations pursue the colonial inheritance argument, together with the persistence of the initial institutions. Among others, de Ferranti et al. (2004) highlight the combined role played by factor endowments and institutions. These authors explain that factor endowments, technology, and the relative scarcity of resources had important implications for the initial inequality. In Latin America, the characteristics of the colonies favored the establishment of large plantations (such as sugar) and mining activities that employed forced labor. As a result, a social structure emerged where a privileged few were in control of the most profitable activities and where most of the population was excluded from access to land, education, and political power. In contrast, the colonial powers in North America soon learned that there was no gold, there were few indigenous peoples to exploit, and soils and climates would not support the production of crops based on large slave plantations. Interestingly, Argentina is very different compared to most Latin American countries. In Argentina, there were no large plantations and mining activities that employed forced labor. Like in North America, land was cheap and labor scarce, while fertile soil and good weather conditions attracted migrants. These conditions might have explained the success of Argentina up to 1913.
Why did inequality persist over time? In answering this question, de Ferranti et al.
(2004) argue that the persistence of inequality during the colonial and early independence period took place because the initial “nexus” of institutions continued to exist, as did the
76 justification for these institutions. The elites that had benefited from colonial disparities were able to quickly gain effective control of the independent countries and determine the general structure of the institutions in ways that favored their interests (Engerman and
Sokoloff, 2006).
For many scholars, explaining the persistence of inequality over the 20 th century
is more challenging, because significant social, economic, and political changes occurred
during the 1900s. Moreover, for some authors the increase in urbanization rates should
have somewhat mitigated the relevance of the highly inegalitarian pattern of land
ownership and its impact on income inequality. Also, modernization moved most of the
Latin American countries in the direction of more open and democratic societies.
However, de Ferranti et al. (2004) believe that the most important causes for the
persistence of inequality over the 20 th century are the low quality of education, a development strategy based on import substitution and isolation from world markets, and imperfect financial markets that may have prevented the poorest from taking advantage of economic opportunities, by restricting their access to credit.
Unfortunately, there is no quantitative estimation of long-run inequality authenticating these arguments for Latin America. A good example is provided by
Bourguignon and Morrisson (2002), who investigated the historical trends in world income inequality. In their studies, conventional wisdom and lack of empirical evidence led them to assume that no changes in income distribution had taken place in Latin
America from independence to the mid-20th century.
Some authors claim that it is possible to infer the evolution of inequality since
1950 on the basis of direct income distribution observations. Table 3 in Appendix E
77 reports Gini coefficients for several Latin American countries. The table indicates that inequality continued to be essentially constant from 1950 to 2000, with a Gini between
0.51 and 0.55. There is, however, significant country heterogeneity. For instance, the
Gini coefficient noticeably increased in Argentina, from 0.396 to 0.477 between 1950 and 1990, but it may have declined in Venezuela, from a high of 0.613 in the mid-20th century to 0.459 in 1990. Likewise, El Salvador may have experienced a major deterioration in inequality over the 1960–1990 period, while Peru saw some progress.
For the pre-1950 period, data availability prevents direct inequality assessments.
One can still empirically investigate the evolution of income inequality using indirect indicators and ranges of country studies follow this approach. Bértola (2005) provides rough estimates of income distribution and Gini coefficients for Uruguay that go back to the late 1800s. Williamson (1999) looked at the consequences for inequality of the early phase of globalization (1870–1914). This author showed an increase of within-country inequality for Argentina and Uruguay over that period, on the basis of the evolution of the wage–land rental ratio. Bértola and Williamson (2003) claim that inequality trends reverted in the interwar period, when the observed abrupt decline in the wage-rental ratio stopped. This ratio increased somewhat after the 1930s. Calvo, Torre, and Szwarcberg
(2002) suggest that the level of inequality changed little during the 20 th century in
Argentina, while Londoño (1995) claims that the inequality levels observed in Colombia
during the 1990s were probably similar to those observed in 1938.
Prados de la Escosura (2005) builds on Williamson (2002) to explore the
historical evolution of the ratio of GDP per worker to the unskilled wage between 1850
and 1950 (or earliest possible date) for Argentina, Brazil, Chile, Mexico, and Uruguay.
78 The justification for this selection is that such a ratio compares the returns to unskilled labor with the returns to all factors of production. Since unskilled labor is the more equally distributed factor of production in developing countries, an increase in the ratio suggests that inequality is rising.
On this basis, Prados de la Escosura (2005) infers that in Argentina, Chile, and
Uruguay income inequality does not seem to have changed much over the period,
whereas Brazil and Mexico may have suffered some worsening in the distribution of
income. On the whole, the evidence that emerges from these studies indicates that, on
average, Latin America started the 20th century with a very high level of inequality,
which continued for the rest of the century, regardless of large variations by country in
special periods.
Table 4 in Appendix E compares the changes in inequality measured by the Gini
coefficient, using household surveys for 18 Latin American countries. By focusing on
the performance of inequality between 1989 and 2004, Gasparini, Gutierrez and
Tornarolli (2007) find that Argentina, Costa Rica, the Dominican Republic, Uruguay, and
Venezuela consistently rank as the most equal economies in the region, while Bolivia,
Brazil, Ecuador, Panama and Paraguay occupy the last positions in the inequality ladder.
However, Argentina and Colombia stand out as the countries that experienced the largest
increases in inequality, with Gini changes of around 6 percentage points. Brazil and
Mexico are the only countries that have experienced a drop in income inequality.
How do these tendencies contrast to those observed in developed economies?
Spain, for instance, experienced an important decline in income inequality between the
1970s and the 1990s, when the Gini coefficient went down by more than 10 percentage
79 points (Table 3 in Appendix E). Prados de la Escosura (2005) finds indirect indicators that suggest that income inequality has been declining in Spain since the 1950s, when
Spain may have had inequality levels comparable to (if not higher than) those observed in
Latin America. For 1950, Prados de la Escosura (2005) estimates a Gini coefficient for
Spain above 0.50.
In the United States, at the beginning of the century inequality was very high,
with a Gini of approximately 0.60 in 1920 (Plotnick et al., 1996). Inequality reached its
pre-World War II high in 1929. Then, from 1929 to 1951, income inequality fell
dramatically, to a Gini of about 0.40.
The United Kingdom experienced a similar pattern. Acemoglu, Johnson, and
Robinson (2002) show that the Gini coefficient for the United Kingdom might have been
more or less 0.55 in the 1890s. After that, for most of the 20th century, inequality seems
to have weakened. The authors also conclude that most of the decline of the United
Kingdom’s inequality took place between 1940 and the late 1970s. Atkinson (2002)
relies on income tax statistics to show that, in the early 1900s, the richest 1 percent in the
United Kingdom shared almost 20 percent of total personal income; in the late 1970s, this
group got 6 percent of this total.
Inequality in France evolved in about the same way. In the early 1900s, the share
of income of the wealthiest percentile in France was about 20 percent, whereas in the
1980s it was approximately 7 percent. The main difference between France and the
United Kingdom is that most of the decline in French income inequality took place
between the 1920s and 1950.
80 Thus, while inequality in Latin America has been persistent and steady over the
last century, inequality in Europe and the United States seems to have declined
significantly over the 20th century. If other countries have managed to break with their
histories on both the growth and income distribution fronts, then why Latin America
cannot also break with its history? This question goes beyond the scope of this
dissertation.
4.6 Regional Inequality in Argentina
Figure 3 in Appendix E shows the Gini coefficient for 23 provinces in Argentina, for the period 1991-2002. This coefficient ranges between 0.40 and 0.50 (See Maps 7 and
8 in Appendix B).
Figure 4 in Appendix E compares the regional Gini coefficient for six regions in
Argentina (See Maps 4 and 5 in Appendix B). Two interesting conclusions emerge. First, during the 1991-2002 period, the regions that experienced some positive per capita GDP growth (Pampeana and Patagonia) showed relatively less inequality, while the regions that experienced negative per capita GDP growth (Northwest, Northeast, and Cuyo) showed relatively higher inequality levels.
Second, the capital of Argentina, Greater Buenos Aires (BA) showed the highest per capita growth rate and also the highest level of inequality (0.482).
To test for differences in the Gini coefficient among regions, I use the General
Linear Model (GML) procedure that tests the null hypothesis that there is no difference in the mean of the Gini coefficient among the provinces in each region. The test rejects the null at the 1 percent level. So, after finding that differences exist among the means of the
Gini coefficient across regions, I use the Bonferroni’s and Tukey’s tests to determine
81 which means differ .13 Table 5 in Appendix E shows that both tests found that there are
differences among Gini coefficient among regions. So, I can consider regions 2, 6, and 3
(Pampeana, Patagonia and Cuyo) as one cluster with lower Gini coefficients, and regions
5, 1 and 4 (Northwest, Capital City, and Northeast) as another cluster of higher Gini
coefficients .
In order to analyze the development of inequality within each region in Argentina,
Table 6 in Appendix E presents the percentage change of four measures related to the income distribution. Column 1 compares the percentage change in the Gini coefficient for the six regions of Argentina between 1991 and 2002. It shows that the Pampeana Region had the largest increment in inequality (19.3 percent, which represents a 0.079 points increment), followed by Greater Buenos Aires (17.1 percent, which represents a 0.078 points increment). Patagonia showed the smallest increase in inequality. Column 2 shows the changes in the share of the third Quintile (Q3), which accounts for a “middle class consensus” and the role of the median voter. 14
Partridge (2005) explains that the Gini is used to control for the overall
distribution, while the share of the third Quintile (Q3) can be used to account for that
specific group in the population. Comparing across the regions of Argentina, Table 3
indicates that the share of total income that the “middle class group” earned during this
13 The Bonferroni test, based on Student's t statistic, adjusts the observed significance level for the fact that multiple comparisons are made. Tukey's honestly significant difference test uses the Studentized range statistic to make all pairwise comparisons between groups and sets the experiment wise error rate to the error rate for the collection for all pairwise comparisons. When testing a large number of pairs of means, Tukey's honestly significant difference test is more powerful than the Bonferroni test. For a small number of pairs, Bonferroni is more powerful. 14 A quintile is any of the four values which divide the sorted data set into five equal parts, so that each part represent one fifth or 20 percent of the sample population. The third quintile represents the group of population between the 40 and 60 percent of income levels.
82 period has been getting smaller in all regions. The decline amounts between 6 percent
(0.059 points) in Cuyo to 15 percent (0.067 points) in the Northeast region of Argentina.
Finally, I compare the changes in the shares of the richest 10 percent of the
population and poorest 20 percent of the population, in each region. In column 3, with the
exception of Pampeana, in all the other regions the richest 10 percent of the population
gained, up to additional 14.6 percent (4.8 percentage points) of total income in the
Northeast. In contrast, in all the regions, the share of the poorest 20 percent of the
population declined. The worst negative effect was suffered by Buenos Aires, with minus
44.3 percent (-2.3 percentage points) while the poorest lost the least in Patagonia.
4.7 Spatial Autocorrelation of Income Inequality
As in the previous chapter, here I want to test if spatial autocorrelation characterizes the measures of inequality among the provinces of Argentina. Figure 5 in
Appendix E displays the Moran’s I statistic for the provincial Gini coefficients in
Argentina between 1991 and 2002. 15 It shows that the Moran’s I statistic has been fluctuating during this period.
Table 7 in Appendix E presents the estimates for the Moran’s I statistic. For the
1991-2002 period, I estimated the coefficients using the EPH. The table shows that there
is evidence of spatial dependence, as the statistics are highly significant during this
period.
15 The reason why I am considering the period 1991-2002 in order to calculate the Moran’s I statistic for the provincial Gini Coefficient is because only during those years the “ Encuesta Permanente de Hogares ” (EPH) includes 23 provinces from a total of 24 provinces, and only the province of Rio Negro is not included in the sample.
83 The Moran’s I statistic corroborates that positive spatial autocorrelation exists.
That is, the value taken by the Gini coefficient at each province i tends to be similar to the values taken by the Gini coefficient at spatially contiguous provinces.
4.7.1 Local Spatial Autocorrelation for Income Inequality
Figures 6 and 7 in Appendix E offer a more disaggregated view of nature of the
spatial autocorrelation for the initial (Figure 6) and final (Figure 7 and 8) years. Each
figure contains a Moran scatterplot for the Gini coefficient. The slope of the regression
line equals Moran’s I = 0.015 for 1991, and Moran’s I = -0.111 for 2002. The Moran
scatterplot is divided into four quadrants:
• The upper right quadrant represents spatial clustering of a high-Gini province
with high-Gini neighbors (HH-quadrant I). In general, these locations are
associated with positive values of the local Moran Ii .
• The upper left quadrant represents spatial clustering of a low-Gini province
surrounded by high-Gini neighbors (LH-quadrant II). In general, these
locations are associated with negative values of the local Moran Ii .
• The lower left quadrant represents spatial clustering of a low-Gini province
surrounded by low-Gini neighbors (LL-quadrant III). In general, these
locations are associated with positive values of the local Moran Ii ; and
• The lower right quadrant represents spatial clustering of a high-Gini province
with low-Gini neighbors (HL-quadrant IV). In general, these locations are
associated with negative values of the local Moran Ii .
84 Viewing Figures 6 and 7 together corroborates the lack of stability in the measures of local spatial dependence for the Gini coefficient. While in Figure 6 there is no sign of local spatial autocorrelation at all in 1991, in Figure 7 there seems to be some negative spatial autocorrelation in 2002. However, the local Moran’s I in those years,
1991 and 2002, is not statistically significant. I then consider the local Moran’s I in 2001; in Figure 8, there is a strong sign of positive local spatial autocorrelation. In summary, the relationship is not stable.
In Chapter 3, I observed a clearer pattern of clustering, given by a positive spatial autocorrelation for the provincial real per capita GDP. Despite some fluctuation, the
Moran’s I is statistically significant throughout the whole period. In contrast, from
Figures 6, 7, and 8, it can be concluded that there is not a clear pattern of clustering for the Gini coefficient. More specifically, only in half of the years, a pattern given by a positive spatial autocorrelation for the provincial Gini coefficient can be observed. Thus, only for the years 1992 to 1994, 1997 and 1998, and 2001, provinces have the local indicators that significant fall in either quadrant I or III of the scatterplot, reflecting HH and LL clustering, respectively.
Concentrating only on those years where the Moran’s I shows statistically significant local spatial autocorrelation, I identify two clusters. First, there is the cluster of high-Gini province with high-Gini neighbors, represented mainly by the provinces of the Northeast region like Chaco, Formosa, Misiones and some provinces from the
Northwest region, including Catamarca, Jujuy, Tucuman and Santiago del Estero, each of which appears in quadrant I. The other main cluster of a low-Gini province surrounded by low-Gini neighbors (LL) includes provinces from the Pampeana region, Cuyo and
85 Patagonia, such as Buenos Aires, Capital City, and La Pampa (Pampeana); Mendoza, San
Luis, and San Juan (Cuyo); Chubut, Santa Cruz and Tierra del Fuego (Patagonia), all of which fall in quadrant III, the vast majority of the years. These results corroborate the findings using Bonferroni and the Tukey’s tests in section 4.6 (see Map 12 in Appendix
B).
Finally, as a measure of robustness of these results, I estimated the global and local measures of spatial autocorrelation while changing the W matrix. I obtained all the
previous results using wij as an element of a spatial weights matrix W such that wij =
distance in kilometers between each provincial capital city (location i) to all the others provincial capital cities (location j), using a cutoff point of 800 km and the actual routes
available in Argentina. Alternatively, I calculated wij as an element of a spatial weights
matrix W such that wij = number of hours that it takes to drive from location i to location j, using seven hours as the cutoff point. I obtained the same significant results.
Comparing maps 11 and 12 in Appendix B, I can draw my main conclusion at this
point. The cluster of provinces with high per capita income coincides with the cluster of
provinces with low Gini coefficient, specifically those provinces in the Patagonia. Also,
the cluster of provinces with low per capita income coincides with the cluster of
provinces with high Gini coefficient, specifically those provinces in the North East and
North-West regions (see Map 12 in Appendix B).
Much research has been conducted after the macroeconomic crisis that severely
affected inequality at the end of 2001. For example, Corbacho et al. (2007) analyze which
households were more vulnerable to the Argentine macroeconomic crisis during 1999-
2002. Their results suggest that households with more children and whose head was
86 male, less educated, and employed in the private sector were the most vulnerable, suffering a larger than the average decline in income. Moreover, shocks to labor income were significant, with both unemployment rates and unemployment spells increasing throughout the period, particularly during the peak of the crisis, towards the end of 2001.
Finally, these authors find that individuals with low levels of human capital (proxied by education and experience), males and the self-employed were more likely to lose their jobs. In contrast, public sector employees were more protected from the impact of the crisis on employment.
Rozada and Menendez (2002) found that in Argentina unemployment accounts for a large part of the increase in income inequality that the country experienced between
1991 and 2002.
Santos (2004) studied the most relevant determinants of inequality in the urban areas of Argentina, in the 1998-2003 period. The paper’s main results are that the high level of inequality in Argentina during 1998-2003 was determined by the labor market, through a combination of high unemployment rates and higher returns to education. Also, they found that urban areas with higher percentages of people with unsatisfied basic needs have higher Gini coefficients.
87 CHAPTER 5
REGRESSION MODELS AND RESULTS
This chapter has two main goals. First, it presents a basic framework for understanding spatial econometrics, and the different methods for including it in the regression analysis. Second, it contrasts the results of the regression analysis when using spatial econometrics techniques and with those from using other econometric methods.
The chapter first discusses the challenges presented by spatial autocorrelation and the comparative advantages of incorporating it in regression models either through a spatial lag or spatial error specification. As a first step in the actual implementation of the empirical strategy, a parsimonious long-run model is specified to relate growth in per capita income to the Gini coefficient at the beginning of the period as the only control variable. A second, base long-run model incorporates a ser of additional control variables. A third step explored the possibility of spill over effects from levels of inequality in neighboring provinces. A fourth step takes advantage of the cross-sectional and time series dimensions of the data, to specify a pooled OLS model, in both a parsimonious and a base version with a time fixed effect. Tests for heteroskedasticity and clustering of the error term lead to several additional approaches, including the Huber-
White Sandwich estimator and the Cluster-Robust Variance-Covariance estimator as well as Levene (1960) Robust Test statistic and a Feasible Generalized Least Squares
88 estimator. The empirical results to be reported are very robust and all confirm a negative statistically significant effect of inequality on per capita income growth in the provinces of Argentina. Similar results are obtained from various specifications using Panel Data.
5.1 Spatial Econometrics
The evidence in chapters 3 and 4 indicates that, in Argentina, there is positive spatial autocorrelation in the provincial per capita GDP and the provincial Gini coefficients. These results indicate that the values taken by these two series at each province i tend to be similar to the values taken by these two series at spatially
contiguous locations. Therefore, in order to analyze any relationship between inequality
and growth, it is important to determine whether or not this relationship is also affected
by spatial autocorrelation.
Attention to serial correlation has been the domain of time series analysis, while
the typical focus in the specification and estimation of models for cross-sectional data has
been heteroskedasticity. Until recently, spatial autocorrelation was largely ignored. In
other disciplines, dependence across space has been more central. For example, Tobler’s
(1979) first law of geography states that “everything is related to everything else, but
closer things more so,” suggesting spatial dependence to be the rule rather than the
exception. 16
The traditional emphasis of econometrics on heterogeneity in cross-sectional data
is not necessarily misplaced, since the distinction between spatial heterogeneity and
spatial autocorrelation is not always obvious. In particular, in a single cross section, the
16 Many techniques have been developed to deal with such dependence. For a comprehensive review, see Cressie (1993); other classical references are Cliff and Ord (1973, 1981), Ripley (1981, 1988), and Upton and Fingleton( 1985, 1989).
89 two may be almost equivalent. This problem is known in the literature as “true contagion versus apparent contagion” (Johnson and Kotz, 1969). The approach taken in spatial econometrics is to impose structure on the problem through the specification of a model, attached with extensive specification testing for potential departures from the null model.
This emphasis on the “model” distinguishes spatial econometrics from the broader field of spatial statistics (see Anselin, 1988, for further discussion of the distinction).
5.1.1 The Problem of Spatial Autocorrelation
Anselin and Bera (1998) define spatial autocorrelation as “the coincidence of value similarity with locational similarity” (p. 241). That is, high or low values for a random variable tend to cluster in space (positive spatial autocorrelation) or locations tend to be surrounded by neighbors with very dissimilar values (negative spatial autocorrelation). Negative spatial autocorrelation implies a checkerboard pattern of values and does not always have a meaningful interpretation (Whittle, 1954). Positive spatial autocorrelation implies that a sample contains less information than an uncorrelated counterpart. To carry out statistical inference, this loss of information must be explicitly acknowledged. This is the core of the problem of spatial autocorrelation in applied econometrics. The notion of “locational similarity” refers to the determination of those locations for which the values of the random variable are correlated. Such locations are referred to as “neighbors.” 17 More formally, spatial autocorrelation may be expressed
by the following moment condition:
Covyy(,)ij= Eyy ( ij ) − Ey ()*()0 i Ey j ≠ for i≠ j (1)
17 Locations are referred as “neighbors”, but strictly speaking this does not mean that they have to be collocated. For a more formal definition of neighbors in terms of a conditional density function, see Anselin (1988) and Cressie (1993).
90
where yi and y j are observations on a random variable at locations i and j in space. There is nothing spatial, per se, to the nonzero covariance in (1). It only becomes spatial when the pairs of i, j locations for which the correlation is nonzero have a meaningful interpretation in terms of spatial structure, spatial interaction or spatial arrangements of observations.
For the set of N observations on cross-sectional data, it is impossible to estimate the potentially ( N x N) covariance terms or correlations directly from the data. This is the fundamental problem in dealing with spatial autocorrelation and it requires the imposition of structure. More explicitly, in order for the problem to become tractable, it is indispensable to impose sufficient constraints on the ( N x N) spatial interaction
(covariance) matrix, allowing that a finite number of parameters characterizing the correlation can be estimated (Albert and McShane 1995).
There are two main approaches to imposing constraints on the interaction. In geostatistics, all pairs of locations are sorted according to the distance that separates them, and the strength of covariance between them is expressed as a continuous function of distance (Cressie, 1993). A second approach is called “lattice perspective” in which, for each data point, a relevant “neighborhood set” must be defined, consisting of those other locations that potentially interact with it. For each observation i, this yields a spatial
ordering of locations j∈ S i , where Si is the neighborhood set, which can then be exploited to specify a spatial stochastic process (Anselin and Bera, 1998).
The neighborhood set for each observation can be constructed by using the
“spatial weights matrix”, which is an ( N x N) positive and symmetric matrix W that
expresses for each observation (row), those locations (columns) that belong to its
91 neighborhood set as nonzero elements. More properly, wij =1 when i and j are neighbors
and wij = 0 otherwise. By convention, the diagonal elements of W are set to zero and the
W is often standardized such that the elements of a row add up to one. Thus, the elements
s of a row-standardized weights matrix equal wij= w ij/ ∑ w ij . This ensures that all weights j
are between 0 and 1 and facilitates the interpretation of operations with W as an averaging of neighboring values. It also guarantees that the spatial parameters are comparable across models. 18 A side effect of row standardization is that the resulting
matrix is likely to become asymmetric (since ∑wij≠ ∑ w ji ), even though the original j i matrix may have been symmetric. This complicates computational matters considerably in the calculation of several estimators and test statistics.
The specification of which elements are nonzero in the spatial weights matrix is arbitrary. The traditional approach is based on the geography of the observations, assigning area units as “neighbors” when they have a border in common or are within a
given distance of each other (Cliff and Ord, 1973, 1981). For example, wij =1 for dij ≤ δ ,
where dij is the distance between units i and j, and δ is the distance cutoff value. Cliff-
Ord weights are a function of the relative length of the common border, adjusted by the
inverse distance between two observations. Formally, Cliff-Ord weights can be shown as:
β bij wij = α (2) dij
18 This relates to constraints imposed in a maximum likelihood estimation framework, which implies that
the spatial autoregressive parameters must be constrained to lie in the interval 1/ wmin to 1/ wmax , where
wmin and wmax are respectively the smallest and the largest eigenvalues of the matrix W (Anselin 1982). For a row-standardized weights matrix, the largest eigenvalue is always +1 (Ord 1975), which facilitates the interpretation of the autoregressive coefficient as a “correlation” (Kelejian and Robinson 1995).
92 where bij is the share of the common border between units i and j in the perimeter
of i (where bij does not necessarily equal bji ), and α and β are parameters. Usually, the weights may be expressed as any measure of “potential interaction” between units i and j
(Anselin, 1998). Typically, the parameters of the distance function are set a priori (such
as α = 2 to reflect a gravity function) rather than estimated jointly with the other coefficients in the model. When such parameters are estimated together with other coefficients in the model, the resulting specification is highly nonlinear (Anselin, 1980;
Ancot et al., 1986; Bolduc et al., 1992).
5.1.2 Spatial Lag Operator
In time series analysis, the values for “neighboring” observations can be
expressed by using a backward or forward-shift operator on the one dimensional time
axis. This yields lagged variables yt− k or yt+ k , where k is the desired shift. By contrast,
there is no equivalent and explicit spatial shift operator; only a regular grid structure is a
potential solution. For instance, there is the rook criterion for contiguity, in which each
grid cell or vertex on a regular lattice, (i,j), has four neighbors: (i+1,j) (east), (i-1,j)
(west), (i,j+1) (north), and (i,j-1) (south). Corresponding to this framework there are four
spatially shifted variables: yi+1, j , yi−1, j , yi, j + 1 , and yi, j − 1 , each of which may require its own parameter in a spatial process model. Another example is the queen criterion, in which each observation has eight neighbors, yielding eight spatially shifted variables; the
four from the rook criterion, as well as yi−1, j + 1 , yi−1, j − 1 , yi+1, j + 1 , and yi+1, j − 1 , again each possibly with its own parameter (Cressie 1993).
93 These formal notions of spatial shift are unfeasible on an irregular spatial structure, since the number of shifts would differ by observation. To solve this problem, the spatial lag operator is used, which consists of a weighted average of the values at neighboring locations (Anselin and Bera, 1998). The weights are fixed and exogenous, comparable to a distributed lag in time series. Formally, a spatial lag operator is obtained as the product of a spatial weights matrix W with the vector of observations on a random
variable y, or Wy . Each element of the resulting spatially lagged variable equals w y , ∑ j ij j
which is a weighted average of the y values in the neighbor set Si , since wij = 0 for j∉ S i .
Row standardization of W guarantees that a spatial lag operation yields a smoothing of the neighboring values, since the positive weights add up to one.
5.1.3 Spatial Autocorrelation in a Regression Model
There are several approaches in order to incorporate spatial autocorrelation in a regression model. Anselin and Bera (1998) discuss two main approaches. These authors explain that spatial autocorrelation is modeled by means of a functional relationship between a variable, y, or an error term, ε , and its associated spatial lag, Wy for the
spatially lagged dependent variable and Wε for a spatially lagged error term. The resulting specifications are referred to as spatial lag and spatial error models . 19
In the spatial lag regression model, the spatial lag dependence is similar to the
inclusion of a serially autoregressive term for the dependent variable ( yt−1 ) in a time
19 S ee Anselin (1992), Anselin and Florax (1995), Anselin and Rey (1997) and Anselin, Florax and Rey (2004).
94 series context. In spatial econometrics, this is referred to as a mixed regressive, spatial autoregressive model (Anselin 1988). Formally,
y=ρ Wy + X β + ε (3)
where y is an ( N x 1) vector of observations on the dependent variable, Wy is the corresponding spatially lagged dependent variable for weights matrix W, X is a ( N x K) matrix of observations on the explanatory (exogenous) variables, ε is an ( N x 1) vector of error terms, ρ is the spatial autoregressive parameter, and β is a ( K x 1) vector of
regression coefficients.
The presence of the spatial lag term Wy on the right-hand side of (3) will provoke
a nonzero correlation with the error term, similar to the presence of an endogenous
variable, but different from a serially lagged dependent variable in the time-series case, in
which yt−1 is uncorrelated with εt in the absence of serial correlation in the errors. In
contrast, (Wy ) i is always correlated with εi , independently of the correlation structure of
the errors. Furthermore, the spatial lag for a given observation i is not only correlated with the error term at i but also with the error terms at all other locations. Therefore, not like the time-series case, an ordinary least-squares estimator will not be consistent for this specification (Anselin, 1988). In a reformulation of the same model:
yI=−(ρ WX )−1 β +− ( I ρε W ) − 1 (4)
The inverse matrix (I− ρ W ) −1 is a full matrix and not triangular as in the time
series case, where dependence is only one-directional. As a consequence, equation (4)
yields an infinite series that involves terms at all locations, given by
22 33 (IW++ρρ W + ρ W + ...) ε . Therefore, it follows that (Wy ) i contains the element εi
95 as well as other ε j , j≠ i .The implication of this particular structure is that the
simultaneity implied by the Wy term must be explicitly accounted for, either in a
maximum likelihood estimation framework or by using instrumental variables. When a
spatially lagged dependent variable is ignored in a model specification, but it is present in
the underlying data generation process, the resulting specification will suffer from an
omitted error variable, which explains why the use of OLS estimates will be biased and
inconsistent.
The interpretation of a significant spatial autoregressive coefficient ρ is not always easy. Two situations may be distinguished. First, the significant spatial lag term indicates true contagion or substantive spatial dependence. For example, it may measure the extent of spatial spillovers or diffusion. This interpretation is valid when the actors under consideration match the spatial unit of observation and the spillover is the result of a theoretical model. Alternatively, the spatial lag model may be used to deal with the spatial autocorrelation that results from a mismatch between the spatial scale of the phenomenon under study and the spatial scale at which it is measured. For example, when data are based on administratively determined units such as census tracts or blocks, there is no good reason to expect economic behavior to conform to these units. Thus, the inclusion of a spatially lagged dependent variable in the model specification is a way to correct for this loss of information. The model then allows for the proper interpretation of the significance of the exogenous variables ( Xs ), after the spatial effects have been corrected for or filtered out. More formally, the spatial lag model may be expressed as:
(I−ρ Wy ) = X β + ε (5)
96 where (I− ρ W ) y is a spatially filtered dependent variable. This means that the
effect of the spatial autocorrelation has been taken out.
A second way to incorporate spatial autocorrelation in a regression model is to
specify a spatial process for the disturbance term. This method is called the Spatial Error
Model . The resulting error covariance will be no spherical, and thus OLS estimates, while still unbiased, will be inefficient. The most common specification is a spatial autoregressive process in the error terms:
y= X β + ε (6)
which is a linear regression with error vector ε , and
ε= λW ε + ξ , (7)
where λ is the spatial autoregressive coefficient for the error lag Wε and ξ is an uncorrelated and homoskedastic error term. Alternatively, this may be expressed as:
yX=β +( I − λ W ) −1 ξ (8)
Spatial error dependence may be interpreted as a nuisance and the parameter λ as a nuisance parameter, in the sense that it reflects spatial autocorrelation in measurement errors or in variables that are otherwise not crucial to the model, like the “ignored” variables spillover across the spatial units of observation. The spatial autoregressive error model can also be expressed in terms of spatially filtered variables. In this case, it can be written as:
(I−λ Wy )( =− I λβξ WX ) + (9)
This is a regression model with spatially filtered dependent and explanatory
variables and with an uncorrelated error term ξ .
97 Some authors have suggested processes that combine spatial lag with spatial error dependence, though such specifications have seen only limited applications. The most general form is the spatial autoregressive, moving average (SARMA) process outlined by
Huang (1984). Formally, a SARMA (p, q) process can be expressed as:
y=ρρ11 Wy + 22 Wy + ρ 33 Wy ++... ρp WyX p ++ βε (10)
for the spatial autoregressive part, and
ελξλξ=11W + 22 W ++... λξξq W q + (11)
for the moving average part.
These spatial regression models can be solved using maximum likelihood and
instrumental variables estimators.
5.2 Inequality-Economic Growth Models
For the most part, the income-distribution/growth literature is associated with
countries. Partridge (1997) and Panizza (2002) show, however, that many of the
hypotheses should apply to provinces or states within countries, because they are
fundamentally small open economies with individual histories and institutions. There also
seems to be sufficient variation in income distribution to produce differential economic
outcomes.
From the income-distribution/growth literature that applies to provinces or states
within developed countries, only few of the studies related to income convergence and
inequality have used spatial econometrics (Rey and Montouri, 1999; Anselin, Florax and
Rey, 2004). The same can be observed in Latin America, only a small portion of the
literature regarding income convergence applies spatial econometrics tools: Aroca and
98 Bosch (2000) for Chile; Aroca, Bosch and Maloney (2005) for Mexico; Bosch et al.
(2003) and De Vreyer and Spielvogel (2005) for Brazil, and Madariaga, Montou and
Ollivaud (2005) for Argentina.
There are advantages to using provinces to examine growth issues. For example,
as Partridge (2005) has noted, large factor flows across provinces should highlight how
small disparities in initial conditions affect economic growth. In contrast, legal and
informational barriers limit the flow of resources between countries, especially for low-
income ones, which dampens the factor flows that produce larger growth differentials.
Hence, any income-distribution/growth relationship should be much easier to detect when
using provinces.
The debate about the potential tradeoff between equity and growth permeated
neoclassical economic thought (Okun, 1975). For some, income inequality generates
incentives for resources to be channeled into more efficient uses and it encourages
savings and capital accumulation (Kaldor, 1956). Inequality creates market signals for
resources to reallocate across industries, occupations, and regions, and it encourages
greater labor specialization and human capital accumulation (Edin and Topel, 1997).
Superior incentives associated with inequality have long been associated with
Schumpeterian factors such as entrepreneurship, risk taking, and innovation (Siebert,
1998) as well as greater work effort (Bell and Freeman, 2001). Self-selection among
inter-provincial migrants can reinforce the incentive effects. For example, Borjas,
Bronars, and Trejo (1992) find that higher- (less-) skilled or more- (less-) able workers
tend to migrate to states with more (less) inequality because of relatively higher market
rewards for their skills. Therefore, these ‘‘classical’’ economic approaches explain
99 greater future growth by stressing how inequality enhances incentives that increase efficiency and capital accumulation.
Recent theories dispute the validity of the equity and growth trade off on several
grounds, because greater inequality may cause policy distortions inefficient credit
constraints, and social and/or political conflict (Kucera, 2002). One important example is
that when there are credit constraints, inequality implies a differential ability of
households to make investments in human or physical capital, reducing subsequent
economic growth (Aghion, Caroli, and Garcıa-Peñalosa, 1999). Political economy
models predict that greater inequality increases corruption, crime, and social
disturbances, which intensify pressures for distorting government redistribution policies
(Persson and Tabellini, 1994). Such models are also applicable to wealthy oligarchies that
expropriate rents. Welch (1999) cautions that inequality may be harmful when low-
income individuals view society as unfair and when upward movement is limited by
insufficient income mobility.
5.3 Regression Specification
The sample covers 22 provinces and the Ciudad Autónoma de Buenos Aires, over
the 1991-2002 period. Assuming that measurement errors are smaller than across
countries, one empirical advantage of using provinces is that provinces should have
relatively similar growth mechanisms and institutions. As Partridge (2005) suggests, the
idea of using the same specification across provinces is more appealing than pooling very
different countries in one model.
100 The empirical testing will proceed as follows. First, a long-run growth model will
be estimated for provinces i during the 1991-2002 period, using the annual percentage
change in real per capita GDP as the dependent variable:
i i i Growth19912002− =+αβi 1 Gini 1990 + β 2 X 1990 + ε it (12)
where Growth is an (n × 1) vector of province i growth rates of real per capita
GDP in the period, and Gini is the Gini coefficient for province i in 1990. The Gini represents the inequality of the overall income distribution, rather than concentration measures for segments of the population (e.g. quintiles). Partridge (2005) finds that the income distribution has an entirely separate effect at the middle versus the tails of the distribution. So, the third quintile (Q3) share accounts for a “middle class consensus” and the role of the median voter, while the Gini controls for the overall distribution. Easterly
(2001) argues that a middle-class consensus promotes growth by encouraging stability, mass education, and better public services. At this point, I am only interested in the overall income distribution; later on, I will consider the effect of Q3 as well. Finally, the vector X represents a set of control variables.
Equation (12) is a long-run model in which growth during the 1991-2002 period is regressed on the initial values of the explanatory variables. One advantage of this type of specification is that, with the exception of the income level for the current period t,
there should be no direct endogeneity.
5.3.1 Spatial Econometric Model Specification
When models are estimated for cross sectional data on spatial units, ignoring the
lack of independence across units can cause serious problems of misspecification
(Anselin, 1988). Three simple types of spatial econometric models are typically used to
101 deal with the spatial dependence of observations: the spatial lag model, the spatial error model, and the SARMA model (Anselin, 1988; Anselin and Bera, 1998; Florax and
Folmer, 1992).
5.3.2 Spatial Lag Model
The spatial lag model to be estimated is given by:
i i ii 2 Growth19912002−=ρ WGrowth 19912002 − ++ αβi 1 Gini 1990 ++ βε 2 X 1990 , ε∼ N(0, σ I ) (13)
where WGrowth is a spatially lagged dependent variable for a spatial weights
matrix W, ρ is a spatial autoregressive parameter, and ε is a vector of spherical error
terms.
From a spatial filtering perspective, the spatial lag model may be expressed as:
i i i 2 (I -ρ W)Growth19912002− = αβi + 1 Gini 1990 + βε 2 X 1990 + , ε∼ N(0, σ I ) (14)
where (I − ρW)y is a spatially filtered dependent variable; i.e., the effect of spatial
autocorrelation has been filtered out.
From equation (14), the spatial lag model allows for the proper interpretation of
the significance of the exogenous variables, after the spatial effects have been corrected
for or filtered out.
The presence of the spatially lagged dependent variable WGrowth on the right-
hand side of equation (13) will induce a nonzero correlation with the error term. The
spatially lagged dependent variable [WGrowth] i is always correlated with εi , irrespective of the correlation structure of the errors. The spatial lag for a given observation i is not
102 only correlated with the error term at i but also with the error terms at all other locations.
Therefore, ordinary least squares (OLS) estimation of the spatial lag model specification yields biased and inconsistent estimates for the coefficients, due to the simultaneity between the error terms and the spatially lagged dependent variable. Instead, alternative estimators based on maximum likelihood (ML) and instrumental variables (IV) have been suggested as consistent estimators (Anselin, 1988; Kelejian and Robinson, 1993; Anselin and Bera, 1998; Kelejian and Prucha, 1998; Conley, 1999).
5.3.3 Spatial Error Model
The spatial error model to be estimated is given by:
i i i Growth19912002− =+αβi 1 Gini 1990 + β 2 X 1990 + ε it (15)
2 where ε = λ W ε + ζ , ζ∼ N(0, σ ζ I ) (16)
Alternatively, from a spatial process perspective, the spatial error specification
(16) may be expressed as
i i i -1 2 Growth19912002− =+αβi 1 Gini 1990 + β 2 X 1990 + (I - λζ W) , ζ ~ N (0, σ ζ I ) (17)
Equation (17) indicates that a random shock in a specific province will not only
affect growth in that province, but it will also affect the growth performance of other
areas through the inverse spatial transformation (I -λ W) -1 . The effects of the random shock will diffuse throughout the entire regional system through the spatial multiplier effect, which yields a Leontief expansion:
(I -λ W)-1 = I + λλ W + 2 W 2 + λ 3 W 3 + · · · (Anselin and Bera, 1998).
OLS estimation in the presence of non-spherical errors yields unbiased estimates of the parametersbut a biased estimate of the parameter’s variance. Thus, inference based
103 on the OLS estimates may be misleading. Instead, inferences should be based on the spatial error model estimated by ML or generalized method of moments (GMM)
(Anselin, 1988; Anselin and Bera, 1998; Conley, 1999; Kelejian and Prucha, 1999).
I do not believe that the SARMA model is useful here for two main reasons. Even though it may appear convenient to combine both the spatial lag and the spatial error dependence, it is difficult to disentangle which one is more relevant. It is also more difficult to interpret the spatial coefficients. Therefore, I plan to check for the appropriateness of the SARMA model through a suitable statistic and, if it is relevant, I will include it in my analysis.
5.4 Empirical Results
5.4.1 Parsimonious Long Run Model
In the literature there is evidence of a long-run relationship between inequality
and growth. For example, Partridge (2005) finds a long-run relationship between
inequality and growth in the 48 U.S. states over the 1960-2000 period. For Argentina, the
data show a persistent and increasing level of inequality. To identify a long-run
relationship, however, one important limitation is that I have data on inequality only for
1991 to 2002. Given this limitation, I will determine whether a 12-year long-run model is
sufficient to find any significant relationship between inequality and growth in Argentina.
I will compare these results to those from a 5-year model and a 3-year model. The longer
the spell of the model, the better the possibility to capture a long-run relationship.
In a first step, I present the results for a parsimonious long-run model. In Table 3,
I compare three long-run sub-models: a 12-year estimation, a 5-year estimation, and a 3-
104 year estimation. In these parsimonious models, I consider the Gini coefficient as the only control variable. For each sub-model, I present the results of the estimation of the OLS regression in the first column, results of the spatial lag model in the second column, and results of the spatial error model in the third column.
As Table 5.1 shows, all regressions yield negative coefficients for the Gini at the beginning of the period, for the 3 sub-models. The estimate coefficients are not significant, except 12-year estimation. In this case, inequality at the beginning of the period negatively affects average provincial economic growth over the period. Next, I tested for spatial autocorrelation by using the robust Lagrange Multiplier, and I found robust evidence of spatial autocorrelation in all the sub-models, particularly when the longer period is considered. The diagnostic for the presence of spatial effects shows results for the two robust Lagrange Multiplier tests and the SARMA test. The null hypothesis is the absence of spatial autocorrelation, which means that knowing the location associated with the units of analysis does not add any information (i.e., the pattern observed is equal to spatial randomness).
Here, the Robust Lagrange Multiplier for both the spatial lag model and the
spatial error model are statistically significant, particularly for the 12-year sub-models.
Thus, the null hypothesis of spatial randomness in can be rejected. In turn, the statistic for
the presence of spatial effects in the form of SARMA (AIC) is not statistically
significant. Thus, spatial autocorrelation can be best interpreted in terms of the spatial lag
or spatial error models.
105 These results mean that it would be a mistake to consider each individual province as an independent unit of analysis. They corroborate in a simple econometric way my findings in chapters 3 and 4.
My hypothesis is that, in Argentina, the spatial lag model is a good model to capture the spatial spillovers or diffusion of per capita GDP growth across provinces. A spillover of per capita GDP growth exists across the clusters generated by the positive spatial autocorrelation found in chapter 3. However, the effects of inequality on per capita GDP growth are multifaceted and go beyond the spillover of per capita GDP growth on the neighboring provinces. Thus, the spatial error model would be a better specification to capture this complex relationship. Moreover, the spatial error model reflects spatial autocorrelation in measurement errors and/or in variables that have been
“ignored” in the model because they are difficult to measure or for which data are not available. The Gini may affect economic growth through different linkages such as incentives, market signals, risk taking, innovation, social and/or political conflict, credit barriers, corruption, crime, disturbances, distorting government redistribution, income mobility and migration, among others.
Thus, in order to account for spatial autocorrelation, I run the spatial lag model and the spatial error model for the three sub-models. A simple W matrix was used, namely the spatially contiguous one that assigns a nonzero spatial weight as:
wij = 0 if dij ≤ lb or dij > ub (18)
1 wij = f if lb< dij ≤ ub (19) dij
106 where ( i,j ) denotes the location pair, dij denotes the distance between locatios i and j, lb denotes the lower bound of the specified distance band, ub denotes the upper bound of the specified distance band, and f denotes a positive friction parameter (by
default f=1).
Thus, after controlling for spatial autocorrelation by using the spatial lag model
and comparing the results with the spatial error model, I can draw three main
conclusions.
First, spatial autocorrelation best works through the error term, instead of the
spatial lag, as indicated by the Akaike info criterion (AIC). Even though, for the 12-year
long-run mode,l both spatial models show statistically significant estimates for lambda,
the lag growth variable, and the robust Lagrange Multiplier statistic, the spatial error
model performs better overall.
Second, the only model that significantly describes the relationship between long-
run growth and the Gini at the beginning of the period is the 12-year model. In this
model, after controlling for spatial autocorrelation by using the spatial lag model, the
coefficient for the Gini becomes statistically significant at 10 percent. Using the spatial
error model, the coefficient of the Gini becomes statistically significant at 5 percent.
107 Long Run Model (12 years) Long Run Model (5 years) Long Run Model (3 years) Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Model 9
Spatial Spatial Spatial Spatial Spatial Spatial Lag Error OLS Lag Error OLS Lag Error OLS Growth Growth Growth Growth Growth Growth Growth Growth Growth gini -0.2385 -0.2961* -0.3205** -0.1894 -0.1987 -0.2007 -0.1019 -0.1039 -0.1002 0.2011 0.1743 0.1638 0.1735 0.1618 0.1642 0.1457 0.1323 0.1318 Lambda 0.442** 0.323 0.343
0.207 0.2568 0.2486 W_Growth 0.4113** 0.2949 0.3408 0.2062 0.2505 0.2484 Moran's I (error) 2.713 *** 1.725*
Lagrange Multiplier (lag) 4.174 ** 3.646** 3.3876* 2.282 3.0471* 2.4575
Lagrange Multiplier (error) 5.1161** 4.342** 4.2902** 2.7595* 3.1552* 2.506 Observations 24 24 24 24 24 24 24 24 24 R-squared 0.0601 0.2299 0.2579 0.0529 0.1588 0.1794 0.0222 0.1449 0.147 AIC -23.89 -25.53 -28.23 -31.76 -32.04 -34.52 -39.8 -40.26 -42.31 Standard errors in second row * significant at 10%; ** significant at 5%; *** significant at 1%
Table 5.1: Results from parsimonious long-run models
Further, the R-squared indicates that differences in the Gini at the beginning of the period explain almost 26 percent of the variation in the real per capita income growth across provinces for the 12-year model.
Finally, while these results seem to be consistent with the Partridge (2005) findings that there is a long-run relationship between the overall distribution of inequality and growth, for the case of the US, the opposite result is actually observed for Argentina.
That is, for the provinces of Argentina the long-run relationship between inequality and growth is negative and not positive, as it is the case for the US. Thus, while in the US
108 inequality encourages growth, in Argentina inequality is an obstacle to growth. A different conceptual framework will be needed, therefore, to interpret these results.
In all cases, I diagnosed for heteroskedasticity using the Breusch-Pagan test, the
Koenker-Bassett test, and the White test and, in all the cases for all sub-models, I always failed to reject the null hypothesis of spherical errors (homocedasticity).
5.4.2 Base Long-Run Model
Next, I present the results for a “Base long-run model,” using as the benchmark
the 12-year parsimonious long-run model from the previous section.
This is a more complete 12-year long-run model, because I consider a set of
control variables, which are part of the vector X in equation (12). Specifically, I run the
full equation in which the 1991-2002 growth rate of per capita GDP is regressed on the
initial values of the explanatory variables. Growth is the province i average growth rate
of real per capita GDP for the period. Gini is the Gini coefficient for province i at the
beginning of the period, and the vector X includes the following variables: Initial income,
Density, Population, Private credit, Agglomeration, Distance, and Higher education.
Initial Income is the level of real per capita income in 1991. As Partridge (2005)
argues, in the empirical literature there are two main perspectives. On the one hand, in
neoclassical models the initial level of income proxies for the initial level of development
and it is inversely related to subsequent growth (Barro, 2000). Using this approach, Barro
and Sala-i-Martin (1991a) find that developed countries slowly converged over much of
the 1800s and 1900s. This is mainly due to the fact that neoclassical convergence relies
on differentials in returns to capital per worker and diminishing marginal returns, which
109 cause factors of production to relocate to developing countries where returns are higher
(Ventura, 1997).
On the other hand, other researchers believe that the initial level of income should
be omitted, because both endogenous and neoclassical growth models suggest that the
initial income term eventually drops out, when economies are near enough their steady-
state (limit) growth paths (Durlauf and Quah, 1999). This suggests that the income term
can be omitted because, when provinces are very close to their steady state, deviations
primarily reflect transitory cyclical and structural shocks rather than neoclassical
convergence. If income is included and cyclical conditions dominate, instrumental-
variable (IV) approaches are required.
Partridge (2005) claims that initial cyclical shocks would negatively bias the
initial income coefficient, making more likely the finding of a stronger ‘‘convergence’’
effect. That is, a negative transitory shock at the beginning of the period in a particular
province would reduce the initial income level and produce a faster subsequent growth
rate, as the province recovers. Another gain to omitting the initial level of income is that
it is probably very correlated with other characteristics (e.g., education), such that it may
take away some of the true effects of other variables when it is included in the model.
Similarly, some have suggested that the convergence effect may be an example of
Galton’s fallacy, in which this effect could simply be spurious (Durlauf and Quah, 1999).
In fact, Scully (2002) argues that, for these reasons alone, the initial income term should
not be included, because it can introduce severe bias in the results.
Here, I decided to include the initial level of income because I do not believe that
in the case of Argentina the provinces are close to their steady state level and because this
110 variable proxies for the initial level of development. I expect to find a negative relationship between initial income and growth, indicating convergence in income per capita among the provinces of Argentina.
The size of the population and the endowment of natural resources are important
ingredients that determine the initial conditions for economic growth. Density , the
number of people who lived in each province per squared kilometer in 1991, is
introduced to control for the size of the population relative to the size of each province.
Also, I include the variable Population , which is the rate of population growth between the censuses of 1895 and 1914. This variable is expected to control for the productivity of a labor force. I assume that the earlier the observation for the productivity of the labor force, the better the chances of avoiding endogeneity. I also run the regression using the rate of population growth in 1980 and the results did not changed. I expect to find positive relationships between “Density” and “Population” and per capita income growth.
Levine (2004) surveys the theoretical and empirical research on the connections between the operation of the financial system and economic growth. He finds evidence indicating that finance (both financial intermediaries and capital markets) has a positive impact on long-run economic growth. Moreover, Beck, Demirgüç-Kunt and Levine
(2004), using a sample of 52 countries over the 1960 to 1999 period, also find a positive relationship between financial intermediaries development and economic growth. They use credit by financial intermediaries to the private sector divided by GDP as a measure of financial intermediary development, an indicator commonly used in the finance and growth literature. Beck, Demirgüç-Kunt, and Levine (2004) and Levine (2004) show a robust causal link from Private Credit to per capita GDP and per capita productivity
111 growth. Private Credit , the value of private credits from financial institutions to the
private sector divided by the provincial GDP for province i, at the beginning of the
period, accounts for the effect of financial deepening on growth. This is an important
variable to consider for Argentina, as during the 1991-2002 period, a major process of
financial deepening took place. Therefore, I expect a positive sign for the coefficient of
Private Credit .
Following Krugman (1991), many papers have highlighted the importance of
agglomeration in the process of economic growth. The core model of geographical
economics attempts to explain why some regions attract more manufacturing firms than
other regions. It claims that a new firm has an incentive to locate where the majority of
other firms are. This process of agglomeration increases local incomes and leads to a
higher increase in demand. In this process, one can observe “the tendency of increasing
returns industries, other things equal, to concentrate near their larger markets and to
export to smaller markets” (Helpman and Krugman, 1985, p. 197). This effect is caused
by the interaction of external and internal economies of scale and it is known as the
“home-market effect”.
I believe that this process is important in Argentina, because of the agglomeration
of firms in big cities in regions such as Buenos Aires, Mendoza, Tucuman, Santa Fe,
Cordoba, Parana, and Ushuaia. Agglomeration represents the share of industrial
production of province i relative to the total for the country. I expect a positive sign for this variable.
Another important feature of the core model of geographical economics is the
consideration of distance as an important determinant of the concentration of firms,
112 which is the main characteristic of the “home-market effect”. The concept of distance is introduced in the core model of geographical economics as of cost of transportation, which is a function of the distance among cities or regions. Distance represents the
distance in kilometers between provincial capital cities weighted by the size of the
population in each province. This variable tries to control for the differences in geography and transportation costs. I expect a negative sign for the Distance variable.
Another important determinant of economic growth is the accumulation of human
capital. Higher Education represents the percentage of adults with college degrees for province i at the beginning of the period, in order to control for human capital accumulation. I expect a positive sign for Higher Education .
Table 4 shows the results of the OLS specification for model (12). The OLS estimation shows a negative coefficient for the Gini coefficient, which in this case is statistically significant and higher than in the case in the simpler parsimonious long-run model, where the coefficient was negative but not statistically significant. The coefficient has the same sigh and is even more significant in the spatial error and spatial lag models.
Furthermore, the coefficient for the initial level of income is negative, but it is statistically significant in the OLS model. This coefficient is negative and statistically significant in the spatial error model, but not in the spatial lag model. Thus, the justification for its inclusion does not seem to find strong support. This suggests
“convergence” in per capita income across provinces.
The coefficient of the variable Private Credit is positive and statistically significant in all three sub-models. The variables Population Growth , Agglomeration,
Higher Education, and Distance, all of which have the expected sign, are not statistically
113 significant. The model explains 30 percent of the per capita income growth for the period considered.
The Lagrange Multiplier tests show robust indication of spatial autocorrelation. I
have also considered the Lagrange Multiplier for the presence of spatial effects in the
form of SARMA, but it is not statistically significant. Therefore, I have not considered
this specification further. Following the same procedures of the previous section, I run
the spatial error model and the spatial lag model in addition to OLS. In Table 4, Model 11
is the spatial error model, which proves to be, one more time, the best specification to
explain per capita income growth for the provinces of Argentina during the 1990s.
Table 4 also shows that, for the spatial error model, the variables Initial Gini,
Initial Income, and Private Credit are statistically significant, as is also lambda. The
spatial error specification explains 65 percent of the variation of the real per capita
income growth from 1991 to 2002. Likewise, after comparing this model to the spatial
lag model (Model 12 in Table 4), the spatial error model is the best specification,
according to the AIC criteria, as I had anticipated in the previous section.
114 Model 10 Model 11 Model 12 Spatial Error Spatial Lag OLS Growth Growth Growth Initial Gini -0.5702* -0.6953*** -0.5657** 0.3187 0.2349 0.2341 Initial Income -0.1155 -0.1061* -0.092 0.0825 0.0632 0.0613 Density 9.4E-05 5.4E-05 8.8E-05** 5.5E-05 3.5E-05 4.0E-05 Population Growth 0.0219 0.0173 0.016 0.0166 0.0116 0.0123 Private Credit 0.2489* 0.2531* 0.2177* 0.1414 0.1076 0.1050 Agglomeration 0.5048 0.5513 0.4777 0.6750 0.3853 0.4937 Higher Education 0.0109 0.0150 0.0084 0.0142 0.0113 0.0106 Distance -4.8E-05 -2.5E-03 -7.7E-04 0.0456 0.0293 0.0333 Lambda 0.5721*** 0.1769 W_Growth 0.3084** 0.1589 Lagrange Multiplier (lag) 2.6205 3.0950 0.1055 0.0785* Lagrange Multiplier (error) 1.8916 4.0163 0.1690 0.0451** Observations 24 24 24 Adjusted R-squared 0.3031 0.6529 0.6109 AIC -27.3239 -31.3401 -28.4189 Standard errors in second row * significant at 10%; ** significant at 5%; *** significant at 1%
Table 5.2: Base long-run models
115 Finally, diagnostics for heteroskedasticity, like the Breusch-Pagan test, the
Koenker-Bassett test, and the White test, in all the cases for all the sub-models fail to
reject the null hypothesis of spherical errors (homocedasticity).
5.4.3 Gini Spillover Effect Model
In the third step of the analysis, I test for the hypothesis that in the provinces of
Argentina there is a spillover effect, caused by the inequality of the neighboring
provinces on the economic growth of province i. In the previous two sections, I focused on within inequality in a long-run framework. Now, I focus on the Gini coefficient of the neighboring provinces to province i in a short-run framework. To accomplish this, I use
the following specification:
i j1 j 2 j3 j 4 j5 j 6 Growthti=+αβ1 Gini t + β 2 Gini t + β 3 Gini t + β 4 Gini t + β 5 Gini t + β 6 Gini tit + ε (20)
where real per capita income growth in province i at time t is a function of the
level of inequality of its six closest provinces, instead of just its own level of inequality.
In Model 1 of Table 5, the coefficients for all the Gini are negative and all, except for the
second and the last one, are statistically significant. This “Growth Model with Gini
Spillover” explains 22 percent of the average variation of real per capita income growth
of province i in time t.
Thus, this simple model shows some evidence that in the provinces of Argentina
there is some kind of spillover effect, caused by the inequality of the neighboring
provinces on the economic growth of province i.
116
Model 1
Growth Model w/ Gini Spillover gini_1 -0.2680* -0.1429 gini_2 -0.1197 -0.128 gini_3 -0.4013** -0.1558 gini_4 -0.3546* -0.1867 gini_5 -0.2070* -0.1185 gini_6 -0.1185 -0.1334 Observations 276 Adjusted R-squared 0.2204
Standard errors in second row* significant at 10%; ** significant at 5%; *** significant at 1%
Table 5.3: Growth model with Gini spillover effects.
5.4.4 Pooled-OLS Models
In a fourth step of my analysis, I present the results of a “pooled-OLS Model,” in order to take advantage of the cross-sectional and time series characteristics of my data.
In this section, I combine the own province i level of inequality and the level of inequality of the neighboring provinces in their effects on province i income growth in the same model. Therefore, I pooled the data for 22 provinces and the Capital City for 12 years (1991 to 2002), in order to estimate the following OLS regression:
117 i iiiii i Growtht=+βββββ01 Gini tttt + 2 Q3 +++ 3 Y 4 X REG ++ Year tt ε (21)
i where Growth t is an (n × 1) vector of province i growth rates of the real per
i capita GDP at time t , and Gini t is the Gini coefficient for province i at time t, which represents the income inequality of the whole income distribution in the same period.
Partridge (2005) argues that the income distribution might have a separate effect at the middle versus the tails of the distribution. Easterly (2001) argues that a middle-class consensus promotes growth. This ‘‘consensus’’ seems to be closely related to the social-
capital literature that stresses cohesiveness and trust (Bowles and Gintis, 2002; Henry,
2002). Glaeser, Scheinkman, and Shleifer (2002) argue that a vibrant middle class and
secure property rights are often found together.
Therefore, in equation (21), I include the Gini, which controls for the overall distribution,
as well as the third quintile (Q3) share, which would control for the strength of the
middle-class and the role of the median voter in each province. When the Q3 is included
in the model, the Gini controls for the overall distribution, especially at the tails.
Partridge (2005) highlights that the influence from the tails of the distribution may reflect
the influence of imperfect credit markets or of incentives from classical models. Iyigun
and Owen (2004) show how the influence of credit-market constraints is affected by the
size of the tails of the distribution.
Thus, I use the Gini and the Q3 as two complementary distribution measures,
while most studies use only one measure at a time. These two variables thereby have the
advantage of being the most widely used, allowing for comparability. I conducted
experiments adding squares of inequality measures, but the squares were insignificant
118 and there was no discernible pattern to the results. Other international studies find that different distribution measures produce similar results (e.g., Clarke, 1995).
I find that the Gini and the Q3 are negatively correlated (-0.73 on average, in this sample). This implies that, over time, any deterioration in the distribution of income would be accompanied by a reduction of the share of income received by the third quintile. As shown in Chapter 4, in Argentina, the growing inequality in the distribution of income worsened the participation of the middle-class and of the poorest in the total income, while it benefited the richest quintile. Thus, as Partridge (2005) explains, any specification that does not include both measures may produce misleading results. For instance, if Q3 were omitted, some of its effects would be picked up by the Gini regression term, reducing it towards zero.
i The variable Yt is the real income per capita at time t in province i. Following the debate about including this variable or not, as a robustness test I decided to compare different specifications of equation (21), by not including real per capita income, using the level of real per capita income, using the log of per capita income, and using the lag- log of per capita income. 20
i Considering the vector of control variables, X t , at time t and province i, I use a
“parsimonious” or “small” model to test for robustness (Perotti, 1996; Panizza, 2002;
Partridge, 2005). With only a few time-variant factors, an implicit assumption is that this is a reduced-form model, in which income distribution and a few other control variables are the causal forces behind other factors. Besides reducing multicollinearity, this allows
20 Examples of international studies that use the level of income include Alesina and Perotti (1994), Alesina and Rodrik (1994), Persson and Tabellini (1994), Clarke (1995), Perotti (1996), Li and Zou (1998), Mo (2000), and Partridge (1997) the latter looking at states. Barro (2000) and Castello and Domenech (2002) are recent examples that use the log. Finally, Partridge (2005) uses the log and lag-log on income.
119 the income-distribution regression terms to capture direct effects as well as indirect
i effects through other variables that are not included. The parsimonious-model X t vector includes two human-capital controls: (i) the share of the population who are high-school graduates, but without a college bachelor’s degree; and (ii) the share of the population with a college bachelor’s degree. Also, it includes the share of the population who are over 64 years of age.
i The base-model X t vector includes the two education variables and, in addition, to account for industry mix and the relative demand for tradeables and nontradeables, it includes employment shares in agriculture, manufacturing, construction, business, transportation and housing. This is a mix variable that reflects employment growth in each province, predicted by the growth of the national level of the sectors of economic activity present in the province. Because national growth should be exogenous to the sector’s growth in a given province, it is routinely used as an instrument for local job growth and local demand shifts (Blanchard et al., 1992).
These coefficients are measured relative to the effect of the (omitted) service/trade sector. Also, considered is the variable Distance as in the long-run model, in order to control for differences in geography and transportation costs, which are the determinants of the “home-market effect” and agglomeration process. To account for slow-moving labor market adjustment, including lagged agglomeration effects, job growth from the immediately preceding year is included. 21 Finally, to account for
21 I would be ideal to combine the parsimonious and base models in a very long run way, taking their effect at the beginning of one, two or three decades. Unfortunately, the data for Argentina does not allow me to do that.
120 spillover effects, I include the lag of the Gini for the four closest provinces to province i in t-1.
Both the parsimonious and the base models include a time fixed effect, Year t , to
account for national factors, which may include common business cycles, fiscal and
monetary policies, productivity growth, and demographic trends in Argentina. Card and
DiNardo (2002) show that skill-biased technological affects both growth and inequality.
For example, in the Galor and Tsiddon (1997) model, greater inequality in matured
development stages can accelerate growth. Yet, assuming that technological change is
common to all provinces, any effects on inequality and growth are accounted for in the
time fixed effects and do not affect the Gini and Q3 regression terms.
Partridge (2005) clarifies that, with the time fixed effects, prudence should be
exercised in extrapolating the provincial findings to the national level. It is possible that
changes in income distribution (or other variables) may produce different responses when
considering country versus provincial data. For example, greater national inequality could
reduce aggregate growth (e.g., through social conflict), which would be reflected in the
time fixed effects. It is also possible that provinces with higher inequality may grow
faster than other provinces through factor migration (i.e., the Gini has a positive effect).
Finally, the variable REG i represents the region and provincial effects, which is
specified by including five major-region indicators for the Pampeana, Cuyo, Northeast,
Northwest, and Patagonia (Gran Buenos Aires is the omitted group). With region
indicators, the regression coefficients are interpreted as the response to within-region
changes in the variables relative to the Gran Buenos Aires. The pooled-OLS approach
with major-region dummies is then estimated (e.g., Persson and Tabellini, 1994). This
121 approach is most beneficial when most of the variation is cross-sectional, as for the income-distribution measures, or if there is measurement error (Griliches and Mairesse,
1995). When a variable mostly varies cross-sectionally, its pooled-OLS regression coefficient likely reflects long-run effects.
Table 5.6 reports various pooled-OLS results. Column 1 reports the 1991-2002
descriptive statistics, whereas models (1) to (5) contain results using the parsimonious X
vector (without job growth and inequality in the neighboring provinces), and models (6)
to (12) contain results using the base X vector (without job growth and inequality in
neighboring provinces), using different specifications about per capita income.
The 1991-2002 models show that the Gini and the Q3 coefficients are similar
whether the parsimonious or the base models are used. The coefficients for both income
distribution variables are negative and significant. The coefficient for the Gini coefficient
is larger in absolute value and significant at 1 percent, in all the models except models
(2), (11) and (12), for which the coefficient is significant at 5 percent.
Due to the debate on using per capita income in growth models, I compare both the parsimonious and the base models by omitting the income term in models (1) and (6), including the level of income in models (2) and (7), using the log of income in models (3) and (8), using the lag of income in models (4) and (9), in order to avoid direct endogeneity from cyclical and structural shocks, and, finally, considering the log-lag of the income term in models (5) and (10). Direct endogeneity from cyclical and structural shocks may affect the results. Thus, models (4) and (9) follow Banerjee and Duflo (2003)
and lag the per capita income one period, to avoid this direct cyclical endogeneity. This procedure is analogous to a simple IV estimator. The results in models (4) and (9) are
122 consistent with the expected negative bias of the initial income term, with the magnitude of the lagged income terms slightly declining. 22 In all the cases, the magnitude of the coefficients for the Gini and the Q3 slightly decline in both the parsimonious and the base models but maintaining their statistical significance. As a result, the income-distribution findings appear to be very robust, regardless of whether the income terms are included in levels, log, lagged or omitted altogether.
In model (11), I consider the full X vector of the base model, including the lagged job growth, which accounts for labor market adjustment, and including lagged agglomeration effects. The results are shown to be similar to those of the previous models.
Finally, in model (12), I include the lagged Gini from the closest four provinces to
each province i, to control for the spillover of inequality between provinces in a long-run framework. 23 Hence, equation (21) becomes:
i=4 Growthi=ββββ Gini i ++++++ Q3 i Y i X i REG i Year βε Gini ji + i (22) ttttt1 2 3 4 t∑i=1 5 tt− 1
ji where Gini t−1 accounts for the spillover effect caused by inequality from the
closest neighboring provinces to province i.
As Table 5.4 shows, in model (12) the spillover of inequality across provinces reduces the magnitude of the coefficient for the Gini and the Q3 in province i, but these coefficients preserve their significance at 5 percent and 10 percent, respectively.
All the year and regional dummies, which are omitted in table 5.4, are shown to be relevant to the specification because most of them are statistically significant.
22 Barro and Sala-i-Martin (1991) and Li and Zou (1998) noticed that if the lag is not too long (e.g., 5–10 years), the lag income term will still control for the level of economic development. 23 OLS models capture how persistent cross-sectional differences in inequality affect long-run growth rates, which Barro (2000) argues is more relevant to understanding growth disparities.
123 The overall pattern is not consistent with a long-run classical/incentive interpretation but rather with a political economy interpretation, where different distorting redistribution policies and social/political conflict are triggered by the differences in inequality among provinces. Moreover, there is sufficient evidence to distinguish the effect of the level of inequality in each particular province i, and the spillover effect of inequality in the neighboring provinces. Furthermore, even thought a separate long-run response related to the Q3 is presented, the pattern that emerges is not consistent with the hypothesis of the vibrancy of the middle class that is characteristic of the US (Partridge, 2005). I believe that the main difference in the pattern of influence of the share of the Q3 on provincial growth between the US and Argentina is that, in
Argentina, the middle class is not as productive as the same group is in the US, nor does its political behavior contribute to the adoption of economic policies conducive to growth. The main reason is the strong propensity of the middle class in Argentina to diverge resources from productive to other less productive activities, such as rent- seeking.
My interpretation of the results about the share of the third quintile (Q3) in
Argentina is as follows. Higher levels of inequality in a particular province and its neighbors increase the levels of corruption, crime, and social disturbances. These social pathologies accentuate the pressures for the adoption of distorting government redistribution to the extent to which resources are channeled toward rent-seeking activities, with their directly unproductive consequences. These redistribution processes threaten the relative well-being of the middle class, which then utilizes its own resources to “survive” and to seek “middle-class rents,” thereby redirecting resources away from
124 productive uses and reducing the rate of economic growth. This is a political economy interpretation consistent with the arguments of Persson and Tabellini (1994) and with the directly-unproductive profit-seeking literature (DUP), following Bhagwati and Srinivasan
(1982) and Mauro (2002).
A very perplexing result is the negative and statistically significant effect of
education on economic growth. The results for the two coefficients basically indicate
that, in the provinces of Argentina, the higher the share of the population with at least
high-school diplomas and the higher the share of the population with a college bachelor’s
degree, the lower the rate of economic growth. These puzzling results may reflect the
distortions and inefficiency caused by the rent-seeking activities. Specifically, the more
educated belong to the middle and upper class, and in many provinces the local
governments generate the largest share of the demand for labor. Moreover, during the
period under analysis, market-oriented policies and the privatization of almost all firms
owned by the state produced an increase in the rate of unemployment. According to the
INDEC, in May of 1991 the unemployment rate was 6.9 percent, this rate then reached
18.4 percent in 1995, and finally reached its peak of 21.5 percent in 2002. During this
difficult period, a major brain drain of highly qualified, trained and talented individuals
emigrated from the country. Many others who could not find jobs, were driving taxis or
running “ maxi kioskos ” or “ locutorios ” (candy shops or telephon kiosks). Even though
the process of brain drain was important during the dictatorship (Portes and Ross, 1976),
Argentina has the highest percentage of scientists emigrating from Latin America to the
United States, according to a study by the Economy Commission for Latin America and
the Caribbean (ECLAC, 2006).
125 Model 1 Model 2 Model 3 Model 4 Model 5 Model 6
Parsim. Parsim. Parsim. Parsim. Parsim. OLS Base OLS OLS OLS OLS OLS Growth Growth Growth Growth Growth Growth Log Lag No Income Mean (SD) No Income Income Log Income Lag Income Income No nginis Real Per-Capita 0.00116 Income Growth 0.0708 Gini Share 0.4596 -0.4486*** -0.4409** -0.4493*** -0.4316*** -0.4200*** -0.5002*** 0.0341 -0.1709 -0.173 -0.1728 -0.1312 -0.1324 -0.174 Q3 Share 13.82 -0.0078* -0.0079* -0.0078* -0.0066** -0.0070** -0.0075* 1.28 -0.0041 -0.0041 -0.0041 -0.0031 -0.0031 -0.0041 High School 30.3 -0.001 -0.001 -0.001 -0.0012** -0.0016*** -0.0008 10.3 -0.0006 -0.0006 -0.0006 -0.0005 -0.0005 -0.0006 College Degree 12.49 -0.0022* -0.0022* -0.0022* -0.0004 0.0004 -0.0024* 4.94 -0.0013 -0.0013 -0.0013 -0.0009 -0.001 -0.0013 Population > 60 0.11 0.0538 0.068 0.0526 -0.054 -0.0379 years 0.039 -0.1013 -0.1113 -0.1105 -0.0796 -0.0801 ypc 6662.72 6.06E-07 4647.13 1.95E-06 Log ypc 8.63 -0.0005 0.5435 -0.0185 Lag ypc -8.75e-06*** 1.36E-06 Log Lag ypc -0.0805*** -0.0134 Employment -0.0047 Growth Shocks 0.0569 Distance 7.12 -0.0390* 0.292 -0.021 Lag Gini 1 0.46 0.03421 Lag Gini 2 0.4596 0.0327 Lag Gini 3 0.4661 0.0309 Lag Gini 4 0.4459 0.0311 Year & Region Dummies - Y Y Y Y Y Y Mix Sectors - N N N N N Y Observations 276 274 274 274 252 252 275 Adjusted R- squared - 0.449 0.447 0.4468 0.6388 0.6315 0.4474 Standard errors in second row * significant at 10%; ** significant at 5%; *** significant at 1%
Continued
Table 5.4: Pooled-OLS Models
126
Table 5.4: Pooled-OLS Models (continued)
Model 7 Model 8 Model 9 Model 10 Model 11 Model 12
Base OLS Base OLS Base OLS Base OLS Base OLS Base OLS Growth Lag Growth Lag Growth Growth Log Growth Lag Growth Log Income Income Income Income Income Lag Income Gemp Gemp No nginis No nginis No nginis No nginis No nginis nginis Gini Share -0.4821*** -0.4901*** -0.4393*** -0.4556*** -0.5125** -0.4786** -0.174 -0.1747 -0.1337 -0.1345 -0.2258 -0.2258 Q3 Share -0.0072* -0.0074* -0.0076** -0.0077** -0.0081* -0.0077* -0.0041 -0.0041 -0.003 -0.0031 -0.0044 -0.0043 High School -0.0008 -0.0008 -0.0012** -0.0014*** -0.0013** -0.0011** -0.0006 -0.0006 -0.0005 -0.0005 -0.0005 -0.0005 College Degree -0.0027** -0.0026** -0.0006 -0.0001 -0.0003 0.001 -0.0013 -0.0013 -0.0009 -0.001 -0.001 -0.001 ypc 3.26E-06 2.21E-06 Log ypc 0.0135 -0.0191 Lag ypc -8.36e-06*** -5.76e-06* -7.60e-06 ** 1.59E-06 3.11E-06 3.24E-06 Log Lag ypc -0.0700*** -0.014 Employment -0.0265 -0.0314 Growth Shocks -0.0273 -0.0277 Distance -0.0617** -0.0462** -0.0034 -0.0256 -0.0076 0.0014 -0.026 -0.0234 -0.0196 -0.0176 -0.02 -0.0197 Lag Gini 1 -0.3379*** -0.107 Lag Gini 2 -0.0487 -0.1058 Lag Gini 3 -0.0297 -0.1307 Lag Gini 4 -0.3577*** -0.1152
Year & Region Dummies Y Y Y Y Y Y Mix Sectors Y Y Y Y Y Y Observations 275 275 253 253 128 128 Adjusted R- squared 0.45 0.4463 0.6359 0.6318 0.6333 0.6564 Standard errors in second row * significant at 10%; ** significant at 5%; *** significant at 1%
127 Next, I test my results under the assumptions of heteroskedasticity and clustering of the
error term.
First, there is the case of pure heteroskedasticity, in which ∑ε is a diagonal
matrix.
Where all non-diagonal elements are zeros and the diagonal elements are different and
2 2 2 equal to σ1 , σ 2 , …, σ N . This case violates the identically distributed assumption about
the errors and allows the variance of ε , conditional on X, to vary across observations. In this dissertation, using data from a household survey, I model the level of inequality as a function of household income. As a result, one might expect the error variance of high- income individuals to be much greater than for low-income individuals because high- income individuals have more discretional income.
Second, I separate the observations into several groups or clusters, within which
the errors are correlated. For instance, I estimate the growth in per capita GDP as a
function of the level of inequality, measured by the provincial Gini coefficient.
Consequently, the errors may be correlated over the households within a province or a
group of provinces with similar characteristics (economic region). The clustering
correlation of errors within a cluster of observations causes the ∑ε matrix to be block- diagonal, as the errors in different groups are independent of one another. This case drops the independently distributed assumption in a particular way. Since each cluster of observations may have its own error variance, the identically distributed assumption is relaxed as well.
128 ∑10 0 ... 0 :: : : : ε = 0 0m ... 0 (23) ∑ ∑ :: : : : 0 0 0 ... ∑ M
For this reason, the diagonal elements of ∑ε differ, all non-diagonal elements are zeros, and the diagonal elements are different and equal to ∑1, ∑ m , …, ∑ M . In this
notation, ∑ m represents a cluster covariance matrix. For cluster m with τ m observations,
∑ m will be ( τ m x τ m ). Zero covariance between observations in the M different clusters gives the covariance matrix ∑ε a block-diagonal form.
Model (13) considers the case where there is pure heteroskedasticity. As the errors are conditionally heteroskedastic and I want to apply a robust approach to estimate the variance-covariance matrix, I use the Huber-White-sandwich estimator of the variance of the linear regression estimation. After comparing models (12) and (13), the estimated Huber-White-sandwich estimator of the variance affects only the coefficients’ standard errors (and the interval estimates not shown in the tables) and it does not affect the coefficient point estimates βˆ . In model (13), the robust standard errors are smaller for all the statistically significant variables, compared to model (12). Such robust standard errors can deal with a collection of important concerns about failure to meet the assumptions about normality, heteroskedasticity, or some observations that exhibit large residuals, leverage or influence (Baum, 2006).
The second case, the clustering correlation of errors, is appropriate here, since I have observations for urban agglomerates and provinces. It is likely that the errors would
129 be correlated by clusters. To deal with this problem, I estimate the variance-covariance
matrix that is robust to the correlation of disturbances within groups and not identically
distributed disturbances. This method is known as the cluster-robust variance-covariance
estimator. (Baum, 2006). If the within-cluster correlations are meaningful, ignoring them
leads to inconsistent estimates of the variance-covariance matrix. In model (14) in Table
7, shows the cluster-robust variance-covariance estimator for clusters at the provincial
level. The cluster-robust variance-covariance estimator does not affect the coefficient
point estimates βˆ but only modifies the estimated standard errors of the estimated parameters. Model (14) shows that the overall pattern of income distribution and, in particular, the effects of the inequality of province i and the inequality of province i’s
neighbors are robust after considering the presence of heteroskedasticity and clustering
correlation of error among provinces.
In order to analyze whether or not the clustering correlation of errors would be a
valid assumption, I test for heteroskedasticity in the error distribution. In cross-sectional
databases representing individuals or households, the disturbance variances are often
related to some measure of scale. Two of the most common tests for heteroskedasticity
are the Breusch-Pagan test and the White’s general test. Based on model (12) in Table 6,
I run the Breusch-Pagan test and the White’s general test, failing to reject the null
hypothesis of homoskedasticity. It is important to keep in mind that these tests rest on
the specification of the disturbance variance given by:
2 α σ i⇒ z i (24)
130
2 α That is, the σ i is proportional to zi , where zi is some scale-related measure for the ith unit. Therefore, a failure to reject the tests’ null hypotheses of homoskedasticity does not indicate an absence of heteroskedasticity but simply implies that the heteroskedasticity is not likely to be of the specified form.
In turn, between-group heteroskedasticity is often associated with pooling data across what may be non-identically distributed sets of observations. This situation often happens when data from different provinces (groups) are pooled together as is the case in my database. Therefore, to test for groupwise heteroskedasticity, where I assume that intra-province (or more generally, intragroup) disturbance variance is constant but may differ among provinces, I conduct the test carried out by Levene (1960). First, I fit a linear trend model to real per capita GDP growth by regressing that variable on the year.
Then, the residuals are tested for equality of variables across provinces, using the
Levene's robust test for equality of variance. The Levene's robust test statistic (W_0) for the equality of variances between the groups defined by provinces and regions is rejected, while the residuals for the Patagonia region (.0904) show a standard deviation considerably larger than for other regions (.0474). Among the provinces, Catamarca has the highest standard deviation (.075), compared with an average of .034.
Finally, if different provinces have different error variances, in model (15) I calculate the
Feasible Generalized Least Squares (FGLS) estimator using analytical weights. In this analysis, I define the analytical weight (aw) series as a constant value for each observation in a province. That value was calculated as the estimated variance of that province’s OLS residuals. The FGLS model still confirms the overall negative
131 statistically significant effect of inequality on economic growth. However, it shows that its absolute magnitude decreases in both cases. Finally, the coefficient for the Q3 share, even though it is still negative and statistically significant, becomes slightly smaller in absolute value.
132 Model 13 Model 14 Model 15 Model 16 Model 17 Base OLS Base OLS Growth Lag Growth Lag Income Base FGLS Income Gemp ngin is Growth Lag Gemp nginis Robust Income Growth Robust Cluster-ID Gemp nginis Growth (IV) (GMM) Gini Share -0.4786** -0.4786*** -0.3439** -0.8981*** -0.8559*** 0.201 0.1696 0.1512 0.3317 0.3093 Q3 Share -0.0077** -0.0077* -0.0053* -0.0136** -0.0129*** 0.0036 0.0041 0.0029 0.0054 0.005 High School 0.000029 0.000029 -0.0002 -0.0007 -0.0008 0.00093 0.0011 -0.0006 0.0006 0.0006 College Degree -0.0013 0.0013 -0.0019 0.0012 0.0013 0.0016 0.0012 0.0012 0.0009 0.0009 Lag Ypc -9.68e-06 *** -9.68e-06 *** -7.40e-06*** -9.79e-06*** -8.65e-06*** 1.93E-06 2.60E-06 1.74E-06 1.78E-06 1.65E-06 Employment 0.0081* 0.0081* 0.0066** 0.0090** 0.0086** Growth Shocks 0.0042 0.0043 0.0032 0.0039 0.0036 Distance -0.0362 -0.0362 -0.0338 -0.0554* -0.0546* 0.0318 0.0303 0.0277 0.0329 0.0321 lag_gini_1 -0.4796** -0.4796** -0.3840*** -0.3410*** -0.3323*** 0.2031 0.1883 0.1434 0.108 0.1039 lag_gini_2 0.0298 0.0298 -0.1188 -0.0782 -0.0692 0.1842 0.1813 0.1564 0.0812 0.078 lag_gini_3 -0.0829 -0.0829 -0.0293 0.0184 -0.0362 0.2148 0.2148 0.1661 0.1316 0.1218 lag_gini_4 -0.3640* -0.3640* -0.3335** -0.3063*** -0.3237*** 0.2001 0.2001 0.1421 0.1185 0.1096 Year & Region Dummies Y Y Y Y Y Mix Y Y Y Y Y Sectors Observations 128 128 128 128 128
Adjusted R-squared 0.7403 0.7403 0.8406 0.7268 1.7268 AIC -438.75 -458.75 -587.1006 - - Robust standard errors in second row * significant at 10%; ** significant at 5%; *** significant at 1%
Table 5.5: Robust, Clusters, FGLS, IV, and GMM Models
133 5.4.5 Instrumental Variables Model
One important concern in this analysis is the existence of endogeneity in the determination of the Gini coefficient and per capita GDP.
To assess the relationship between the income distribution and growth in per capita income, I use Instrumental Variable (IV) regressions in order to address the endogeneity problem. The results of the OLS regressions may also be biased due to reverse causation and simultaneity bias. To assess the robustness of the results, I therefore use IV regressions and extract the exogenous component of income distribution.
Following the growth literature, I use the absolute value of the latitude of the capital city in each province in Argentina as well as the effect of the income distribution of the neighboring provinces (Levine, 2004).
5.4.5.1 Durbin-Wu-Hausman Tests for Endogeneity
Table 5.5, shows the estimation results of the specification using Instrumental
Variables (IV) in model (16), and the “Generalized Method of Moments” (GMM) specification in model (17). Both models use as instrumental variables the latitude of the capital city in each province in Argentina as well as the spillover effect of the income distribution of the neighboring provinces.
I used the Durbin-Wu-Hausman test to check for the exogeneity of the Gini coefficient:
Tests of endogeneity of: gggigiiinnnniiii H0: Regressor is exogenous Wu-Hausman F test: 444.4...66660000555500009999 F(1111,2222220000) P-value = 000.0...00003333222299997777 Durbin-Wu-Hausman chi-sq test: 555.5...11118888777722227777 Chi-sq(1111) P-value = 000.0...000022222277775555
134 The Hausman test statistics reject the null for exogeneity of the Gini variable.
Consequently, using the IV model is essential to solving the problem of endogeneity of
the Gini. I also test for the exogeneity of the Q3 share, the two educational variables, the
variable distance, the lag of income per capita, and employment shocks using the C test
under the null hypothesis that these regressors are exogenous and satisfy orthogonality
conditions. The C test fails to reject the null, indicating that these sets of regressors can
be considered exogenous and satisfy orthogonality conditions.
To test the appropriateness of the instruments, I use two tests. First, to test
whether the instrumental variables are valid, I used the Hansen test of the overidentifying
restrictions (OIR), which assesses whether the instrumental variables are associated with
the dependent variable beyond their ability to explain cross-province variation in income
distribution. Under the joint null hypothesis that the excluded instruments (i.e., the
instruments not included in the second-stage regression) are valid instruments, i.e.,
uncorrelated with the error term, and that the excluded instruments are correctly excluded
from the estimated equation, the Hansen test is distributed as a χ 2 in the number of overidentifying restrictions:
Hansen J statistic (overidentification test of all instruments): 777.7...777799994444 Chi-sq(6666) P-val = 000.0...2222555533336666
The test fails to reject the null, indicating that the instruments that I have chosen are appropriately uncorrelated with the disturbance process.
Second, I test for the joint significance of the instruments chosen. In this instance,
I run the Anderson Likelihood test under the null hypothesis that the IV are jointly insignificant:
135 Underidentification tests: Chi-sq(777) 7 P-value AAAnAnnnddddeeeerrrrssssoooonnnn cccacaaannnnoooonnnn.... cccocooorrrrrr.... llliliiikkkkeeeelllliiiihhhhoooooodddd rrraraaattttiiiioooo ssststttaaaatttt.... 55535333....77774444 000.0...0000000000 CCCrCrrraaaagggggg----DDDDoooonnnnaaaalllldddd NNN*N***mmmmiiiinnnnEEEEvvvvaaaallll ssststttaaaatttt.... 666666....777777 000.0...0000000000 Ho: matrix of reduced form coefficients has rank=K-1 (underidentified) Ha: matrix has rank>=K (identified)
Anderson canon. corr. LR statistic (identification/IV relevance test): 77707000....111133337777 Chi-sq(7777) P-val = 000.0...0000000000
I reject the null hypothesis. Hence, this test also confirms that the instruments that I have chosen to IV the Gini are appropriate.
Finally, another potential problem that might arise in this analysis is the existence of heteroskedasticity in cross-section data, where the scale of the dependent variable and the explanatory power of the model tend to vary across observations (Greene, 2005). 24
The existence of heteroskedasticity has an important consequence for the IV estimator, which is consistent but inefficient in the presence of heteroskedasticity. I run four tests under the null hypothesis of homoskedasticity:
IV heteroskedasticity test(s) using levels of IVs only Ho: Disturbance is homoskedastic Pagan-Hall general test statistic : 44424222....777700005555 Chi-sq(10) P-value = 000.0...0000000000 Pagan-Hall test w/assumed normality : 44484888....666688887777 Chi-sq(10) P-value = 000.0...0000000000 White/Koenker nR2 test statistic : 33303000....999977773333 Chi-sq(10) P-value = 000.0...000000006666 Breusch-Pagan/Godfrey/Cook-Weisberg : 88828222....5555888888 Chi-sq(10) P-value = 000.0...0000000000
I reject the null hypothesis in the entire test. One approach when facing heteroskedasticity of unknown form is to use the generalized method of moments (GMM), introduced by
Hansen (1982), which is my approach in model (17) in Table 5.5. 25
Comparing these two models, one more time they confirm the overall negative statistically significant effect of the own inequality and that of neighboring provinces on
24 In the presence of heteroskedasticity, the OLS estimator is still unbiased, consistent, and asymptotically normally distributed. However, the OLS estimator is inefficient and it needs some correction to the estimated asymptotic covariance matrix for appropriate inference. 25 In the presence of heteroskedasticity, the IV estimator is inefficient but consistent, whereas the standard estimated IV covariance matrix is inconsistent, preventing valid inference. The GMM estimator makes use of the orthogonality conditions to allow for efficient estimation in the presence of heteroskedasticity of unknown form. Consequently, if heteroskedasticity is present, the GMM estimator is more efficient than the simple IV estimator, whereas if heteroskedasticity is not present, the GMM estimator is no worse asymptotically than the IV estimator. Thus, the test to check for the presence of heteroskedasticity is key.
136 the provincial rate of growth of per capita GDP. The coefficient is statistically significant at 1 percent in both models. One key difference in the IV specification is that the coefficient for the own inequality is greater in absolute value (given by the coefficient of the Gini in province i) than the coefficient reflecting the influence of inequality in neighboring provinces (given by the coefficient of the Gini in the neighboring provinces).
Furthermore, the coefficient of the Q3 share is negative and statistically significant at 1 percent for the GMM model.
5.4.6 Spatial Pooled-OLS Models
Using the same weight matrix ( W) than in section 5.4.2, next I test for the presence of spatial autocorrelation in the pooled-OLS estimation.
Diagnostics
Test Statistic df p-value
Spatial error: Moran's I ---0-000....5555000000 111 111.1...333388883333 Lagrange multiplier 000.0...666633330000 111 000.0...444422227777 Robust Lagrange multiplier 000.0...0000002222 111 000.0...999966662222
Spatial lag: Lagrange multiplier 111.1...222233331111 111 000.0...222266667777 Robust Lagrange multiplier 000.0...666600004444 111 000.0...444433337777
Table 5.6: Spatial autocorrelation test statistic for the spatial lag and error models.
137
The Moran’s I statistic fails to reject the null hypothesis of spatial randomness.
Also, the Lagrange Multiplier statistic (LM) does not show evidence of the
appropriateness of using either the spatial error model or the spatial lag model, because
there is no longer need to control for spatial autocorrelation.
This is an important result, as it shows that I have developed a better model
specification which controls for the original spatial autocorrelation through the regressors
chosen, specially through the decoupled effect of inequality into the effects of inequality
from the own province i and the inequality from the neighboring provinces to province i.26 Thus, both sources of inequality affect the economic growth in province i.
5.4.7 Panel Data Models
In the fifth step of my analysis, I take advantage of the panel structure of the data.
Panel data are defined as repeated measures of one or more variables on one or more provinces. One of the limitations of my long-run models is that there are only 24 provinces in Argentina, which puts some constraints on the degrees of freedom available.
Taking advantage of the panel data framework, I consider 23 provinces (including the
Ciudad Autonoma de Buenos Aires ) and 10 years (1993-2002), which generates 230 observations. Another important advantage of working with panel data is that the repeated cross-sectional time series data are more informative, have more variability and less collinearity, the estimates are more efficient and, what is maybe the main advantage, panel data make it possible to control for individual unobserved heterogeneity.
26 As an experiment, I run both the spatial error and the spatial lag models, and I find that the coefficients for lambda and the WGowth are not statistically significant. Moreover, the Akaike Information Criterion (AIC) indicates that the full base model is the preferred model over the spatial models.
138 In my analysis, I consider the same equation (22):
i=4 Growthi=ββββµ Gini i ++++++ Q3 i Y i X i i Year β Gini ji + ε i (25) t1 tttt 2 3 4 t∑i=1 5 tt− 1 but with three important modifications. First, the term µ i controls for the individual unobserved heterogeneity. Hence, µ i varies over provinces. Every province has a fixed
value on this latent variable (fixed-effect). Therefore, ui represents the province-specific
unobserved heterogeneity. This term replaces the REG i term in equation (22).
Second, the error term is decomposed into two components: a province-specific
i i error u and an idiosyncratic error εt ,
i i (u + εt ) (26)
The combination of the two error terms in (26) is sometimes referred to as the composite-
error term (Baum, 2006).
Third, in model (26), the constant term is omitted because it would be collinear
with ui , the provincial specific error term that does not change over time. The rest of the control variables are the same as the specification in equation (22).
Table 5.7 shows the descriptive statistics for the variables included in the analysis. The period considered is between 1993 and 2002.
139 Variable Mean Std. Dev. Min Max Observations
ggdppc overall ---.-...00001111000000333300008888 ...0.0005555999955558888881111 ---.-...2222444466668888444400001111 ...2.22233334444000000007777 N = 230 between ...0.000111166661111777799997777 ---.-...0000444499994444000022228888 ...0.0002222333333888800002222 n = 23 within ...0.000555577774444333399991111 ---.-...2222000077774444666688881111 ...2.2221111117777888800006666 T = 10
gini overall ...4.444666644445555888899991111 ...0.000333333333311112222 ...3.333999911118888333322225555 ...5.555888899998888444455554444 N = 230 between ...0.0002222223333777733332222 ...4.4441111777755551111222222 ...5.555000088886666000088881111 n = 23 within ...0.000222255551111000011114444 ...4.444000099996666999933339999 ...5.555888833331111222244448888 T = 10
q3 overall 11131333....77771111555533337777 111.1...2222666600007777776666 11101000....66669999 11181888....1111 N = 230 between ...8.88822229999333333332222 11121222....333300008888 11151555....333333 n = 23 within ...9.999666633337777333399991111 11101000....99993333888833337777 11171777....44440000222233337777 T = 10
hs overall 33323222....3333666600002222 999.9...2222888888333399996666 999.9...00006666 44474777....33339999 N = 230 between 222.2...44443333888800009999 22262666....777733338888 33373777....111144442222 n = 23 within 888.8...999977775555777722225555 777.7...3333444411111199996666 44424222....66660000888811119999 T = 10
cd overall 11131333....4444999933337777 444.4...6666668888333311113333 333.3...88883333 222222....777777 N = 230 between 222.2...2222777766668888999999 11101000....222244445555 11171777....88885555 n = 23 within 444.4...1111000000333322221111 ...3.3335555333366669999555555 11191999....44442222777777 T = 10
mix_agri overall ---.-...0000000044441111113333 ...0.0000022227777111188883333 ---.-...0000111155554444 ...0.000111100002222 N = 230 between ...0.00000004444888800008888 ---.-...000000111155557777 ...0.000000055557777 n = 23 within ...0.00000222266667777771111 ---.-...0000111144442222444411113333 ...0.0000099992222111188887777 T = 10
mix_in~y overall ---.-...00000011110000555500009999 ...0.0000088881111888833335555 ---.-...0000444422228888 ...0.000222288882222 N = 230 between ...0.0000011114444000033335555 ---.-...00000044441111 ...0.00000222211114444 n = 23 within ...0.000008888000066667777 ---.-...0000333399997777555500009999 ...0.0002222555500000099991111 T = 10
mix_co~t overall ---.-...0000000022225555222222 ...0.0000022225555222244445555 ---.-...00001111333333 ...0.00000888888 N = 230 between ...0.000000044444466665555 ---.-...0000001111117777 ...0.000000088886666 n = 23 within ...0.0000022224444888866663333 ---.-...00001111222233338888222222 ...0.0000077778888222277778888 T = 10
mix_bu~s overall ---.-...00000000888800007777 ...0.00000666666000033336666 ---.-...0000333355553333 ...0.0002222333333 N = 230 between ...0.00000111111555522221111 ---.-...0000003333334444 ...0.000001111888888 n = 23 within ...0.0000066665555000066663333 ---.-...000033332222777766667777 ...0.00022220000666611113333 T = 10
mix_tr~t overall ---.-...000000005555666644448888 ...0.000004444111111117777 ---.-...0000222233337777 ...0.000111155556666 N = 230 between ...0.00000007777333366669999 ---.-...000000222233335555 ...0.000000099996666 n = 23 within ...0.0000044440000444477778888 ---.-...0000222211119999111144448888 ...0.000111144440000777755552222 T = 10
Note: Variables are ggdppc : growth of GDP per capita; gini : Gini coefficient; q3 : share of the third quintile; hs : the share of the population who are high-school graduates, but without a college bachelor’s degree; cd : the share of the population with a college bachelor’s degree; mix agri, mix industry, mix construct, mix business, and mix transport : represent the employment shares in agriculture, manufacturing, construction, business and transportation.
Continued
Table 5.7: Descriptive Statistics of the Panel Data Variables.
140
Table 5.7: Descriptive Statistics of the Panel Data Variables (continued).
Variable Mean Std. Dev. Min Max Observations
mix_ho~g overall ---.-...00000011110000888811113333 ...0.00000777744444400003333 ---.-...0000444411116666 ...0.000222277774444 N = 230 between ...0.0000011113333111100004444 ---.-...000000444411119999 ...0.00000111166667777 n = 23 within ...0.0000077773333222288886666 ---.-...0000333388884444999911113333 ...0.000222244446666444488887777 T = 10
mix_fi~e overall ---.-...0000000033334444448888 ...0.00000222288880000333333 ---.-...0000111166668888 ...0.000111111 N = 230 between ...0.00000005555111166668888 ---.-...000000111166661111 ...0.000000077774444 n = 23 within ...0.0000022227777555577772222 ---.-...00001111555555333344448888 ...0.00000999999111155552222 T = 10
mix_pu~c overall ---.-...000000004444222255552222 ...0.0000022227777444400002222 ---.-...0000111144449999 ...0.0000099998888 N = 230 between ...0.0000000444477775555 ---.-...000000111155554444 ...0.000000055556666 n = 23 within ...0.00000222277770000003333 ---.-...0000111133337777888855552222 ...0.00000888888111144448888 T = 10
mix_so~l overall ---.-...000000006666333377778888 ...0.0000044441111444422224444 ---.-...0000222233337777 ...0.000111155556666 N = 230 between ...0.0000000777744441111 ---.-...000000222244443333 ...0.000000088882222 n = 23 within ...0.0000044440000777788882222 ---.-...0000222211119999000077778888 ...0.0001111444411114444222222 T = 10
Ln_dis~m overall 777.7...11111177772222111111 ...2.222999922221111666699991111 666.6...888800008888111199997777 888.8...000033334444666677775555 N = 230 between ...2.222999988880000888855554444 666.6...888800008888111199997777 888.8...000033334444666677775555 n = 23 within 000 777.7...11111177772222111111 777.7...11111177772222111111 T = 10
Lnlag_~c overall 888.8...66665555777711118888 ...5.5554444888844444477779999 777.7...8888889999222299998888 11101000....11110000888899996666 N = 207 between ...5.555556666667777999999 777.7...9999555544448888333333 999.9...99998888111133339999 n = 23 within ...0.00055554444111199996666 888.8...444477772222000099994444 888.8...8888111166663333666666 T = 9
lag_gi~1 overall ...4.444555599997777666666 ...0.00033333322222244445555 ...3.333999911118888333322225555 ...5.555888899998888444455554444 N = 230 between ...0.000222233330000111177775555 ...4.444111144442222000099991111 ...5.555000055554444555577771111 n = 23 within ...0.000222244443333999900002222 ...4.44400000077772222229999 ...5.5558888555555000099995555 T = 10
lag_gi~2 overall ...4.444666600002222777766667777 ...0.0003333222233331111119999 ...3.3339999888844440000888888 ...5.555888899998888444455554444 N = 230 between ...0.000111188883333444477775555 ...4.444333300004444999955554444 ...5.555000055554444555577771111 n = 23 within ...0.00022226666888844445555 ...4.4440000111122223333337777 ...5.555888866660000222200002222 T = 10
lag_gi~3 overall ...4.4446666661111666622221111 ...0.000222299995555000022226666 ...3.333999911118888333322225555 ...5.555444400005555999944449999 N = 230 between ...0.000111177778888555522223333 ...4.444111144442222000099991111 ...4.444888888555533333333 n = 23 within ...0.000222233337777555533335555 ...4.4442222221111888899998888 ...5.5553333222288888877778888 T = 10
lag_gi~4 overall ...4.44444666677777733339999 ...0.000222299995555999955551111 ...3.3339999888844440000888888 ...5.555888899998888444455554444 N = 230 between ...0.000111177770000111155552222 ...4.444333300004444999955554444 ...4.4448888666699995555666666 n = 23 within ...0.00022224444444488885555 ...3.3338888777777333300008888 ...5.555777722225555111177773333 T = 10
Note: Variables are mix housing mix, finance, mix public, and mix social: represent the employment shares in Housing, finance, the public sector and for social programs; Ln_distkm: log of distance, Lnlag_ypc: log of lag of real per capita income; lag_gini_1, lag_gini_2, lag_gini_3, and lag_gini_4: represent the lag of the Gini coefficient of the first, second, third, and fourth closest provinces to province i.
141
In Table 5.7, the variable “Distance” (Ln_distkm) by construction does not vary
within provinces (it is time invariant). Therefore, it would not be included in the model.
All the other variables do vary within provinces.
5.4.7.1 Fixed Effect Model
As with the pooled-OLS model, here I estimate both a parsimonious model and a base model, with different specifications according to the interpretation of the per-capita income variable. Due to the debate on the incorporation of the per capita income in growth models, I compare the parsimonious and the base models by omitting the income term in models (1) and (6), including the level of income in models (2) and (7), using the log of income in models (3) and (8), using the lag of income in models (4) and (9) in order to avoid direct endogeneity from cyclical and structural shocks, and finally, considering the log-lag of the income term in models (5) and (10).
Table 5.7 reports several Fixed-Effect (FE) results. Models (1) to (5) contain results using the parsimonious X vector (without job growth and the inequality from the neighboring provinces), and models (6) to (12) contain results using the base X vector
(without the job growth and the inequality from the neighboring provinces), using different specifications for the per capita income.
The 1993-2002 models show that the Gini and the Q3 share coefficients are similar whether using the parsimonious or the base models. The Gini that controls for within inequality is negative for all the models but only significant at 10 percent in
142 models (4), (5), (9) and, (10). In turn, the coefficient of the Q3 share is negative, but it is
not significant in all the models.
All the models in Table 5.8 show a negative and statistically significant
coefficient for the lagged Gini coefficient. This implies that for each province, the level
of inequality in the previous year ( t-1) has a negative effect on the rate of economic growth in the current year. The coefficient for the other variables have the expected sign, except for per capita private credit, which shows a negative sign.
At this point, it is important to test whether the individual-specific heterogeneity given by the ui is necessary and if it is a good assumption. Therefore, I run an F test for
the null hypothesis that the constant terms are equal across units. The test gives the
following result:
F test that all u_i=0: F(222222, 111515554444) = 333.3...00009999 Prob > F = 000.0...0000000000
The result of the F test soundly rejects the null hypothesis; therefore, the individual-specific heterogeneity given by the ui should be included in the specification
of the model.
In model (11) of Table 5.9, the “job growth” variable is included to control for a slow-moving labor market adjustment, including lagged agglomeration effects, and in model (12) the Gini coefficient of the four closest provinces to province i controls for the
influence of inequality from the neighboring provinces on growth in province i. The
results with respect to the Gini and the Q3 share are similar to those from previous
models. The coefficients for the Gini from the neighboring provinces are, all negative,
but they are not statistically significant.
143 5.4.7.2 Random Effects Model
The Random Effects model specifies the individual effect as a random draw that is uncorrelated with the regressors and the overall disturbance term. In equation (26), the
i i i (u + εt ) is a composite error term and the u are the individual effects. A crucial
assumption of this model is that the ui are uncorrelated with the regressors. This orthogonality assumption is used by the RE estimator to construct a more efficient estimator. If the regressors are correlated with the ui , they are correlated with the composite error term and the RE estimator is inconsistent.
Model (13) in Table 5.9 estimates the RE model by the application of the Feasible
Generalized Least Squares (FGLS) estimator. Like the pooled OLS, the GLS- RE estimator is a matrix-weighted average of the within and between estimators. The optimal weights applied are based:
2 σ ε 2 λ=2 2 =(1 − θ ) (27) σε +T σ u where λ is the weight attached to the covariance matrix of the between estimator. To the extent to which λ differs from unity, pooled OLS will be inefficient, as it will attach too
much weight on the between-units variation, attributing it all to the variation in X rather
i than apportioning some of the variation to the differences in εt across units.
2 The setting of λ=1( θ = 0) is appropriate if σ u = 0 ; that is, if there are no RE, then a pooled OLS model is optimal. If θ = 1, λ = 0 and the FE estimator is appropriate.
To the extent to which λ differs from zero, the FE will be inefficient, in that it applies
zero weight to the between estimator. The GLS RE estimator applies the optimal λ in the
144 unit interval to the between estimator, whereas the FE estimator arbitrarily imposes λ = 0 .
i This imposition would be appropriate only if the variation in εt was trivial in comparison with the variation in ui .
In order to test for the appropriateness of the RE model, I should test the crucial assumption of the RE model that the ui are uncorrelated with the regressors. I use a
Hausman test for the null hypothesis that the extra orthogonality conditions imposed by
the RE estimator are valid. It is important to keep in mind that if the regressors are
correlated with the ui , the FE estimator is consistent but the RE estimator is not consistent. If the regressors are uncorrelated with the ui , the FE estimator is still consistent; however, it is inefficient, whereas the RE estimator is consistent and efficient.
Test: Ho: difference in coefficients not systematic
chi2(666) 6 = (b-B)'[(V_b-V_B)^(-1)](b-B) = 11131333....22226666 Prob>chi2 = 000.0...0000333399992222 (V_b-V_B is not positive definite)
After comparing the differences in the point estimates between the FE and the RE models, the Hausman test’s null hypothesis that the RE estimator is consistent is rejected at 5 percent. In my analysis, the provincial-level individual effects do appear to be correlated with the regressors. Therefore, the FE model seems to be the preferred specification. Model (13) in Table 5.9 shows that the coefficient of both the Gini and the
Q3 share are negative, but now the coefficient for the Gini is not significant and the coefficient for Q3 is significant at 10 percent.
In conclusion, in Argentina, using pooled-OLS, Feasible General Least Squares,
Instrumental Variables, General Methods of Moments, and Spatial Lag and Error
specifications, inequality and growth are negatively related. Moreover, there is sufficient
145 evidence to decouple the effect of inequality into the inequality from the own province i and the inequality from the neighboring provinces. Furthermore, the share of income going to the middle class, measured by the middle-quintile income share (Q3), is also negatively related to growth This result is not consistent with the effect of the vibrancy of the middle class that is a characteristic of the US. These overall patterns are consistent with a political economy interpretation, where distorting redistribution policies, rent- seeking and the associated DUP, and social/political conflict are triggered by differences in inequality across provinces.
146
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7
Fixed Effect Fixed Effect Fixed Effect Fixed Effect Fixed Effect (Pars (Pars (Pars Fixed Effect Fixed Effect (Pars (Pars Log Lag Log Lag (Base (Base No Income) Income) Income) Income) Income) No Income) Income) Gini -0.3014 -0.2866 -0.2839 -0.3167* -0.3175* -0.3029 -0.2833 -0.205 -0.2021 -0.2027 -0.1777 -0.1679 -0.1814 -0.1766 Q3 Share -0.0056 -0.0054 -0.0056 -0.0053 -0.0049 -0.0062 -0.0058 -4.60E-03 -4.40E-03 -4.20E-03 -4.70E-03 -4.80E-03 -4.00E-03 -3.80E-03 High School -0.0002 -0.0002 -0.0003 -0.0014* -0.0012* -0.0002 -0.0002 -0.0012 -0.0013 -0.0013 -0.0007 -0.0007 -0.0011 -0.0012 College Degree 0.0003 0.0001 -0.0004 0.0003 0.0014 0.0005 0.0003 -0.0024 -0.0024 -0.0025 -0.0017 -0.0017 -0.0025 -0.0025 Population > 60 0.0407 0.0627 0.0478 -0.0329 0.0044 years -0.0332 -0.0396 -0.043 -0.043 -0.0383 ypc 4.75E-06 5.04E-06 4.53E-06 5.13E-06 Log ypc 0.1211 -0.0849 Lag ypc -1.4e-05** 5.19E-06 Log Lag ypc -0.2609*** -0.0346 Year Dummies Y Y Y Y Y Y Y Mix Sectors N N N N N Y Y Observations 229 229 229 206 206 230 230 Number of id 23 23 23 23 23 23 23 Robust standard errors in second row * significant at 10%; ** significant at 5%; *** significant at 1% Y= Yes and N= No
Table 5.8: Panel Data Models of Real Per Capita Growth
147 Model 8 Model 9 Model 10 Model 11 Model 12 Model 13
Fixed Effect Random Fixed Effect (Base Effect Base Fixed Effect Fixed Effect Fixed Effect (Base Log Lag Log Lag (Base (Base (Base Log Lag Income Income Log Lag Log Lag Income Gemp Gemp Income) Income) Income) Gemp) nginis) nginis) Gini -0.2783 -0.3269* -0.3259* -0.3258* -0.3313* -0.2033 -0.1754 -0.1651 -0.16 -0.1602 -0.1644 -0.1337 Q3 Share -0.0059 -0.0064 -0.0058 -0.0058 -0.0055 -0.0056* -3.60E-03 -4.00E-03 -4.20E-03 -4.30E-03 -4.10E-03 -3.20E-03 High School -0.0004 -0.0015** -0.0014* -0.0014* -0.0014* -0.0012 -0.0012 -0.0006 -0.0008 -0.0008 -0.0008 -0.0009 College Degree -0.0002 0.0008 0.0017 0.0017 0.0015 0.0001 -0.0026 -0.0018 -0.0017 -0.0017 -0.0015 -0.001 Log ypc 0.123 -0.0893 Lag ypc -1.2e-05** 4.84E-06 Log Lag ypc -0.2486*** -0.2489*** -0.2567*** 0.0033 -0.0348 -0.0349 -0.0315 -0.0068 Employment 0.0199 0.0098 -0.1163 Growth Shocks -0.1386 -0.1394 -0.2005 Lag Gini 1 -0.0489 -0.0136 -0.0777 -0.1041 Lag Gini 2 -0.059 -0.0265 -0.0757 -0.1024 Lag Gini 3 0.0515 -0.1776 -0.2126 -0.115 Lag Gini 4 -0.1556 -0.0934 -0.1529 -0.1126
Year Dummies Y Y Y Y Y Y
Mix Sectors Y Y Y Y Y Y Observations 230 207 207 207 207 207 Number of id 23 23 23 23 23 23 Robust standard errors in second row * significant at 10%; ** significant at 5%; *** significant at 1% Y= Yes
Table 5.9: Fixed Effects & Random Effects Models of Real Per Capita Growth
148 CHAPTER 6
CONCLUSIONS
6.1 Summary
This dissertation analyzes whether or not inequality in the income distribution in the provinces of Argentina affects real per capita income growth in those provinces. One important objective, from a spatial perspective, is to consider both the effect of within inequality, namely the own province i level of inequality, as well as the spillover of inequality in the closest provinces into growth in province i. Another objective is to consider the influence of inequality on growth both in the long run and in the short run.
The main hypothesis is that, in the provinces of Argentina, inequality in the income distribution negatively affects economic growth. In addition, the effect of inequality on growth goes beyond the borders of each province. These hypotheses have been verified.
To accomplish this, the analysis builds on a framework and empirical strategy used by Partridge (2005), which starts by considering a “parsimonious” model, with a few key explanatory variables. A set of control variables is then added, in order to get a more fully specified “base” model. The parsimonious model not only reduces multicollinearity, but it also offers a test for robustness of the relationship between inequality and growth (Perotti, 1996; Panizza, 2002; Partridge, 2005).
149
6.2 The Main Results
Prior to the econometric estimations, the dissertation explores the record of growth and of differences in inequality in the income distribution in the provinces of
Argentina.
Looking at data since 1980, over the following 22 years the poorer provinces grew faster than the initially richer provinces. This would lend some support to the hypothesis of convergence of incomes across the provinces of Argentina during this period. However, if the pattern of per capita GDP dispersion across provinces during the
1980-2002 period is analyzed more closely, a distinction must be made. In effect, measured by the coefficient of variation of the natural log of real provincial per capita incomes, I find two distinct trends: on the one hand, a trend of declining per capita income dispersion until 1990, and, on the other hand, a trend of increasing dispersion after 1990. This interruption of the convergence trend coincided with rapid economic growth during the Menem period.
Income growth shows substantial spatial connections across the provinces.
Estimates for the Moran’s I statistic for the provincial per capita incomes of Argentina, for the 1980-2002 period, show strong evidence of positive spatial autocorrelation.
Indeed, the statistic is highly significant for all the years at p = 0.01. Thus, the value taken by per capita income at each province i tends to be similar to the values taken by per capita income at spatially contiguous locations.
Moreover, two main regional clusters persist throughout the period. The first cluster, of high-income provinces with high-income neighbors, is given by the provinces of Patagonia (Chubut, Neuquén, Santa Cruz, and Tierra del Fuego) plus La Pampa, each
150 of which appear in quadrant I in a Moran scatterplot. All other provinces, except for
Buenos Aires and Corrientes, constitute a second cluster of low-income provinces surrounded by low-income neighbors, each of which fall in quadrant III the vast majority of the years.
Over the period under analysis, a major increase in inequality took place as measured by the Gini coefficient for Argentina. The Gini increased from 0.45 in 1992 to reach a peak of 0.537 in 2003. Since then, the level of inequality has slightly declined.
There are important regional differences in inequality, as well. The Gini coefficient for
23 provinces ranged between 0.40 and 0.50 for the 1991-2002 period. After comparing the Gini coefficient for six regions in Argentina, I find that the regions that experienced some positive per capita GDP growth (Pampeana and Patagonia) showed relatively less inequality, while the regions that experienced negative per capita GDP growth
(Northwest, Northeast, and Cuyo) showed higher inequality levels. At the same time, the share re of total income that the “middle class” earned was getting smaller in all regions during this period, while the richest 10 percent of the population gained.
Spatial dependence is less pronounced for inequality than with respect to provincial growth. The estimates for the Moran’s I statistic for the provincial Gini coefficients of Argentina, for the 1980-2002 period, show that there is not a clear pattern of clustering. Only about half of the time, a pattern of positive spatial autocorrelation for the provincial Gini coefficient can be observed. For most of the 1990s, however,
Pampeana, Patagonia and Cuyo form one cluster with low Gini coefficients, while
Northwest, Capital City, and Northeast form another cluster of high Gini coefficients.
These results corroborate the findings using the Bonferroni and Tukey tests.
151
With the parsimonious and the base long run models, I find robust evidence of a long-run negative relationship between inequality (given by the Gini of the own province i) and growth. The best results are obtained for the 12-year model, the longest period that
the data allowed.
In general, it would be a mistake to consider each individual province as an independent unit of analysis. I find robust evidence of spatial autocorrelation in both the parsimonious and the base long-run models, particularly when the 12-year model is estimated. The Robust Lagrange Multiplier for both the spatial lag model and the spatial error model are statistically significant. This supports the rejection of the null hypothesis of spatial randomness in all the models. In turn, the statistic for the presence of spatial effects in the form of SARMA is not significant. Thus, spatial autocorrelation is best interpreted in terms of the spatial lag or spatial error models. These simple models help to specify the long-run relationship between inequality and per capita income growth in the presence of spatial autocorrelation. After controlling for spatial autocorrelation using both models, I find that the spatial error model performs better overall, as indicated by the Akaike info criterion (AIC).
My hypothesis is that, in Argentina, the spatial lag model captures the spatial spillovers or diffusion of per capita GDP growth among the provinces. The effects of inequality on per capita GDP growth are complex, however, and they go beyond the spillover of per capita GDP growth. For this reason, the spatial error model is even a better specification, which can capture this complex relationship, as it reflects spatial autocorrelation in measurement errors and/or in variables that have been “ignored” in the model.
152
Inequality in neighboring provinces affects not only inequality in province i but also its economic growth. I find robust evidence that, on average, the inequality of the neighboring provinces negatively affects economic growth in the provinces of Argentina in the 1991-2002 period.
Pooled-OLS models are used to further verify the robustness of the negative relationship between, on the one hand, the own province i inequality and the inequality of the neighboring provinces and, on the other, economic growth in the provinces of
Argentina in the 1991-2002 period.
The overall pattern of these results is not consistent with a long-run classical/incentive interpretation of the relationship between inequality and growth.
Rather, it lends itself to a political economy interpretation, in which distorting redistribution policies and social and political conflict are generated by the differences in inequality among provinces and where these factors, in turn, reduce the rate of economic growth. Moreover, even thought there exists a separate long-run response related to the share of the third quintile (Q3), the pattern identified for Argentina is not consistent with the vibrancy of the middle class, which is a characteristic of the US (Partridge, 2005). I believe that the main difference in the pattern of influence of the Q3 share between the
US and Argentina is that, in Argentina, the middle class is not as productive as in the US.
One important reason is the diversion of effort and resources to other less productive activities, such as rent-seeking. Higher levels of inequality in a province and its neighbors increase the levels of corruption, crime, and disturbances and in turn disrupt growth.
These events intensify pressures for distorting government redistributions, with the accompanying rent-seeking activities and their directly unproductive consequences. In
153 particular, these processes threaten the relative well-being of the middle class, which utilizes its own resources to “survive” and seek “middle-class rents,” thereby redirecting resources away from productive uses and reducing the rate of economic growth in the province. This is a political economy interpretation in light of Person and Tabellini
(1994) and the directly unproductive profit-seeking literature (Bhagwati and Srinivasan,
1980; Mauro, 2002).
The results also indicate that, in the provinces of Argentina, the higher the share of the population with at least high-school diplomas and with a college bachelor’s degree, the lower the rate of provincial economic growth. I believe that these puzzling results reflect the distortions produced by the rent-seeking activities, particularly among the educated, as well as the consequences of the labor market shocks experienced during this period.
Panel Fixed Effect models were used to examine whether or not the effect of inequality in each particular province i, and the spillover effect of inequality in the neighboring provinces, on economic growth in province i suffers from sign reversals. For the provinces of Argentina, I did not find sign reversals. The results of the fixed effect models seem to confirm that both types of inequality negatively affect economic growth in the provinces of Argentina.
6.3 Contributions
It is not surprising that the literature on inequality and growth is very rich and that the empirical evidence is mixed. This dissertation reaches conclusions similar to those of some of the existing literature on inequality and growth. Like this dissertation, some
154 studies do find a negative coefficient for initial inequality, when this dimension is included as an explanatory variable in empirical growth models. In regressing the average growth rate during 1960–1985 on the Gini coefficients for the income and land distributions around 1960, Alesina and Rodrik (1994) find statistically significant coefficients for both. Persson and Tabellini (1994) employ an alternative measure: they use the share of income accumulating to the middle fifth of the income distribution as a proxy for equality. The coefficient for this variable is statistically significant and positive, which is consistent with the Alesina and Rodrik results. Although Perotti (1996) uses a larger set of countries and tests for additional specifications, the results remain significant.
More recently, this potential consensus has been challenged by Forbes (2000), based on the analysis of a new data set known as the Deininger and Squire “high-quality” data set. Forbes found a positive relationship between lagged inequality and growth.
Moreover, using pooled OLS and two-stage least squares models (2SLS), Partridge
(2005) finds that inequality and growth are positively related, and he also finds that a more-vibrant middle class, measured by the middle-quintile income share (Q3), is positively related to growth. Yet, with fixed-effect (FE) models, Panizza (2002) finds that small specification changes can reverse the results. While this dissertation uses
Partridge’s analytical framework to examine the relationship between inequality and growth, for the case of Argentina it finds the opposite results.
Other studies have also found significant negative coefficients for measures of wealth and asset inequality in growth regressions. For example, Birdsall and
155
Londoño (1997) use a subset of the Deninger-Squire data set and conclude that “initial inequalities in the distribution of land and human capital have a clear negative effect on economic growth, and the effects are almost twice as great for the poor as for the population as a whole.” (p. 35)
López (2003) uses both an econometric approach similar to Forbes and the
Deininger-Squire data. Taking into account the simultaneous nature of the determination of growth and inequality dynamics, he finds no sign of an effect of growth on inequality, in keeping with work by Ravallion and Chen (1997) and Dollar and Kraay (2002).
Rather, Lopez finds statistically significant evidence that initial inequality reduces growth.
In addition, the harmful effects of high inequality on economic development are not restricted to growth. A broader view of development includes more than output per capita. Rodrik (1999) suggests that countries suffering from more pervasive social divisions, whether ethnic and racial in nature or income and class-based, seem not to adjust to large aggregate shocks as well as other, more egalitarian and cohesive societies.
The implicit mechanism is that the institutions responsible for sharing the burdens of adjustment work less well in economies in which distributions are more unequal.
This dissertation contributes to the inequality and growth literature in Latin
America and, particularly, in Argentina. In this area, this dissertation is close to similar studies by de Ferranti et al. (2004), Perry et al. (2006) and Gasparini, Gutierrez and
Tornarolli (2007) for Latin America; Williamson (1999) for Argentina and Uruguay;
Calvo, Torre, and Szwarcberg (2002) for Argentina; and Prados de la Escosura (2005) for
Argentina, Brazil, Chile, Mexico, and Uruguay.
156
The uniqueness of this dissertation is that it combines the inequality and growth literature with the regional economic literature, by integrating the role of spatial
inequality in the growth relationship. This is accomplished by considering both the effect
of the own province i level of inequality and the effect of the spillover of inequality from the closest provinces.
6.4 Policy Implications
Even though this dissertation is only concerned with the relationship between inequality and growth (and not with the further effects on poverty), I believe that it has important policy implications for Argentina. First, it shows that programs to reduce inequality may have long-run and short-run effects on economic growth. Second, reducing inequality in a given province may not only induce economic growth in that province but it may also accelerate economic growth in neighboring provinces. Reducing inequality will likely induce pro-poor growth.
6.5 Limitations and Future Research
This section identifies some limitations of this dissertation and, at the same time, indicates interesting directions for future research and extensions.
The focus has been on the effects of inequality in the own province i and of the neighboring levels of inequality on the provincial economic growth. Unfortunately, this analysis does not provide information on the linkages that produce the significant negative relationship between inequality and growth or between the Q3 share and growth.
Likewise, this dissertation does not provide evidence on the final effect on regional
157 poverty. For the policymaker, it is important to know not only the sign of the relationship between inequality and growth but also the determinants of this relationship, in order to implement policies that help reduce inequality and poverty.
A second important limitation is given by the available data. In Argentina, the
INDEC gradually started to produce household surveys in 1974, but at that time it included only three conglomerates, corresponding to the province of Buenos Aires
(Ciudad de Buenos Aires, Gran Buenos Aires, and Partidos del Gran Buenos Aires ).
Different sets of surveys make the comparison of the results difficult. First, 22 conglomerates were gradually added to the list between 1982 and 1990, while two more conglomerates were added between 1991 and 1994. Finally, four more conglomerates were added to the list between 1995 and 2003. Although these changes are welcome, they pose significant comparison problems. Since regions differ in their economic and social situations, adding a new region usually significantly affects the national statistics. In a recent paper, Gasparini, Gutierrez and Tornatolli (2007) find that these types of problems are characteristic of the whole region, but they have been particularly acute in Argentina.
Moreover, the content and definitions have been revised several times. 27
Another important limitation of the surveys conducted in Argentina is that they
only cover the urban population, which nonetheless represents over 85 percent of the total
population. 28 Some studies suggest, however, that including the rural areas would not
27 The sample includes data for Argentina, Bolivia, Brazil, Chile, Colombia, Costa Rica, Dominican Republic, Ecuador, El Salvador, Honduras, Jamaica, Mexico, Nicaragua, Panama, Paraguay, Peru, Uruguay and Venezuela. The sample covers all countries in mainland Latin America (except for Guatemala), and two of the largest countries in the Caribbean (Dominican Republic and Jamaica). In each period, the sample represents around 92 percent of LAC total population (98 percent in Latin America and 29 percent in the Caribbean). 28 Gasparini, Gutierrez and Tornatolli (2007) also found that in Latin American just the household surveys of Argentina and Uruguay cover only the urban population.
158 substantially affect poverty estimates in Argentina (Haimovich and Winkler, 2005;
Winkler, 2005).
An important and imminent limitation, which could affect future research in
Argentina, is when the whole INDEC loses its credibility. This situation has been happening in Argentina since 2007, when the government has been manipulating the official estimates regarding the rate of inflation (which directly affects the household surveys). This is a very serious problem, which has been criticized by economists and international organizations. The potential consequences of massaging the estimates are enormous and go beyond the scope of this dissertation.
Barro (2000) explains that four broad theories can be constructed to assess the macroeconomic relations between inequality and economic growth. These theories are credit-market imperfections, political economy, social unrest, and savings rates. I have explained my conclusions following the political economy and social unrest theories. An interesting agenda for future research would be to test hypotheses from the sociopolitical unrest theory as applied to Argentina. This argument would go like this: inequality of wealth and income motivates, on the one hand, the poor to engage in crime, riots, and other disruptive activities and, on the other hand, the middle class try to “survive” by seeking “middle-class rents.” 29 Both groups redirect economic resources, time and energy from more productive uses, which affects economic growth.
Some reflection can help understand how the results of the dissertation would help in explaining related issues about the reality of Argentina.
29 See, for example, Hibbs (1973), Venieris and Gupta (1986), Gupta (1990), Alesina and Perotti (1996), and Benhabib and Rustichini (1996).
159
Continuous struggles for power, in order to obtain rents that would benefit small groups, have characterized the political and economic history of Argentina. This dynamics have been repeated many times in the history of the country. Some of those fights to gain economic power were between the provinces from the Northeast,
Northwest and Pampeana regions and Buenos Aires, in the eighteenth and nineteenth centuries. 30 The struggles have involved the Argentine oligarchy (especially large landowners) and the middle class, Radicals and Peronists, unions and enterprises, the
Military and the Communists, Liberals and Conservatives during the nineteenth and twentieth centuries.
History helps us discover that this characteristic of continuous fights for power to obtain rents was the way in which the Spanish Empire settled the region. The main center of the Viceroyalty of Peru was the city of Lima, where silver mining became its main economic activity, and Indian forced labor was its primary workforce. Peruvian gold provided revenue for the Spanish Crown and fueled a complex trade network that extended as far as Europe and the Philippines. Potosí was founded in 1546 as a mining town; it soon produced fabulous wealth, becoming one of the largest cities in the
Americas and the world, with a population exceeding 200,000 people. According to official records, 45,000 tons of pure silver were mined from Cerro Rico between 1556 and 1783. After 1800, the silver mines became depleted, and this eventually led to a slow economic decline.
From 1542 to 1776, Argentina was part of the Viceroyalty of Peru and this shaped the political economy of the region. Two situations caused that, at that time, the provinces from the Northeast, Northwest and Pampeana regions were more important
30 These set of provinces are called from the “interior” of the country in opposition to Buenos Aires.
160 than Buenos Aires. First, the importance of Lima and Potosí and its large flow of commerce created a large demand for food, labor, and many other resources. Waves of conquistadors came to Argentina from Peru, Chile and Paraguay. From Peru came the founders of Tucumán in 1565, Córdoba in 1573, Salta in 1582, Jujuy, and La Rioja in
1591. From Chile came the founders of Santiago del Estero in 1553, San Juan and
Mendoza in 1562, and San Luis in 1596. From Paraguay came the founders of Corrientes and Paraná in 1558, Santa Fe in 1573 and, Buenos Aires in 1580.
Second, from 1561 to 1739, the trade between Spain and its colonies was implemented through the system called “ flotas y galeones ”, which gave exclusivity to trade between the ports of Cádiz and San Lúcar de Barrameda (in Spain) and Cartagena,
Portobelo and San Juan de Ulúa in America. Any other port was excluded from trade.
This policy was changed in 1776, when the Viceroyalty of Río de la Plata was founded, and in 1778 the Spanish Crown adopted the “Free Trade Laws”. This policy change gave the ports of Buenos Aires and Montevideo strategic importance. 31 This was the
beginning of a fiery dispute between Buenos Aires and the rest of the provinces, in order
to gain the benefits of trade and the incomes from taxes from trade. Despite the
liberalization of trade, however, the obligation that Spaniards should undertake all trade
was kept. This meant that the owners of all the ships, their captains and officers, and 2/3
of the crew should have been Spaniards (Ferrer, 2005). This is an example of the
31 The geographic importance of the “Río de la Plata” was given by the fact that it was the best access to the heart of the Spaniard colonial power in the south of Peru. From Buenos Aires to Potosí there were 1,750 Km of plain road and it took two months to get there. However, from Lima to Potosí, there were 2,500 Km of mountain roads and it took four months to get there. These differences between Lima and Buenos Aires, and other provinces to trade with Potosí increased the importance of the creation of the Viceroyalty of Río de la Plata ( Ferrer, 2005 ).
161 concentration of wealth among selected groups, which has continued throughout the years in Argentina’s history.
This tension between the provinces and Buenos Aires can be observed today by the persistence of the “caudillo system,” which shapes the politics of many provinces, where their leaders come from a few traditional families. It also shapes the way in which each province obtains the resources from the collection of federal taxes. Even today, an important source of revenues for the federal government is to levy taxes on exports. Most recently, additional taxes on exports have been used to address the food price crisis.
Farmers have responded with strikes and road blockades.
Moreover, today the disputes about centralized or decentralized policies are part of the regular debate regarding education, health care, and retirement plans. These struggles are sources of potential inequality, because they provide chances for opportunistic rent-seeking behavior and give politicians opportunities to patronize some regions. An interesting future research agenda would be to study the connection between centralized policies regarding education, health care, and poverty and their consequences on inequality and growth.
History also highlights the way in which the state divided and sold the country’s land. From the beginning, the distribution of land was characterized by large amounts of land owned by few people. Properties were sold to them right after the land was expropriated from indigenous peoples. For example, in 1826, the law of “ Enfiteusis ” distributed over eight million hectares of land in Buenos Aires among 293 people. In
1857, another three million hectares of land were distributed among 300 people (Ferrer,
2005).
162
The distribution of land ended in 1884, by the so-called “ Conquista del desierto”
(Conquest of the Desert). This was a military campaign directed mainly by General Roca in the 1870s, which established Argentine dominance over Patagonia, until then inhabited by indigenous peoples. 32 In 1882, millions of hectares were sold in London and Paris by
the Argentinean embassies. The maximum piece of land sold was 40,000 hectares per
person. In 1885, the government gave 8,000 hectares to every military member and 100
hectares to every soldier who participated in the Conquest of the Desert. By 1884, almost
all the land from the Pampeana region had an owner (Rapoport, 2000).
I believe that the distribution of land has been one of the most influential sources
of inequality in Argentina, which has had important consequences on the social, political,
and economic development of the country. Specifically, the concentration of land has
characterized agricultural production. Indeed, by 1937, about 95 percent of the population
working in agriculture did not own the land they cultivated. At the same time, one
percent of the population working in agriculture owned 70 percent of the total area, with
at least 3,000 hectares per farm.
Another important consequence of this bimodal concentration in the distribution
of land was the impossibility to integrate the large waves of immigrants that came from
Europe into farming. As a result, 75 percent of all immigrants went to the big cities,
making Argentina a very urbanized country despite the large size of its territory and low
density of population. 33 This situation affected salaries in both the urban and the rural
32 Some author like Andermann (http://www.bbk.ac.uk/ibamuseum/texts/Andermann02.htm), Rock (2002) and Fernández (2005) claim that the “Conquest of the Desert” was a genocide by the Argentine government against the indigenous tribes. 33 Between 1830 and 1950, 65 million immigrants came from Europe to America. Of these, 61 percent went to the US and 10 percent to Argentina. Most of those who came to Argentina were from Spain and Italy (Cornblit, Gallo and O’Connell, 1962).
163 areas. The increase in the supply of labor in the cities reduced the wage rate there and increased the rate of unemployment. Ferrer (2005) claims that, even in good times, with exports booming like in 1913, the rate of unemployment was higher than 5 percent and, in difficult times, like in World War I in 1914 or during periods of economic contraction, the rate of unemployment was around 20 percent. The concentration of land ownership also deteriorated the salaries in the rural areas, because the lack of access to land reduced the reservation wage of rural workers. This is an important difference of Argentina compared to the US, where the distribution of land was more oriented to the immigrant willing to work on it. Among other things, in the US, the “Homestead Act”, approved in
1862 by Lincoln, produced a different result.
Today, I believe that opportunistic rent-seeking behavior arises when specific groups, like judges, do not pay income taxes, politicians obtain privileges in wages and retirement, and civil servants and union members become very wealthy after a few years of service. These people represent the bulk of the Q3 quintile. Their behavior, rather than contributing to economic growth, explains the inverse relationship between the Q3 share and growth.
Another interesting research agenda for the future would be to study the connection between today’s distribution of land and wealth and income inequality by province. A similarly attractive research initiative would be to study the differences in the policies of distribution of land between Argentina and the US and their consequences on inequality. In this dissertation, I have focused on inequality in the income distribution, but there are other ways in which inequality may affect economic growth; for example,
164 inequality of opportunities in society and inequality created by the rule of law or the democratic system may have major implications on growth.
Finally, history could help in discovering a third characteristic of the continuous fights for power in order to obtain rents. These struggles are reflected in the recurrent fiscal deficits, the weaknesses of the monetary system, and the country’s dependence on external funds. From the early nineteenth century, all the provinces had serious problems in obtaining fiscal resources to balance their expenses. For example, in 1839, the budget of Jujuy, one of the poorest provinces, was 9,040 pesos annually. Of these, 2,860 pesos belonged to the provincial government, including 1,500 pesos for the salary of the governor, while the budget for education was 480 pesos (Ferrer, 2005). In Buenos Aires, customs duties provided between 80 and 90 percent of all of the country’s fiscal resources. However, constant fiscal deficits were the norm for the provinces and the nation, and domestic money creation and external credit were the solution to balance the fiscal accounts. Since 1822, money issue was very frequent, when trade deficits (due to fluctuations in commodity prices, internal conflicts, and external shocks) were accompanied by liquidity shortages. This led to inflation and depreciation of the peso, which made it difficult for the government to purchase gold in order to pay the interest and amortization of the external debt (Ferrer, 2005). This was another source of growing inequality in the distribution of income. 34 Inflation hurt the poor. In contrast, landlords
with political power had large debts in pesos with the national banks. The depreciation of
the peso made them better off, because it reduced the real value of their debt obligations.
Thus, this group lobbied for a more expansive monetary policy, and they were opposed to
34 Ferrer (2005) claims that t.he depreciation of the national currency increased the prices of export in pesos and because they were agricultural and meat products which were also consumed in the internal market. Therefore, the depreciation of the peso also increased the internal prices.
165 any policy to balance the budget. A regressive tax system, in which between 70 and 80 percent of total collection came from indirect taxes levied on consumers, contributed to inequality.
At the same, since the second half of the nineteenth century, Argentina received an important amount of foreign investment (most of it coming from the United Kingdom and from the US after World War I). Foreign investment helped the country develop its infrastructure in railroads, ports, public transportation, gas, electricity, water, financial services and cold storage chambers for meat. 35 This promoted growth, but eventually
repayment became difficult during the Great Depression.
The persistent fiscal deficits, the weaknesses of the monetary system, and
dependence on external funds produced the first external debt crises in 1890, in which
Argentina defaulted in the payment of its obligations with the British financial institution
called the “Baring”. The combination of fiscal deficit, a weak monetary system, and
dependence on external funds was also characteristic of the debt crisis in 2001.
Another future research agenda would be to study the relationship between the
financial system and inequality in Argentina. One interesting possibility would be to test
the credit-market imperfections channel for the influence of inequality on growth. Since
the early 1980s, Argentina had a very low rate of bancarization and lack of confidence of
its population in the banking system. The level of bancarization was 47 percent during
the 1990s, while in many developed countries the figure is higher (over 90 percent for
35 Ferrer (2005) shows that, in 1913, foreign capital in Argentina represented 50 percent of the total fixed capital at that time. Even in 1929, foreign capital represented 32 percent of the total.
166
Spain and over 80 percent in the US). 36 The level of bancarization of Argentina is low even when compared to other Latin American countries. For example, Chile’s level of bancarization is seven-fold and of Brazil is three-fold the levels of Argentina. During the
1990s, the ratio of private credit to GDP was 25 percent, which is very low compared to other countries. It was over 50 percent in Bolivia during the same period. Argentina had many years of financial chaos and irresponsibility, characterized by inflation, hyperinflation, and bank deposit confiscation, during the 1980s and 1990s. I believe that this exercise could be undertaken using the Levine (2004) empirical framework to analyze the connections between the operation of the financial system and economic growth.
I would like to conclude by remembering that, even though Argentina suffered its worse economic collapse in 2001 and its history shows an unequal distribution of land and of opportunities that affected its income distribution, in 2007 it showed the second lowest level of poverty, after Chile, and the lowest level of inequality (by the Gini coefficient) among the seven largest economies in Latin America. By its fascinating economic history, Argentina is still a puzzling country, which makes it very attractive for more scholarly future research in several different areas.
36 According to the Asociación de Bancos de la Argentina (ABA), the population in Argentina that is considered “economically active” represents 16 million people between 18 and 65 years old. From that group, 10 million have problems in getting access to the banking system.
167
BIBLIOGRAPHY
Acemoglu, D., S. Johnson, and J. Robinson (2002). Reversal of Fortune: Geography and Institutions in the Making of the Modern World Income Distribution. The Quarterly Journal of Economics, 117 (4), 1231-1294.
Adams, D. (1983). Mobilizing Household Savings through Rural Financial Markets. In D.W Adams, J.D. Von Pischke, and D. Gordon (ed.), Rural Financial Markets in Developing Countries. Their Use and Abuse . Baltimore: The Johns Hopkins University Press.
Aghion, P., E. Caroli, and C. García-Peñalosa. (1999). Inequality and Economic Growth: The Perspective of the New Growth Theory. Journal of Economic Literature, 37 , 1615–1660.
Albert, P., and M. McShane (1995). A Generalized Estimating Equations Approach for Spatially Correlated Binary Data: Applications to the Analysis of Neuroimaging Data. Biometrics, 51 , 627-638.
Alesina, A., and R. Perotti. (1994). The Political Economy of Economic Growth: A Critical Survey of the Literature. World Bank Economic Review, 3 , 351–371.
Alesina, A., and R. Perotti. (1996). Income Distribution, Political Instability, and Investment. European Economic Review, 40 , 1203-1228.
Amos, O. M. J. (1988). Unbalanced Regional Growth and Regional Income Inequality in the Latter Stages of Development. Regional Science Urban Economics, 18 , 554- 566.
Alesina, A., and D. Rodrik. (1994). Distributive Politics and Economic Growth. The Quarterly Journal of Economics, 109 (2), 465-490.
Amos, O. M. J. (1989). An Inquiry Into the Causes of Increasing Regional Income Inequality in the United States. Review of Regional Studies, 19 , 1-12.
Ancot, J. P., J.Paelinck, and J. Prins. (1986). Some New Estimations in Spatial Econometrics. Economics Letters, 21 , 245-249.
168 Anselin, L. (1980). Estimation Methods for Spatial Autoregressive Structures (Vol. Series 8). Ithaca, New York: Cornell University.
Anselin, L. (1982). A Note on Small Sample Properties of Estimaitors in a First-Order Spatial Autorregressive Model. Enviroment and Planning 14 , 1023-1030.
Anselin, L. (1988). Spatial Econometrics: Methods and Models . Dordrecht:Kluwer
Anselin, L. (1992). SpaceStat Tutorial: A Workbook for Using SpaceStat in the Analysis of Spatial Data. National Center for Geographical Information and Analysis . Santa Barbara, CA.: University of California.
Anselin, L. (1995). Local Indicators of Spatial Association - LISA. Geographical Analysis 27 , 93–115.
Anselin, L. (2001). Spatial Econmetrics. In B. Baltagi (ed.), A Companion to Theoretical Econometrics (pp. 310-330). Oxford, UK: Basil Blackwell.
Anselin, L. (2002). Under the Hood: Issues in the Specification and Interpretation of Spatial Regression Models. Agricultural Economics, 17 , 247-267.
Anselin, L. (2003). Spatial Externalities, Spatial Multipliers, and Spatial Econometrics. International Regional Science Review, 26 (2), 153-166.
Anselin, L., and A. Bera. (1998). Spatial Dependence in Linear Models with an Introduction to Spatial Econometrics (Vol. 155). New York: Marcel Dekker.
Anselin, L., and R. Florax. (1995). New Directions in Spatial Econometrics . Berlin: Springer-Verlag.
Anselin, L., and S. Rey (1991). Properties of Tests for Spatial Dependence in Linear Regression Models. Geographical Analysis, 23 , 112-131.
Anselin, L., and S. Rey (1997). Spatial Econometrics. Special Issue, International Regional Science Review, 20 .
Anselin, L., R. Florax and S. Rey (2004). Advances in Spatial Econometrics: Methodology, Tools and Applications . Berlin: Springer-Verlag.
Aroca, P., and M. Bosch. (2000). Crecimiento, Convergencia y Espacio en las Regiones Chilenas. Estudios de Economía, 27 (2), 199-224.
169
Aroca, P., M. Bosch, and W. Maloney. (2005). Spatial Dimensions of Trade Liberalization and Economic Divergence: Mexico 1985–2002 . Washington, DC.: World Bank.
Atkinson, A. (1970). On the Measurement of Inequality. Journal of Economic Theory, 2 , 244-263.
Atkinson, A. (2003). Top Incomes in the United Kingdom over the Twentieth Century . U.K.: University of Oxford.
Azariadis, C. (1996a). The Economics of Poverty Traps Part One: Complete Markets Journal of Economic Growth, 1 , 449-486
Azariadis, C. (1996b). The Economics of Poverty Traps. Part Two: Incomplete Markets. Mimeo, Los Angeles : UCLA .
Azzoni, C. R. ( 2001). Economic Growth and Regional Income Inequality in Brazil. Annals of Regional Science 35 (1), 1.
Banerjee, A., and E. Duflo (2003). Inequality and Growth: What Can the Data Say? Journal of Economic Growth 8 (3), 267-299
Banerjee, A., and A. Newman. (1993). Occupational Choice and the Process of Development. Journal of Political Economy 101 (2), 274–298.
Barro, R. J. (2000). Inequality and Growth in a Panel of Countries. Journal of Economic Growth 5 (1), 5-32.
Barro, R., and X. Sala-i-Martin. (1991a). Economic Growth in a Cross Section of Countries. Quarterly Journal of Economics, 106 , 407-443.
Barro, R., and X. Sala-i-Martin. (1991b). Convergence Across States and Regions Brookings Papers on Economic Activity, 1 , 107-182.
Barro, R., and X. Sala-i-Martin. (1992). Convergence. Journal of Political Economy, 100 , 223-251.
Barro, R., and X. Sala-i-Martin. (2004). Economic Growth . Cambridge, MA: MIT Press.
170
Baum, C. (2006). An Introduction to Modern Econometrics Using Stata . Texas: Stata Press.
Baumol, W. J. (1986 ). Productivity Growth, Convergence and Welfare: What the Long Run Data Show? . American Economic Review 76 , 1072–1085.
Beck, T., A. Demirgüç-Kunt, and R. Levine. (2004). Finance, Inequality and Poverty: Cross Country Evidence. World Bank Policy Research Working Paper (3338).
Bell, L., and R. Freeman. (2001). The Incentive for Working Hard: Explaining Hours Worked Differences in the US and Germany. Labour Economics, 8 , 181–202.
Benabou, R. (1996). Inequality and Growth. National Bureau of Economic Research , Working Paper 5658.
Benhabib, J., and A. Rustichini. (1996). Social Conflict and Growth. Journal of Economic Growth, 1 , 129-146.
Bernard, A., and C. Jones. (1996). Comparing Apples to Oranges: Productivity Convergence and Measurement Across Industries and Countries The American Economic Review, 86 (5), 1216-1238.
Bertola, G. (1993). Factor Shares and Savings in Endogenous Growth. American Economic Review, 83 , 1184-1198.
Bértola, L. (2005). A 50 años de la curva de Kuznets, una reivindicación sustantiva: distribución del ingreso y crecimiento económico en Uruguay y otros países de nuevo asentamiento desde 1870. Instituto Laurentano Figuerola de Historia Económica, Working Paper Series 05-04 (Universidad Carlos III de Madrid).
Bértola, L., and J.G. Williamson (2003). Globalization in Latin America before 1940. National Bureau of Economic Research , Working Paper 9687. Cambridge, MA.
Besley, T., and Robin Burgess. (2003). Halving Global Poverty. Journal of Economic Perspectives, 17 , 3-22.
Bhagwati, J. N. (1982). Directly Unproductive, Profit-Seeking (DUP) Activities. The Journal of Political Economy, 90 (5), 988-1002.
171
Blanchard, O., L. F. Katz, R,.Hall, and B.Eichengreen. (1992). Regional Evolutions. Brookings Papers on Economic Activity, 1992 (1), 1-75.
Bolduc, D., R. Laferriere, and G. Santarrosa. (1992). Spatial Autorregressive Error Components in Travel Flow Models. Regional Science and Urban Economics, 22 , 371-385.
Borjas, G. J., S. G. Bronars, and S. J. Trejo. (1992). Self-Selection and Internal Migration in the United States. Journal of Urban Economics, 32 , 159–185.
Bosch, M., P. Aroca, I. Fernandez, and C. Azzoni. (2003). Growth Dynamics and Space in Brazil. International Regional Science Review, 26 (3), 393–418.
Bourguignon, F. (2003). The Growth Elasticity of Poverty Reduction; Explaining Heterogeneity across Countries and Time Periods. In T. Eichler and S. Turnovsky (ed.), Inequality and growth: Theory and Policy Implications . Cambridge: The MIT Press.
Bourguignon, F. (2004). The Poverty-Growth-Inequality Triangle. The World Bank Policy Research Working Paper (28102).
Bourguignon, F., and C. Morrisson (2002). Inequality among World Citizens: 1820-1992. The American Economic Review, 92 (4), 727-744.
Bourguignon, F., and S.R. Chakravarty (2003). The Measurement of Multidimensional Poverty Journal of Economic Inequality, 1 (1), 25-49.
Bourguignon, F., F. Ferreira, and M. Menéndez. (2003). Inequality of Outcomes, Inequality of Opportunities, and Intergenerational Educational Mobility in Brazil . Washington, DC: The World Bank.
Bowles, S., and H. Gintis. (2002). Social Capital and Community Governance. Economic Journal, 112 , F419–F436.
Calvo, E., J. Torre, and M. Szwarcberg (2002). The New Welfare Alliance . Buenos Aires: Universidad di Tella, Department of Political Science.
Caramuta, D. (2005). A Dynamic Approach to the Relationship between Inequality, Social Capital and Institutions. Paper presented at the CONICET, Universidad Nacional del Sur, Argentina
172 Card, D., and John E. DiNardo. (2002). Skill-Biased Technological Change and Rising Wage Inequality: Some Problems and Puzzles. Journal of Labor Economics, 20 , 733–783.
Carlino, G. A., and L. O. Mills (1993). Are the US Regional Incomes Converging? A Time Series Analysis. Journal of Monetary Economics 32 , 335–346.
Castello, A., and R. Domenech. (2002). Human Capital Inequality and Economic Growth: Some New Evidence. Economic Journal, 112 , 187–200.
CEDLAS. (2003). Estadísticas Distributivas de la Argentina Centro de Estudios Distributivos, Laborales y Sociales . La Plata: Universidad Nacional de La Plata.
Clarke, G. (1995). More Evidence on Income Distribution and Growth. Journal of Development Economics, 47 , 403–427.
Cliff, A., and J.K.Ord (1973). Spatial Autocorrelation . London: Pion.
Cliff, A., and J.K.Ord (1981). Spatial Processes:Models and Applications . London: Pion.
Coatsworth, J.H. (1998). Economic and Institutional Trajectories in Pre-Modern Latin America. In J. H. Coatsworth and A. M. Taylor, (ed.), Latin America and the World Economy Since 1800 . Cambridge, Mass.: Harvard University Press.
Conley, T. G. (1999). GMM Estimation with Cross Sectional Dependence. Journal of Econometrics 92 , 1-45.
Corbacho, A., M. Garcia-Escribano, and G. Inchauste. (2007). Argentina: Macroeconomic Crisis and Household Vulnerability. Review of Development Economics, 11 (1), 92–106.
Cornblit, O., E. Gallo and A. O'Connell. (1962). La generación del 80 y su proyecto: antecedentes y consecuencias. Desarrollo Económico, 1 (4).
Coughlin, C., and T.B. Mandelbaum. (1988). Why Have State Per-Capita Incomes Diverged Recently? Review of Federal Reserve Bank of St. Louis, 70 , 24-36.
Cowell, F. (1998). Measurement of Inequality . Paper presented at the London School of Economics and Political Science, London.
173 Cowell, F. A., and A. Jenkins. (1995). How Much Inequality Can We Explain? A Methodology and an Application to the United States. Economic Journal, 105 , 421-430.
Cressie , N. (1993). Statistics for Spatial Data . New York: Wiley.
Daseking, C. (2004). Lessons from the Crisis in Argentina. Washington, DC: International Monetary Fund.
De Ferranti, D., G. E. Perry, F. Ferreira, and M. Walton. (2004). Inequality in Latin America and the Caribbean: Breaking with History? Washington D.C.: World Bank.
De la Fuente, A. (1997). The Empirics of Growth and Convergence: A Selective Review. Journal of Economic Dynamics and Control 21 , 23–73.
De Pablo, J. C. (2005). La Economía Argentina en la Segunda Mitad del Siglo XX (Vol. Tomo 1). Buenos Aires: La Ley.
De Vreyer, P., and G. Spielvogel. (2005). Spatial Externalities between Brazilian Municipios and their Neighbors . France: University of Lille.
Della Paolera, G., and A.M. Taylor. (2003). A New Economic History of Argentina . New York: Cambridge University Press.
Di Tella, G. (1986). Economic Controversies in Argentina from the 1920s to the 1940s. In Platt (ed.), The Political Economy of Argentina 1880-1946 . New York: St. Martin's Press.
Díaz Alejandro, C. (1974). Some Characteristics of Recent Export Expasion in Latin America. In H. Giersch (ed.), The International Division of Labor: Problems and Perspectives . Tubingen' Mohr.
Díaz Alejandro, C. (1984). The 1940s in Latin America. In L. T. M. Syrquin, and L. E. Westphal (ed.), Economic Structure and Performance: Essays in Honor of Hollis B. Chenery . Orlando, Fla.: Academic Press.
Dobson, S., and C. Ramlogan. (2002). Economic Growth and Convergence in Latin America. Journal of Development Studies, 38 (6), 83-104.
174 Dollar, D., and A. Kraay. (2002). Growth is Good for the Poor. Journal of Economic Growth 7 (3), 195-225.
Doreian, P. (1980). Linear Models with Spatially Distributed Data, Spatial Disturbances or Spatial Effects. Sociological Methods and Research, 9 , 29-60.
Durlauf, S. N., and D.Quah. (1999). The New Empirics of Economic Growth. In J. Taylor and M. Woodford (ed.), Handbook of Macroeconomics (Vol. 1A). Amsterdam: North-Holland.
Easterlin, R. (1960a). Interregional Differences in Per Capita Income, Population and Total Income, 1840-1950. In National Bureau of Economics Research Trends in the American Economy in the Nineteenth Century . Princeton, NJ: Princeton University Press.
Easterlin, R. (1960b). Regional Growth of Income. In M. Adams and S. Kuznets (ed.), Population Redistribution and Economic Growth in the United States, 1870-1950 . Philadelphia.: American Philosophical Society.
Easterly, W. (2001). The Middle Class Consensus and Economic Development. Journal of Development Economics, 6 , 317–335.
ECLAC. (2006). Human Rights and Development to the Centre of Debate on Migration in Latin America and the Caribbean (No. LC/G.2303(SES.31/11) ). Montevideo, Uruguay: ECLAC Document:.
ECLAC. (2007). Social Panorama of Latin America : La Plata, Argentina: United Nations.
Edin, P.-A., and R. Topel. (1997). Wage Policy and Restructuring: The Swedish Labor Market since 1960. In R.B. Freeman, R.H. Topel, and B. Swedenborg (ed.), The Welfare State in Transition, Reforming the Swedish Model (pp. 155–201). Chicago: University of Chicago Press.
Engerman, S., and K. Sokoloff. (2006). The Persistence of Poverty in the Americas: The Role of Institutions. In S. Bowless, S. Durlauf, and K. Hoff (ed.), In Poverty Traps : Forthcoming.
Fallah, B., and M. Partridge. (2007). The Elusive Inequality-Economic Growth Relationship: Are there Differences between Cities and the Countryside? The Annals of Regional Science, 41 (2), 375-400.
175 Fanan, C., and E Casetti. (1994). The Spatial and Temporal Dynamics of US Regional Income Inequality, 1950-1989. Annals of Regional Science, 28 .
Ferreira, F. (2001). Education for the Masses? The Interaction Between Wealth, Educational and Political Inequalities. Economics of Transition 9 (2), 533–552.
Ferreira, F., and R. Paes de Barros. (1998). Climbing a Moving Mountain:Explaining the Decline of Income Inequality in Brazil from 1976 to 1996. Washington, D.C.: Inter-American Development Bank.
Ferrer, A. (2005). La Economía Argentina: Desde sus Orígenes hasta Principios del Siglo XXI. (Primera Reimpresión, 2005 ed.). Buenos Aires: Fondo de Cultura Económica.
Ferreres, O. J. (2005). Dos Siglos de Economía Argentina, 1810-2004. Historia Argentina en Cifras . Buenos Aires: El Ateneo y Fundación Norte y Sur.
Figueras, A., J. L. Arrufat, and P.J. Regis. (2003). El Fenómeno de la Convergencia Regional: Una Contribución . Paper presented at the Annual Meetings of the Asociacion Argentina de Economía Política (AAEP).
Fingleton, B. (2003). European Regional Growth . New York: Springer-Verlag.
Fingleton, B., and E. López-Bazo. (2006). Empirical Growth Models with Spatial Effects Regional Science 85 (2), 177–198.
Florax, R., and H. Folmer. (1992). Specification and Estimation of Spatial Linear Regression Models: Monte Carlo Evaluation of Pre-Test Estimators. Regional Science and Urban Economics 22 , 405-432.
Forbes, K. J. (2000). A Reassessment of the Relationship Between Inequality and Growth American Economic Review, 90 , 869–887.
Galor, O., and D. Tsiddon. (1997). Technological Progress, Mobility, and Economic Growth. American Economic Review, 87 , 363–382.
Galor, O., and J. Zeira. (1993). Income Distribution and Macroeconomics. Review of Economic Studies, 60 , 35-52.
176 García-Verdú, R. (2002). Income Dynamics Across States in Mexico .Unpublished manuscript, University of Chicago.
Garrido, N., A. Marina, and D. Sotelsek. (2000). Crecimiento y Convergencia. Un Ejercicio Empírico sobre las Regiones Españolas y las Provincias Argentinas : Asociacion Argentina de Economía Política (AAEP).
Gasparini, L., and W. Sosa Escudero. (2001). Assessing Aggregate Welfare: Growth and Inequality in Argentina. Cuadernos de Economía 38 (113).
Gasparini, L., F.Gutierrez, and L.Tornarolli. (2005). Growth and Income Poverty in Latin America and the Caribbean: Evidence from Household Surveys . Washington, DC.: World Bank.
Gasparini, L., F.Gutierrez, and L.Tornarolli. (2007). Growth and Income Poverty in Latin America and the Caribbean: Evidence from Household Surveys. Review of Income and Wealth, 53 (2), 209-245.
Geary, R. (1954). The Contiguity Ratio and Statistical Mapping. The Incorporated Statistician 5 , 115–145.
Gerber, J. (2002). Are Incomes Converging Along the US-Mexico Border? Unpublished manuscript, San Diego State University.
Getis, A., and J. K. Ord. (1992). The Analysis of Spatial Association by Use of Distance Statistics. Geographical Analysis 24 , 189–206.
Getis, A., J. Mur and H. Zoller (2004). Spatial Econometrics and Spatial Statistics : Palgrave Macmillan.
Glaeser, E., J. Scheinkman, and A. Shleifer. (2002). The Injustice of Inequality. National Bureau of Economic Research , Working Paper 9150.
Gupta, D. (1990). The Economics of Political Violence . New York: Praeger.
Haimovich, F., and H. Winkler. (2005). Pobreza rural y urbana en Argentina: un análisis de descomposiciones . La Plata, Argentina: CEDLAS, Universidad Nacional de La Plata.
Helpman, E., and P. Krugman. (1985). Market Structure and Foreign Trade . Cambridge, Mass.: MIT Press.
177 Henry, M. S. (2002). Civic Community Approaches to Rural Development in the South: Discussion. Journal of Agricultural and Applied Economics, 34 , 343–347.
Hibbs, D. (1973). Mass Political Violence: A Cross-Sectional Analysis. New York: Wiley.
Huang, J. S. (1984). The Autorregressive Moving Average Model for Spatial Analysis. Australian Journal of Statistics, 26 , 169-178.
Islam, N. (2003). What Have We Learnt from the Convergence Debate? Journal of Economic Surveys, 17 (3), 309-362.
Iyigun, M. F., and A. L. Owen. (2004). Income Inequality, Financial Development, and Macroeconomic Fluctuations. Economic Journal 114 (352–376).
Johnson, J., and S. Kotz. (1969). Distribution in Statistics: Discrete Distributions . Boston: Houghton Miffin.
Jones, C. (1997). On the Evolution of the World Income Distribution. Journal of Economic Perspectives, 11 , 19-36.
Kaldor, N. (1956). Alternative Theories of Distribution. Review of Economic Studies, 23 83–100.
Kawachi, I., B.P. Kennedy , K. Lochner , and D. Prothrow-Stith (1997). Social Capital, Income Inequality, and Mortality. American Journal of Public Health, 87 (9), 1491-1498.
Kelejian, H. H., and I.R.Prucha. (1998). A Generalized Spatial Two-Stage Least Squares Procedure for Estimating a Spatial Autoregressive Model with Autoregressive Disturbances. Journal of Real Estate Finance and Economics 17 , 99-121.
Kelejian, H. H., and D.P. Robinson. (1993). A Suggested Method of Estimation for Spatial Interdependent Models with Autocorrelated Errors, An Application to a County Expenditure Model. Papers in Regional Science 72 , 297-312.
Kelejian, H. H., and D.P. Robinson. (1995). Spatial Correlation: A Suggested Alternative to the Autoregressive Model. In L. Anselin and R. Florax (ed.), New Directions in Spatial Econometrics (pp. 75-95). Berlin: Spriger-Verlag.
178 Knack, S., and P. Keefer. (1997). Does Social Capital Have an Economic Payoff? A Cross-Country Investigation. Quarterly Journal of Economics, 112 , 1251–1288.
Kraay, A. (2005). When Is Growth Pro-Poor? Evidence from a Panel of Countries. Journal of Development Economics, Forthcoming .
Krugman, P. (1987). The Narrow Moving Band, the Dutch Disease, and the Competitive Consequences of Mrs Thatcher. Journal of Development Economics, 45 , 1-28.
Krugman, P. (1991). Increasing Returns and Economic Geography. The Journal of Political Economy, 99 (3), 483-499.
Kucera, D. (2002). The Effects of Wealth and Gender Inequality on Economic Growth: A Survey of Recent Empirical Evidence. International Institute for Labour Studies, Working Paper DP/136/2002.
Kuznets, S. (1955). Economic Growth and Income Inequality. The American Economic Review, 45 (1), 1-28.
Leon, G. (2007, September 2007). The Military, Wealth and Strategic Redistribution . Paper presented at the Northeast Universities Development Consortium Conference (NEUDC), Harvard University.
Levene, H. (1960). Robust Tests for Equality of Variances. In I. Olkin (ed.), Contributions to Probability and Statistics . Palo Alto, CA: Stanford University Press.
Levine, R. (2004). Finance and Growth:Theory and Evidence. NBER Working Paper Series (10766).
Li, H., and Hengfu Zou. (1998). Income Inequality is Not Harmful for Growth: Theory and Evidence. Review of Development Economics, 2 , 318–334.
Londoño, J. L. (1995). Distribucióon del Ingreso y Desarrollo Económico: Colombia en el Siglo XX . Bogotá.
Lopez, H. (2003). Macroeconomics and Inequality . Washington, DC: The World Bank.
Lopez, H., and L. Serven. (2006). A Normal Relationship? Poverty, Growth, and Inequality. World Bank Policy Research Working Paper, 3814.
179 Madariaga, N., S. Montout and P. Ollivaud. (2005). Regional Convergence and Agglomeration in Argentina: a Spatial Panel Data Approach.
Madariaga, N., S. Montout and P. Ollivaud. (2004). Regional Convergence, Trade Liberalization and Agglomeration of Activities: An Analysis of NAFTA and MERCOSUR Case.
Maddison, A. (1995). Monitoring the World Economy . Paris: OECD.
Magrini, S. (2004). Regional (Di)Convergence. In T. Henderson and J.F. Thisse (ed.), Handbook of Regional and Urban Economics (Vol. 4). Amsterdam: Elsevier B.V.
Marina, A. (2002). Economic Convergence of the First and Second Moment in Provinces of Argentina. Estudios de Economía, 27 , 259-277.
Marshall, A., and D. Orlansky. (1983). Inmigración de países limítrofes y demanda de mano de obra en la Argentina, 1940-1980. Desarrollo Económico, 23 , 35-57.
Mauro, P. (2002). The Persistence of Corruption and Slow Economic Growth . Washington, DC.: IMF
Mo, P. H. (2000). Income Inequality and Economic Growth. Kyklos, 53 , 293–316.
Moran, P. (1948). The Interpretation of Statistical Maps. Journal of the Royal Statistical Society, Series B 10 , 243–251.
Moreno, R., and B. Trehan (1997). Location and the Growth of Nations. Journal of Economic Growth, 2 (4), 399-418.
Murdoch, J. C., M. Rahmatian, and M.A. Thayer. (1993). A Spatially Autoregressive Median Voter Model of Recreation Expenditures. Public Finance Quarterly, 21 , 334-350.
Nan, L. (2000). Inequality in Social Capital Contemporary Sociology 29 (6), 785-795.
Neutel, M., and A. Heshmati. (2006). Globalization, Inequality and Poverty Relantionships: A Cross Country Evidence . Bonn: Institute for the Study of Labor (IZA).
Okun, A. M. (1975). Equality and Efficiency: The Big Tradeoff . Washington: Brookings Institution.
180
Ord, J. K. (1975). Estimation Methods for Models of Spatial INteraction. Journal of the American Statistical Association, 70 , 120-126.
Panizza, U. (2002 ). Income Inequality and Economic Growth: Evidence from American Data. Journal of Economic Growth 7 (1), 25-41
Partridge, M. (1997). Is Inequality Harmful for Growth? Comment. The American Economic Review, , 87 , 1019–1032.
Partridge, M. (2005). Does Income Distribution Affect U.S. State Economic Growth? Journal of Regional Science, 45 (2), 363–394.
Perotti, R. (1993). Political Equilibrium, Income Distribution, and Growth. Review of Economic Studies, 60 , 755-776.
Perotti, R. (1996). Growth, Income Distribution, and Democracy: What the Data Say. Journal of Economic Growth, 2 , 149–187.
Perry, G., O. Arias, H. López, W. Maloney, and L. Servén. (2006). Poverty Reduction and Growth: Virtuous and Vicious Circles. Washington, D.C.: The World Bank.
Persson, T., and G. Tabellini. (1994). Is Inequality Harmful for Growth? American Economic Review, 84 , 600–621.
Plotnick, B., E. Smolensky, E. Evenhouse, and S. Reilly. (1996). The Twentieth-Century Record of Inequality and Poverty in the United States. In G. Engerman and R.Gallman (ed.), The Cambridge Economic History of the United States (Vol. II). Cambridge, U.K.: Cambridge University Press.
Portes, A., and A. Ross. (1976). Modernization for Emigration: The Medical Brain Drain from Argentina. Journal of Interamerican Studies and World Affairs, 18 (4), 395- 422.
Prados de la Escosura, L. (2005). Growth, Inequality, and Poverty in Latin America: Historical Evidence, Controlled Conjectures. Economics History and Institutions Working Papers, 54104(Universidad Carlos III, Departamento de Historia Económica e Instituciones, Madrid).
Quah, D. (1996). Empirics for Economic Growth and Convergence. European Economic Review, 40 (6), 1353-1375.
181
Rapaport, M. (2000). Historia económica, politica y social de la Argentina (1880-2000). Buenos Aires: Ediciones Macchi.
Ravallion, M. (1994). Poverty Comparisons. Fundamentals of Pure and Applied Economics, 56 .
Ravallion, M. (2001). Growth, Inequality and Poverty: Looking Beyond Averages. World Development, 29 , 1803-1815.
Ravallion, M. (2004). Looking beyond Averages in the Trade and Poverty Debate. World Bank Policy Research Working Paper (3461).
Ravallion, M., and S. Chen. (1997). What Can New Survey Data Tell Us about Recent Changes in Distribution and Poverty? World Bank Economic Review 11 (2), 357– 382.
Rey, S., and M. Janikas (2005). Regional Convergence, Inequality and Space. Journal of Economic Geography, 5 , 155-176.
Rey, S. J., and B.D. Montouri. (1999). US Regional Income Convergence: A Spatial Econometric Perspective. Regional Studies, 33 (2), 143-156.
Ripley, B. D. (1981). Spatial Statistics . New York: Wiley.
Ripley, B. D. (1988). Statistical Inference for Spatial Processes . Cambridge: Cambridge University Press.
Rodrik, D. (1999). Where Did All the Growth Go? External Shocks, Social Conflict, and Growth Collapses. Journal of Economic Growth 4 (4), 385-412.
Roemer, J. E. (1998). Equality of Opportunity . Cambridge, MA: Harvard University Press.
Rozada M., and A. Menendez. (2002). Why Have Poverty and Income Inequality Increased so Much? Argentina 1991-2002. Centro de Investigación en Finanzas, Documento de Trabajo 01/2002, 1-31.
Santos, M. E. Factors Influencing Inequality and Poverty Across Argentina's Cities. Journal of Economic Literature, 31 (121).
182 Scully, G. W. (2002). Economic Freedom, Government Policy and the Tradeoff between Equity and Economic Growth. Public Choice, 113 , 77–96.
Sen, A. (1992). Inequality Reexamined . Cambridge, MA: Harvard University Press.
Serra, M. I., M.F.Pazmino, G. Lindow, B. Sutton, and G. Ramirez. (2006). Regional Convergence in Latin America. International Monetary Fund Working Paper, 06(125), 1-29.
Siebert, H. (1998). Commentary: Economic Consequences of Income Inequality . Paper presented at the Symposium of the Federal Reserve Bank of Kansas City on Income Inequality: Issues and Policy Options.
Taylor, A. (1992). External Dependence, Demographic Burdens and Argentine Economic Decline After the Belle Epoque. Journal of Economic History, 52 (December), 907-936.
Taylor, A. (1994). Mass Migration to Distant Southern Shores: Argentina and Australia, 1870-1939. In W. Hatton and J.G. Williamson (ed.), Migration and the International Labor Market, 1850-1939 . London: Routledge.
Temple, J. (1999). The New Growth Evidence. Journal of Economic Literature, 37 , 112– 156.
Tobler, W. (1979). Cellular Geography. In S. Galle and C. Olsson (ed.), Pilosophy in Geography (pp. 379-386). Dordrecht: Reidel.
Upon, G., and J., B. Fingleton. (1985). Spatial Data Analysis by Example (Vol. 1: Point Pattern and Quantitative Data). New York: Wiley.
Upon, G. J., and B. Fingleton. (1989). Spatial Data Analysis by Example (Vol. 2: Categorical and Directional Data). New York: Wiley.
Utrera, G. E., and J.A. Koroch (1998). Convergencia: evidencia empirica para las provincias argentinas (1953-1994). Anales de la XXXIIIa Reunion Anual de la Asociacion Argentina de Economia Poltica. Argentina.
Venieris, Y., and D. Gupta. (1986). Income Distribution and Sociopolitical Instability as Determinants of Savings: A Cross-Sectional Model. Journal of Political Economy, 94 , 873-883.
183 Ventura, J. (1997). Growth and Interdependence. Quarterly Journal of Economics, 112 , 57–84.
Welch, F. (1999). In Defense of Inequality. American Economic Review, 89 , 1–17.
Whittle, P. (1954). On Stationary Processes in the Plane. Biometrika, 41 , 434-449.
Williamson, J. (1965). Regional Inequality and the Process of National Development. Economic Development and Cultural Change, 4 , 3-47.
Williamson , J. G. (1999). Real Wage Inequality and Globalization in Latin America before 1940. Revista de Historia Económica, 17 , 101-142.
Williamson , J. G. (2002). Land, Labor, and Globalization in the Third World, 1870- 1940. Journal of Economic History, 62 (1), 55-85.
Winkler, H. (2005). Monitoring the Socio-Economic Conditions in Uruguay. La Plata, Argentina: CEDLAS Universidad Nacional de La Plata.
184
APPENDIX A
TABLES AND FIGURES FOR CHAPTER 1
185 0.6
0.5
0.4
0.3
0.2
0.1
0 1950 1960 1970 1980 1990 2000
Source: Perry et al. (2006).
Figure A.1: Gini coefficient for the income distribution in Latin America, 1950-2000.
186 0.6
0.5
0.4
0.3
0.2
0.1
0 1950 1960 1970 1980 1990 2000
Note: Measured using a poverty line of US$2 a day.
Source: Perry et al. (2006) for 1950-1980 and Gasparini, Gutierrez and Tornarolli (2005) for 1990 and 2000.
Figure A.2: Poverty rates in Latin America, 1950-2000.
187 Share of poor people Rural Share of poor living in population $1.08 a day Millions of people in the total rural areas as share of poverty line (a) people population (%) (%) (b) total (%) Region 1990 2001 1990 2001 2001 2001 East Asia 472 271 30 15 80 63 Eastern Europe and Central Asia 2 17 1 4 53 37 Latin America and Caribbean 49 50 11 10 42 24 Middle East and North Africa 6 7 2 2 63 42 South Asia 462 431 41 31 77 72 Sub-Saharan Africa 227 313 45 46 73 67
Note: (a) Poverty lines set in 1993 US$ adjusted for purchasing power parity 37 . (b) Calculated as rural poverty rate × (100 – urbanization rate) / national poverty rate.
Source: Columns 1–4: Chen and Ravallion, 2004. Columns 5–6: Calculated from World Bank, 2004.
Table A.1: Population living below the US$ 1 poverty line, 1990 and 2001
37 Target one of the Millennium Development Goals is formulated in terms of an extreme poverty line of one dollar a day, adjusted for purchasing power parity (PPP). The World Bank has used this line since 1990, and it represents a minimum international measurement of poverty in any country of the world. Its value has been set at the median of the 10 lowest per capita national poverty lines in the world, which are used for countries in Africa and Asia. The exact amount is US$ 1.08 per day, or US$ 32.74 per month, expressed in terms of 1993 purchasing power parity.
188 Share of poor people in the total population $2.15 a day poverty (a) Millions of people (%) Region 1990 2001 1990 2001 East Asia 1,116 865 70 47 Eastern Europe and Central Asia 23 93 5 20 Latin America and Caribbean 125 128 28 25 Middle East and North Africa 51 70 21 23 South Asia 958 1,064 86 77 Sub-Saharan Africa 382 516 75 77
Notes: (a) Poverty lines set in 1993 US$ adjusted for purchasing power parity.
Source: Chen and Ravallion, 2004.
Table A.2: Population living below the US$ 2 poverty line, 1900 and 2001.
189 Share of top 10 Share of bottom Ratio of Gini percent in total 20 percent in incomes of 10th Coefficient income (%) total income (%) to 1st decile Brazil (2001) 59.0 47.2 2.6 54.4 Guatemala (2000) 58.3 46.8 2.4 63.3 Colombia (1999) 57.6 46.5 2.7 57.8 Chile (2000) 57.1 47.0 3.4 40.6 Mexico (2000) 54.6 43.1 3.1 45.0 Argentina (2000) 52.2 38.9 3.1 39.1 Jamaica (1999) 52.0 40.1 3.4 36.5 Dominican Republic (1997) 49.7 38.6 4.0 28.4 Costa Rica (2000) 46.5 34.8 4.2 25.1 Uruguay (2000) 44.6 33.5 4.8 18.9 United States (1997) 40.8 30.5 5.2 16.9 Italy (1998) 36.0 27.4 6.0 14.4
Source: Altimir (1987) and Londoño and Szeleky (2000) based on the World Bank Development Indicators Database.
Table A.3: Indicators of inequality for selected Latin American countries, the United States, and Italy, late 1990s.
190 0.64
0.62
0.6
0.58
0.56
0.54
0.52
0.5
0.48
0.46
0.44
0.42
0.4 Mid-80s Early-90s Mid-90s Early-00s Mid-00s
Argentina Bolivia Brazil Chile Colombia Uruguay Venezuela Mexico
Source: Constructed using Gini coefficients from SEDLAC.
Figure A.3: Trends in inequality in major Latin American countries from the early 1980s to mid-2000s (Gini coefficients).
191
APPENDIX B
MAPS OF ARGENTINA
192
Map B.1: Satelital map of Argentina using Google Earth.
193
Mapa B.2: Satelital map of Argentina in South America using Google Earth
194
Mapa B.3: Satelital map of the provinces of Argentina using Google Earth
195
Source: http://209.15.138.224/argentina_mapas/m_rArgeninaPolitic.htm
Mapa B.4: Map of the provinces of Argentina
196
Note: Metopolitana is the Ciudad Autonoma de Buenos Aires, Noroeste is Northwest, Nordeste is Northeast.
Source: http://209.15.138.224/argentina_mapas/m_rArgentinaZonas.htm
Mapa B.5: Map of the regions of Argentina
197 Per Capita Income Growth by province Argentina, 1992-2002
0 - .25 -.05 - 0 -.1 - -.05 -.21 - -.1 -.42 - -.21
Mapa B.6: Per capita income growth by province (1992-2002) using STATA
198 Map Gini Coefficient by province Argentina, 1991
.44 - .47 .43 - .44 .41 - .43 .38 - .41
Mapa B.7: Gini Coefficient by province in 1991 using STATA
199 Map Gini Coefficient by province Argentina, 2002
.52 - .54 .49 - .52 .48 - .49 .47 - .48 .44 - .47
Mapa B.8: Gini Coefficient by province in 2002 using STATA
200 Size by province thousand Km Sq
213.85 - 986.42 146.13 - 213.85 101.12 - 146.13 88.93 - 101.12 62.64 - 88.93 0.20 - 62.64
Mapa B.9: Area of provinces in Argentina using STATA
201 Density of Population by Province Argentina, 2001
22.6 - 13679.6 10.6 - 22.6 5.9000001 - 10.6 2.7 - 5.9000001 .1 - 2.7
Mapa B.10: Density of population in the provinces of Argentina using STATA
202 Low Income
High Income
Mapa B.11: Two main clusters for the real per capita income in the provinces of
Argentina using the Moran I, 1980-2002.
203 High Gini
Low Gini
Mapa B.12: Two main clusters for the Gini in the provinces of Argentina using the
Moran I, 1991-2002.
204
APPENDIX C
GEARY-KHAMIS METHODS OF AGGREGATION
205 METHODS OF AGGREGATION
This appendix begins with the Geary-Khamis method of aggregation, explaining some of its advantages and disadvantages. It then compares the G-K results with several other methods of aggregation, to illustrate some of the differences.
A. The Geary system
The valuation of a country's output in international prices can be written as:
(1) and where the p is are the international prices for each of the basic headings and rgdpj is
GDP of country j valued at those prices. The particular contribution of Geary was to define the international prices in such a way that they would produce an overall PPP for a country that was consistent with the prices. The definition of the PPP in the ICP is:
(2)
206 where Eij is the expenditure in national currency on basic heading i by country j . That is,
the purchasing-power parity over GDP is the ratio of the GDP of a country in national
currency to its GDP in international prices.
For Geary there were actual quantities and prices associated with the agricultural
output that he was concerned with valuing across countries. The international prices
would be in a numeraire currency, such as the dollar, and the international prices would
be so many dollars per unit quantity, say, ton of rice. In the ICP, there are basic heading
parities, PPijS , that have been generated by EKS or CPD. These basic heading parities
have the dimension of units of currency of country j to the numeraire currency for the
basic heading.
This means that the interpretation of quantity and price at the basic heading level are not
tons and rupees per ton. Rather, the quantity in the G-K technique as used in the ICP is
what is termed a notional quantity. It is defined as:
(3)
Each country's expenditure for a basic heading is converted to the currency of the
numeraire country; it is termed a notional quantity because it serves the function of a
quantity with its values at numeraire country prices.
One might ask why one cannot simply add up the notional quantities for each basic
heading for a country to get a GDP in a common currency. The answer is that the result
would use the relative prices between each basic heading that prevailed in the numeraire
country. This means that the total would depend on which country was chosen as
numeraire, and the result would not be base country invariant. In the G-K system, the
international price for heading i is defined as:
207 (4)
Equation (4) has been written as a weighted sum of the ratios of the heading parities to
the aggregate PPP. The weights used to obtain the international prices typically are the
notional quantities. Usually, the expenditures (EijS) entering into equation (3) are the
total expenditures of a country, though alternative weights have been used. a/ For each
country this is a ratio that will centre on 1.0 because in the Geary system the PPPj is a
weighted average of the basic heading parities, where the weights are the notional
quantities.
An important feature of the G-K system is illustrated in equation (5), where the
denominator of equation (4) is brought to the left-hand side:
(5)
Each side of equation (5) is a measure of the contribution of output of a basic heading to regional or world GDP. It is only in the G-K system that the valuation of quantities at international prices is consistent with their basic heading parities and expenditures, as well as the overall purchasing-power parity of each country.
Equations (1) and (4) represent the complete G-K system when PPPj and qij are
defined as in equations (2) and (3). When m is over 150 and n is over 60, this appears to
be a large system to solve. However, it turns out that the easiest way to solve the system
208 is by iteration; and it also turns out that the iterative procedure is itself instructive, as the following discussion is intended to show.
The basic data are the expenditures ( EijS ) and parities ( ppijS ) at the basic heading
levels, and from these the qijs can be derived. Consider an iteration that begins by
initially setting each PPPj equal to the exchange rate. For example, if the United States
were the base country, then its initial PPP would be 1.0 and the initial PPPs for the other
countries would be their exchange rate relative to the dollar. Then a set of international
prices can be estimated using equation (4). These p is can then be plugged into equation
(1) and then equation (2) to estimate a set of PPPjS . The process can then be repeated beginning with the new PPPiS . The iteration will be complete when the difference between the initial set of PPPjS and the end set is very small. Typically, in eight iterations the differences will only be observed at the fourth decimal place. It is unlikely that when the last iteration is complete the new PPP for the United States will equal 1.0.
The system is then normalized so that each new PPP is adjusted so that the United States value will be 1.0, and the p is appropriately scaled so that, for the United States, gdp and
rgdp as obtained from equation (1) are equal.
While one can begin the iteration with any set of values, there is another way to
begin that is also instructive. Consider setting each of the initial international prices ( p is)
equal to 1.0. The same loop can then be followed, estimating the PPPjs when the p is are
all 1.0, and work back through the system to obtain a new set of international prices, and
a new set of PPPs and so on. A normalization as described in paragraph 9 would also be
carried out to make the PPP of the base country 1.0. Beginning with all international
209 prices equal to 1.0 is equivalent to using the relative price structure of the numeraire country. The fact that the final set of international prices will differ substantially from
1.0, no matter which country is numeraire, again illustrates why one cannot simply sum up the notional quantities given in equation (3).
This discussion should also make clear that the international prices of the ICP centre around 1.0 and are used to value a quantity that has no natural dimension, such as a
b kilogram, but has a notional character depending on the numeraire currency. / The iteration procedure also illustrates how the Geary system achieves additivity across countries and basic headings to achieve matrix consistency.
As discussed in the text the major advantage of the Geary system is that the
international prices are analogous to the prices used to generate the national accounts of
an individual country. In the Geary formulation, large rich countries receive more weight
in determining international prices used to value quantities in each country. This means
that the structure of international prices will tend to be closer to those of rich countries.
There is also usually an inverse relationship between price and quantity across countries,
so that items that are expensive in poor countries, for example, will be consumed in
relatively small quantities and vice versa. The G-K price structure will tend to value the
large quantities of relatively inexpensive items in poor countries, such as services, at
higher prices. Conversely, those items are relatively cheap in rich countries, such as
transport equipment, will be valued at international prices closer to their national value.
This effect is present in all of the aggregation systems since it is part of the world
economic structure that the ICP is attempting to represent.
210 However, the international price systems that are explicit or implicit in other systems are usually closer to middle-income countries because the weights used are not in proportion to country GDP. As a consequence, the G-K system tends to lower the income of rich countries relative to poor countries more than the other aggregation methods.
Some regard this as a desired result stemming from the national accounts basis of the G-
K system, while others regard it as a drawback. c/
a/ For the world comparisons in phases I-IV, countries were assigned additional
weight to reflect the importance of countries not included in the benchmark
comparisons. The total expenditures of a country were termed its supercountry
weight, and the sum of all supercountry expenditures would be world GDP. One
reason for using supercountry weights was to estimate the international prices that
were implicit in world GDP. Since the G-K result does depend on the number of
countries in the calculation, the use of supercountry weights was designed to
approximate the international prices if all countries in the world were participating in
the ICP. This in turn should, in principle, make the results from earlier benchmark
ICP comparisons, when relatively few countries participated, better approximate later
comparisons involving more benchmark countries.
In the Geary system, it is also possible to use per capita expenditure weights or other
weighting systems, For example, one could assign equal weights to each country over
all expenditures and in effect use the percentage expenditure for each basic heading
as the country weight. The discussion in this annex assumes that the overall weight
for each country is their GDP, or supercountry GDP.
211 b/ The international prices will depend on the numeraire country chosen. This point is discussed in Kravis, Heston and Summers (1982, pp. 94-95). Two other technical points may be noted. First, some regions have chosen to use a numeraire currency outside the region, as, for example, Africa. In the African comparisons, all prices and expenditures are initially converted into United States dollars at exchange rates. In the
African study, no single country is used as the base, but rather the average of all countries is used. While the results of the African study are presented in dollars, this does not make them comparable with other countries, such as the United States, because dollar conversions have only been carried out at exchange rates.
A second point is that when an average of a group of countries is used, as in Africa or the European Communities, there will still be a set of international prices implicit in the calculation. In the African case, the system would be normalized to make the sum of expenditures of all headings and all countries converted at exchange rates equal to the sum of all notional quantities valued at international prices. For any particular basic heading, this equality would not hold, and the ratio of the sum across all countries of the basic heading notional quantities valued at international prices to their value at exchange rates would be the international price for that basic heading. c/ Usually, the G-K results are criticized because they depart from Fisher binary results, being closer to the Laspeyres than the Paasche estimate for poor countries.
However, the binary comparisons being used as a reference weight each country the same. The EKS system, which is an indirect least squares type of estimate from the binaries, naturally comes closer to the Fisher result than does G-K. However, Prasada
Rao has shown that, if a binary is done using the GDP weights of the G-K system,
212 then the multilateral G-K is a direct least squares estimate based on the binaries and,
of course, comes much closer to the G-K binaries than does EKS . The point, then, is
that it is really the weighting system that produces more difference between methods
than other factors (see Prasada Rao (1972)). 38
38 See the HANDBOOK OF THE INTERNATIONAL COMPARISON PROGRAMME http://unstats.un.org/unsd/methods/icp/ipc7_htm.htm
213
APPENDIX D
TABLES AND FIGURES FOR CHAPTER 3
214 Time span Argentina Brazil Chile Colombia Mexico Peru Venezuela 1810-1850 0.38 NA NA NA NA NA NA 1850-1870 0.94 0.20 0.08 NA 0.31 NA -0.06 1870-1890 2.48 0.21 0.10 NA 1.68 NA 0.13 1890-1900 0.34 -0.23 2.50 NA 0.39 NA -0.03 1890-1913 2.02 0.36 5.80 1.05 1.72 NA 1.24 1913-1929 0.83 1.74 1.56 1.24 0.09 2.82 7.33 1929-1938 -0.76 1.73 -0.87 2.28 -0.84 0.91 2.14 1938-1950 1.71 2.18 1.67 1.05 3.50 2.13 4.98 1950-1960 1.10 3.39 1.18 1.69 2.93 2.94 2.74 1960-1970 2.77 2.77 1.94 2.31 3.10 2.33 1.08 1970-1980 0.94 5.78 0.96 3.23 3.51 1.00 -0.65 1980-1990 -2.48 -1.11 1.10 1.23 -0.31 -3.47 -1.97 1990-1997 4.85 1.57 5.94 1.58 0.84 3.41 1.37 1997-2000 -1.20 0.39 0.94 -2.05 3.56 -0.51 -2.78 2000-2002 -5.87 0.11 1.03 -0.05 -0.96 0.54 -2.93 1870-1929 1.85 0.68 2.68 0.74 1.26 1.17 2.47 1938-1980 1.63 3.46 1.45 2.02 3.27 2.10 2.15 1980-2000 0.22 0.04 2.74 0.86 0.66 -0.67 -0.93 1870-1980 1.55 1.82 1.92 1.35 1.85 1.50 2.32 1870-2000 1.34 1.54 2.04 1.28 1.67 1.16 1.81 1810-2004 1.06 NA NA NA NA NA NA
Note: NA: Not Available. Per capita GDP is measured in constant Geary Khamis US$ Million of 1990.
Source: Constructed by the author using the Ferreres (2005) dataset.
Table D.1: Average per capita income growth rates for the seven major Latin American economies, 1810-2004.
215 12000
10000
8000
6000
4000
2000
0 1900 1903 1906 1909 1912 1915 1918 1921 1924 1927 1930 1933 1936 1939 1942 1945 1948 1951 1954 1957 1960 1963 1966 1969 1972 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002
Argentina Brasil Chile Colombia México Perú Venezuela
Note: NA: Not Available. Per capita GDP is measured in constant Geary Khamis US$ Million of 1990.
Source: Constructed by the author using the Ferreres (2005) dataset.
Figure D.1: Per Capita GDP for seven major Latin America economies, 1900-2004.
216
10000
9000
8000
7000
6000
5000
4000
3000
2000
1000
0 1810 1818 1826 1834 1842 1850 1858 1866 1874 1882 1890 1898 1906 1914 1922 1930 1938 1946 1954 1962 1970 1978 1986 1994 2002
Note: NA: Not Available. Per capita GDP is measured in constant Geary Khamis US$ Million of 1990.
Source: Constructed by the author using the Ferreres (2005) dataset.
Figure D.2: Per Capita GDP for Argentina, 1810-2004.
217 2.2500
2.0000
1.7500
1.5000 Per capita growth, % Pergrowth, capita
1.2500
1.0000 6 6.5 7 7.5 8 8.5 Initial income, log
Note: Per capita GDP is measured in constant Geary Khamis US$ Million of 1990.
Source: Constructed by the author using the Ferreres (2005) dataset.
Figure D.3: Average annual GDP growth rates experienced by the seven largest Latin American economies between 1900 and 2004, with their corresponding (logged) initial per capita income level in 1900.
218 0.650
0.550
0.450
0.350
0.250
0.150
0.050
-0.050 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2004
Note: Per capita GDP is measured in constant Geary Khamis US$ Million of 1990.
Source: Constructed by the author using the Ferreres (2005) dataset.
Figure D.4: Cross-country standard deviation of per capita GDP for the seven largest
Latin America economies, 1900-2004.
219 0.090
0.080
0.070
0.060
0.050
0.040
0.030
0.020
0.010
0.000 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2004
Note: Per capita GDP is measured in constant Geary Khamis US$ Million of 1990.
Source: Constructed by the author using the Ferreres (2005) dataset.
Figure D.5: Coefficient of variation of log of per capita GDP for the seven largest Latin
America economies, 1900-2004.
220 North Main Northen Latin Time span America Spain Europe Europe Oceania America Argentina Asia 1820-1850 1.18 0.25 0.37 -0.11 2.35 0.13 0.49 -2.51 1850-1870 1.38 0.91 2.75 1.99 4.05 0.58 0.94 3.27 1870-1890 1.65 1.48 0.29 1.10 1.07 1.80 2.48 0.41 1890-1900 1.95 1.00 3.28 1.70 0.06 7.81 1.36 9.65 1900-1913 2.58 0.77 1.39 1.82 1.67 2.18 2.53 1.63 1913-1929 1.29 1.69 1.52 1.66 -0.18 2.03 0.83 0.20 1929-1938 -1.26 -4.10 -0.02 2.21 1.76 0.43 -0.76 1.49 1938-1950 3.96 1.43 1.09 2.30 2.15 2.73 1.71 -2.03 1950-1960 1.77 3.67 4.14 2.83 1.39 2.18 1.10 2.80 1960-1970 3.04 7.81 3.94 3.82 2.48 2.11 2.77 3.69 1970-1980 2.44 2.67 2.48 2.61 1.36 1.55 0.94 4.66 1980-1990 1.95 2.55 1.71 2.07 1.43 -1.06 -2.48 4.79 1990-2004 1.85 2.15 1.24 1.89 1.95 1.53 1.89 3.92 1850-1900 2.32 1.16 2.09 1.51 3.48 2.48 1.64 3.35 1870-1900 1.75 1.32 1.27 1.30 0.73 3.76 2.10 3.40 1870-1929 1.80 1.30 1.37 1.51 0.69 2.94 1.85 2.13 1938-1980 2.85 3.75 2.82 2.86 1.86 2.17 1.63 2.04 1980-2004 1.89 2.32 1.44 1.97 1.73 0.44 0.05 4.28 1870-1980 1.95 1.77 1.80 2.08 1.22 2.44 1.55 2.04 1870-2004 1.94 1.87 1.74 2.06 1.31 2.08 1.28 2.44
Note: Per capita GDP is measured in constant Geary Khamis US$ Million of 1990.
Source: Constructed by the author using the Ferreres (2005) dataset.
Table D.2: Average annual rates of growth of per capita GDP for regions of the world,
1820-2004 (percentage).
221 34,000.0
32,000.0
30,000.0
28,000.0
26,000.0
24,000.0
22,000.0
20,000.0
18,000.0
16,000.0
14,000.0
12,000.0
10,000.0
8,000.0
6,000.0
4,000.0
2,000.0
0.0
0 4 0 6 2 4 0 8 4 6 2 6 2 8 4 7 8 98 40 70 182 18 18 188 189 18 190 191 1916 1922 192 193 19 194 195 1958 1964 19 197 198 198 199 2000
US UK Germany Norway Australia Japon Argentina
Note: Per capita GDP is measured in constant Geary Khamis US$ Million of 1990.
Source: Constructed by the author using the Ferreres (2005) dataset.
Figure D.6: Per capita GDP for the United States, United Kingdom, Germany, Norway,
Australia, Japan and Argentina, 1820-2004.
222
Region Provinces Gran Buenos Aires Capital Federal and the main suburban areas around it Pampeana rest of Buenos Aires, Cordoba, Entre Rios, La Pampa and, Santa Fe Cuyo Mendoza, San Juan and, San Luis NorthEast Corrientes, Chaco, Formosa and, Misiones NorthWest Jujuy, Salta, Catamarca, La Rioja, Tucuman and, Santiago del Estero Patagonia Chubut, Neuquen, Rio Negro, Santa Cruz and,Tierra del Fuego
Table D.3: Economic Regions and Provinces in Argentina. EPH
223 D Province Urban Conglomerate Region 1 Buenos Aires Bahía Blanca - Cerri- 03 Pampeana Gran La Plata -02 Mar del Plata y Batán - 34 San Nicolás-V.Constitución -38 2 Catamarca Gran Catamarca- 22 NW 3 Chaco Gran Resistencia - 08 NE 4 Chubut Comodoro Rivadavia- Rada Tilly- 09 Patagonia Rawson-Trelew -91 5 Gran Buenos Aires Ciudad de Buenos Aires- 32 Gran Bs As Gran Bs As-01 Partidos del conurbano- 33 6 Córdoba Gran Córdoba- 13 Pampeana Río Cuarto- 36 7 Corrientes Corrientes - 12 NE Curuzú Cuatiá- 21 Goya - 24 8 Entre Ríos Gran Paraná -06 Pampeana Concordia -14 Gualeguaychú- 16 9 Formosa Formosa- 15 NE 10 Jujuy S.S.de Jujuy - Palpalá- 19 NW 11 La Pampa Santa Rosa - Toay- 30 Pampeana 12 La Rioja La Rioja- 25 NW 13 Mendoza Gran Mendoza- 10 Cuyo 14 Misiones Posadas- 07 NE 15 Neuquén Neuquén - Plottier- 17 Patagonia 16 Rio Negro Viedma-Carmen de Patagones- 93 Patagonia 17 Salta Salta-23 NW 18 San Juan Gran San Juan- 27 Cuyo 19 San Luis San Luis - El Chorrillo- 26 Cuyo 20 Santa Cruz Río Gallegos- 20 Patagonia 21 Santa Fe Gran Rosario- 04 Pampeana Gran Santa Fe - 05 22 Sgo.del Estero Sgo. del Estero - La Banda- 18 NW 23 T.del Fuego Ushuaia - Río Grande- 31 Patagonia 24 Tucumán G.S.M.de Tucumán - Tafí Viejo- 29 NW
Table D.4: Urban Conglomerates by Provinces in Argentina. EPH.
224 4
3
2
1
0 Average growth Average rates -1
-2
-3 0 0.5 1 1.5 2 2.5 3 3.5 Initial per capita (log)
Source: Author’s calculation using data from ECLAC39.
Figure D.7: Average annual GDP growth rates experienced by 23 provinces of Argentina, 1980-2002, with their corresponding (logged) initial per capita income level in 1980
39 I would like to thank Ricardo G. Martínez from the office of ECLAC in Buenos Aires, who constructed this database and has allowed me to work with it
225 0.073
0.071
0.069
0.067
0.065
0.063
0.061
0.059
0.057
0.055
0.053
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 8 8 8 8 8 8 8 8 8 8 9 9 9 9 9 9 9 9 9 9 0 0 0 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 20 20 20
Source: Author’s calculation using data from ECLAC.
Figure D.8: Coefficient of variation of the log of provincial real per capita GDP for 23 provinces and the capital city ( Ciudad Autónoma de Buenos Aires ) in Argentina, 1980-
2002.
226
0.755
0.655
0.555
0.455
0.355
0.255
0.155
0.055
-0.045 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 8 8 8 8 8 8 8 8 8 8 9 9 9 9 9 9 9 9 9 9 0 0 0 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 19 20 20 20
Source: Author’s calculation using data from ECLAC.
Figure D.9: Moran's I statistic for the provincial real per capita GDP of Argentina, 1980-
2002.
227 Year Moran's I E(I) sd(I) z p-value* 1980 0.684 -0.045 0.124 5.869 0.000 1981 0.695 -0.045 0.124 5.978 0.000 1982 0.703 -0.045 0.123 6.066 0.000 1983 0.716 -0.045 0.126 6.050 0.000 1984 0.659 -0.045 0.112 6.270 0.000 1985 0.700 -0.045 0.114 6.540 0.000 1986 0.500 -0.045 0.107 5.092 0.000 1987 0.486 -0.045 0.105 5.054 0.000 1988 0.512 -0.045 0.107 5.204 0.000 1989 0.551 -0.045 0.116 5.140 0.000 1990 0.552 -0.045 0.126 4.755 0.000 1991 0.484 -0.045 0.112 4.737 0.000 1992 0.509 -0.045 0.120 4.634 0.000 1993 0.523 -0.045 0.125 4.564 0.000 1994 0.538 -0.045 0.126 4.630 0.000 1995 0.579 -0.045 0.128 4.890 0.000 1996 0.513 -0.045 0.126 4.451 0.000 1997 0.450 -0.045 0.123 4.030 0.000 1998 0.407 -0.045 0.121 3.738 0.000 1999 0.352 -0.045 0.117 3.396 0.001 2000 0.334 -0.045 0.115 3.289 0.001 2001 0.380 -0.045 0.119 3.581 0.000 2002 0.338 -0.045 0.114 3.354 0.001 *two-tail test
Source: Author’s calculation using data from ECLAC.
Table D.5: Moran’s I statistics for the provincial per capita GDP of Argentina, 1980-
2002.
228 0.750 0.075
0.700
0.070 0.650
0.600
0.065 0.550
0.500 0.060
0.450
0.400 0.055
0.350
0.300 0.050
80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 00 01 02 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2
Moran CV
Source: Author’s calculation using data from ECLAC.
Figure D.10: Provincial coefficient of variation of the log of provincial real per capita
GDP for 23 provinces and the capital city of variation and the Moran's I statistic,
Argentina, 1980-2002.
229
Moran scatterplot (Moran's I = 0.624) gdppc_80
3
4 22 2 19
1 Wz
1 15 0 11 5 18 8 13 21 10 149 20 6 12 2 3 7 17 23 16
-1 -1 0 1 2 3 4 z
Source: Author’s calculation using data from ECLAC.
Figure D.11: Local Moran’s I statistic for the provincial real per capita GDP in 1980.
230 Moran scatterplot (Moran's I = 0.329) gdppc_02
2
4 22
1 19
1 Wz
13 11 0 15 5 17 18 12 8 6 21 7 20 2 3 23 16 9 1014
-1 -1 0 1 2 3 4 z
Source: Author’s calculation using data from ECLAC.
Figure D.12: Local Moran’s I statistic for the provincial real per capita GDP in 2002.
231
Province ID 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 Buenos Aires 1 2 2 2 2 2 2 2 2 2 2 2 2 Catamarca 2 3 3 3 3 3 3 3 3 3 3 3 3 Chaco 3 3 3 3 3 3 3 3 3 3 3 3 3 Chubut 4 1 1 1 1 1 1 1 1 1 1 1 1 Corrientes 5 1 1 1 1 1-4 1-4 1-4 1-4 1-4 1-4 1-4 4 CBA 6 3 3 3 3 3 3 3 3 3 3 3 3 Córdoba 7 3 3 3 3 3 3 3 3 3 3 3 3 Entre Ríos 8 3 3 3 3 3 3 3 3 3 3 3 3 Formosa 9 3 3 3 3 3 3 3 3 3 3 3 3 Jujuy 10 3 3 3 3 3 3 3 3 3 3 3 3 La Pampa 11 4 4 4 4 4 4 4 4 4 4 4 1 La Rioja 12 3 3 3 3 3 3 3 3 3 3 3 3 Mendoza 13 3-4 3-4 3-4 3-4 3-4 3-4 3-4 3-4 3-4 3-4 3-4 2 Misiones 14 3 3 3 3 3 3 3 3 3 3 3 3 Neuquén 15 1 1 1 1 1 1 1 1 1 1 1 1 Salta 16 3 3 3 3 3 3 3 3 3 3 3 3 San Juan 17 3 3 3 3 3 3 3 3 3 3 3 2 San Luís 18 3 3 3 3 3 3 3 3 3 3 3 4 Santa Cruz 19 1 1 1 1 1 1 1 1 1 1 1 1 Santa Fe 20 3 3 3 3 3 3 3 3 3 3 3 3 Santiago 21 3 3 3 3 3 3 3 3 3 3 3 3 Tierra del fuego 22 1 1 1 1 1 1 1 1 1 1 1 1 Tucumán 23 3 3 3 3 3 3 3 3 3 3 3 3
Continued
Table D.6: Summary of local Moran statistic as a measure of spatial association: real per capita GDP by quadrants, 1980-2002.
232 Table D.6 continued
Province ID 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 Buenos Aires 1 2 2 2 2 2 2 2 2 2 2 2 Catamarca 2 3 3 3 3 3 3 3 3 3 3 3 Chaco 3 3 3 3 3 3 3 3 3 3 3 3 Chubut 4 1 1 1 1 1 1 1 1 1 1 1 Corrientes 5 4 4 4 4 4 4 4 4 4 4 1 CBA 6 3 3 3 3 3 3 3 3 3 3 3 Córdoba 7 3 3 3 3 3 3 3 3 3 3 3 Entre Ríos 8 3 3 3 3 3 3 3 3 3 3 3 Formosa 9 3 3 3 3 3 3 3 3 3 3 3 Jujuy 10 3 3 3 3 3 3 3 3 3 3 3 La Pampa 11 1 1 1 1 1 1 1 1 1 1 1 La Rioja 12 3 3 3 3 3 3 3 3 3 3 3 Mendoza 13 2 2 2 2 2 2 2 2 2 2 2 Misiones 14 3 3 3 3 3 3 3 3 3 3 3 Neuquén 15 1 1 1 1 1 1 1 1 1 1 1 Salta 16 3 3 3 3 3 3 3 3 3 3 3 San Juan 17 2 2 2 2 2 2 2 2 2 2 3 San Luís 18 4 4 4 4 4 4 4 4 4 4 3 Santa Cruz 19 1 1 1 1 1 1 1 1 1 1 1 Santa Fe 20 3 3 3 3 3 3 3 3 3 3 3 Santiago 21 3 3 3 3 3 3 3 3 3 3 3 Tierra del fuego 22 1 1 1 1 1 1 1 1 1 1 1 Tucumán 23 3 3 3 3 3 3 3 3 3 3 3
233 40.00%
35.00%
30.00%
25.00%
20.00%
15.00%
10.00%
5.00%
0.00% 1980 1991 2002
Buenos Aires CBA Córdoba Santa Fe M endoza
Source: Author’s calculation using data from the INDEC.
Figure D.13: GDP participation of the five richest provinces in Argentina, 1980, 1991 and 2002.
234
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0 1980 1991 2002
Buenos Aires CBA Santa Fe Córdoba M endoza
Source: Author’s calculation using data from the INDEC.
Figure D.14: Population participation among the five largest regions.
235
60.0%
50.0%
40.0%
30.0%
20.0%
10.0%
0.0% CBA Pampeana Cuyo Norteast Northwest Patagonia
GDP 1980 GDP 2002
Source: Author’s calculation using data from the INDEC.
Figure D.15: Population shares of the five largest provinces in Argentina, 1980, 1991 and
2002.
236
4.00
3.50
3.00
2.50
2.00
1.50
1.00
0.50
0.00 CBA Salta Jujuy Chaco Chubut La Rioja La Córdoba Misiones Formosa Luís San Santa Fe Santa Neuquén Tucumán Mendoza San Juan San La Pampa La RíoNegro Entre Ríos Entre Corrientes Catamarca Santa Cruz Santa Buenos Aires Buenos Tierra del fuego del Tierra Santiago del Estero del Santiago
1980 1991 2002
Source: Author’s calculation using data from the INDEC.
Figure D.16: Comparison of provincial per capita income relative to the country’s average, for 1980, 1991 and 2002.
237
APPENDIX E
TABLES AND FIGURES FOR CHAPTER 4
238
Source: Bourguignon (2004).
Figure E.1: Decomposition of a change in distribution and poverty into growth and distributional effects.
239
Share of deciles Income ratios 1 2 3 4 5 6 7 8 9 10 10/1 90/10 95/80 EPH-15 cities 1992 1.8 3.0 4.1 5.1 6.2 7.6 9.4 12.0 16.5 34.1 19.0 7.9 2.0 1993 1.7 3.0 4.1 5.2 6.4 7.9 9.6 12.3 16.6 33.1 19.9 8.1 1.9 1994 1.7 2.9 4.0 5.1 6.3 7.7 9.5 12.1 16.4 34.2 19.7 8.2 1.9 1995 1.4 2.7 3.7 4.8 5.9 7.3 9.1 11.6 16.7 36.7 25.8 9.6 2.1 1996 1.4 2.6 3.6 4.7 5.9 7.3 9.2 11.9 17.0 36.5 26.5 10.1 2.0 1997 1.4 2.6 3.6 4.7 6.0 7.3 9.2 12.0 17.2 36.1 26.7 10.5 2.1 1998 1.2 2.4 3.4 4.5 5.7 7.0 9.0 12.0 17.1 37.7 30.2 11.2 2.1 EPH - 28 cities 1998 1.3 2.4 3.4 4.5 5.7 7.1 9.0 11.9 16.9 37.8 29.9 11.1 2.1 1999 1.3 2.5 3.5 4.6 5.8 7.3 9.2 12.0 17.0 36.8 28.0 10.9 2.1 2000 1.2 2.3 3.3 4.4 5.6 7.2 9.1 12.2 17.4 37.4 32.3 11.9 2.1 2001 1.0 2.1 3.1 4.1 5.4 6.9 9.0 12.0 17.4 39.0 40.0 13.9 2.2 2002 1.0 2.0 3.0 4.1 5.4 6.9 8.7 11.6 17.2 40.3 39.4 14.3 2.3 2003 1.1 2.1 3.0 4.0 5.2 6.8 8.8 11.9 17.3 39.8 34.8 13.5 2.2 EPH-C 2003-II 1.0 2.1 3.1 4.1 5.3 6.7 8.8 11.9 17.1 39.8 38.1 13.7 2.2 2004-I 1.2 2.3 3.3 4.3 5.5 7.1 9.0 11.9 16.8 38.6 32.7 11.8 2.1 2004-II 1.1 2.3 3.3 4.4 5.7 7.2 9.1 12.0 17.0 37.9 33.0 12.0 2.0 2005-I 1.2 2.4 3.4 4.4 5.7 7.3 9.1 11.9 16.9 37.8 32.5 11.7 2.1 2005-II 1.2 2.3 3.4 4.5 5.8 7.3 9.1 11.9 16.8 37.6 32.7 11.8 2.1
Note: Income distribution for the population in major urban cities of Argentina.
Source: Constructed by the author using Socio-Economic Database for Latin America and the Caribbean (CEDLAS and The World Bank).
Table E.1: Distribution of household per capita income in Argentina (deciles shares and income ratios), 1992-2005.
240
Gini Theil CV A(.5) A(1) A(2) E(0) E(2) EPH-15 cities 1992 0.450 0.370 1.101 0.165 0.299 0.510 0.355 0.606 1993 0.444 0.359 1.077 0.162 0.297 0.517 0.352 0.580 1994 0.453 0.378 1.112 0.168 0.303 0.510 0.361 0.618 1995 0.481 0.430 1.205 0.190 0.340 0.569 0.416 0.726 1996 0.486 0.442 1.260 0.194 0.349 0.607 0.429 0.793 1997 0.484 0.422 1.146 0.190 0.346 0.586 0.424 0.656 1998 0.502 0.471 1.300 0.207 0.369 0.608 0.461 0.845 EPH - 28 cities 1998 0.502 0.472 1.307 0.207 0.368 0.605 0.458 0.854 1999 0.491 0.443 1.213 0.197 0.356 0.606 0.440 0.735 2000 0.504 0.464 1.231 0.208 0.377 0.647 0.474 0.757 2001 0.522 0.497 1.264 0.224 0.404 0.675 0.517 0.798 2002 0.533 0.530 1.356 0.233 0.412 0.657 0.530 0.920 2003 0.528 0.519 1.343 0.227 0.401 0.637 0.512 0.902 EPH-C 2003-II (*) 0.537 0.625 3.056 0.244 0.417 0.673 0.539 4.671 2003-II 0.529 0.532 1.457 0.231 0.407 0.672 0.522 1.061 2004-I 0.510 0.507 1.714 0.216 0.380 0.621 0.478 1.469 2004-II 0.506 0.499 1.550 0.213 0.379 0.624 0.476 1.201 2005-I 0.502 0.473 1.306 0.208 0.373 0.624 0.466 0.853 2005-II 0.501 0.480 1.418 0.209 0.373 0.624 0.467 1.005
Note: (*) this calculation uses the EPH weights corresponding to the 28 major provincial cities. CV=coefficient of variation. A(e) refers to the Atkinson index with a CES function with parameter e. E(e) refers to the generalized entropy index with parameter e. E(1)=Theil.
Source: Constructed by the author using Socio-Economic Database for Latin America and the Caribbean (CEDLAS and The World Bank).
Table E.2: Inequality Indices from household surveys in major provincial cities in
Argentina, 1992-2005.
241
0.540
0.520
0.500
0.480
0.460
0.440
0.420
0.400 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003-II * 2003-II 2004-I 2004-II 2005-I 2005-II
Note: (*) this calculation uses the EPH weights corresponding to the 28 major provincial cities.
Source: Constructed by the author using EPH.
Figure E.2: Gini Coefficient for Argentina, from the distribution of per capita household income, 1992-2005.
242
Country 1950 1960 1970 1980 1990 Argentina 0.396 0.414 0.412 0.472 0.477 Bolivia 0.53 0.534 0.545 Brazil 0.57 0.571 0.571 0.573 Chile 0.482 0.474 0.531 0.547 Colombia 0.51 0.54 0.573 0.488 0.503 Costa Rica 0.5 0.445 0.485 0.46 Dominican Republic 0.455 0.421 0.481 El Salvador 0.424 0.465 0.484 0.505 Honduras 0.618 0.549 0.57 Mexico 0.55 0.606 0.579 0.509 0.531 Panama 0.5 0.584 0.475 0.563 Paraguay 0 0.451 0.57 Peru 0.61 0.485 0.43 0.464 Uruguay 0.37 0.428 0.436 0.406 Venezuela 0.613 0.462 0.48 0.447 0.459 LAC 4 0.505 0.532 0.531 0.491 0.507 LAC 6 0.548 0.548 0.532 0.542 LAC 15 0.539 0.519 0.532 Spain 0.457 0.363 0.347
Note: LAC 4 = population-weighted average of Brazil, Chile, Mexico and Venezuela. LAC 6 = population-weighted average of LAC 4 + Argentina and Uruguay. LAC 15 = population-weighted average of LAC 6 + Colombia, Cuba, Ecuador, Peru, Costa Rica, El Salvador, Guatemala, Honduras, and Panama.
Source: Constructed by the author using Perry (2006); Altimir (1987); Lodoño and Szekely (2000).
Table E.3: Inequality in Latin America between 1950 and 2000. Measured by Gini coefficients.
243
Change Change in in Gini Gini Country Period points Country Period points Argentina 1992-1998 0.05 El Salvador 1991-2003 -0.02 1998-2002 0.03 Honduras 1997-2003 0.01 2002-2004 -0.02 Jamaica 1990-1999 -0.02 1992-2004 0.06 1990-2002 0.02 Bolivia 1993-1997 0 Mexico 1992-1996 -0.02 1997-2002 0.03 1996-2002 -0.03 1993-2002 0.02 1992-2002 -0.04 Brazil 1990-1995 -0.01 Nicaragua 1993-1998 -0.02 1995-2003 -0.02 1998-2001 0 1990-2003 -0.03 1993-2001 -0.02 Chile 1990-1996 0 Panama 1995-2002 0.01 1996-2003 0 Paraguay 1997-2002 0.01 1990-2003 -0.01 Peru 1997-2002 0.01 Colombia 1992-2000 0.07 Uruguay 1989-1998 0.02 2000-2004 0 1998-2003 0.01 Costa Rica 1992-1997 0 Venezuela 1989-1995 0.04 1997-2003 0.04 1995-2003 0 1992-2003 0.04 1989-2000 0.02 Dominican Republic 2000-2004 -0.01 1989-2003 0.04 Ecuador 1994-1998 0.02
Source: Constructed by the author using Gasparini, Gutierréz and Tornarolli (2007).
Table E.4: Changes in inequality measured by percentage points of Gini Coefficient using household surveys in each country.
244 0.500
0.400
0.300
0.200
0.100
0.000
t a co A s s a ja za n lt ís o rca a bu e ío o o é Fe g Aires a h u CB nt R Jujuy Ri qu Sa Lu ta C rie e nd u n n fue s tam Ch r tr PampLa ta Cruz el no Córdoba Formosa a Me MisionesNe San JuanSa n Sa Tucumán e Co En L u Ca Sa ra d B r Tie Santiago del Estero
Source: Constructed by the author using EPH.
Figure E.3: Provincial Gini coefficients for Argentina. Averages for 1991-2002
245 0.490 0.484 0.482
0.480
0.469 0.470
0.460
0.454
0.450 0.445 0.442
0.440
0.430
0.420 Pampeana Cuyo NE NW Patagonia BA
Source: Constructed by the author using EPH.
Figure E.4: Regional Inequality in Argentina, as shown by Gini coefficients. Averages for 1991-2002.
246
Subset Region N 1 2 3 2 60 .44047753 6 48 .44218363 3 36 .45430459 .45430459 Tukey 5 HSD(a,b,c) 72 .46939949 .46939949 1 12 .48243993 4 48 .48440541 Sig. .418 .317 .323
Notes: Means for groups in homogeneous subsets are displayed. Based on Type III Sum of Squares. The error term is Mean Square(Error) = .001. a Uses Harmonic Mean Sample Size = 32.727. b The group sizes are unequal. The harmonic mean of the group sizes is used. Type I error levels are not guaranteed. c Alpha = .05. Regions: 1) Capital City, 2) Pampeana, 3) Cuyo, 4) Northwest, 5)Northeast and 6) Patagonia.
Table E.5: Bonferroni and the Tukey’s tests to determine means differ in Gini coefficient among regions in Argentina, 1991-2002.
247 Change Change Change Change Bottom Gini Q3 Top 10% 20% Region 91-02 91-02 91-02 91-02 Buenos Aires 17.06 -11.26 12.52 -44.32 Pampeana 19.28 -10.42 -2.59 -29.71 Cuyo 14.21 -6.23 12.45 -28.39 Northeast 15.23 -15.05 14.59 -33.54 Northwest 11.94 -7.49 12.32 -19.21 Patagonia 9.32 -12.07 9.77 -18.61 Argentina 16.22 -7.82 15.30 -35.56
Source: Constructed by the author using EPH.
Table E.6: Changes in Gini coefficient, third quantile (Q3), top 10 percent and bottom 20 percent shares in income of the population by region, between 1991 and 2002
(percentage).
248
0.655
0.555
0.455
0.355
0.255
0.155
0.055
-0.045 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002
Source: Author’s calculation using the EPH.
Figure E.5: Moran’s I statistic for the provincial Gini coefficients of Argentina, 1980- 2002.
249 Year Moran's I E(I) sd(I) z p-value* 1991 0.015 -0.045 0.164 0.372 0.71 1992 0.325 -0.045 0.164 2.25 0.024 1993 0.25 -0.045 0.169 1.75 0.08 1994 0.246 -0.045 0.167 1.746 0.081 1995 0.057 -0.045 0.168 0.608 0.543 1996 0.105 -0.045 0.168 0.894 0.371 1997 0.333 -0.045 0.169 2.244 0.025 1998 0.327 -0.045 0.168 2.221 0.026 1999 0.152 -0.045 0.166 1.188 0.235 2000 0.082 -0.045 0.16 0.795 0.427 2001 0.59 -0.045 0.167 3.801 0.000 2002 -0.114 -0.045 0.165 -0.417 0.676 *Two-tail test
Source: Author’s calculation using the EPH.
Table E.7: Estimates of the Moran’s I statistic for the provincial Gini coefficients of
Argentina, 1991-2002.
250
Moran scatterplot (Moran's I = 0.015) gini_91
1 9
7 16 21 10 14 3 2 23 0 19 20 1 17 13 8 12 11 18 6 4 22 -1 Wz 15
-2 5
-3 -3 -2 -1 0 1 2 z
Source: Author’s calculation using the EPH.
Figure E.6: Local Moran’s I statistic for the Gini coefficients provincial in 1991.
251
Moran scatterplot (Moran's I = -0.114) gini_02
2
9 10 7 1
1 21 16 20 0 19 2 23 3 17 15 618 1113 14 Wz 12 8 -1 5
-2 4 22
-3 -3 -2 -1 0 1 2 z
Source: Author’s calculation using the EPH.
Figure E.7: Local Moran’s I statistic for the Gini coefficients in 2002.
.
252 Moran scatterplot (Moran's I = 0.590) gini_01
2
9 10
1 14 7 16 3 23
20 21 1 0 2 8 6 15 18 17 13 12 Wz 11 19 -1 5
-2 22 4
-3 -3 -2 -1 0 1 2 z
Source: Author’s calculation using the EPH.
Figure E.8: Local Moran’s I statistic for the Gini coefficients provincial in 2001.
253