Essays on Development Economics

By Paola Andrea Barrientos Quiroga

A PhD thesis submitted to

School of Business and Social Sciences, Aarhus University,

in partial fulfilment of the requirements of

the PhD degree in

Economics and Business

August 2013

CONTENTS

Preface ii

Summary iv

Chapter 1

Convergence Clubs determined by Economic History in Latin America 1

Chapter 2

External Flows and Development in Bolivia 39

Chapter 3

Income Convergence and the Flow out of Poverty in India, 1994-2005 81

i

PREFACE

This dissertation was written during my PhD enrolment at Aarhus University at the De-

partment of Economics and Business. I am grateful to the Department of Economics and

Business for giving me the opportunity to do research in the topics that I am interested in

and for providing me the excellent facilities and environment for this purpose.

I would like to thank my main supervisor, Prof. Bent Jesper Christensen for his guidance

and support through all this process and to my second supervisor, Prof. Martin Paldam,

from whom I have learnt very much and with whom I have had discussions which became

the main inspiration for this dissertation.

I am thankful to have worked with my co-authors, Prof. Nabanita Datta Gupta and

Prof. Niels-Hugo Blunch because their insights and discussions were very enlightening.

I am specially grateful to Kenneth D. Petersen, my husband and co-author, for his great

contribution to a chapter in this dissertation and for his comments, suggestions and support

through all this process.

During the second semester of 2010, I visited the Department of Economics at Cornell

University and I would like express my gratitude for their hospitality during my stay. I

appreciate very much the great discussions I had with Prof. Ravi Kanbur, Prof. Eswar

Prasad and Prof. Gary Fields.

During the first semester of 2011, I also visited The Institute for Advanced Studies in

Bolivia. My special thanks to Osvaldo Nina, Lykke Andersen for their hospitality, and to

Gustavo Machicado, Walter Valdivida and Pablo Selaya for their excellent comments on my

work. I also benefited greatly from the guidance of Tatiana M. Quiroga.

ii

I enjoyed very much the nice environment at Aarhus University, especially to have shared

office with Torben, Jonas, Nisar, Juan Carlos, Sanni, and Jannie, and to have shared nice

times at the department with Firew, Zhenjian, Anders, Paolo, Mette, Christian and to

Johannes T.K. extra credit for his invaluable help with latex.

Finally and most importantly, I am infinitely grateful for the unlimited love from my

children, Fraya and Matias, my husband, my parents, my sister, Mafe, Ulla, my parents in

law and Daniel.

Thank you all very much.

Paola Andrea Barrientos Quiroga

Aarhus, April 2013

UPDATED PREFACE

The pre-defense took place the 14th of June in Aarhus. I want to express my gratitude to

the members of the assessment committee, Frederic Warzynski, Ingvild Alms and Ingrid

Henriksen for their very useful comments and suggestions.The present version of the disser-

tation benefited greatly from their insights.

Paola Andrea Barrientos Quiroga

Aarhus, August 2013

iii

SUMMARY

This dissertation studies the dynamics and persistence of income inequalities in the devel-

oping world. I am interested in answering the question, Will the poor remain poor? This

question is very broad and has many dimensions, ranging from theoretical to empirical as-

pects, from a macroeconomic viewpoint - at country level, to a microeconomic viewpoint -

at a family level, and from policy perspectives, as to what can be done to overcome poverty.

This dissertation covers many of these dimensions in three self-contained chapters. First,

it asks whether the poor will remain poor in a country-level setting, by looking into GDP

convergence among Latin American countries. Then, it asks if the poor will remain poor, in

a family-level setting, by analysing household income convergence and flows out of poverty

in rural India. Finally, it asks if the external mechanisms of overcoming poverty have worked

for Bolivia, by looking into the role of external financing in determining GDP. The results

entail important policy implications and also lay groundwork for future research.

The first chapter deals with club convergence in Latin America. Despite the fact that

Latin America is considered a club itself due to many common characteristics, such as

language, geography, religion and history, it still exhibits differences across countries and

actually the discrepancies are growing in time. The concept of club convergence has been

widely used to group countries in clubs with similar development paths. However, there is

no unified agreement on how to identify the clubs in the first place. In this chapter, I argue

that economic history can guide us to identify clubs. The argument is that economic history

helps us understand when, where, and how institutions are formed, and since institutions

determine the way scarce resources are used by their chosen policies, it allows us to under-

iv

stand economic growth. I study a period of more than 100 years, from 1900 to 2007, where

first, I identify two main common external shocks to the region: the Great Depression in the

1930s and the oil price shock in 1974. Second, I classify countries in clubs according, first,

to their natural resources endowments, and then, after each shock, to their policy-response

to the shocks. Lastly, I test convergence within each club. I find significant and positive

convergence speeds within each club, implying that this way of identifying clubs should not

be ruled out.

The message with respect to whether the poor will remain poor is that poor countries

will not stay poor, since I find convergence dynamics within each club. Regarding policy

implications, I find that the clubs to which countries appertain, are determined by policy

makers but also by external shocks and natural resources endowments. I cannot conclude

that one club is superior than another, because successful countries belong to different clubs.

Therefore, I cannot suggest how to jump to a superior club, as is suggested in a traditional

approach, where clubs are defined by income or capital thresholds which imply that a

significant transfer of money would help a country jump to a superior club. However, I do

believe that external transfers may have a potential to help a country develop. Therefore

the next chapter explicitly tests for this, for the Bolivian economy.

The second chapter, co-authored with Kenneth Dencker Petersen, examines the empiri-

cal impact of the most important external resources on the Bolivian economy for the period

1950-2009, where external resources are understood as FDI, private and public loans, and

aid. This is a complicated task since there are many inter-relations to take into account.

One is the endogeneity between the flows and the Bolivian GDP. Another is the relation

between the flows, which are determined by sender countries characteristics and the fact

that the flows themselves can complement or substitute each other. A further challenge is

to distinguish between the long-run from the short-run relations. We take all these relations

into account by using a CVAR(2) model and letting the data tell us how the variables be-

have for the Bolivian economy. Our results indicate that loans are positive for the Bolivian

economy in the long run, through an increase in exports and a decrease in purchasing power

v

of exports, however, we did not include accumulated debt. The effect of aid is negative for

the Bolivian GDP per capita, which is explained by the temporary increase in purchasing

power of exports, which lead to a permanent loss in the export markets. Foreign direct

investment instead, was found to be exogenous to the system. The effects of investments

lead to a positive response of Bolivian GDP per capita and exports and a reduction of

purchasing power of exports.

The message is that certain external resources have the potential to overcome poverty

in terms of increasing GDP per capita in Bolivia, especially FDI. Regarding policy impli-

cations, we suggest that policy makers in Bolivia should attract FDI to productive sectors

by providing the necessary incentives and property rights, and that aid changes its nature

since the way it is currently works is harmful for the economy. We suggest that aid should

be directed into more productive and lasting sectors. For example by promoting Bolivian

exports through the opening of donors markets instead of adding trade barriers.

The last chapter, co-authored with Nabanita Datta Gupta and Niels-Hugo Blunch,

explores the dynamics of the income situation and poverty status of rural Indian households

over the period 1994 to 2005. The estimation strategy consists of a convergence analysis to

test whether poor households are catching-up with rich ones in terms of income, followed by

a transition analysis to test whether poor households are exiting poverty faster than staying

in it. The identification strategy explicitly addresses issues pertaining to the potential

endogeneity and measurement error of initial income and poverty. We find evidence of both

income convergence and poverty reduction: poorer households are catching up to richer

households over time and the bottom of the distribution is being lifted up. We find that

the key variables are education and asset ownership.

The message here is that the poor families will not stay poor. One policy recommen-

dation would be to provide access to productive assets to families and at the same time

increase public expenditures on education in rural areas, perhaps emphasizing training of

usage of such assets.

vi

1

Convergence Clubs determined by

Economic History in Latin America

Paola A. Barrientos Quiroga

Abstract

The concept of club convergence has been widely used in empirical analysis to

group countries in clubs with similar development paths. However, there is no

unified agreement on how to identify the clubs in the first place. In this paper, I

argue that economic history can guide us to identify clubs. The argument is that

economic history helps us understand when, where, and how institutions are formed

and since institutions determine the way scarce resources are used by their chosen

policies, it allows us to understand economic growth. Even though Latin America is

typically considered a club itself, due to common characteristics, such as language,

geography, religion and history, it still exhibits differences across countries. I study

a period of more than 100 years, from 1900 to 2007, where first, I identify two main

common external shocks to the region: the Great Depression in the 1930s and the

oil price shock in 1974. Second, I classify countries in clubs according, first, to their

natural resources endowments, and then, after each shock, to their policy-response

to the shocks. Lastly, I test convergence within each club. I find significant and

positive convergence speed within each of the clubs, implying that this way of

finding clubs should not be ruled out.

2

1 Introduction

The detection of income disparities across clubs of economies can help determine how to speed up

the process of economic development and understand the sources of differences in growth

performances. In theory, the reasons behind club convergence could be several, among these: the

existence of some threshold level in the endowment of strategic factors of production, non-

convexities or increasing returns, similarities in preferences and technologies, and government

policies and institutions (Canova, 2004; and Azariadis, 1996). Empirically, there is no unified

agreement on how to identify clubs. Most researchers (e.g. Durlauf and Johnson, 1995; Paap and

Van Dijk, 1998; Desdoigts, 1998; Hansen, 2000; Canova, 2004; Owen, et al., 2009) lean towards

the approach of letting the data decide the clubs. They usually study the shape of the distribution of

income (or capital) and focus on finding an income (or capital) threshold to divide countries into

clubs, or the thresholds are determined beforehand. However, the division of clubs by income (or

capital) is not very informative with respect to the forces behind the heterogeneity in income (or

capital) in first place.

Although Latin America is typically considered a club itself, due to its common characteristics,

such as language, geography, religion, history and policies, it exhibits differences across countries

(see Figure 1). Dispersion in GDP per capita has been increasing on average over the period 1950-

2005 in Latin America, whereas it has been decreasing among the OECD countries. Then a relevant

question is, Why diversity in growth trajectories in a region with so many common roots? Some

candidates for an explanation come to mind: commodity lottery/geography, poor market integration,

colonial heritage, and differences in economic policies, among others.

Some researchers have gone far in time to explain the diversity in development paths in Latin

America and the connection to institutions. Acemoglu, Johnson and Robinson (2001) find that there

is a strong correlation between early institutions and institutions today. In the specific case of the

Americas they distinguish between regions that were settled by Europeans and regions that, due to

high settler mortality, the Europeans established “extractive states” instead. The latter model paved

the way for extractive states even after political independence in the nineteenth century. In a later

work, Acemoglu and Robinson (2012, pp. 114-115) restate their point that the extractive political

3

and economic institutions of the conquistadors have endured and condemned much of the region to

poverty. There are, however exceptions. Argentina and Chile have fared better than most. Because

they had few indigenous people or mineral riches (exploitable at the time) they were “neglected” by

the Spanish. Consequently, there are differences even in this dismal picture of colonial heritage.

Similarly, Engermann and Sokoloff (2002) argue that institutions are endogenous, and that the roots

of the disparities in the extent of the inequality that we observe today lay in the initial factor

endowments. Through comparative studies of suffrage, public land and schooling policies they

document systematic patterns by which the societies in the Americas, that began with more extreme

inequality or heterogeneity in the population were more likely to develop institutional structures that

greatly advantaged members of the elite by providing them with more political influence and access

to economic opportunities.

Figure 1. GDP per capita dispersion in the World, OECD, Latin America and eight Latin American countries

(LA8 - Argentina, Brazil, Chile, Colombia, Mexico Peru, Uruguay and Venezuela). Standard deviation of the

logarithm of GDP per capita.

Empirical research on convergence in Latin America is still scarce compared to other regions1,

and only one other study incorporates economic history features into the analysis: Astorga et al.

(2005). They do a time-series analysis for each of their six countries of study, during 1900-2000,

1 There are only nine cross-country studies on convergence: Blyde, 2005 and 2006; Holmes, 2005; Astorga et. al., 2005; Dobson and Ramlogan, 2002a and 2002b; Utrera, 1999; Dabus and Zinni, 2005; and Madariaga et.al, 2003.

.2

.4

.6

.8

1

1.2

Sigma

1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005year

LA 8 OECDWorld LA

4

where they find different breaks for each of them. They conclude that there are two external shocks

affecting the six economies simultaneously: in the 1930s due to the Great Depression and in the

1980s due to the shift in US monetary policy and debt crises2. They find convergence between the

six countries by using panel data and error correction models, and conclude, among other things,

that there is convergence among the six countries but it seems there is divergence between the rest,

forming two distinct convergence clubs (they do not test for this). They say that the "rest" show an

inferior pattern of growth compared to the six, due to lower growth rate and greater vulnerability,

which may possibly relate their greater vulnerability to the external shocks.

This study pretends as well, to answer the question posted earlier of why diversity in growth

trajectories in a region with common roots and fill in the gap in the literature by analyzing,

empirically, the existence of club convergence in 32 Latin American countries over more than 100

years, more than any other study.

I analyze club formation in a different way than the conventional procedures (income or capital

threshold determination) and incorporate the institution link to growth. I examine sources of

heterogeneity on the basis of economic history, which informs us on the different initial

endowments, the most important common external shocks and the policy-responses. Such historical

events shape the way institutions are formed, and institutions determine the way scarce resources

are used. As mentioned by Easterly (2003), technology is endogenous to the institutions that make

adoption of better production techniques likely. The institution link in terms of Acemoglu and

Robinson (2012) and Engerman and Sokoloff (2002) enters in the analysis when I, first, divide the

clubs according to their factor endowments. However, after the external shocks hit the region, a mix

of countries of different endowments try to change their pattern of development while others remain

in the same path. In other words, I recognize that there is a legacy from the colonizers but I also

accept that external shocks and certain circumstances can also change this legacy by the decision

and possibilities of the policy makers.

The criteria of division of clubs follow three steps. First, I identify the main common external

shocks that changed the development patterns in the region: the Great Depression in 1930s and the

2 However, from my point of view, the shift in US monetary policy and debt crises are consequences of the exogenous

shock of the increase in oil prices in 1974.

5

oil price shock in 1974. Then, I classify countries in clubs according to first, their initial endowment

of natural resources and then according to their policy-response to the shocks. Here, I focus on

information of the policies rather than outcomes so that the results are not driven by the selection of

the club thresholds in the first place. Finally, I test for convergence within each club.

Before 1930, I define two clubs according to their exporting product: the mineral and

agricultural producers. After the Great Depression I follow Diaz (1984) classification of clubs,

according to passive or reactive countries, where the reactive responded autonomously to protect

themselves, while the passive did/could not. After the oil price shock, I classify the clubs according

to the Lora index (Lora, 2001), which describes to what extent countries applied structural reforms

to liberalize their economies. I also have the Caribbean countries as a separate club.

In connection to growth theory, under multiple-equilibria models, each historical watershed

represents an opportunity to modify the set of initial conditions and to escape from a development

trap, whereas in the Solow model approach, those historical points represent critical changes in

policy parameters and a redefinition of the steady state. The empirical analysis cannot distinguish

between these two kinds of models.

The following section discusses the background of the paper. First it discusses the theoretical

aspects behind club convergence, then it summarizes the prior research in Latin America, and

finally it reviews the common economic history events in the region. Section 3 describes the

empirical specification of the paper, which consists of the division of clubs and the econometric

specification. Section 4 presents the results and Section 5 discusses the strengths and weaknesses of

the approach. Finally, I present the conclusions. The data details are presented in the appendix.

2 Background

2.1 Connection to theory: from the Solow-Swan model to multiple

equilibrium models

The concept of economic convergence has been discussed through many years since Ramsey

(1928) until now. This section does not pretend to do an exhaustive summary and analysis of all

6

growth theories3, but instead it discusses and compares, in general terms and briefly, the two most

important theories behind club convergence that emerge from the most basic version of Solow-

Swan model and the multiple equilibria models.

The neoclassical growth models, of which the simplest version is Ramsey-Solow model, arrive

at a growth equation where convergence can be estimated by4,

���� . log(���/���) = � − ��������� ∙ log(��∗/���) + ��� (1)

where the average growth rate in the interval from 0 to T for country i, ���� . log(���/���), is

related negatively to the initial output ��� in relation to the steady state output ��∗, and positively to

the technology growth x, while keeping β (speed of convergence) and T (period) constant.

Equation (1) describes conditional β-convergence in the sense that a poor country A will grow

faster than rich country B, understanding that country A is poorer because it is further away from its

own steady-state than country B is. In contrast, absolute β-convergence assumes that ��∗is the same

for country A and B, ��∗ = �∗. We cannot know exactly what the steady state output looks like, but

we know it is related to structural characteristics such as technologies, preferences, propensity to

save, institutions, policies, etc.

Empirically, one can estimate β-conditional convergence by finding proxies for the steady state5

or by grouping economies that we assume have the same steady state (Sala-i-Martin, 1996). So, if

we gather countries that have or we expect to have similar steady-states we are finding (or testing

for) convergence in different groups, namely group convergence.

On the other hand, the multiple equilibria models starting with Azariadis and Drazen (1990)

(and followed by many others, see Durlauf and Quah, 1999) advocate for multiple regime in which

different economies obey different linear models when grouped according to different initial

conditions. For example, there exists a range of human or physical capital levels over which the

3 Durlauf and Quah (1999) offer a great summary of economic growth theories and empirics

4 Basically, from a Cobb-Douglas production function for the economy, and a Utility function for a representative agent, the

economy will eventually arrive to a steady state, where the economy cannot grow anymore. Equation (1) is the resulting

equation after optimization and log-linearlization (see Barro and Sala-i-Martin (2004) for the derivations). 5 However, the problem of adding controls as proxies for the steady-state, is that these will probably be endogenous

(Durlaug and Quah, 1999).

7

aggregated production function is not concave which will lead to different long-run steady-states. In

this way, initial conditions can “trap” countries into not reaching the rich countries.

Azariadis (1996) and Canova (2004) suggest that the potential causes of traps are several, like

technologies, preferences, market structures, fertility patterns and public policies. These variables

preserve and augment initial inequality in per capita income among otherwise identical national

economies. This concept is called club convergence.

Unfortunately, economic theory does not guide us on the number of clubs or the way in which

the different variables defining initial conditions interact in determining the clubs. To address this

issue, most researchers (e.g. Durlauf and Johnson, 1995; Bai, 1997; Hansen, 2000; Pesaran, 2006;

Paap van Dijk,1998; and Desdoigts, 1998) lean towards the approach of letting the data decide the

clubs. They usually study the shape of the distribution of income per capita and focus on finding an

income threshold to divide clubs; however they may not be able to explain the differences in income

or capital in first place.

The difference between group convergence and club convergence lies on their assumption about

stratification. Group convergence assumes the stratification is due to different steady states, while

club convergence assumes that the stratification comes from interactions on initial conditions with

different variables. Empirically, both can be estimated from Equation (1).

2.2 Prior research in Latin America

In Latin America, there are only nine cross-country empirical studies6 on convergence (Blyde,

2005 and 2006; Holmes, 2005; Astorga, et.al 2005; Dobson and Ramlogan, 2002a and 2002b;

Utrera, 1999; Dabus and Zinni, 2005; and Madariaga et.al,2003). Although they analyze the same

region, they study different countries and periods, and apply different methodologies.

Some of the authors use methodologies that do not measure a specific speed of convergence,

such as Blyde (2006), who studies 21 countries during 1960-2004 and uses a distribution dynamics

approach. He finds that countries are converging to two clubs; one large for low and low-middle

6 The number of studies within a given country is higher than across countries, and usually concentrated in few countries,

such as Chile, Argentina, Brazil and Colombia (e.g. Marina (2001), Azzoni et al.(2001), Anriquez and Fuentes (2001),

Cardenas and Ponton (1995), Magalhaes, Hewings and Azzoni (2005), Serra et al.(2006)).

8

income countries and another small for rich-income countries. The high-income countries are

Uruguay, Argentina, Chile, and Mexico, and the remaining 17 countries are in the other club.

Dobson, Goddard and Ramoglan (2003) study the case of 24 countries during 1965-1998 by

using cross-section analysis and unit root with panel data tests, and find convergence but not a

specific speed nor different clubs. Other researchers find concrete results but no clubs. For

example, Dobson and Ramlogan (2002a and b) study 19 countries and 28 and 30 years, respectively

(1970-1998 and 1960-1990), using cross-section regression and panel data analysis, and find speeds

of convergence of 0.02% to 2%7. Helliwell (1992) analyzes 18 Latin American countries over the

period 1960-1985 and finds convergence at a speed of 2.5%8.

The only other study that incorporates economic history features into their analysis is Astorga et

al., 2005. They first do a time series analysis for each of their six countries of study, during 1900-

2000, where they find different breaks for each of them. They conclude that there are two external

shocks affecting the six economies simultaneously: in the 1930s due to the Great Depression and in

1980s due to the shift in US monetary policy and debt crises9. Later, they find convergence between

the six using panel data and error correction models, at a speed between 1% and 1.9%, where the

oscillation comes from the addition or subtraction of explicative variables that proxy for the steady

state10. They conclude, among other things, that there is convergence among the six countries but it

seems there is divergence between the rest, forming two distinct convergence clubs (they do not test

for this). They say that the "rest" show an inferior pattern of growth compared to the six, due to

lower growth rate and greater vulnerability, which may possibly relate their greater vulnerability to

the external shocks.

In stark contrast to these findings of relatively low speeds of convergence, Dabus and Zinni

(2005) analyze 23 countries from 1960 to 1998, and find absolute and very high conditional

convergence rates. The authors argue that once controls are introduced and extremely high speeds of

7 Their studies include, as proxies for the steady state, sectorial decomposition variables, country dummies, population

growth, savings, and human capital. 8 He includes variables such as investments, population growth, human capital, and scale effects.

9 However, from my point of view, the shift in US monetary policy and debt crises are consequences the exogenous shock

of the increase in oil prices in 1974. 10

They include human capital, external, institutional, and economic variables, together with dummy variables related to

external events, such as the Great Depression and the Debt Crises. The countries are Argentina, Brazil, Chile, Colombia,

Mexico, and Venezuela.

9

conditional convergence are found, compared to absolute convergence, then it is a signal of

divergence. This is a good point, since when controlling for many characteristics, a hypothetical

speed of convergence is being calculated, while the real speed of convergence would be the one

closest to absolute convergence11. They conclude that convergence of any type is absent in Latin

America.

2.3 Economic history of the region

To analyze the economic history of 32 countries during more than 100 years is a complicated

task, and even more so when one wants to focus on the common factors of the region as a whole

rather than country specific sets of events. Historians face this task, and one of the main references

for my analysis in this section is Thorp (1998), who captures in depth the comparative reality within

Latin America.

Below, I describe the common events and focus on two very important external shocks that have

changed the development patterns in the region: the Great Depression of 1930 and the oil crisis in

1974. The first shock changed the political economy of the region, and as a result many of the

countries underwent a process of import substitution industrialization. The second shock, too,

changed the political economy of the region, resulting in a debt crisis, to which the response in

many cases was to adopt structural reforms to liberalize the economies. Thus, the pattern changed

from initially exporting, to substituting imports with the state playing an important role, and finally

liberalizing and a lowering the role of the state.

1900-1930: The Exporting Phase

There is no doubt that during the first phase of the 20th century the economics of the region was

characterized by being dependent on exports, which were primary products with low value added.

The region was vulnerable to world income and to fluctuations in primary products prices.

The first phase is characterized by the world export demand being high and the capital flows

being fluent to the region. These two facts determined the way Latin America developed. The

11

In this regard, Durlauf and Quah (1999) mention that the choice of the steady-state proxies depends on the interest of

the researcher and this can lead to wrong results.

10

region exported the needed primary products and at the same time imported more elaborated goods

produced in the "center".

WWI (1914-1918) accelerated the shift in trade and investment structures in the region. The

demand for Latin America’s exports increased, and according to Furtado (1981) the war stimulated

the industrial growth in the region, especially in the mineral countries. The economic pattern and the

political economy did not change after WWI, but they did when the Great Depression hit the region.

1930: The Great Depression

In 1929 the US stock market crashed and provoked a fall in economic activity in the

industrialized countries, which in turn reduced their demand for primary products and reversed the

capital flows to Latin America. This situation deteriorated the terms of trade of all primary products,

leading to an increase of the Latin American real import prices. The natural mechanism of

adjustment is a decrease in real export prices so that demand is stimulated again, but due to the

extreme circumstances of the Great Depression, world demand could not recover. Instead, Latin

American demand went from imported manufactured goods to domestic manufactured products.

This process stimulated the import substitution phase of Latin America. Cardoso and Helwege

(1992) call it "import substitution by default".

The process of industrialization via import substitution was reinforced by WWII (1939-1945).

Although WWII brought an increase of Latin American exports, there were constraints on imports.

Consequently, the scarcity of imports and the deterioration of terms of trade of primary products

encouraged new efforts to substitute imports, but these efforts were in turn limited by scarcity of

imported inputs and capital goods. National governments promoted industries and restricted

imports, mainly by lowering interest rates, giving easy credits, and controlling prices. Capital

inflows were attracted through loans to the public sector. Moreover, governments applied multiple

exchange rates, protective tariffs, import licenses, and different import quotas that could favor the

essential goods imports and reduce final goods imports.

As a result of the protection of the national markets, the exporting sectors in many countries in

Latin America were discouraged due to high cost of domestic intermediate products, and the

restriction on imports demand overvalued the exchange rates, making prices less competitive.

11

Moreover, fiscal revenues from the commodity product sector went down and public spending rose,

creating a fiscal gap, which in some cases was monetized and later created persistent inflation. The

result was detrimental for sectors that were not intensive in capital, like the agricultural sectors and

the artisans. Finally, the low interest rates given by the government to promote investments

discouraged saving even when helping inefficient firms and corruption increased greatly. However,

for those countries where industrialization was strong, innovations were made in terms of

organization, technology and R&D (together with investment in education), like in Brazil,

Argentina and Mexico. Another positive side was that some enterprises were ready to export.

Overall, more manufactured goods were produced.

1974: Oil Price Shock

Later, in 1974, the shock of the increase in oil prices led Latin America to become highly

indebted, which led to a debt crises in the region. The mechanism is described by Cardoso &

Helwege (1992) as follows: "..Oil exporters deposited their earnings in the commercial banks of

developed countries, but higher oil prices caused a recession in OECD countries and reduced the

demand for credit. Left with excessive liquidity bankers eagerly lent to the Third World at very low

interest rates.." .

The debt crises started in 1979 and 1981 when the Unites States and other OECD countries kept

their money supply tight and increased interest rates radically. Since countries acquired loans at

floating interest rates, their debt obligations increased very much12. The adjustment of the debt

crises was costly for all countries in the region, mainly due to the massive capital outflow.

Governments were not able to continue their policies and had to make drastic changes. In general,

governments printed more money to cover or keep their fiscal deficits constant. With all the

borrowed money, governments were used to spending more than their incomes. Since printing

money can cause inflation pressures and damage real wages, some governments indexed the

nominal wages to prices to keep real wages constant. Speculators, trying to earn from the

indexation, raised prices at higher rates than salaries. Sooner or later inflation exploded into

hyperinflation and governments were no longer able to manage it.

12

The average real interest rate on LDC debt rose from -6% in 1981 to 14.6% in 1982 (Thorp, 1998).

12

Countries were desperate to stabilize and gain access to foreign credit again, and the

"Washington Consensus policy package" was an option to reach stabilization. The package was a

set of structural reforms to liberalize the economy. The specific policies were to cut budget deficits

(by reducing expenses and increasing taxes), privatize, liberalize imports, impose exchange controls

(devaluate), eliminate price controls (to reflect the real costs), and increase interest rates (Cardoso &

Helwege, 1992). Some countries took the package as such, and others took some elements of it.

However, in general the adjustment left behind common problems that reinforced each other, such

as capital outflows, fiscal deficits, inflation, overvaluation, and balance of payment crises.

Later, in the 90’s, some trends of thought support the idea that good institutions create

complementarities between productivity growth and equality. Others maintain that policies that are

linked to the political constituency will create a combination of economic and social development.

When the population participates in the process of making decisions, the feeling of ownership helps

to monitor and accomplish their obligations better. Thorp calls these new currents "the New

Paradigm Shift", which started by the mid 1990s, as a response to the poor welfare results. Thorp

points out that the rise of the paradigm shift is a result of the increasing capital flows, the debt

crises, and the costly adjustment process. However, it is hard to attribute the results to either

globalization or policy shifts alone.

3 Empirical Specification

The previous section described how the political economy changed from initially exporting, to

substituting imports with a great role played by the state, and finally to liberalizing and a

diminishing the role of the state. These changes are clearly radical, and according to multiple

equilibria models, each historical watershed is an opportunity to modify the set of initial conditions

and to escape from a development trap and according to Solow-Swan model, each political change

will result in different steady-states.

I focus on a criterion to divide countries into clubs that describe the initial conditions after the

shocks, as under the multiple-equilibria models. The criterion is based on the policies adopted at the

beginning of each phase, as a response to the shock. I focus on information of policies rather than of

13

outcomes so the results are not driven by the selection of the club thresholds in the first place13. I

explain first the club division and later the econometric specification.

3.1 Division of Clubs

Mineral and Agricultural Clubs: 1900-1930

For the first phase, the initial conditions are defined in terms of type of natural resource

endowment. Due to lack of data in this phase, I divide countries into groups according to mineral vs.

agricultural countries, rather than a more extensive type of division by product.

Agricultural countries’ production was vulnerable to natural disasters, and minerals were

vulnerable to recessions in the "center", because minerals were used in construction, machinery, and

chemicals production. Moreover, the two types of production had different spillovers. For instance,

the mining sector was characterized by using less land and labor with more capital and

technological intensity, and having different transport needs than the agricultural sector. Acemoglu

et.al (2001) also points out that the mining countries set more extractive institutions.

The agricultural countries are: Brazil, Colombia, El Salvador, Nicaragua, Costa Rica,

Guatemala, Honduras, Ecuador, Cuba, Argentina, and Uruguay. They were mainly producing

coffee, bananas, cacao, sugar, meat, and/or wheat. Those mainly producing coffee were Brazil,

Colombia14, El Salvador and Nicaragua. Costa Rica and Guatemala were mainly producing coffee

and bananas, while Honduras was producing bananas and precious metals. Cuba mainly produced

sugar, but also tobacco. Argentina and Uruguay were mainly producing meat and wheat.

The mineral countries numbered four: Chile, Mexico, Peru, and Venezuela. They exported

mainly petroleum and copper. Petroleum was produced by all except Chile, and copper was

produced by all except Venezuela. Before 1917, Venezuela was mainly producing coffee and cacao,

but after that year petroleum became the most important source of revenue15. Mexico was the most

13

In Barrientos (2010), I actually analyzed different clubs in the region based on the outcomes rather than the policies. 14

Colombia also exported gold (Antioquia region) besides coffee but I keep it in the agricultural group because coffee has

been more traditional. 15

It is debatable whether Venezuela is among the mineral countries, since its oil was discovered more in the middle than at

the beginning of the phase. Still, I decided to keep it in the mineral club, because since its oil discovery, Venezuela has been

dependent on its petroleum.

14

diversified export country in Latin America, also exporting lead, zinc, silver, gold, coffee, rubber,

and cotton. It discovered its oil in 1910.

Reactive and Passive Clubs : 1931-1974

After the onset of the Great Depression in 1930, countries responded in different ways. Díaz

(1984) divides countries into reactive and passive. The reactive countries had policy autonomy in

the sense that they could, for example, depreciate their exchange rate and thereby speed up the

relative price adjustment to recover faster, while the passive countries had to stay tied to the dollar.

Also monetary and fiscal policies were employed. Some countries were not included in Díaz (1984),

so I use Taylor (1999) to complete the clubs. Those countries that did some sort of exchange rate

control and market activity control were included in the reactive club16.

Díaz (1984) classified as reactive countries: Argentina, Brazil, Chile, Colombia, Mexico, Peru,

and Uruguay. I added to this group six countries that were not mentioned by Diaz (1984) but by

Taylor (1999): Bolivia, Costa Rica, Nicaragua, Paraguay, and Venezuela. According to Table 3 in

Taylor (1999), these countries exerted some sort of exchange rate control and/or some sort of

control of capital market activity.

Diaz (1984) has the following passive countries: Cuba, Dominican Republic, Honduras, Haiti,

Panama, and Puerto Rico. I added Ecuador, Guatemala, and El Salvador, following Taylor (1999).

Low and High Reformers Clubs: 1975-2007

After the oil prices shock in 1974, countries went into debt crises, and the policy decision was

whether to follow the structural reforms proposed by the Washington consensus or not. The change

in policies was very radical in the region. Many countries went from protection of national markets

and great control by the state to policies that facilitate the operation of markets and reduction of the

distorting effects of state intervention in economic activities. Lora (2012) develops an index that

tries to capture how deep the reforms went (rather than outcomes). The higher the index, the more

market friendly the reforms. The index summarizes the status of progress in policies within trade,

financial, tax, privatization, and labor areas. By using the Lora index, I classify countries into two

16

During the second phase, a natural way of dividing clubs seems to be according to whether countries were industrialized

or not. However, this approach would divide clubs by result more than policy, and would not reflect the initial condition for

the phase, so I rule out this possibility.

15

groups: the high reformers, whose indices are above the average, and the low reformers, whose

indices are below average17.

According to the Lora index the high-reformers are: Argentina, Bolivia, Chile, Panama,

Paraguay, and Uruguay. I added to this group Panama and Puerto Rico, given that both have close

relations with USA who promoted the Washington Consensus package.

According to Lora's index, the low-reformers are: Brazil, Colombia, Costa Rica, Dominican

Republic, Ecuador, Guatemala, Honduras, Mexico, Nicaragua, Peru, and El Salvador. I added Cuba

for obvious reasons.

Caribbean Club

Finally, the Caribbean countries are treated as one club, due to its own characteristics. They are

small, dependent on USA, and are characterized by their vulnerability to capital flight and

international interest rate changes. They are quite open18 and primary products producers.

Additionally, Caribbean countries are exposed to natural disasters. I include the Caribbean club in

each phase except the first due to lack of data.

The Caribbean group consists of many islands and English speaking countries, mainly part of

the trade union CARICOM: The Bahamas, Barbados, Belize, Dominica, Grenada, Guyana, Haiti,

Jamaica, Saint Lucia, St. Kitts and Nevis, St. Vincent and the Grenadines, and Trinidad and Tobago.

3.2 Econometric Specification

The model setup follows Barro and Sala-i-Martin (2004), where the main interest is in

measuring the non-linear relation between initial output and growth. Although the setup is

developed for neoclassical growth models, it is used when measuring club convergence as well.

I start from the absolute convergence definition:

��� = − �����!"#" ∙ ��$" + ��� (2)

17

The Lora index of structural reforms is taken from the year 1985, while phase 3 starts in 1974. I considered starting the

last phase in 1985, but then I would be inconsistent with the previous phase, where the break was determined by an

external shock. One could argue that the increase in US interest rates in 1979 was the external shock, but this in turn was a

response to the oil price shock earlier. Also from 1974 to 1979 the results should not change significantly. 18

In the 1990s, 19 of the 26 Caribbean states had a ratio of Exports and Imports to GDP of more than 100 percent (Thorp,

1998).

16

where subscript i refers to countries, i=1,...N and t refers to periods, t=1,...T. Each period has a

length of %�, which is determined by the availability of data19, ��� is the growth rate of GDP per

capita over the period, ��$" is the initial output per capita (measured in logarithms), is a

constant20, β is the speed of convergence if β>0 (or divergence if β<0), and ��� is the disturbance

term.

Equation (2) tells us that if β is positive, the relation between initial income and growth is

negative, so that the poorer the country is at the beginning, the faster the growth rate, which implies

that the differences at the beginning of the phase tend to disappear.

Galor (1996) mentions that by adding empirically significant elements to the neoclassical growth

model, one can analyze club convergence under the same framework, he suggests inequality

measures as an example. So, in order to control more adequately for the initial differences, I add

two more variables at the beginning of each period: size and ranking of each country. Size could

give a certain advantage in the growth process, since size is associated with economic power. In a

similar fashion, position in the distribution of income captures the relative ranking at the beginning

of the period:

��� = − 1 − '�(!"%� ∙ ��$" +)*+ ∙ �+�$" +

,

+-����

(3)

where ���$" is the size of a country measured by the logarithm of population and �,�$" is the

ranking measured by the relation of each country’s output per capita to the highest output of the

same year.

Now, I introduce the two main external shocks discussed earlier:

��� = − �����!"#" ∙ ��$" +∑ *+ ∙ �+�$" +,+-� ∑ /0 ∙ 10$" +,0-� ��� (4)

19

Details on τ are in the appendix. 20

The constant is capturing the common effects for being in the same region as language, culture, religion, etc., and the

common steady-state. I could have included a dummy for each country, as it is usually done in panel data studies, in order

to include somehow each of their steady-states but that would lose the essence of the idea of the paper, which is that

inside each club , we expect convergence to occur. Moreover, after including country-specific characteristics, we would

probably find higher rates of convergence, which would be artificial (more on this in the discussion section).

17

where 1� is a dummy for the first phase 1900-1930, 1, for 1931-1974, and 12 for1975-2007.

The next step is to introduce a dummy for each club. I create dummies where I combine phase

and club characteristics. I replace the phase dummies with club dummies:

��� = − �����!"#" ∙ ��$" + ∑ *+ ∙ �+�$" +,+-� ∑ 34 ∙ 54$" +64-� ��� (5)

where 54$" is the club dummy. In total we have eight dummies (c=1,...8) that represent the clubs

mentioned in the previous section. In Phase 1 we have the agricultural and mineral clubs, in phase 2

reactive and passive clubs, and in phase 3, high-market friendly and low-market friendly countries.

Moreover, we have the Caribbean countries as a club and included for phases 2 and 3.

Equations (2) to (5) describe general aspects for the region as a whole. Only one common β

coefficient is included. So, after controlling for the different dummy characteristics, we get one beta

for the entire region. Additionally, we can see the significance of each club in the overall growth

and compare their contribution.

Next, I focus on finding different β coefficients, one for each club. I first calculate a similar

version of Equations (2) and (3) with a different β for each group:

��� = − ��7�∑ 89∙:9;<=9>? ∙@<

A< ⋅ y�;< + uEF (6)

��� = − ��7�∑ 89∙:9;<=9>? ∙@<

A< ⋅ y�;< +∑ *+ ∙ �+�$" +,+-� uEF (7)

and then adding the phase dummies to Equations (6) and (7):

��� = − ��7�∑ 89∙:9;<=9>? ∙@<

A< ⋅ y�;< + ∑ /0 ∙ 10$" +20-� uEF (8)

18

��� = − ��7�∑ 89∙:9;<=9>? ∙@<

A< ⋅ y�;< + ∑ *+ ∙ �+�$" +,+-� ∑ /0 ∙ 10$" +20-� uEF (9)

Since it is a costly model in terms of parameter estimation, I restrict the parameters *+ to be

equal across clubs. I also restrict the model to have only three (phase specific) constants instead of

eight different (club) constants. Moreover, the interest lies in the initial income coefficients, and

here the club effect is allowed. I prefer not to do a separate regression for each club since the panel

data sample for each club becomes too small.

4 Results

The econometric tool employed is non-linear pooled OLS regressions for 32 countries for the

period 1900 to 2007. The data description is in the Appendix. I report the coefficients,

heteroskedasticity consistent standard errors (White, 1980) and other descriptive estimates, from

Equations 2-5 in Table 1 and from Equations 6-9 in Table 2.

From Table 1 we can see that the initial income coefficient, β, has almost the same rate in

Equation 2 as in Equation 3: around 0.15%. However, in both cases we fail to reject the null

hypothesis that the β coefficient is zero (no convergence). When adding size and position, Equation

2, the coefficient of the variable size is significant and negative, while the coefficient for the

variable position is positive but insignificant.

Equation 4 shows that each of the phase dummies is significantly different from the last phase

(the omitted dummy). I also test whether both phase dummies are jointly significant in the equation

(H₀: c₁=c₂=0). The test statistic is F=3.25, and we reject the null (at a level of 95% of confidence).

The β coefficient is -0.65%, showing overall divergence. The coefficient of size remains negative,

while position has changed to negative. Both variables are significant.

Equation 5 substitutes the phase dummies for the club dummies, since the last ones include the

first ones. Results are in the last column of Table 1. Regarding β, there is a significant negative

coefficient, supporting divergence among all Latin American countries. This means that the relation

between initial income and growth is positive once we take into account the effect of the different

19

clubs and initial conditions. All coefficients of club dummies are significant, except for the mineral

countries. This is clear evidence that the club division is successful. The coefficients of all dummies

are negative, which is just showing the differences in the constant term of the growth equation

according to the clubs. Size and position retain negative signs.

So far, I have shown that the division of phases and clubs is very important, and that there is

significant divergence among all countries, after controlling for differences in initial conditions and

membership in different clubs.

The next task is to see whether contrary to the overall divergence picture there is in fact

convergence inside each club. We proceed to calculate β convergence for each group. Table 2

shows the results from estimations of Equations 6 to 9.

Equation 6 is a similar version to Equation 2, in the sense that no controls are included. The

results are presented Table 2. The βs for all clubs are significant and positive. This supports again

the basic idea behind the paper, that there is club convergence, and the coefficients are significant.

The positive sign of β means that there is a non-linear negative relation between initial income and

growth. All coefficients are low and similar to each other, so I test whether the club dummy

coefficients are significantly different from each other, and whether they are jointly significant. The

F test is 3.54 for the first test indicates that we can reject the null with 95% confidence, and

similarly, F=3.48 for the second test, which means that the dummies are jointly significant and

different from each other. To control for more initial conditions, Equation 7 is estimated. The results

in column 2 show that the βs remain similar, all positive and significant. I do the same tests as for

Equation 5, and the results show that all βs are jointly significant and significantly different from

each other. The coefficient for size is still negative and for position positive, but both insignificant.

In general, the results show that the division by historical phases and clubs is important. When

the club dummies were introduced in the specification, where it was assumed that β was common in

the clubs, in Column 4 Table 1, the club dummies were significant, so that their inclusion was

correct, and the β coefficient that relates initial income to growth was negative, which means

divergence among all countries (confirming the impression in Figure 1). After allowing for

heterogeneity in the non-linear relation between growth and initial income, in Table 2, there is

enough evidence that the clubs show convergence. The βs for all clubs are significant and positive

20

as expected (Columns 1 and 2 in Table 2). When adding more controls, two of them become

insignificant, the Caribbean and the high-reformers (Columns 3 and 4 in Table 2) in the last phase.

The two variables besides income, used to control for differences in the initial conditions, size in

terms of population, and position in the income distribution, show a negative relation with growth

when significant (Columns 2 to 4 in Table 1). The rates of speeds of convergence are all around

0.5% which is lower than the typical 2% found in the literature. The reason for this difference may

lie in that I do not have as many controls for the steady state.

21

Table 1. Common β. Econometric results from estimations of Equations 2 to 5. Standard errors in brackets,

*** significant with p<0.01, ** p<0.05, *p<0.10, ++p<0.15 and +p<0.20. Phase 3 is omitted

Variable Equation 2 Equation 3 Equation 4 Equation 5

Initial GDP 0.001 0.001 -0.006+ -0.007***

[0.003] [0.003] [0.005] [0.003]

Size -0.002** -0.002** 0.000

[0.001] [0.001] [0.001]

Position 0.003 -0.014* -0.020***

[0.006] [0.008] [0.006]

Phase 1 0.012**

[0.006]

Phase 2 0.010**

[0.004]

Agricultural -0.024***

[0.008]

Mineral -0.019+

[0.014]

Caribbean 2 -0.011**

[0.006]

Reactive -0.024***

[0.007]

Pasive -0.027***

[0.006]

Caribbean 3 -0.029***

[0.010]

Low-reformers -0.042***

[0.008]

High-reformers -0.031***

[0.008]

Constant 0.027 0.054* -0.003

[0.021] [0.029] [0.042]

N 257 257 257 257

rss 0.118 0.116 0.113 0.108

R2 0.002 0.020 0.047 0.419

22

Table 2. Different β rates. Econometric results of estimations of Equations 6 to 9. Same description as Table 1.

Variable Equation 6 Equation 7 Equation 8 Equation 9

Initial GDP

Mineral 0.004* 0.004* 0.000 -0.001

[0.002] [0.002] [0.006] [0.007]

Agricultural 0.004*** 0.004*** 0.001 0.000

[0.001] [0.001] [0.006] [0.008]

Caribbean 2 0.002* 0.002* 0.002** 0.002**

[0.001] [0.001] [0.001] [0.001]

Reactive 0.004*** 0.004*** 0.005*** 0.004***

[0.001] [0.001] [0.001] [0.001]

Pasive 0.004*** 0.004*** 0.005*** 0.005***

[0.001] [0.001] [0.001] [0.001]

Caribbean 3 0.003*** 0.004** 0.001 0.000

[0.001] 0.002 0.004 0.005

Low-reformers 0.005*** 0.005*** 0.003 0.001

[0.001] [0.001] 0.004 0.006

High-reformers 0.003*** 0.004*** 0.001 0.000

[0.001] [0.001] 0.004 0.005

Size -0.001 -0.001

0.002 0.002

Position 0.003 -0.002

0.006 0.007

First Phase 0.021 0.029

0.041 0.064

Second Phase 0.055*** 0.064**

0.010 0.029

Third Phase 0.027 0.028

0.036 0.045

Constant 0.046*** 0.056**

0.009 0.028

N 257 257 257 257

rss 0.109 0.109 0.109 0.109

R2 0.076 0.078 0.415 0.416

23

5 Strengths and Weaknesses of Approach

Even though the results are quite satisfactory, there are caveats regarding the approach that need to

be discussed. In this section I discuss the flaws of the current division of clubs, other possible ways

of finding clubs, omitted variables, unbalanced panel data, and measurement errors.

5.1 Division of Clubs

The most controversial characteristic of this paper is the division of clubs. There could have been

superior alternative ways to approach the division.

Ideally, I could have determined structural breaks for the given time period of data for all 32

countries. One way of doing this is following Bai (1997), who develops a method for finding

multiple breaks. Another option is to analyze breaks for each country and see if there were common

breaks. Astorga et al. (2005) do this for six countries over 100 years, using the Chow test. Table 1 in

their paper shows the different structural breaks by country. These shocks account for external and

internal events, like revolutions, dictatorships, and country specific characteristics. At the end, the

authors do a panel data analysis where they recognize that the major events for all countries were

the crisis of 1929 and its aftermath in 1930, along with the debt crises in the 1980s. Instead, I let

historians decide the breaks, and, after all, the breaks are similar to the ones in Astorga et al. (2005).

Moreover, the significance statistical tests on the phase dummies prove that the breaks are relevant.

Similarly, regarding the clubs, I could have chosen many other ways of dividing countries into

clubs. Canova (2004) argues that the initial distribution of income per capita, the initial level of

human capital, and human capital within the country could be used as economic causes of

heterogeneity. In addition, he says, geography/location can be used to measure the neighborhood

externalities, and policy variables could measure national effects.

Given that the data are limited, I am not able to use more variables than the ones I have already

used. I could have had more variables but for fewer countries, which would change the essence of

the paper. I did try to group countries according to geography: Caribbean, Central and South

American clubs. The results showed divergence. I also tried to divide the countries according to

economic integration and had no success (in Barrientos, 2010) because integration in Latin America

is not yet well developed. This is not enough evidence to claim superiority over other ways of

24

dividing countries into clubs, but it is appealing to have another way of diving into clubs than those

already known.

It is worth noting that the division of clubs by economic history has flaws. The clubs in the

paper are presented as independent from each other. However, clubs in each phase depend on clubs

in previous phases. Many countries may not be able to make a "fresh start" at every historical

juncture. Moreover, the division of phase one, which is by resources, is still very important in later

phases, as noted by Acemoglu et.al (2001) and Engerman and Sokoloff (2002).

5.2 Omitted Variables

This paper studies more countries and years than any other study. However, this imposes

restrictions in terms of the possibility of adding more variables. I could have restricted the analysis

into fewer countries, fewer years, and more variables. However, the essence of this paper is the

inclusion of as many countries and as many years as possible to analyze historical events and use

these events in a way that maybe variables would inform. Still missing variables is a problem in this

approach, which means that the results may be biased and inconsistent.

Nevertheless, as mentioned before, including proxies for the steady state introduce endogeneity

problems and the results can be hard to interpret in the sense that the inclusion of more controls, will

tell us less about the true convergence. β convergence tells us about poor countries growing faster

than rich ones, conditional on the controls. So intuitively, when adding controls, we will most likely

find high rates of convergence but these will probably be artificial.

5.3 Unbalanced Panel and Measurement error

The data is an unbalanced panel, where some countries do not have information, especially for the

first years. This can be a problem if the reason for missing information is related to the error term,

but since the reason here is connected to the regressor (initial output per capita), having unbalanced

panel data is not a problem.

Another concern is the temporal measurement error that can lead to inflated convergence rates.

Barro and Sala-i-Martin (1992) show in their appendix that measurement error is unlikely to be

important, results seem to be similar. Here I use the same setting as in their article with the

25

difference that I use more homogenous countries. So I rely in their results and arguments for not

worrying for measurement errors, as they do.

6 Conclusions

This article investigates and connects the economic history of Latin America, reflected into the

analysis of external shocks, trends and ideologies, as sources of heterogeneity in the growth process

and club formation in Latin America. First, I identify two main common external shocks to the

region: the Great Depression in the 1930s and the oil price shock in 1974. Then I classify countries

in clubs according to their policy-response to the shocks. I focus on a criterion to divide countries

into clubs that describe the initial conditions after the shocks. The criterion is based on the policies

adopted at the beginning of each phase, as a response to the shock. I focus on information on

policies rather than outcomes so the results are not driven by the selection of the club thresholds in

the first place.

Before 1930, I define two clubs: the minerals and agricultural. After the Great Depression I

follow Diaz (1984) classification of clubs according to passive or reactive, where the reactive

responded autonomously to protect themselves, while the passive did/could not. After the oil price

shock, I classify the clubs according to the Lora index, which describes how far countries applied

structural reforms to liberalize their economies. I also include the Caribbean countries as a separate

club.

In general, the results show that the division of phases and clubs is important. When the club

dummies were introduced in the specification with a common β coefficient, Column 4 in Table 1,

the club dummies were significant, so that their inclusion was correct, and the β coefficient that

relates initial income to growth was negative, which means divergence among all countries

(confirming the impression from Figure 1). When allowing for heterogeneity in the β coefficients,

there is evidence that the clubs show convergence. The βs for all clubs are significant and positive

as expected (Columns 1 and 2 in Table 2). When adding more controls, two of the club βs become

insignificant, the Caribbean and the high-reformers (Columns 4 and 5 in Table 2), in the last phase,

but the overall impression is still one of club convergence, and the βs are jointly significant.

26

Regarding policy implications, I find that the clubs to which countries appertain, are determined

by policy makers but also by external shocks and natural resources endowments. I cannot conclude

that one club is superior than another, because successful countries belong to different clubs.

Therefore, I cannot suggest how to jump to a superior club, as is suggested in a traditional approach,

where clubs are defined by income or capital thresholds which imply that a significant transfer of

money would help a country jump to a superior club.

27

Appendix

Data

The analysis covers 32 countries, listed in Table 3, for the period 1900-2007. The potential number

of observations is 3,456, but due to incomplete data for some countries, the number of real

observations is reduced to 2,209.

The main variable is the GDP per capita measured in constant 1990 International (Geary-

Khamis) dollars. This measure allows for comparison of standards of living of the countries; it takes

into account the purchasing power parity of currencies and the international commodity prices. The

sources are the Madison database (2003) and the World Bank (2004). The final data base has

information from the Madison database (M) (from 1900 until 1989) and from the World Bank

database (W) (from 1990 to 2007).

A converter factor (C) is calculated as: C₍₁₉₉₀₎=M₍₁₉₉₀₎/W₍₁₉₉₀₎ for each year and is kept

constant from 1995. Then C is multiplied by the existent W. In the case of ten small Caribbean

countries, M has no data, so C is taken constant, for the year 1995, from another country that

heavily influenced these economies and is assumed to have a similar C. The one from USA is used

for The Bahamas; from Great Britain for Barbados and Belize; from Haiti for Dominica St.Kitts and

Nevis, St. Lucia, St.Vincent and the Grenadines; from Colombia for Guyana, and finally from The

Dominican Republic for Grenada. In the case of Cuba, the available GDP from W was measured in

constant 2000 local currency. Here, C was calculated with that kind of data and kept constant for the

year 2001. The transformed data go from 2001 to 2007.

The panel data were created by taking averages or the values of variables in subperiods of

different lenght. The choice for different lengths is to take advantage of the data and coincide with

the phase years.

28

Table 3. Description of observations in data set.

Country ObservationsMissing

observations Starting

yearEnding

year

Argentina 108 0 1900 2007

The Bahamas 28 80 1975 2002

Belize 33 75 1975 2007

Bolivia 63 45 1945 2007

Brazil 108 0 1900 2007

Barbados 25 83 1975 1999

Chile 108 0 1900 2007

Colombia 108 0 1900 2007

Costa Rica 88 20 1920 2007

Cuba 76 32 1929 2004

Dominica 31 77 1977 2007

Dominican Republic 58 50 1950 2007

Ecuador 69 39 1939 2007

Grenada 28 80 1980 2007

Guatemala 88 20 1920 2007

Guyana 33 75 1975 2007

Honduras 88 20 1920 2007

Haiti 63 45 1945 2007

Jamaica 64 44 1913 2007

St. Kitts and Nevis 31 77 1977 2007

St. Lucia 28 80 1980 2007

Mexico 108 0 1900 2007

Nicaragua 88 20 1920 2007

Panama 63 45 1945 2007

Peru 108 0 1900 2007

Puerto Rico 52 56 1950 2001

Paraguay 69 39 1939 2007

El Salvador 88 20 1920 2007

Trinidad and Tobago 58 50 1950 2007

Uruguay 108 0 1900 2007

St. Vincent and the Grenadines 33 75 1975 2007

Venezuela 108 0 1900 2007

Total 2,209 1,247

29

countries1 ly lppl pos countries1 ly lppl pos

arg mean 8.57 16.64 0.78 hnd mean 7.33 14.39 0.20

sd 0.36 0.60 0.20 sd 0.20 0.82 0.08

max 9.27 17.49 1.00 max 7.62 15.79 0.42

min 7.91 15.36 0.47 min 6.91 13.12 0.10

obs 108 108 108 obs 88 108 88

bhs mean 9.43 12.30 0.96 hti mean 6.88 15.11 0.09

sd 0.13 0.30 0.06 sd 0.15 0.54 0.03

max 9.54 12.72 1.00 max 7.17 16.09 0.20

min 9.05 11.66 0.80 min 6.61 14.26 0.04

obs 28 47 28 obs 63 108 63

blz mean 8.08 12.02 0.24 jam mean 7.94 14.18 0.27

sd 0.32 0.34 0.05 sd 0.41 0.43 0.05

max 8.57 12.65 0.33 max 8.33 14.80 0.39

min 7.64 11.45 0.17 min 6.41 13.49 0.16

obs 33 47 33 obs 64 108 64

bol mean 7.67 15.07 0.20 kna mean 8.13 10.70 0.26

sd 0.17 0.51 0.04 sd 0.47 0.06 0.09

max 7.96 16.07 0.33 max 8.74 10.83 0.39

min 7.36 14.34 0.14 min 7.33 10.60 0.14

obs 63 108 63 obs 31 47 31

bra mean 7.61 17.92 0.30 lca mean 7.62 11.72 0.15

sd 0.74 0.74 0.07 sd 0.30 0.20 0.04

max 8.76 19.06 0.43 max 7.92 12.03 0.20

min 6.52 16.70 0.20 min 7.03 11.38 0.10

obs 108 108 108 obs 28 47 28

brb mean 9.13 12.42 0.73 mex mean 8.02 17.33 0.44

sd 0.12 0.03 0.05 sd 0.57 0.71 0.07

max 9.33 12.47 0.81 max 8.94 18.47 0.64

min 8.90 12.35 0.63 min 7.21 16.43 0.29

obs 25 47 25 obs 108 108 108

chl mean 8.32 15.74 0.60 nic mean 7.45 14.19 0.23

sd 0.49 0.54 0.13 sd 0.30 0.83 0.09

max 9.48 16.63 1.00 max 8.12 15.54 0.40

min 7.58 14.91 0.37 min 6.91 13.08 0.07

obs 108 108 108 obs 88 108 88

col mean 7.77 16.43 0.34 pan mean 8.25 13.79 0.35

sd 0.57 0.74 0.05 sd 0.42 0.77 0.07

max 8.79 17.61 0.46 max 9.01 15.02 0.45

min 6.88 15.20 0.24 min 7.52 12.48 0.23

obs 108 108 108 obs 63 108 63

cri mean 8.01 13.88 0.37 per mean 7.70 16.06 0.32

sd 0.50 0.87 0.06 sd 0.55 0.66 0.06

max 8.89 15.31 0.47 max 8.49 17.17 0.47

min 7.26 12.60 0.24 min 6.71 15.15 0.21

obs 88 108 88 obs 108 108 108

cub mean 7.63 15.54 0.24 pri mean 8.76 14.58 0.62

sd 0.26 0.59 0.07 sd 0.57 0.43 0.23

max 8.02 16.23 0.43 max 9.66 15.19 1.00

min 6.88 14.32 0.15 min 7.67 13.77 0.29

obs 76 108 76 obs 52 108 52

dma mean 7.55 11.16 0.14 pry mean 7.72 14.30 0.23

sd 0.29 0.05 0.03 sd 0.31 0.79 0.07

max 7.90 11.22 0.19 max 8.16 15.63 0.44

min 6.93 11.02 0.08 min 7.31 12.99 0.15

obs 31 47 31 obs 69 108 69

30

Table 4. Description of observations in data set by country, where ly is the logarithm of GDP per capita, lppl is the logarithm of

population and pos is the position of country with respect to the richest country.

countries1 ly lppl pos countries1 ly lppl pos

dom mean 7.63 14.74 0.18 slv mean 7.44 14.67 0.21

sd 0.41 0.92 0.04 sd 0.39 0.67 0.04

max 8.40 16.10 0.27 max 7.99 15.62 0.30

min 6.93 13.15 0.13 min 6.71 13.55 0.14

obs 58 108 58 obs 88 108 88

ecu mean 7.97 15.19 0.28 tto mean 9.08 13.40 0.78

sd 0.40 0.75 0.04 sd 0.41 0.53 0.16

max 8.50 16.41 0.36 max 9.95 14.10 1.00

min 7.17 14.15 0.21 min 8.21 12.50 0.46

obs 69 108 69 obs 58 108 58

grd mean 8.02 11.48 0.22 ury mean 8.40 14.56 0.66

sd 0.28 0.04 0.05 sd 0.37 0.39 0.17

max 8.35 11.54 0.30 max 9.14 15.02 1.00

min 7.50 11.39 0.14 min 7.70 13.73 0.40

obs 28 47 28 obs 108 108 108

gtm mean 7.81 15.09 0.31 vct mean 7.52 11.51 0.14

sd 0.31 0.75 0.11 sd 0.33 0.09 0.03

max 8.22 16.41 0.66 max 8.06 11.60 0.18

min 7.15 14.08 0.18 min 6.85 11.32 0.09

obs 88 108 88 obs 33 47 33

guy mean 8.04 13.50 0.23 ven mean 8.40 15.75 0.70

sd 0.11 0.07 0.03 sd 0.93 0.81 0.28

max 8.23 13.57 0.28 max 9.33 17.13 1.00

min 7.84 13.28 0.18 min 6.68 14.75 0.23

obs 33 47 33 obs 108 108 108

Total mean 7.96 14.68 0.38

sd 0.67 1.70 0.24

max 9.95 19.06 1.00

min 6.41 10.60 0.04

obs 2209 2907 2209

31

Table 5. Clubs in phase 1. ly is the logarithm of GDP per capita, lppl is the logarithm of population and pos is the position of country

with respect to the richest country.

Mineral Agricultural

countries1 ly lppl pos countries1 ly lppl pos

chl mean 7.80 15.09 0.70 arg mean 8.17 15.86 1.00

sd 0.15 0.12 0.08 sd 0.13 0.28 0.01

max 8.13 15.29 1.00 max 8.38 16.29 1.00

min 7.58 14.91 0.59 min 7.91 15.36 0.95

obs 31 31 31 obs 31 31 31

mex mean 7.44 16.53 0.49 bra mean 6.75 17.02 0.24

sd 0.10 0.06 0.05 sd 0.16 0.19 0.03

max 7.60 16.66 0.64 max 7.05 17.33 0.30

min 7.21 16.43 0.38 min 6.52 16.70 0.20

obs 31 31 31 obs 31 31 31

per mean 6.98 15.31 0.31 col mean 7.09 15.53 0.34

sd 0.19 0.10 0.05 sd 0.11 0.21 0.03

max 7.39 15.50 0.43 max 7.32 15.88 0.46

min 6.71 15.15 0.25 min 6.88 15.20 0.30

obs 31 31 31 obs 31 31 31

ven mean 7.09 14.88 0.37 cri mean 7.40 12.88 0.42

sd 0.46 0.07 0.17 sd 0.05 0.15 0.04

max 8.14 15.01 0.80 max 7.50 13.12 0.47

min 6.68 14.75 0.23 min 7.33 12.60 0.36

obs 31 31 31 obs 11 31 11

cub mean 7.36 14.76 0.36

sd 0.06 0.26 0.02

max 7.40 15.16 0.38

min 7.32 14.32 0.35

obs 2 31 2

gtm mean 7.30 14.23 0.38

sd 0.10 0.09 0.02

max 7.48 14.39 0.41

min 7.15 14.08 0.36

obs 11 31 11

slv mean 6.89 13.87 0.25 hnd mean 7.20 13.43 0.34

sd 0.06 0.19 0.01 sd 0.10 0.19 0.03

max 6.97 14.18 0.27 max 7.35 13.76 0.37

min 6.82 13.55 0.22 min 7.04 13.12 0.28

obs 11 31 11 obs 11 31 11

ury mean 7.99 14.03 0.84 nic mean 7.22 13.28 0.35

sd 0.17 0.20 0.07 sd 0.12 0.11 0.03

max 8.37 14.35 1.00 max 7.47 13.43 0.40

min 7.70 13.73 0.67 min 7.08 13.08 0.30

obs 31 31 31 obs 11 31 11

32

Table 6. Clubs in phase 2. ly is the logarithm of GDP per capita, lppl is the logarithm of population and pos is the position of country with respect to

the richest country.

Reactive Pasive

countries1 ly lppl pos countries1 ly lppl pos

arg mean 8.55 16.70 0.77 cub mean 7.50 15.61 0.27

sd 0.23 0.23 0.17 sd 0.23 0.27 0.06

max 9.03 17.06 1.00 max 7.79 16.05 0.43

min 8.17 16.31 0.52 min 6.88 15.18 0.18

obs 44 44 44 obs 44 44 44

bol mean 7.53 14.96 0.21 dom mean 7.22 14.75 0.15

sd 0.12 0.19 0.04 sd 0.18 0.43 0.02

max 7.79 15.35 0.33 max 7.63 15.45 0.20

min 7.36 14.70 0.16 min 6.93 14.08 0.13

obs 30 44 30 obs 25 44 25

bra mean 7.51 17.89 0.27 ecu mean 7.64 15.09 0.27

sd 0.38 0.35 0.04 sd 0.27 0.37 0.03

max 8.31 18.48 0.39 max 8.13 15.72 0.36

min 6.91 17.35 0.20 min 7.17 14.51 0.21

obs 44 44 44 obs 36 44 36

chl mean 8.27 15.70 0.58 gtm mean 7.71 14.98 0.34

sd 0.22 0.26 0.13 sd 0.21 0.38 0.13

max 8.64 16.14 0.81 max 8.10 15.61 0.66

min 7.73 15.30 0.42 min 7.21 14.41 0.22

obs 44 44 44 obs 44 44 44

col mean 7.71 16.38 0.33 hnd mean 7.20 14.30 0.20

sd 0.23 0.34 0.07 sd 0.14 0.36 0.07

max 8.19 16.97 0.46 max 7.40 14.92 0.42

min 7.28 15.90 0.24 min 6.91 13.79 0.14

obs 44 44 44 obs 44 44 44

cri mean 7.75 13.78 0.34 hti mean 6.91 15.03 0.12

sd 0.33 0.45 0.07 sd 0.07 0.22 0.03

max 8.40 14.51 0.46 max 7.01 15.44 0.20

min 7.26 13.14 0.24 min 6.76 14.71 0.08

obs 44 44 44 obs 30 44 30

mex mean 7.87 17.23 0.38 pan mean 7.87 13.76 0.30

sd 0.35 0.37 0.06 sd 0.28 0.34 0.06

max 8.52 17.87 0.49 max 8.35 14.33 0.41

min 7.22 16.68 0.29 min 7.52 13.17 0.23

obs 44 44 44 obs 30 44 30

nic mean 7.51 14.05 0.27 pri mean 8.28 14.60 0.44

sd 0.33 0.45 0.05 sd 0.40 0.17 0.14

max 8.08 14.81 0.40 max 8.90 14.89 0.69

min 6.91 13.44 0.20 min 7.67 14.28 0.29

obs 44 44 44 obs 25 44 25

per mean 7.81 15.95 0.36 slv mean 7.31 14.62 0.22

sd 0.34 0.30 0.06 sd 0.32 0.34 0.03

max 8.34 16.51 0.47 max 7.80 15.24 0.30

min 7.05 15.51 0.28 min 6.71 14.19 0.17

obs 44 44 44 obs 44 44 44

pry mean 7.44 14.26 0.23

sd 0.09 0.34 0.09

max 7.67 14.82 0.44

min 7.31 13.71 0.16

obs 36 44 36

ury mean 8.38 14.64 0.65 ven mean 8.79 15.58 0.96

sd 0.19 0.15 0.15 sd 0.47 0.43 0.08

max 8.59 14.85 0.94 max 9.28 16.32 1.00

min 7.92 14.37 0.46 min 7.87 15.02 0.74

obs 44 44 44 obs 44 44 44

33

Table 6. Clubs in phase 3. ly is the logarithm of GDP per capita, lppl is the logarithm of population and pos is the position of country with respect to

the richest country.

Low Reformers High Reformers

countries1 ly lppl pos countries1 ly lppl pos

bra mean 8.55 18.82 0.38 arg mean 8.98 17.30 0.59

sd 0.09 0.17 0.03 sd 0.11 0.13 0.08

max 8.76 19.06 0.43 max 9.27 17.49 0.78

min 8.34 18.50 0.30 min 8.77 17.07 0.47

obs 33 33 33 obs 33 33 33

col mean 8.48 17.32 0.35 bol mean 7.80 15.73 0.18

sd 0.15 0.19 0.04 sd 0.09 0.21 0.03

max 8.79 17.61 0.44 max 7.96 16.07 0.24

min 8.19 16.99 0.31 min 7.64 15.38 0.14

obs 33 33 33 obs 33 33 33

cri mean 8.55 14.95 0.38 chl mean 8.88 16.41 0.54

sd 0.15 0.24 0.04 sd 0.34 0.15 0.13

max 8.89 15.31 0.46 max 9.48 16.63 0.73

min 8.35 14.53 0.31 min 8.37 16.16 0.37

obs 33 33 33 obs 33 33 33

cub mean 7.84 16.17 0.20 pan mean 8.59 14.71 0.39

sd 0.15 0.06 0.03 sd 0.17 0.20 0.03

max 8.02 16.23 0.24 max 9.01 15.02 0.45

min 7.52 16.06 0.15 min 8.32 14.36 0.31

obs 30 33 30 obs 33 33 33

dom mean 7.93 15.82 0.21 pri mean 9.21 15.08 0.79

sd 0.22 0.19 0.03 sd 0.24 0.08 0.16

max 8.40 16.10 0.27 max 9.66 15.19 1.00

min 7.66 15.48 0.17 min 8.85 14.91 0.61

obs 33 33 33 obs 27 33 27

ecu mean 8.32 16.13 0.30 pry mean 8.03 15.27 0.23

sd 0.08 0.20 0.03 sd 0.10 0.24 0.03

max 8.50 16.41 0.35 max 8.16 15.63 0.28

min 8.15 15.75 0.23 min 7.71 14.85 0.15

obs 33 33 33 obs 33 33 33

gtm mean 8.11 16.02 0.25 ury mean 8.83 14.95 0.50

sd 0.07 0.23 0.04 sd 0.15 0.05 0.06

max 8.22 16.41 0.32 max 9.14 15.02 0.62

min 7.98 15.64 0.18 min 8.60 14.86 0.40

obs 33 33 33 obs 33 33 33

hnd mean 7.54 15.41 0.14

sd 0.05 0.26 0.02

max 7.62 15.79 0.17

min 7.36 14.95 0.10

obs 33 33 33

mex mean 8.75 18.23 0.47

sd 0.10 0.18 0.04

max 8.94 18.47 0.55

min 8.55 17.89 0.36

obs 33 33 33

nic mean 7.46 15.24 0.14 slv mean 7.80 15.49 0.18

sd 0.28 0.21 0.06 sd 0.12 0.11 0.03

max 8.12 15.54 0.30 max 7.99 15.62 0.24

min 7.18 14.84 0.07 min 7.64 15.26 0.14

obs 33 33 33 obs 33 33 33

per mean 8.22 16.89 0.28 ven mean 9.11 16.80 0.68

sd 0.13 0.19 0.05 sd 0.11 0.23 0.16

max 8.49 17.17 0.40 max 9.33 17.13 1.00

min 7.96 16.53 0.21 min 8.85 16.36 0.43

obs 33 33 33 obs 33 33 33

34

Table 7. Description of subperiods of panel data. First line is the initial year, second line is the ending year,

and the last line is M.

1 7

t0: 1900 t0: 1959

t1: 1919 t1: 1965

τ: 20 τ: 7

2 8

t0: 1920 t0: 1966

t1: 1930 t1: 1974

τ: 11 τ: 9

3 9

t0: 1931 t0: 1975

t1: 1937 t1: 1981

τ: 7 τ: 7

4 10

t0: 1938 t0: 1982

t1: 1944 t1: 1988

τ: 7 τ: 7

5 11

t0: 1945 t0: 1989

t1: 1951 t1: 1996

τ: 7 τ: 8

6 12

t0: 1952 t0: 1997

t1: 1958 t1: 2007

τ: 7 τ: 11

Subperiods

35

References

Astorga, P., Bergés, A. R., & FitzGerald, V. 2003. “Productivity growth in Latin America during

the twentieth century”. Economics Group, Nuffield College, University of Oxford.

Astorga, P., Berges, A., and Fitzgerald, V. 2005. "Endogenous growth and exogenous shocks in

Latin America during the twentieth century". University of Oxford. Discussion Paper in

Economic and Social History. No. 57, March.

Acemoglu, Robinson, Johnson. 2001. "The Colonial Origins of Comparative Development: An

Empirical Investigation". American Economic Review, 91, pp. 1369-1401.

Acemoglu, D. and J.A. Robinson. 2012. “Why Nations Fail. The Origins of Power, Prosperity,

and Poverty”. New York: Crown Business 2012

Azariadis, Costas, and Allan Drazen. 1990. "Threshold externalities in economic development”.

The Quarterly Journal of Economics 105.2 501-526.

Azariadis,C. and Stachurski, J. 2005. "Poverty Traps". In Handbook of Economic Growth . S.

Durlauf and P. Aghion, eds, North-Holland

Bai, J. 1997. "Estimating Multiple Breaks One at a Time," Econometric Theory, Cambridge

University Press, vol. 13(03), pages 315-352, June.

Barrientos Q., Paola A. 2010. “Convergence Patterns in Latin America”. Economic Working

paper 2010-15. School of Economics and Management, Aarhus University.

Barro, Robert J., and Xavier Sala-i-Martin. 1992. "Convergence." Journal of Political Economy:

223-251.

Barro, R. and Sala-i-Martin, X. 2004. Economic Growth. MIT Press, Cambridge.

Blyde, J. 2006. "Latin American Clubs: Uncovering Patterns of Convergence". Inter-

American Development Bank.

Cardoso, E., and Helwege, A. 1992. Latin America’s Economy. MIT press.

Canova, F. 2004."Testing for Convergence Clubs in Income per Capita: A Predictive

Density Approach". 2004. International Economic Review, 45 (1), 2004, 49-77

Dabus, C., Zinni, B. 2005. "No convergencia". Política Económica en Argentina.

36

Díaz Alejandro, C. F. 1983. Stories of the 1930s for the 1980s. In Financial Policies and the

World Capital Market: The Problem of Latin American Countries , edited by P. Aspe

Armella, R. Dornbusch and M. Obstfeld. Chicago: University of Chicago Press.

Desgoits, A.1998. "Pattern of Economic Development and the formation of Clubs".

Université d’ Evry-Val D’ Essone EPPE, working paper.

Dobson, S., and Ramlogan, C. 2002a. "Convergence and divergence in Latin America,

1970-1998". Applied Economics, 34:4, 465 - 470.

Dobson, S. and Ramlogan, C. , 2002b "Economic Growth and Convergence in Latin

America’ , Journal of Development Studies, 38:6, 83 - 104

Dollar,D., Kraay. 2003. "Institutions, Trade and Growth". Journal of Monetary Economics

50(1), 133-162.

Durlauf, S., Johnson, P. 1995. "Multiple Regimes and Cross-Country Growth

Behaviour".Journal of Applied Econometrics, Vol. 10, No. 4. (Oct. - Dec., 1995), pp.

365-384.

Durlauf, S.,and Quah, D. 1999. "The new empirics of economic growth". In Taylor, J., and

Woodford, M., eds., Handbook of Macroeconomics.

Easterly, W. and R. Levine (2003), "Tropics, Germs, and Crops: How Endowments

Influence Economic Development", Journal of Monetary Economics, 50, 1, p. 3-39.

Engermann and Sokoloff, “Factor endowments, Inequality, and Paths of Development Among

New World Economies.” NBER Working Paper no 9259, 2002

Furtado, C. 1981. Economic Development of Latin America: Historical Background and

Contemporary problems. Cambridge University Press.

Galor, O.1996. "Convergence? Inferences from theoretical models". The Economic

Journal, Vol. 106, No. 437. (Jul., 1996), pp. 1056-1069.

Hansen, Bruce E., 2000. "Sample Splitting and Threshold Estimation," Econometrica,

Econometric Society, vol. 68(3), pages 575-604, May.

Holmes, M. 2005. "New evidence on long-run output convergence among Latin American

countries". Journal of Applied Economics. Vol VIII, No. 2 (Nov 2005), 299-319.

37

Islam, N. 1995. "Growth Empirics: A Panel Data Approach". The Quarterly Journal of

Economics, Vol. 110, No. 4. (Nov., 1995), pp. 1127-1170.

Lora, Eduardo. 2001 "Structural reforms in Latin America: what has been reformed and how to

measure it." Working paper No 466. Inter-American Development Bank.

Maddison, A. 2003. "The World Economy: Historical Statistics". Development Centre

Studies. OECD.Paris- France

Owen, A., Videras, J., Davis, L. 2009. "Do all countries follow the same growth process?,"

Journal of Economic Growth, Springer, vol. 14(4), pages 265-286, December.

Paap, R; Van Dijk, H. 1998. "Distribution and mobility of Wealth Nations". European

Economic Review 42 1269-93.

Pesaran, H. 2006. "A pair-wise approach to testing for output and growth convergence".

Journal of Econometrics 138 (2007) 312–355.

Quah, Danny. 1997. "Empirics for growth and distribution : stratification, polarization, and

convergence clubs". Journal of economic growth, 2 (1). pp. 27-59.

Quah, Danny. 1996. "Twin Peaks: Growth and Convergence in Models of Distribution

Dynamics", The Economic Journal, Vol. 106, No. 437. (Jul., 1996), pp. 1045-1055.

Rodrik, D. Subramanian A., Trebbi F. 2004. "Institutions Rule: The Primacy of

Institutions over Geography and Integration in Economic Development". Journal of

Economic Growth 9 131-165

Serra, I., Pazmino, M., Lindow, G., Sutton, B., Ramirez, G. 2006. "Regional Convergence

in Latin America". IMF working paper No.125.

Thorp, R. 1998. Progress, Poverty and Exclusion. An Economic History of Latin America

in the 20th Century. Baltimore, Md.

Taylor, A.M., 1998. "Latin America and Foreign Capital in the Twentieth Century: Economics,

Polictics, and Institutional Change," Working Papers e-98-1, Hoover Institution, Stanford

University.

White, Halbert, 1980. "A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct

Test for Heteroskedasticity," Econometrica, vol. 48(4), pages 817-38, May.

38

Williams, R. 1986. Export Agriculture and the Crisis in Central America. University of

North Carolina Press.

Woolcock, Michael , Szreter, Simon and Rao, Vijayendra(2011) 'How and Why Does History

Matter for Development Policy?', Journal of Development Studies, 47: 1, 70 — 96

World Bank. 2009. “World Development Indicators Data Query”, World Bank. URL:

http://devdata.worldbank.org/data-query/.

External Flows and Development in Bolivia

Paola Barrientos Quiroga∗and Kenneth Dencker Petersen†

Abstract

External resources for a low income country have the poten-

tial of accelerating growth but also of creating Dutch Disease

effects. This paper analyses the empirical impact of the most

important external resources, foreign direct investment, aid,

and loans, on the Bolivian economy for the period 1950-2009.

We use a multivariate CVAR(2) model, which directly deals

with endogeneity issues and allows for a more profound insight

into the chain of causality, through the impulse responses. Our

results indicate that loans are positive for the Bolivian economy

in the long run, through an increase in exports and a decrease

in purchasing power of exports however, we did not include

total debt. We find that the effect of aid is negative for the

Bolivian GDP per capita in the long run, which is explained by

the temporary increase in purchasing power of exports, which

lead to a permanent loss in the exports. Finally, foreign direct

investments lead to a positive response of Bolivian GDP per

capita and exports, and a reduction of purchasing power of

exports in the long run.

∗Aarhus University†Jyske Bank

39

1 Introduction

Low-income countries are characterized by having scarce resources to promote their devel-

opment. The most important are the domestic resources, which come from taxation on

economic activity or government borrowing from the savings that people and companies

make. Typically, the main economic activity in a low income country is concentrated in few

primary products, which make its overall resources volatile to the variation in international

prices.

However, domestic resources may not be enough to create investments needed for eco-

nomic development. They need to be supplemented by external financial resources, such

as FDI, aid, loans, debt cancellation, portfolio equity investment, and remittances. These

flows could, in theory, accelerate the economy by accumulating capital, or providing impor-

tant public goods with positive externalities. However, theory also predicts the possibility

of Dutch disease, which result in a damage of the tradable sector (Temple, 2010).

Bolivia is an example of a low-income country, depending on its mining and oil sectors,

and has received all kinds of external resources and relies on these to promote its develop-

ment. The traditional flows have been aid, FDI, and loans. In recent years, there has been

an increase of the non-traditional transfers such as remittances and debt relief 1.

The sources of external financing have varied according to Bolivia’s development. Bolivia

went through many radical economic changes in the period of study (1950-2009): from a

nationalized economy, with state-owned enterprises which relied on public loans until the

debt crises in 1978, to a fully liberalization of the economy from 1985 until 2005, where FDI

played an important role due to the privatizations reforms. From 2005, Bolivia went into

a period of nationalization again. Aid became important from the period of liberalization

while remittances became the most important source after 2001. Debt relief has been given

mainly in the last years but forgiveness and re-schedules have been registered along the

whole period, especially after the debt crises.

1Debt relief is not a transfer itself but instead is a cut on public debt repayment conditioned on publicspending on certain areas (Burnside and Fanizza, 2004)

40

Previous research on external financing to low income countries have focused on aid.

Doucouliagos and Paldam (2008) analyse the results of 100 papers until 2005 relating aid

and growth, and find that in general, the relation is positive but small, insignificant and

falling. However, a recent paper by Juselius et al. (2013) who use time series data to set up

a VAR model for each African country for 30 years, find that for a great majority of their

countries, aid flows are positive for growth in the long run.

In Bolivia, Andersen and Evia (2003) use a computable general equilibrium model to

simulate a temporary increase of foreign aid in the economy. They find a positive effect

during the period of aid, but then this effect dissipates, thus a null effect in the long run.

Similarly, Morales and Cazadilla (2003) find that aid is negative for growth in the long

run by using a dynamic computable general equilibrium model under the context of a

small economy with low physical productivity that receives aid not oriented to address this

problem.

Regarding debt and growth, Paldam (2008) finds a negative relation for a sample of

88 developing countries. On the other hand Cordella et al. (2005) argues that the debt-

growth relationship varies with indebtedness levels and other country characteristics. Their

findings, for a sample of 79 developing countries, suggest that there is a negative marginal

relationship between debt and growth at intermediate levels of debt, but not at very low

levels of debt nor at very high levels. With respect to debt relief, Johansson (2008) analyses

118 developing countries, and finds no general evidence of a growth effect from debt relief

in HIPC countries (Heavily-indebted poor countries countries)2.

In Bolivia, Humerez and Hernaiz (2006) use a VAR setting, with quarterly data from

1990 to 2003, and find a negative relation between external debt and growth in the short

and long run.

However, empirical research on comparison of external funds to low-income countries

is totally absent. Instead, studies generally focus on one flow and many countries at the

same time, commonly in short periods of time. Many of them ignore the endogenous

2See also Hepp (2008) and Depetris Chauvin and Kraay (2005) for debt relief discussion.

41

characteristic of the flows. External resources in a low-income country are a decision from

the sender country based on the recipient’s characteristics. For example, FDI come to a

low-income country if the private agents in the high-income countries have the capital and if

the low-income country has the potential of good returns. On the contrary, aid has different

incentives, high income countries provide assistance if they are doing well and if the low-

income country needs it. This negative relation can be misunderstood as the effect of aid

on growth. On the other hand, loans react in a different manner: a growing country will

be better able to attract loans, while concessional loans are given under certain conditions.

We want to fill in the gap in the literature by analysing the most traditional capital flows

during 60 years, 1950-2009, more than any other study, in Bolivia. We use a cointegrated

VAR(2) model which allows us to account for endogenous relations and to omit strict

assumptions about the model. One of the most important endogenous relation, is the

one between the flows and the Bolivian GDP. Another inter-relation is that between the

flows. These are determined by sender country characteristics, and the flows themselves can

complement or substitute each other. A further contribution is that we distinguish between

the long-run from the short-run relations.

The only other studies using a time-series approach are Juselius et al. (2013) and

Humerez and Hernaiz (2006). Juselius et al. (2013), fit CVAR models to 36 sub-Saharan

countries. These authors only consider aid among the flows and Humerez and Hernaiz

(2006) only consider debt in Bolivia, whereas we are interested in the different roles of the

flows and the relations among them, so we in addition include FDI, loans, and exports.

Furthermore, we consider the possibility of Dutch disease and so include a price series, in

particular the purchasing power of exports, which is not considered by Juselius et al. (2013).

Finally, we assess the dynamic impact of the different variables through generalized impulse

response functions for the endogenous variables and cumulative dynamic multipliers for

the exogenous variables, whereas Juselius et al. (2013) focus on the contribution of aid to

growth and on determining in how many and which of the countries aid is exogenous. The

analysis of multiple countries (whether by time-series, cross-sectional, or panel methods) is

42

more relevant, e.g., for Africa, where there are many poor countries, whereas Bolivia is a

more isolated case in its region, considerably poorer than most of its neighbours (Figure 1).

Figure 1: GDP per capita, Bolivia’s neighbours. In Current US Dollars per person

020

0040

0060

0080

0010

000

1940 1960 1980 2000 2020year

Bolivia Argentina Brasil Chile Paraguay Perú

Our results indicate that loans are positive for the Bolivian economy in the long run,

through an increase in exports and a decrease in purchasing power of exports however, we

did not include total debt. We find that the effect of aid is negative for the Bolivian GDP

per capita in the long run, which is explained by the temporary increase in purchasing

power of exports, which lead to a permanent loss in the export markets. Finally, foreign

direct investments lead to a positive response of Bolivian GDP per capita and exports, and

a reduction of purchasing power of exports in the long run.

Our results indicate that FDI turns out to be exogenous to the system, with insignificant

short run marginal effect and positive long run effect on Bolivian GDP, the cumulative

dynamic multiplier is positive throughout, and there is an increase in exports and a reduction

of purchasing power of exports. Aid is endogenous: when GDP increases, aid is reduced.

The marginal effect of aid on GDP is only positive in the short run and negative in the

long run, the generalized impulse response is negative throughout, and there is a temporary

increase in purchasing power of exports, which lead to a permanent loss in the exports.

Finally, the marginal effect of loans is insignificant in the short run but positive in the

43

long run, the generalized impulse response is positive throughout and there is a decrease in

purchasing power of exports resulting in a increase of exports.

However, our approach has limitations regarding the loans result. Due to lack of detailed

data during the period of study, we do not include total debt at market values, which should

reflect a negative relation in the long run (although more than 50% of the debt has been

forgiven so far). The frequent shifts in the classification of debt because of rescheduling,

arrears, and the assumption of the debt of one sector by another, makes the analysis of debt

very difficult (Morales and Sachs, 1990b).

Our paper is organized as follows. First, we describe the flows in the Bolivian economic

history context. Second, we describe the data, the sources and some descriptive statistics.

Third, we describe the methodology. Fourth, we illustrate and discuss the results. Finally,

we present the conclusions.

2 Description of the flows in the historical context

Figure 2 shows the sum over the years of all the traditional and non traditional external

flows in Bolivia. The most important in value are indeed the traditional because they were

given for many years, first FDI, then loans and finally aid. The rest of the flows became

important in recent years. Right after aid, come remittances, which have gained ground

compared to the other flows. In fact, remittances became the most important foreign flows

in the last years 2008-2009 (see Figure 3). Then, debt relief also show high values at some

years compared to the other flows since it was given only in some years.

The traditional flows were attracted, given and used according to the specific context

of the Bolivian economy. Therefore it is important to describe the development path in

Bolivia during the period of analysis.

From the beginning of our period until 1978, Bolivia’s economy was based on a model of

nationalization, where the state was the main economic actor controlling many important

companies including the oil and mining ones. In that period, the external debt raised in

44

Figure 2: Capital flows to Bolivia. Sum over years. In millions of US dollars

02,

000

4,00

06,

000

8,00

0

FDI LoansAid RemittancesDebtRelief

PortfolioInvestment

order to finance the expenditures of public enterprises (Machicado, 2013). Furthermore, the

external context in this period was beneficial in terms of oil and mining export prices.

Such large inflows of credits from international banks mainly, stopped drastically pro-

voking a crises in the country in 1978 3. As a result, Bolivia experienced one of the highest

hyperinflations of the world, which was attributed to finance the fiscal deficit by printing

money (Machicado, 2013). However, concessional loans replaced regular loans which were

aimed first, at helping reconcile the balance of payments (Prego, 2011).

In order to get access to capital markets again, Bolivia followed the Washington Concesus

policies, which were focused on liberalizing the economy and reducing the size of the state.

Such reforms were carried out from 1985. One of the reforms, carried out in the 90’s, was

to privatize the most important state companies. In this way, FDI was attracted to the

country. However, the external context was not as favourable as before.

Furthermore, in this period, aid, as unilateral transfers started taking place due to the

levels of poverty and the efforts of liberalization, which was a condition from the interna-

tional cooperation. In 1997 Bolivia qualified as a HIPC country - Heavily Indebted Poor

3Which was triggered by the shock of international oil prices in 1974 which provoked the onset of highinterest rates.

45

Country, which signified that could receive debt relief 4. With this program, the IMF has

expected to reduce Bolivia’s total debt by 58% in 2014.

The last period starts in 2006, when Evo Morales assumed the presidency and national-

ized the economy again, where the state plays the leading role by owing the main companies

in strategic sectors such as oil, electricity and telecommunications, under a favourable in-

ternational context so far.

3 Data

We use two different sources to obtain the data for flows: the International Monetary Fund-

International Financial Statistics (IFS) and the United Nations-Economic Commission for

Latin America and the Caribbean (ECLAC), which constantly adjust their data to make

it comparable between countries. The Appendix provides a detailed description of the

merging process. Unfortunately, remittances, debt relief and portfolio investment have very

few observations5. Therefore, we focus on the traditional flows since our approach is to look

into the long run effects. The flows used in the econometric analysis are described below.

Aid, defined as unilateral transfers to the public sector, which comprises international

cooperation. The main characteristic of this flow is that it does not need to be paid back,

contrary to concessional loans, which are considered part of ODA (Official Development

Assistance). We think that concessional loans should go into the loans variable, since they

require repayment. We also exclude debt relief (also included in ODA) from our variable

aid, because we think that it is not entirely correct to consider it a flow itself when there

was no real transfer of money. It should definitely enter in lowering the total debt.

Loans, defined as net loans to the public and private sector excluding debt relief. This

variable includes concessional loans to the public sector, where interest rates are lower than

market rates.

4The debt relief program is based on the notion that high indebted countries channel their resources inservicing its debt rather than spending it on reducing poverty. All information about debt relief is on theIMF website.

5Actually, we have observations for remittances through all the period but most of it is close to zero orzero

46

−1,00001,0002,0003,000

1950195119521953195419551956195719581959196019611962196319641965196619671968196919701971197219731974197519761977197819791980198119821983198419851986198719881989199019911992199319941995199619971998199920002001200220032004200520062007200820092010

Por

tfolio

Inve

stm

ent

FD

ILo

ans

Aid

Rem

ittan

ces

Deb

tR

elie

f

Fig

ure

3:

Cap

ital

flow

sto

Bol

ivia

by

year

.In

mil

lion

sof

curr

ent

US

dol

lars

per

year

.

47

FDI, defined as foreign direct investment.

Data on GDP per capita of Bolivia is taken from the Bolivian Statistical Institute (INE).

As a proxy for senders development situation we build a GDP index, Y ∗, that takes into

account GDP per capita of several relevant countries, where the source is Penn World Tables

version 7.0 (Heston et al., 2011), calculated as:

Y ∗t =∏

GDP ∗pctθ

where GDP∗pc is the real GDP per capita of several countries in constant prices, and θ is

the weight of importance of each country on Bolivia’s GDP 6. We assume θ is constant over

time, and the countries considered are the most important trading partners. Each weight

is calculated as the share of Bolivian exports plus imports in terms of the total Bolivian

GDP averaged from 1980-2010. The resulting weights are shown in Table 1.

Finally, the relation of export to import prices is taken from ECLAC as the ratio be-

tween the export and import deflator indexes with base year in 2005. This relation can be

interpreted as the terms of trade 7.

Table 1: Main trading partners and their weights

Country Weight Country Weight

U.S.A 0.071 Brazil 0.063Argentina 0.058 Japan 0.023Peru 0.020 U.K. 0.017Chile 0.017 Germany 0.013Sweden 0.012 Colombia 0.012

6In a different context, Bruckner (2011) uses a geometric average to construct a country-specific exportprice index, with time-invariant weights reflecting the relative importance of different goods in the country’sexports.

7Time-series figures of each of the variables are in the Appendix

48

3.1 Stationarity of the variables

Table 2 displays descriptive statistics for the variables used in our analysis. Next section

explains how each variable enters in our model and why. Whenever possible we use the log

of the series 8.

Since we have long term data, we need to analyse if the variables are stationary or not.

We conducted augmented Dickey-Fuller tests including a constant in the regression equation

to perform unit root test. The tests were carried out with a maximum lag length of 8 and

for each variable the model with the lowest AIC was chosen. The results are shown in Table

3, and show that at 5% significance we fail to reject the null of a unit root in all the series

but the loans.

Table 2: Summary statistics

Variable Mean Std. Dev. N Label

pib bol00 909.601 133.584 59 Bolivian GDP per capita (US.Doll.2000)ystar1 14.881 1.506 59 GDP per capita of senders(I.Doll.2005)relp05 100.56 26.583 61 Relation of export to import prices (Base:2005)fdi 138.356 261.666 60 FDI(Mill.US.Doll)tr c 98.006 105.708 60 Aid(Mill.US.Doll)debt drt 118.072 181.342 61 Loans(Mill.US.Doll)exp05 927.577 740.931 60 Exports(Mill.US.Doll.2005)

Table 3: ADF Unit Root Test

Variable t-ADFc t-ADFc,t

lpib bol00 -2.757 -3.493lystar1 -1.263 -2.335lrelp05 -2.279 -2.254fdi -0.104 -1.637ltr c -1.157 -1.884debt drt -4.535 -4.515lexp05 1.087 -0.216

t-ADFc includes constant in regression. 5% = -2.92.

t-ADFc,t includes a constant and a trend in regression. 5% = -3.50.

8For example, FDI can be negative due to disinvestments. Loans may also be negative, simply reflectinga net repayment of debt in that year.

49

4 Methodology

We are interested in correcting for the endogeneity characteristic of capital flows, which

will allow us control the interaction between the different variables and isolate the effect

of each of the capital flows on Bolivia’s GDP and other variables. Another challenge is to

distinguish between the long-run from the short-run relations. We focus on the traditional

flows: FDI, loans and aid with data from 1950 to 2009.

Instead of testing a specific theory we will employ a general to specific approach. In

this fashion we remove insignificant variables until we obtain a more parsimonious model

with only significant coefficients. We do however, have some prior expectations about the

relations between some of the variables.

Our intuition tells us that capital flows to low-income countries are determined by their

low-income situation and by the economic possibilities of sender countries. In the case of

aid, the main reason for their presence in Bolivia is non-economic, in the sense that senders

do not expect returns but the level of aid depends on the economic situation of Bolivia. We

have seen that the most important determinants of aid is the lack of growth, poverty, or

higher levels of debt in the recipient country.

In the case of FDI and loans, the driving reasons are more economic ones, in the sense

that senders expect economic returns or payments of interest. Theoretically, FDI is higher

when country prospectives and the investment environment are improving. The case of

debt is less clear, rich countries and poor countries borrow alike.

In all cases, all kinds of flows are likely to be conditioned by having a prosperous

situation in the sender countries. Whenever a sending country is doing badly, the flows to

other countries, especially to low-income countries will probably diminish. Moreover, we

expect that the different forms of capital flows are best seen as complements, not substitutes

(Reinert and Goldin, 2006).

A good approach for solving the endogeneity problem and looking into interactions

between the variables is the VAR(p) model. In the baseline model we will use a VAR(2)

50

with the following variables: FDI, log(aid), total loans, log(Bolivian GDP) and log(Y ∗),

purchasing power of exports, and log(exports)9. Formally the VAR(2) for a system of p

equations is,

xt = µ+ Φ1xt−1 + Φ2xt−2 + εt (1)

We will consider a version of the model above where Bolivian GDP, Y ∗ and exports are

in real terms. Since we do not have a proper deflator for FDI, aid and loans we simply use

current values. This is in line with Bruckner (2011) who investigates the effect of nominal

aid on real GDP in a panel data model. Furthermore one might be interested in the effect

of nominal flows on real GPD.

The data analysis showed that all these variables except for the loans series seem to be

I(1), whereas a VAR model requires stationary variables for proper statistical inference. A

typical solution is to do a VAR in first differences but we prefer not to miss any possibility

of long term relations since presence of unit roots indicates that there might exist one

or more cointegration relations that are important to take into account. We test for the

cointegration rank and the results are shown later. Therefore we reformulate the VAR(2)

as a VECM(1),

∆xt = µ+ Πxt−1 + Γ1∆xt−1 + εt (2)

where Πxt−1 describes the long-term relation. The presence of unit roots implies that

Π cannot be inverted and is therefore of reduced rank. In this case the long term relation

can be written as αβ′ where β is of dimension p× r with r denoting its rank. The matrix

β is such that it renders the linear combination β′xt−1 stationary and β′xt−1 can be given

the interpretation as a stationary long run relation.10 The α matrix may be interpreted

as measuring the speed of adjustment towards the long run stationary relation when the

9The variables in logs will henceforth be referred to without log, i.e. log(aid) will simply be written asaid

10We treat loans like the other variables since they are stationary and so including them do not alter thelong run relation, and similarly for the first differences.

51

system is shocked away from equilibrium.

Inference and estimation of β is carried out using Johansens likelihood based approach.

Unless otherwise stated we used STATA version 11 for estimation. First the rank of β is

determined, and then we impose identifying restrictions by inserting an r×r identity matrix

in the top of β to obtain standard errors and further test down for insignificant coefficients

in β. Thus when the rank was higher than one, we decided to try out different orderings of

the possible combinations, except that we never allowed for Bolivian GDP to be restricted,

since we are interested in identifying long run relations between the flows and Bolivian GDP.

We chose between the different orderings by selecting the model(s) with most significant

coefficients, because this/these model(s) produce higher likehood values.

With the long run relation β properly identified we followed Juselius (2006) chapter 13

and created the stationary relation zt = β′xt−1, with xt =[ xtxt

]and set up the VAR in

differences,

∆xt = αzt + Γ1∆xt−1 + Γ1∆xt−1 + εt (3)

The statistical justification for letting zt = β′xt−1 and treat it as a predetermined

stationary variable is that β is superconsistent and converges with rate T as T →∞ whereas

α converges with the usual rate of√T as T → ∞ (see Juselius (2006)). We then imposed

zero restrictions on insignificant coefficients and estimated the model by maximum likelihood

employing iterative POLS estimations11. Furthermore some variables had only insignificant

coefficients and so did not respond to the long run relations, i.e., the corresponding rows in

α were zero. These are cases of long run exogeneity, see Juselius et al. (2013). Joint short

and long run exogeneity requires in addition that the Γ1 coefficients on the endogenous

variables are zero. In such cases, we decided to treat the variable(s) as exogenous, indicated

by ∆xt−1. Since our sample is rather small the change of a variable from endogenous to

exogenous may have a rather large positive effect on efficiency due to fewer coefficents to

be estimated.

11See Hamilton (1994) chapter 11 for maximum likelihood estimation of the restricted VAR

52

4.1 Generalized impulse responses

To study the effects of a shock in the j’th equation and trace out the dynamics of the

system, we applied the results in Pesaran and Shin (1998) and calculated the generalized

impulse responses for the endogenous variables by rewriting the model back into a levels

representation. Formally the generalized impulse response12 in the cointegrated VAR(p)

from a shock in the jth equation on xt+n, n = 0, 1, 2, ... is,

ψx,j(n) = σ− 1

2jj BnΣej . (4)

Here Σ = E(εtε′t) is the covariance matrix from the VARX-model in differences in (3)

and σjj is the (j, j) element of Σ. The selection vector ej has the number 1 at its jth

position and zeros elsewhere. Bn is the cumulative effects matrix with B0 = Ip and Bn

can be calculated recursively from the VAR coefficients Φi, i = 1, 2, ..., p. For the VAR(k),

Bn = Φ1Bn−1 + Φ2Bn−2 + · · ·+ ΦkBn−k, n = 1, 2, . . .

where Bn = 0 for n < 0.

The advantage of the generalized impulse is its invariance to the ordering of variables,

which orthogonalized impulse responses on the contrary suffer from. The orthogonalized

impulse responses do however solve the problem of a non-diagonal covariance matrix, by

multiplying the model in (3) through by the Cholesky decomposition of the covariance

matrix, forcing the errors to be contemporanously uncorrelated. One could also circumvent

the problem of non-diagonal covariance matrix by imposing proper structure on the VAR,

however we have no theoretical model to suggest a particular model structure. Hence we

decided to use the generalized impulse responses.13

For the exogenous variables we calculated the cumulated dynamic multipliers to obtain

12For a thorough exhibition see Pesaran and Shin (1998). Here we will only reproduce the main results13Juselius et al. (2013) do not consider the generalized impulse responses. They test for exogeneity as a

zero condition on the endogenous variables in the long run impact matrix, i.e., the sum of the coefficientmatrices in the infinite moving average representation of the first differenced system.

53

the effect of a change in the exogenous variable on to the endogenous variables in the system.

5 Results

We set up the VAR(2) model with Y ∗, purchasing power of exports, total loans, exports in

constant 2005 prices, aid, FDI and Bolivian GDP. Before proceeding with the estimations

we determined the rank by the top-bottom procedure. We allowed for a constant restricted

to the cointegration space, and the rank test shows that the rank is r = 2.

Table 4: Johansen Test for Cointegration

Rank Parameters LL Eigenvalue Trace Statistic Critical Value

0 49 -293.05 . 151.34 131.701 63 -271.75 0.54 108.74 102.142 75 -253.97 0.48 73.19* 76.073 85 -239.36 0.41 43.97 53.124 93 -229.92 0.29 25.09 34.915 99 -222.77 0.23 10.79 19.966 103 -219.08 0.13 3.40 9.427 105 -217.38 0.06

Note: Number of observations = 55, Lags = 2. Trend: Restricted constant. Critical value = 5%. Vars: fdi,lpib bol00, lexp05, ltr c, debt drt, lrelp05 , lystar1.

We then estimated the model for the 15 different possible orderings and found that the

order just listed produced a model in which all the variables were significant and we can

thus proceed without making further restrictions. Furthermore the relations were more

intuitive and easier to interpret than the long run relations of other orderings.

Table 5 below shows that in the long run Bolivian GDP is positively correlated with

foreign direct investment, sender GDP and real exports, whereas there is a negative corre-

lation between GDP and aid, reflecting that either aid has a negative impact in the long

run on Bolivian GDP or that aid is reduced as Bolivian GDP rises. The positive correlation

of GDP with purchasing power of exports and loans may be somewhat surprising. The

purchasing power of exports is export prices divided by import prices and a higher value of

this implies a less competitive exporting sector, which should hurt exports. However higher

prices may also be a sign that the value of the exported goods have gone up, in which case

54

one would expect a positive correlation between Bolivian GDP and purchasing power of

prices in the long run. The positive correlation between loans and GDP may be explained

by the fact that as countries grow richer it becomes easier to borrow.

With only one cointegrating relation, OLS would be superconsistent for the cointegrating

vector. We also implemented OLS regressions of GDP on all other variables, as well as on

all variables except purchasing power and except respectively Y ∗. The results were broadly

similar with those in Table 5, obtained from the system estimation, as were results from

regressions in first differences, except that in this case purchasing power showed a negative

effect if no variable was excluded. However, since there are two long run relations, OLS

is strictly inappropriate. Furthermore, the relevant dynamic impacts of different variables

are better measured by impulse responses within the the system, as we do below. The

cointegrating relations only reflect the long run relations not correcting for endogeneity,

i.e., they do not isolate causal effects.

Table 5: Cointegrating relations

lystar1 lrelp05 debt drt lexp05 ltr c fdi lpib bol00 cons

ce1 1 0 .00016995 .09960395 -.07980207 .0001064 -.72581809 1.8611997ce2 0 1 .00185519 .70635567 -.2805189 .00132841 -3.8514971 17.613012

Plotting the cointegrating relations in Figure 4 also shows that both relations are sta-

tionary, thus motivating to proceed with estimation of the VAR model in (3).

Considering the unrestricted cointegrated VAR(1) in differences (see Table A.1 in the

Appendix) we restricted the most insignificant values to zero and continued doing so until

the system only contained significant coefficients. The results for the restricted model, in

Table A.2 in the appendix, show that Y ∗ does not respond to the long run relation, i.e.,

it is long run exogenous. Furthermore, in the short run dynamics Y ∗ is only a function of

its own lagged first difference and lagged first difference of Bolivian exports. That lagged

Bolivian exports should be a determinant for Y ∗ seems pretty odd and counter intuitive,

because Bolivia is a very small economy compared to its trading partners. Thus we believe

it is correct to consider Y ∗ as exogenous (in the short and long run) to the system. In the

55

Figure 4: Cointegrating relations

−.1

−.0

50

.05

.1.1

5P

redi

cted

coi

nteg

rate

d eq

uatio

n

1940 1960 1980 2000 2020year

ce1−

1−

.50

.51

Pre

dict

ed c

oint

egra

ted

equa

tion

1940 1960 1980 2000 2020year

ce2

FDI equation the only significant variable was the exogenous Y ∗ with a positive coefficient,

and the endogenous variables as well as the equilibrium relations were all insignificant. This

shows that also FDI is exogenous, thus yielding an even more parsimonious system.

The final model is thus a VARX in differences as in (3) with five endogenous variables,

purchasing power of exports, loans, real exports, aid, and Bolivian GDP per capita, and

with Y ∗ and FDI as exogenous, and two equilibrium relations zit, i = 1, 2. Treating two

variables as exogenous results in many more significant coefficients.

Employing the general to specific approach in the unrestricted VARX we removed in-

significant coefficients and ended up with a total of 29 zero restrictions. From Table A.3

listing the restricted model we observe that the endogenous variables all responded to either

one or both of the long run relations, indicating that the identified long run structure plays

an important role in the VARX model. Had one proceeded with a model only in differ-

56

ences without taking the cointegrating relations into account one would have lost important

information about not only the long run relation but also about the dynamic adjustment.14

Given our estimated coefficients and covariance matrix from the VARX model in differ-

ences, we calculated the generalized impulse responses. All the shocks we studied were of

one standard deviation. 95% confidence bands were calculated by first running a simulation

of the VARX using the estimated coefficients and covariance matrix to obtain new time se-

ries for the endogenous variables. Next the model was re-estimated using the simulated data

and a new set of coefficients and a covariance matrix was obtained. In total we did 1, 000

repetitions and for each repetition we used the re-estimated parameters to calculate the

generalized impulse responses for all the shocks on all the endogenous variables. Confidence

bands were found by adding and subtracting the absolute deviation of the 2.5 percentile

and 97.5 percentile from the median to the estimated impulse response respectively. In

the graphs for the generalized impulse responses the center-line depicts the expected effect

estimated from the VARX and the shaded area depicts the 95% simulated confidence band.

5.1 The effect of aid

First we consider the effect of a shock in the aid equation on the system displayed in Figure

5. Notice that by Figure 5a the effect on GDP pr. capita of a shock in the aid equation

is statistically significantly negative with the effect becoming larger as time progresses. In

particular GDP pr. capita is affected mostly within the first five to ten years after which

GDP pr. capita stabilizes at a permanently lower level about −0.04 lower than initially.

This is in spite of the positive (0.017) and significant coefficient on aid in the GDP equation

(see Table A.3, equation 5), thus reinforcing the importance of our approach, assessing

dynamic impacts using the generalized impulse responses.

So what causes the impact of aid on GDP pr. capita to be negative? Consider what

happens to real export as a result of the aid shock displayed in Figure 5b. The graphs shows

14Indeed, we implemented the first differenced system without the cointegrating relations. The resultschanged drastically, e.g., in the GDP equation, the effect of Y ∗ turned positive and significant, whereas it isnegative and insignificant in the CVAR, and purchasing power turned negative, whereas it was significantlypositive in the CVAR, consistent with the VAR in differences being misspecified.

57

that despite a positive effect in real exports after a few years the result is that exports is

significantly harmed with a negative effect that after around ten years on expectation will

be lowered by 0.06. Somehow aid seems to affect GDP pr. capita negatively through its

negative impact on real exports. We can however shed even further light on the mechanism

driving the negative relation between GDP pr. capita and aid by considering Figure 5c,

because export prices actually increase after a short fall in the first year with a positive

price effect until around five years after the aid shock. Thereafter the purchasing power of

exports falls to a permanently lower level. So the mechanism may very well be that the

purchasing power of exports increases temporarily implying a less competitive exporting

sector, where the effect actually seems to be permanent on exports. A possible explanation

is that aid creates Dutch decease effects, with an increase in domestic demand raising input

costs in the exporting sector and therefore increasing costs and lowering competitiveness.

5 10 15 20

−0.

05−

0.04

−0.

03−

0.02

−0.

01

(a) Response of GDPpc

5 10 15 20

−0.

08−

0.06

−0.

04−

0.02

0.00

0.02

0.04

(b) Response of export inconstant prices

5 10 15 20

−0.

010.

000.

010.

02

(c) Response of purchas-ing power of exports

Figure 5: Generalized Impulse Responses to One Standard Error Shock in the AidEquation

Still, why haven’t we seen the positive effects of aid either as capital accumulation, or

creation of public goods or goods with externalities? Temple (2010) gives the example of

aid directed into roads which in fact could help the tradable sector and offset the negative

impact of Dutch disease. One could also think about education and health.

Aid in Bolivia has been directed into different areas defined by the donor community.

During the 50’s aid consisted mainly of donations of flour, wheat, and sugar among others

58

(Memorias of 1955, BCB). Later from the 80’s, donations increased gradually due to a com-

promise of the donor community to help poor countries (OECD). The donations consisted

of scholarships, and donations in kind (Memorias of 1983, BCB). While its value increased

over time, the projects became more and more diverse, varying from donor to donor15.

Andersen and Evia (2003) highlight some of the areas where donations were directed in re-

cent years, namely coca eradication, rural development, roads, institutional strengthening

among others.

In most of the cases, aid was required to have a counterpart of Bolivian expenditures and

furthermore had a condition to use the donors country’s resources, as machinery, personal,

etc.(Report VIPFE, 2005). So a lot of the expenditures have been attributed to transporta-

tion costs (OECD). Furthermore, there has been lack of donor coordination between the

donors and also with the Bolivian Government (Danida, 1997). Nowadays, though, there

are intentions of a more coordinated work between the parties (Report VIPFE, 2011). All

evidence, confirms that aid has not gone to the creation of public goods, enough to matter

for growth nor to create capital in any productive sector. Instead it went to several projects

that were costly for the Bolivian economy.

5.2 The effect of loans

Consider next the effect of a shock in the total loans equation in Figure 6. In all the cases

the effect seems to be statistically significant. By Figure 6a the response of GDPpc is a

permanent increase of about 0.03 points after around 10-15 years. This seems to suggest

that total loans is invested into productive sectors of the economy, generating a positive

feedback to production. The figures in Figure 6b and Figure 6c indicate that the exporting

sector becomes more competitive as a result of the shock in the total loans equation due to

a fall in the purchasing power of exports and hence an increase in constant price exports.

The loan variable includes both private and public loans (including concessional loans)

15In 2005, the first attempt to document details of the cooperation, revealed that in Bolvia there are 8multilateral agencies, 14 bilateral agencies (countries), and 11 organizations from the United nations.(ReportVIPFE, 2005)

59

5 10 15 20

0.01

0.02

0.03

0.04

(a) Response of GDPpc

5 10 15 20

0.00

0.02

0.04

0.06

(b) Response of export inconstant prices

5 10 15 20

−0.

09−

0.08

−0.

07−

0.06

−0.

05−

0.04

(c) Response of purchas-ing power of exports

Figure 6: Generalized Impulse Responses to One Standard Error Shock in the TotalLoans Equation

after repayment, and excludes debt relief, being the public debt the more dominant com-

pared to the private sector (Morales and Sachs, 1990b). Our perception is that private

loans were probably intended into productive sectors, because in order to get a loan in first

place, the private agents need to convince borrowers of their ability to repay and therefore

profitability. On the other hand, we observe that public loans were high in a period of

state-owned enterprises, which were capital-intensive16. Among the major recipients, we

have the state oil company, YPFB, followed by the smelting company, CMK (Complejo

Minero Karachipampa), and the state mining company COMIBOL (Morales and Sachs,

1990a).

In the period of liberalization, data shows that the sector with more concentration of

loans is roads (Statistics in BCB). Still in order to get the loans, the government needs to

think about its repayment schedule.

So we think that the positive effects were dominant because loans went into capital

accumulation and public goods, and the destiny of the money is decided by local residents

contrary to aid which destiny is decided by donors, i.e. local people with the right incentives

can make better use of the resources. Another difference between aid and loans is the cost

of borrowing, which gives a different motivation for its use. However, the cost of debt can

16However, a lot of money went into corruption, precisely because the public sector was so powerful

60

be very high for a poor country.

Unfortunately, we did not include total debt, which should reflect the costs of loans in

the long run (as Humerez and Hernaiz (2006) find for the period 1990-2003). The best

way to include the cost of loans is to calculate the present value of debt at each year. We

tried to do this but because we need to take into account the amount of rescheduled debt,

concessional loans, the debt relief program, and market values at each year, we were not

able to do it for such a long period. Instead we accumulated the loans to represent the

cost of having debt but the numbers were far from the official numbers (we do have access

to total debt values but only for recent years, from BCB) and did not reflect the real cost

of debt. Therefore we decided to include repayments at each year, however excluding debt

relief due to its exceptional high values in some years.

Nonetheless the accumulated debt in HIPC countries has been found not to be nega-

tively correlated with growth (Cordella et al., 2005). The author mentions that non HIPC

countries are more depending on international capital markets, and resource flows are more

responsive to indebtedness than HIPC countries, who receive resources from the donor

community to prevent debt repayments from crowding out investments.

5.3 The effect of FDI

The last flow to consider is FDI, which was treated as exogenous where the effect of a change

in FDI can be measured by the cumulative dynamic multiplier. For comparison with the

impulse responses we used a standard deviation change in FDI. The results are shown

in Figure 7 and Figure 8. Not surprisingly by Figure 7a, GDP increases to a permanently

higher level (0.015) above the initial level. FDI is a direct investment into productive capital

and furthermore it may be accompanied by a technology and knowledge transfer by which

the production of the economy can grow. The effect on exporting prices or the purchasing

power of exports of an increase in FDI depicted in Figure 7c is negative which indicates

that FDI increases competitiveness of the exporting sector. However by Figure 7b, export

initially falls to revert back to a positive permanent higher level than initially. Both aid

61

and total loans fall as a direct consequence of an increase in FDI (see Figure 8a and Figure

8b). The likely reason for aid to fall is probably due to the higher level of GDPpc17. FDI

probably reduces the need to borrow money domestically to finance investment projects

hence the decrease in total loans as seen in 8b.

5 10 15 20

0.00

00.

005

0.01

00.

015

0.02

00.

025

(a) Response of GDPpc

5 10 15 20

−0.

03−

0.02

−0.

010.

000.

010.

020.

03

(b) Response of export inconstant prices

5 10 15 20

−0.

08−

0.06

−0.

04−

0.02

0.00

(c) Response of purchas-ing power of exports

Figure 7: Cumulative Dynamic Multiplier to One Standard Change in FDI

5 10 15 20

−0.

15−

0.10

−0.

050.

00

(a) Response of aid

5 10 15 20

−15

0−

100

−50

0

(b) Response of total loans

Figure 8: Cumulative Dynamic Multiplier to One Standard Change in FDI

FDI in Bolivia has been directed into the traditional sectors which are mining, gas

and other natural resources. These kind of investments are capital intensive and therefore

accumulate capital and therefore growth. FDI has increased in the years of liberalization

when state enterprises were privatized but lowered later when there was political unrest

around 2005 and later increased.

17See more on the relation between a shock in the GDPpc equation and aid below

62

5.4 The effect of GDP

Finally we examine the effect of a shock in the GDPpc equation and observe that the

impulses all are significant at 95%. One of the motivations to do a VAR model of the flows

was exactly to take into account how GDPpc affects the flows. For instance we hypothesized

that an increase in Bolivian income would result in less aid. The graphs are depicted in

Figure 9 and Figure 10. Figure 9a shows that aid responds negatively to the GDPpc shock,

thereby supporting our hypothesis that aid decreases when Bolivia grows richer. The effects

on loans is an increase as seen in Figure 9b probably reflecting that it becomes easier to

borrow when income goes up, i.e. the size of the collateral has increased.

5 10 15 20

−0.

25−

0.20

−0.

15−

0.10

(a) Response of aid

5 10 15 20

2030

4050

(b) Response of loans

Figure 9: Generalized Impulse Responses to One Standard Error Shock in the GDPpcEquation

The impulse responses in Figure 10 show that export prices fall and export increases as

a result of a positive GDPpc shock. Thus overall the GDPpc shock seems to be self enforcing

leading to a chain of positive implications for the Bolivian economy.

63

5 10 15 20

0.00

0.01

0.02

0.03

0.04

0.05

0.06

(a) Response of export inconstant prices

5 10 15 20

−0.

035

−0.

030

−0.

025

−0.

020

−0.

015

−0.

010

(b) Response of purchas-ing power of exports

Figure 10: Generalized Impulse Responses to One Standard Error Shock in the GDPpcEquation

6 Conclusions

We study the impact of capital flows to Bolivia through a multivariate CVAR(2) model.

This set-up allows us to correct for the endogeneity issues and study the impacts of shocks

to the system through impulse response analysis. We show that aid, loans, exports and

purchasing power of exports together with Bolivian GDP pr. capita form a system of five

endogenous equations with an additional two exogenous variables, FDI and foreign GDP

and two long run cointegrating relations. The novelty of our study is the combined analysis

of all these capital flows in a dynamic multivariate model which directly deals with the

endogeneity and allows for a deeper investigation through impulse responses to study the

different impact shocks in each flow and on the Bolivian GDP, both in the short run and

long run. The impulse responses also provide a more profound insight into the chain of

causality, through the dynamic patterns displayed.

Our empirical results indicate that loans are positive for the Bolivian economy in the

long run, through an increase in exports and a decrease in purchasing power of exports.

Furthermore, we observe that loans increase with a rise in Bolivian GDP, reflecting that

a growing economy will be able to increase its loans. Overall, loans seem to have been

directed into the capital-intensive sectors of the economy or on public goods that provide

64

externalities, resulting in an increase in both exports and Bolivian GDP per capita. How-

ever, we did not include total debt at market values, which should show a negative relation

with the GDP in the long run, although, great part of it was forgiven so far (more than

50%).

Considering aid, our results based on the generalized impulse responses show that the

effect is negative for the Bolivian GDP per capita, probably explained by the accompanying

fall in exports following a shock in the aid equation. This means the reason could be due

to the temporary increase in purchasing power of exports, which could lead to a permanent

loss in the export markets, i.e., a Dutch disease effect. Furthermore, we observe that aid

decreases with a positive shock in the equation for GDP per capita, revealing that aid from

foreign countries is lowered as the Bolivian economy improves.

Foreign direct investment instead, was found to be exogenous to the system, it seems

to be determined solely by investor characteristics (e.g., current economic condition in the

investing countries). The effects of investments are all in line with theory, leading to a

positive response of Bolivian GDP per capita and exports and a reduction of purchasing

power of exports, overall showing that foreign direct investments are channelled into the

productive sectors of the economy, which accumulates capital.

Regarding policy implications, we suggest that policy makers in Bolivia should attract

FDI to productive sectors by providing the necessary economic incentives and property

rights. On the other hand, aid needs to change its nature, since the way it is now is

harming the economy. Aid should be focused into more productive and lasting sectors. For

example by promoting Bolivian exports through the opening of donors markets instead of

adding trade barriers.

65

Appendix

A.1 Data Merging

The main sources of information for the flows are from ECLA (The Economic Commis-

sion for Latin America, United Nations) and IMF (International Monetary Fund). ECLA

provides data from 1950 until 2008 while IMF-International Financial Statistics (IFS) has

from 1975 until 2009. We decide to use IFS data because is more detailed and has extended

documentation (see Fund (1993)) and then fill in the data from ECLA from 1950-1975. We

merge both sources to obtain a complete data set from 1950 to 2009. All the following

flows are taken from the Balance of Payments(where parenthesis is the name used in the

statistical software).

A.1.1 Aid (tr c)

Defined as current transfers to the public sector. IFS’s definition(used from 1976 to 2009):

general government net current transfers (credit minus debit), which is in the current ac-

count in current transfers. ECLA’s definition(used from 1950 to 1975): unilateral official

transfers, which is in the capital account. We merge them although they are in different

accounts in the Balance of Payment, because we believe the definition has changed over

time. Values are very similar to the IFS values (see Figure A.1). It is worth mentioning

that this variable excludes explicitly debt relief and concessional loans.

A.1.2 Loans (debt drt)

Defined as net liabilities to the public and private sector. IFS’s definition (used from 1976 to

2009) is: net liabilities (liabilities minus assets) of public sector (which is the sum of general

government and monetary authorities - National Bank) and private sector (understood

banks and other sectors such as insurance companies, private non-profit institutions, and

households, among others) . ECLAC’s definition (used from 1950 to 1975) is: total capital

of the public and private sector (short run and long run), which is also interpreted as

66

Figure A.1: Aid from both sources. In millions of current US dollars

010

020

030

040

0

1940 1960 1980 2000 2020year

ECLAC IFS

disbursements and repayments.

We noticed that the data had a large peak in 2006 and after examining its composition

we realized that it was due to the debt relief program in that year. Fund (1993) describes

how to write down a debt relief transaction. It should be treated as a capital transfer

offsetting the reduction of liabilities in the financial account. We think that is not very

reasonable to use debt relief in our analysis as a capital flow because an actual inflow did

not take place, however in order to balance the balance of payments, it needs to be included

as capital inflow. So we decided to exclude explicitly debt relief from loans. .

Figure A.2: Loans from both sources. In millions of current US dollars

−40

0−

200

020

040

060

0

1940 1960 1980 2000 2020year

ECLACIFS

67

A.1.3 FDI (fdi)

Defined as foreign direct investment. IFS : Net Direct Investment in reporting economy in

the financial account. CEPAL: Foreign Direct Investment in the capital account.

Figure A.3: FDI from both sources. In millions of current US dollars

−50

00

500

1000

1940 1960 1980 2000 2020year

ECLACIFS

A.1.4 Debt Relief (trkgg c)

Defined as debt relief. IFS definition is: Debt forgiveness of the General Government under

capital transfers in the capital account.

Figure A.4: Debt Relief. In millions of current US dollars

050

010

0015

0020

00IF

S: D

ebt r

elie

d

1940 1960 1980 2000 2020year

68

A.1.5 Portfolio Investment (pi)

IFS definition is: portfolio investment (Net) in capital account. CEPAL: portfolio invest-

ment.

Figure A.5: Portfolio Investment. In millions of current US dollars

−10

0−

500

50

1950 1960 1970 1980 1990 2000year

ECLACIFS

A.1.6 Remittances (tr1)

IFS definition is: Workers’ net remittances and capital transfers from migrants which is in

current and capital account. CEPAL: Unilateral private transfers, which is in the current

account only.

Figure A.6: Remittances. In millions of current US dollars

020

040

060

080

010

00

1940 1960 1980 2000 2020year

IFSECLAC

69

Finally, Figure A.7 displays the graphs for the rest of the variables.

Figure A.7: Remaining variables used in the model

700

800

900

1000

1100

GD

Ppc

(US

$200

0)

1940 1960 1980 2000 2020year

010

0020

0030

0040

00E

xpor

ts(M

ill.U

S$2

005)

1940 1960 1980 2000 2020year

1214

1618

Y*

(I$2

005)

1940 1960 1980 2000 2020year

6080

100

120

140

160

ToT

(20

05)

1940 1960 1980 2000 2020year

70

A.2 VAR Results

A.2.1 Unrestricted VAR: fdi, log(aid), total loans, log(Bolivian GDP) and

log(senders GDP), purchasing power of exports, and log(exports) in

constant prices

Notation: Prefix ”d” means first difference. Prefix ”L” means first lag. Thus prefix ”L.d”

means the first difference lagged one period. ce1o and ce2o refer to the first and second

cointegrating relation respectively.

Table A.1: Estimation results : var

Variable Coefficient (Std. Err.)

Equation 1 : dlystar1

L.dlystar1 0.674 (0.122)

L.dlrelp05 -0.009 (0.008)

L.ddebt drt 0.000 (0.000)

L.dlexp05 0.009 (0.008)

L.dltr c 0.001 (0.003)

L.dfdi 0.000 (0.000)

L.dlpib bol00 0.013 (0.032)

L.ce1o 0.009 (0.031)

L.ce2o -0.002 (0.006)

Equation 2 : dlrelp05

L.dlystar1 2.460 (2.173)

L.dlrelp05 -0.075 (0.141)

L.ddebt drt 0.000 (0.000)

L.dlexp05 -0.016 (0.140)

L.dltr c 0.034 (0.045)

L.dfdi 0.000 (0.000)

L.dlpib bol00 0.307 (0.574)

L.ce1o 0.143 (0.555)

L.ce2o -0.164 (0.101)

Equation 3 : ddebt drt

L.dlystar1 -201.596 (3248.978)

L.dlrelp05 216.726 (210.794)

L.ddebt drt 0.113 (0.163)

L.dlexp05 34.249 (208.836)

Continued on next page...

71

... table A.1 continued

Variable Coefficient (Std. Err.)

L.dltr c -101.437 (66.745)

L.dfdi 0.477 (0.225)

L.dlpib bol00 -1103.426 (857.686)

L.ce1o 431.490 (830.300)

L.ce2o -594.317 (150.699)

Equation 4 : dlexp05

L.dlystar1 2.496 (2.585)

L.dlrelp05 -0.127 (0.168)

L.ddebt drt 0.000 (0.000)

L.dlexp05 -0.007 (0.166)

L.dltr c -0.002 (0.053)

L.dfdi 0.000 (0.000)

L.dlpib bol00 0.370 (0.682)

L.ce1o 0.824 (0.661)

L.ce2o -0.118 (0.120)

Equation 5 : dltr c

L.dlystar1 1.814 (7.562)

L.dlrelp05 -0.554 (0.491)

L.ddebt drt 0.000 (0.000)

L.dlexp05 -0.871 (0.486)

L.dltr c 0.177 (0.155)

L.dfdi 0.000 (0.001)

L.dlpib bol00 -3.418 (1.996)

L.ce1o 4.737 (1.932)

L.ce2o -0.495 (0.351)

Equation 6 : dfdi

L.dlystar1 4537.652 (2756.065)

L.dlrelp05 -41.236 (178.814)

L.ddebt drt -0.087 (0.138)

L.dlexp05 -24.359 (177.153)

L.dltr c 4.943 (56.619)

L.dfdi -0.006 (0.191)

L.dlpib bol00 491.496 (727.564)

L.ce1o -528.798 (704.333)

L.ce2o 149.876 (127.836)

Equation 7 : dlpib bol00

L.dlystar1 -0.322 (0.420)

Continued on next page...

72

... table A.1 continued

Variable Coefficient (Std. Err.)

L.dlrelp05 0.060 (0.027)

L.ddebt drt 0.000 (0.000)

L.dlexp05 0.067 (0.027)

L.dltr c 0.017 (0.009)

L.dfdi 0.000 (0.000)

L.dlpib bol00 0.271 (0.111)

L.ce1o 0.521 (0.107)

L.ce2o -0.041 (0.019)

A.2.2 Restricted VAR: fdi, log(aid), total loans, log(Bolivian GDP) and log(senders

GDP), purchasing power of exports, and log(exports) in constant prices

Notation: Prefix ”d” means first difference. Prefix ”L” means first lag. Thus prefix ”L.d”

means the first difference lagged one period. ce1o and ce2o refer to the first and second

cointegrating relation respectively.

Table A.2: Estimation Results for Restricted VAR: var

Variable Coefficient (Std. Err.)

Equation 1 : dlystar1

L.dlystar1 0.658 (0.093)

L.dlrelp05 0.000 (0.000)

L.ddebt drt 0.000 (0.000)

L.dlexp05 0.011 (0.006)

L.dltr c 0.000 (0.000)

L.dfdi 0.000 (0.000)

L.dlpib bol00 0.000 (0.000)

L.ce1o 0.000 (0.000)

L.ce2o 0.000 (0.000)

Equation 2 : dlrelp05

L.dlystar1 0.000 (0.000)

L.dlrelp05 0.000 (0.000)

L.ddebt drt 0.000 (0.000)

Continued on next page...

73

... table A.2 continued

Variable Coefficient (Std. Err.)

L.dlexp05 0.000 (0.000)

L.dltr c 0.000 (0.000)

L.dfdi 0.000 (0.000)

L.dlpib bol00 0.000 (0.000)

L.ce1o 0.000 (0.000)

L.ce2o -0.130 (0.054)

Equation 3 : ddebt drt

L.dlystar1 0.000 (0.000)

L.dlrelp05 0.000 (0.000)

L.ddebt drt 0.000 (0.000)

L.dlexp05 0.000 (0.000)

L.dltr c 0.000 (0.000)

L.dfdi 0.305 (0.128)

L.dlpib bol00 0.000 (0.000)

L.ce1o 0.000 (0.000)

L.ce2o -414.572 (72.139)

Equation 4 : dlexp05

L.dlystar1 0.000 (0.000)

L.dlrelp05 0.000 (0.000)

L.ddebt drt 0.000 (0.000)

L.dlexp05 0.000 (0.000)

L.dltr c 0.000 (0.000)

L.dfdi 0.000 (0.000)

L.dlpib bol00 0.000 (0.000)

L.ce1o 0.716 (0.372)

L.ce2o 0.000 (0.000)

Equation 5 : dltr c

L.dlystar1 0.000 (0.000)

L.dlrelp05 0.000 (0.000)

L.ddebt drt 0.000 (0.000)

L.dlexp05 -0.737 (0.408)

L.dltr c 0.000 (0.000)

L.dfdi 0.000 (0.000)

L.dlpib bol00 -4.894 (1.496)

L.ce1o 5.088 (1.506)

L.ce2o -0.588 (0.240)

Continued on next page...

74

... table A.2 continued

Variable Coefficient (Std. Err.)

Equation 6 : dfdi

L.dlystar1 5554.826 (1769.211)

L.dlrelp05 0.000 (0.000)

L.ddebt drt 0.000 (0.000)

L.dlexp05 0.000 (0.000)

L.dltr c 0.000 (0.000)

L.dfdi 0.000 (0.000)

L.dlpib bol00 0.000 (0.000)

L.ce1o 0.000 (0.000)

L.ce2o 0.000 (0.000)

Equation 7 : dlpib bol00

L.dlystar1 0.000 (0.000)

L.dlrelp05 0.042 (0.021)

L.ddebt drt 0.000 (0.000)

L.dlexp05 0.057 (0.022)

L.dltr c 0.024 (0.006)

L.dfdi 0.000 (0.000)

L.dlpib bol00 0.334 (0.085)

L.ce1o 0.471 (0.082)

L.ce2o -0.023 (0.013)

A.2.3 Restricted VARX. Endogenous vars: Purchasing power of exports, to-

tal loans, and log(exports) in constant prices, log(aid) and log(Bolivian

GDP). Exogenous vars: log(senders GDP), fdi

Notation: Prefix ”d” means first difference. Prefix ”L” means first lag. Thus prefix ”L.d”

means the first difference lagged one period. ce1o and ce2o refer to the first and second

cointegrating relation respectively.

Table A.3: Estimation results : var

Variable Coefficient (Std. Err.)

Continued on next page...

75

... table A.3 continued

Variable Coefficient (Std. Err.)

Equation 1 : dlrelp05

L.dlrelp05 -0.075 (0.141)

L.ddebt drt 0.000 (0.000)

L.dlexp05 -0.016 (0.140)

L.dltr c 0.034 (0.045)

L.dlpib bol00 0.307 (0.574)

L.dlystar1 2.460 (2.173)

L.dfdi 0.000 (0.000)

L.ce1o 0.143 (0.555)

L.ce2o -0.164 (0.101)

Equation 2 : ddebt drt

L.dlrelp05 216.726 (210.794)

L.ddebt drt 0.113 (0.163)

L.dlexp05 34.249 (208.836)

L.dltr c -101.437 (66.745)

L.dlpib bol00 -1103.426 (857.686)

L.dlystar1 -201.596 (3248.978)

L.dfdi 0.477 (0.225)

L.ce1o 431.490 (830.300)

L.ce2o -594.317 (150.699)

Equation 3 : dlexp05

L.dlrelp05 -0.127 (0.168)

L.ddebt drt 0.000 (0.000)

L.dlexp05 -0.007 (0.166)

L.dltr c -0.002 (0.053)

L.dlpib bol00 0.370 (0.682)

L.dlystar1 2.496 (2.585)

L.dfdi 0.000 (0.000)

L.ce1o 0.824 (0.661)

L.ce2o -0.118 (0.120)

Equation 4 : dltr c

L.dlrelp05 -0.554 (0.491)

L.ddebt drt 0.000 (0.000)

L.dlexp05 -0.871 (0.486)

L.dltr c 0.177 (0.155)

L.dlpib bol00 -3.418 (1.996)

L.dlystar1 1.814 (7.562)

Continued on next page...

76

... table A.3 continued

Variable Coefficient (Std. Err.)

L.dfdi 0.000 (0.001)

L.ce1o 4.737 (1.932)

L.ce2o -0.495 (0.351)

Equation 5 : dlpib bol00

L.dlrelp05 0.060 (0.027)

L.ddebt drt 0.000 (0.000)

L.dlexp05 0.067 (0.027)

L.dltr c 0.017 (0.009)

L.dlpib bol00 0.271 (0.111)

L.dlystar1 -0.322 (0.420)

L.dfdi 0.000 (0.000)

L.ce1o 0.521 (0.107)

L.ce2o -0.041 (0.019)

77

References

Andersen, L. E. and J. L. Evia (2003, September). The effectiveness of foreign aid in

Bolivia. Development Research Working Paper Series 01/2003, Institute for Advanced

Development Studies.

BCB. Banco Central de Bolivia. URL:www.bcb.gob.bo.

Bruckner, M. (2011). On the simultaneity problem in the aid and growth debate. Journal

of Applied Econometrics.

Burnside, C. and D. Fanizza (2004, November). Hiccups for HIPCs? Working Paper 10903,

National Bureau of Economic Research.

Cordella, T., L. A. Ricci, and M. Ruiz-Arranz (2005). Debt Overhang or Debt Irrelevance?

Revisiting the Debt-Growth Link. SSRN eLibrary .

Danida (1997). Strategy for Danish -Bolivian development cooperation. Technical report,

Ministry of Foreign Affairs-Denmark.

Depetris Chauvin, N. and A. Kraay (2005). What has 100 Billion Dollars Worth of Debt

Relief Done for Low-Income Countries? SSRN eLibrary .

Doucouliagos, H. and M. Paldam (2008). Aid effectiveness on growth: A meta study.

European Journal of Political Economy 24 (1), 1 – 24.

ECLAC (2009). America Latina y el Caribe: Series historicas de estadısticas economicas,

1950-2008. Technical report, Economic Commission for Latin America and the Caribbean.

Fund, I. M. (1993). Balance of payments manual. International Monetary Fund.

Hamilton, J. D. (1994). Time series analysis, Volume 2. Cambridge Univ Press.

Hepp, R. (2008). Can debt relief buy growth? Technical report, Fordham University,

Department of Economics.

78

Heston, A., R. Summers, and B. Aten (2011, June). Penn World Table Version 7.0. Center

for International Comparisons of Production, Income and Prices at the University of

Pennsylvania.

Humerez, J. and D. Hernaiz (2006). Efectos de la deuda externa y otras polıticas macroe-

conomicas sobre el producto. Analisis economico, Vol.21 .

IFS (2011). International financial statistics. URL:www.elibrary-data.imf.org.

INE (2009). Instituto nacional de estadıstica. URL:www.ine.gob.bo.

Johansson, P. (2008, August). Debt relief, investment and growth. Working Papers 2008:11,

Lund University, Department of Economics.

Juselius, K. (2006). The cointegrated VAR model: methodology and applications. Advanced

texts in econometrics. Oxford University Press.

Juselius, K., N. F. Møller, and F. Tarp (2013). The long-run impact of foreign aid in 36

African countries: Insights from multivariate time series analysis. Oxford Bulletin of

Economics and Statistics.

Machicado, G. (2013). Productivity in Bolivia: A development accounting approach be-

tween 1950 and 2010. Mimeo, INESAD.

Morales, J. A. and J. D. Sachs (1990a, August). Aspects of foreign debt accumulation, 1952-

85. In Developing Country Debt and Economic Performance, Volume 2: The Country

Studies – Argentina, Bolivia, Brazil, Mexico, NBER Chapters, pp. 214–225. National

Bureau of Economic Research, Inc.

Morales, J. A. and J. D. Sachs (1990b). Bolivian debt management, 1985-88. In Developing

Country Debt and Economic Performance, Volume 2: The Country Studies–Argentina,

Bolivia, Brazil, Mexico, pp. 252–259. University of Chicago Press, 1990.

Morales, R. and A. Cazadilla (2003). Is foreign aid good for growth and poverty alleviation?

CIESS-Econometrica, Bolivia.

79

OECD (2013). Development assistance committee. Organisation for Economic Co-operation

and Development. URL:www.oecd.org.

Paldam, M. (2008, September). Development and foreign debt: The stylized facts 1970-2006.

Economics Working Papers 2008-10, School of Economics and Management, University

of Aarhus.

Pesaran, H. H. and Y. Shin (1998, January). Generalized impulse response analysis in linear

multivariate models. Economics Letters 58 (1), 17–29.

Prego, F. D. L. C. (2011). Ayuda externa en Bolivia (1985-2003): Auge y caıda del ne-

oliberalismo. Technical report, Catedra de Cooperacion Internacional y con Iberoamerica

(COIBA), Universidad de Cantabria.

Reinert, K. and I. Goldin (2006). Globalization for development: Trade, finance, aid,

migration, and policy.

Temple, J. (2010). Aid and conditionality. Handbook of development economics 5, 4415–

4523.

VIPFE (2005). Informe 2011 sobre cooperacion internacional y financiamiento para el

desarrollo en Bolivia. Technical report, Ministerio de Planificacion del Desarrollo - Bolivia.

VIPFE (2012). La cooperacion internacional en Bolivia. Technical report, Ministerio de

Hacienda - Bolivia.

80

Income Convergence and the Flow out of

Poverty in India, 1994-2005

Paola A. Barrientos Q.∗, Niels-Hugo Blunch† andNabanita Datta Gupta‡

Abstract

This paper explores the dynamics of income and poverty

of rural Indian households over the period 1994 to 2005. The

estimation strategy consists of a convergence analysis to test

whether poor households are catching-up with rich ones in terms

of income, followed by a transition analysis to test whether

poor households are more likely to exit poverty than to remain

poor. The identification strategy explicitly addresses issues

pertaining to the potential endogeneity and measurement error

of initial income and poverty. We find evidence of both income

convergence and poverty reduction: poorer households are

catching up to richer households over time and the bottom of

the distribution is being lifted up. The key variables driving

these results are education and asset ownership.

∗Aarhus University†Washington and Lee University and IZA, Bonn‡Aarhus University and IZA, Bonn

81

1 Introduction

The concept of convergence has been typically used to evaluate inequalities among incomes

of different countries or regions with aggregated data. Recent research, however, has used

household data as well, which allows for a broader perspective of convergence. Ravallion

(2012) uses household data to study not only income but also poverty convergence among 100

developing countries. He claims that income convergence should lead to poverty convergence:

poorer families grow faster and catch-up in terms of their incomes, which makes them

less poor and thus they also catch-up with respect to their poverty levels with higher

proportionate rates of poverty reduction. Nevertheless he finds that there is income but not

poverty convergence under the argument that poverty itself restricts poverty convergence,

also known as ”poverty traps”.

In this context, India is a country characterized by its high growth rates but also by its

widespread inequalities, hinting at the possibility of the presence of poverty traps. India’s

high growth rates are attributed to its liberalization reforms undertaken since the 1990s,

which are associated with an expansion of capital and skill-intensive exports (e.g. software

and business services, pharmaceuticals, vehicles, auto parts, and steel, among others).

This pattern of development contrasts sharply with the Chinese case which has been more

manufacturing-centred (Bardhan, 2007). The kind of development path which India has

followed seems to have been mainly benefited the urban sectors up to now, and in fact,

earlier research has shown that the liberalization process exacerbated regional disparities

between urban and rural areas (Deaton and Dreze, 2002).

Despite the strong economic growth that India has experienced, the results for welfare

have not been as successful either (Deaton and Dreze, 2002). Recent studies show that the

poverty decline in India has been slower than in other LDCs (Lenagala and Ram, 2010)1.

The main explanation seems to be the nature of the development path, which has benefited

1Results on poverty from different studies may differ according to the measure of poverty employed.Indeed there is an extensive debate in India about this issue, with considerable disagreement on how povertyshould be measured. Deaton and Kozel (2005) argue that there is no consensus on what happened to povertyin India in the 1990s; there is good evidence both that poverty fell and that the official estimates of povertyreduction are too optimistic, particularly in rural India.

82

mostly high-skilled people that live in urban areas rather than the majority of people who

live in the rural areas and work in the agricultural sector (Cain et al., 2010). In China

and other Asian countries, on the other hand, the manufacturing-intensive growth pattern

adopted there benefited a much larger share of the population (Kotwal et al., 2011).

Over time, however, the process of liberalization may have generated some positive

externalities for rural areas. Kotwal et al. (2011) mentions that a major contribution of

the liberalization process to agricultural growth is likely to be a diversification effect as

rising incomes lead to greater urban demand for higher-valued edible oils, milk, fruits and

vegetables than staple food cereals in the longer run. Another positive spillover acts through

remittances to rural households from family members migrating to urban areas (Parida and

Madheswaran, 2010).

Traditionally, inequality has been linked to the rigid stratification of Indian society

according to caste and socioreligious background 2. In fact, both Singh (2010) and Desai and

Dubey (2011) find that the caste system generates inequalities. Singh (2010) decomposes

net farm income inequality and finds that caste explains a substantial part of it. Similarly,

Desai and Dubey (2011) find persistent caste disparities in education, income, and social

networks.

Other research shows, however, that intergenerational social mobility can occur despite

the caste system. Hnatkovska et al. (2013) use the household level National Sample Survey

(NSSO) data from 1983-2005 to examine the economic progress of sons compared to their

fathers in terms of income, education, and occupation. They find that when comparing

the most disadvantaged groups, namely the Dalits (SCs) and Adivasis (STs) to the most

advantaged, there is significant intergenerational mobility in all of these dimensions, including

significant occupational upgrading by sons as compared to their fathers. Examining a longer

time-period Azam and Bhatt (2012) find that there has been significant improvement in

educational mobility across generations in India (since the 1940s)at the aggregate level,

across social groups, and across states.

2The caste system is an ancient classification of rigid social stratification of the Indian society. Caste isendogamous, rarely changed, and implies occupational specialization.

83

As a result, several questions arise, What has happened in welfare terms in rural areas

in the decade following liberalization? Are rural households worse off in terms of income

inequality and poverty reduction or did they actually benefit from the liberalization process?

Are there signs of poverty traps? What is the role of socioreligious grouping, if any?

Two earlier papers have attempted to answer some of these questions. A recent article

by Jha et al. (2012) models the vulnerability to poverty by using panel data for rural India

for the period 1999-2006 (3,628 observations in both years). They find evidence for the

existence of a poverty trap in the sense that ex ante vulnerability of being poor in 1999

translates into ex post poverty in 2006. Instead, we use a much larger data set and compare

income to poverty outcomes. On the other hand, Krishna and Shariff (2011) examine the

same data studied here and compare those households that exit poverty to the ones that

enter poverty, treating them as two different independent samples. We do not split the

sample, rather we use the whole sample for our calculations.

We intend to answer all the questions raised above, by using a unique panel dataset

of Indian rural households (more than 13,000 households) observed during 1993-1994 and

2004-2005. This allows us to trace economic welfare over time in the rural sector, paying

particular heed to how vulnerable low caste and religious minority households have fared

over time compared to the high castes. Thus, we are able to provide rich micro-data based

evidence on the longer run effects of liberalization at the level of the household taking the

dynamics into account in a conditional analysis.

We assess inequalities and poverty traps, by studying poverty transitions and income

convergence, which take into account the potential effect of initial conditions on catching-up

in income/exiting poverty. Simultaneously we establish the main income growth determinants

at the household level as well as the determinants of poverty outcomes. We focus on the role

of caste (separating Dalits, Adivasis, OBCs and High-castes) but also on the role of religion.

According to the 2006 Sachar Committee Report, Muslims are also a disadvantaged group

in India, and there is no reservation policy3 for this group currently. Moreover, we explore

3Affirmative action policy has been adopted in a larger scale in India than anywhere else in the world,with a nation-wide program of reservation (quotas) of new jobs, political seats and slots in higher educational

84

whether we capture different welfare concepts when using a continuous variable for income

as compared to the binary variable for poverty.

The paper is organized as follows. We first describe the dataset and the most important

variables in our models. Then we present our empirical strategy. After that we discuss our

findings and finally, we present the conclusions, discuss policy implications, and provide

suggestions for future research.

2 Data

2.1 Data description

Our data set is based on 13,081 households that were interviewed in two different waves.

During the year 1993 to 1994, the survey known as ”Human Development Profile of India”

(HDPI), carried out by the National Council of Applied Economic Research (NCAER),

consisted of 33,230 households in rural areas only. Later, during 2004 and 2005, the survey

denoted ”India Human Development Survey (IHDS)” collected information for a total of

41,554 households situated both in rural and urban areas; this survey was conducted by

NCAER and the University of Maryland (Desai et al., 2008). From the first survey, a

random sample was selected to be re-interviewed in the second survey. However, the original

1993-1994 survey was not randomly selected. With that concern, a new random sample was

chosen from rural areas to be compared to the re-interviewed sample in order to determine

whether the re-interviewed sample was overrepresented among certain segments of the society.

The comparisons suggest that on most variables of interest such as caste, religion, education,

and economic status, the re-interviewed sample does not differ substantially from the fresh

sample (see Table 1).

The re-interviewed sample consists of a total of 13,081 households, from which 82%, or

institutions for the historically discriminated lowest caste groups in Indian society, the scheduled castes(SCs) and the scheduled tribes (STs). Although such quotas had existed earlier, in 1982 they were set at15% of all public sector openings for SCs and 7.5% for STs, though varying according to fluctuations inthe SC/ST population share measured during the (previous) decennial census (with administrative lags inimplementation). In 1990 a further 27% of all public sector new hires were reserved for other backward castes(the OBCs).

85

10,791 households, were contactable for re-interview and the rest, 2,290 households, had

separated from the original household but were still living in the same village, so they were

contacted for an interview, as well.

Table 1: Comparison of new and re-interviewed sample

New Re-interviewedSample Sample

Socioreligious group

Forward Caste Hindu 16 18OBC 38 35Dalit 23 26Adivasi 12 10Muslim 9 9Christian, Sikh, Jain 2 2Place of ResidenceMetro 0 0Other urban 1 1More developed village 50 45Less developed village 49 54Maximum adult educationIliterate 30 291 to 4 Std 10 105 to 9 Std 34 3310 to 11 Std 11 1212 Some college 8 8College graduate 7 8Household incomeNegative to Rs 999 3 31st Quintile (Rs 1,000 to 14,000) 27 232nd Quintile (Rs 14,001 to 22,950) 24 233rd Quintile (Rs 22,951 to 36,097) 19 214th Quintile (Rs 36,098 to 69,000)) 17 185th Quintile (Rs 69,001+) 10 12

Note: Comparison of new and re-interviewed rural sample in districts where any re-interviews took place.Table extracted from Desai et al. (2010)

A main characteristic of the IHDS survey is that the household income variable is built

from 50 different separate components. Although the disaggregation in detail is available

only for 2005, the total income of households is comparable across waves, Shariff (2009).

The 50 components can be aggregated into fewer components, as follows. Farm income:

IHDS asked about crop production and prices, usage of crop residues, animal ownership

and home-produced animal and crop products, expenditures on a variety of farm inputs,

and net payments on agricultural rents; wage and salary income: consist of all wages

in the household, meals, housing benefits, and bonuses; remittances; non-farm business

income, government benefits: include scholarships, pensions, maternity scheme, NGO or

86

other assistance, and other government transfers; and others: include income from rental

property, interest, pensions and other income.

In 2005, some households from our sample reported negative incomes, more specifically,

farming income. These households reported crop failures and high farm expenses. The

survey designers confirm that the negative numbers are correct but they advise working with

positive incomes higher than 1000 Rs (Desai and Vanneman, 2008). Therefore, our sample

is reduced to 12,352 households. In terms of the household composition in our sample, 59%

are adults, 7% older people and 34% are children. As much as 95% of household heads are

male and 60% of them are below 30 years of age, 52% are illiterate and 48% are engaged in

agriculture (additional descriptive statistics can be found in Table A.1).

2.2 Variables

We are interested in measuring the financial situation of households in rural areas. We have

at our disposal the variable income, which is directly related to financial position, at least

in theory. However, in practice, income may be incorrectly measured due to a variety of

reasons (more on this in Section 3.1). Therefore we use food consumption expenditures as a

proxy for income, as well.

Figure (1) shows the household per capita income distribution for both waves4. The

distributions look roughly similar for both years. The 2005 income distribution is somewhat

smoother at the top and also somewhat wider. This is confirmed by the higher standard

deviation in 2005 as compared to 1994 (see Table A.1 in the Appendix). This is a sign

of increasing inequality, since it implies that the distribution became more unequal over

time. However, we have to remember that these are the unconditional distributions without

controlling for other factors.

4A seemingly natural objection here might be that we should have used adult equivalent income insteadof per capita income in order to reflect the lower needs of children. To calculate the adult equivalent incomevariable, however, we need detailed data for all individuals in all households, and unfortunately the individualdata is incomplete. Instead we use income in per capita terms since information on the total household sizeis available. Nevertheless, we include in the analysis information of the household composition by the threeage ranges which are available in the data (share of adults, old people and children) to at least go some waytoward incorporating household composition in the analysis.

87

As for income, also the standard deviation of consumption expenditure has increased

from 1994 to 2005. Figure (1) shows the distribution of food consumption as compared to

income. As expected, the distribution of food consumption is more narrow, since people

tend to fulfil their food needs first5.

Figure 1: Kernel density plot for food consumption expenditure and income per capita

0.2

.4.6

.81

−5 0 5x

kdensity y94_m0 kdensity y05_m0kdensity c94_m0 kdensity c05_m0

Note: All variables are demeaned. We use the Consumer Price Index for Agricultural and Rural Labourers(CPI-AL) to transform the 1994 data to 2005 prices. y is income and c is food consumption per capita.

Turning next to the explanatory variables, our household level variables consist mainly of

characteristics of the head of the household 6. We include education, age, occupation, gender,

and reported caste of the head and asset ownership. Regional (geographic) characteristics

are captured via dummies for each of the states. Finally we include a policy variable, namely

5We were not able to recover expenditure on other consumption goods because of a lack of detailedinformation in the 1994 survey. However, our constructed variable of food consumption expenditures is stillquite complete. It includes the value of purchased and home produced cereals and pulses, sugar, milk, edibleoil, meat and fish, eggs, vegetables, fruits and others. We could have used consumption of cereals only butOldiges (2012) mentions that cereal consumption seems unrelated to per capita income but that other foodexpenditures like fruits or meat do increase with income (normal goods), indicating that richer people do eatbetter.

6We are aware that the information of the head of the household may not be the only relevant informationto include in the analysis. Perhaps information of the person that is most educated is also (possibly evenmore) relevant (Jolliffe, 2002), or that of the main earner of the household (Glewwe, 1991). Unfortunately,we could not recover individual-level data since a large share of individuals is missing. However, informationon the head is complete and likely to be still quite important.

88

the political reservation policy, which benefits most disadvantaged castes (STs and SCs). It

represents the share, by state and year, of parliament seats reserved for these castes (more

details on this variable can be found in Chin and Prakash (2011)).

To calculate poverty estimates, we use the national poverty lines provided by the Planning

Commission of India which in contrast to the World Bank’s 1.25 US dollar a day for any

country give different lines for each state and for rural and urban areas. As Krishna and

Shariff (2011), we do not deflate income from 1994. Instead, we compare current income

with the official poverty line year by year. As mentioned in Krishna and Shariff (2011), the

results are not directly comparable with other estimates that are based on consumption7.

According to our population estimates of poverty incorporating IHDS design weights,

poverty has increased from 37% in 1994 to 41% in 2005 (see Table 2). This does not mean

that all households were worse off in 2005 compared to 1994. For example, about 22% of

the households managed to leave poverty, and 41% stayed non poor. The results are driven

by those that entered poverty, 18%, and by those 19% that remained poor throughout the

entire period.

Table 2: Poverty estimates

Proportion Robust Std. Err. [95% Conf. Interval]

1994 Non-poor 0.63 0.011 0.61 0.65Poor 0.37 0.011 0.35 0.39

2005 Non-poor 0.59 0.010 0.57 0.61Poor 0.41 0.010 0.39 0.43

Note: Std. Err.adjusted for 788 clusters. Design weights are used to get population estimates.

We can also examine poverty flows in terms of socioreligious group. The Brahmin and

Forward castes (jointly denoted as High caste) had greater positive outcomes than negative

ones: 56% remained non poor and 19% left poverty. The minorities had a similar pattern:

63% stayed non poor and 14% left poverty. In contrast, 32% of the Adivasis entered poverty,

and 22% stayed poor. The numbers are similar for the Dalits and Muslims (Table 3).

In regional terms we also have contrasting outcomes. The Northern, Western and

7For a detailed discussion of poverty measures and poverty lines in India, see Deaton and Kozel (2005).

89

Tab

le3:

Pov

erty

flow

s(P

erce

nt)

Tota

l2005

Non

-poor

Poor

1994

Non

-poor

41

18

Poor

22

19

Ob

s.in

mill.

70

2005

Soci

ore

ligio

us

Gro

up

ing

Hig

hca

ste

OB

CD

alit

Ad

ivasi

Mu

slim

Min

ori

ties

Non

-poor

Poor

Non

-poor

Poor

Non

-poor

Poor

Non

-poor

Poor

Non

-poor

Poor

Non

-poor

Poor

1994

Non

-poor

56

14

44

17

32

22

32

20

32

19

63

20

Poor

19

10

24

15

22

25

22

25

25

24

14

4O

bs.

inm

ill.

12.7

24.4

18.6

6.8

6.3

1.2

2005

Reg

ion

sN

ort

her

nU

pp

erC

entr

al

Low

erC

entr

al

Wes

tern

East

ern

Sou

ther

nN

on

-poor

Poor

Non

-poor

Poor

Non

-poor

Poor

Non

-poor

Poor

Non

-poor

Poor

Non

-poor

Poor

1994

Non

-poor

56

19

32

21

36

18

46

14

33

24

55

15

Poor

13

12

26

21

24

22

23

16

18

25

19

11

Ob

s.in

mill.

4.5

16.1

16

11.8

8.7

12.9

Note

:F

low

sca

lcula

ted

as

share

sof

tota

lsa

mple

sof

each

gro

up.

Des

ign

wei

ghts

are

use

dto

obta

innati

onally

repre

senta

tive

popula

tion

esti

mate

s.

90

Southern regions perform better than the rest of the regions. For example, 56% of the

households in the Northern region remain non poor in both years, while in the Upper Central

Region, only 32% are non-poor in both years.

3 Empirical strategy

3.1 Income convergence

We study the dynamics of income by examining income growth and its determinants by

considering the following equation:

gyi = x′iαy + βy · y0i + εyi (1)

εy ∼ N(0, σy)

where gyi is the annual growth rate of income of household i during the period 0 to τ ; y0i

is the income at the beginning of the period measured in logarithms; the vector xi includes

a constant and all the controls (education, occupation, caste, gender and age of the head

plus age composition of the household, assets, a policy variable, state dummies and the split

dummy to control for the households that split from the first survey). Finally εy is the error

term and is assumed to be normally distributed.

The coefficient βy measures the relationship between initial income and growth, after

controlling for household and geographic characteristics. When βy is negative, the poorer the

household the more it will grow in comparison to the richer households, holding everything

else constant.

Some of the explanatory variables in Equation (1) are predetermined, such as caste,

gender and age of the head of the household. However, other variables such as occupation and

education of the head may be simultaneously determined or create some reverse causation

with respect to income growth. Therefore, in order to avoid simultaneous relations, we

91

include all our explanatory variables observed at the beginning of the period8.

On the other hand, we know that our income variable has potential measurement error

problems. Measurement error in self reported surveys is well known and documented. It

may be that the persons interviewed are not able to report the precise value of income. One

could argue that the misreport is caused by age, education, or some other characteristic

of the individual being interviewed, which will necessarily introduce a bias (specifically,

attenuation bias, see Wooldridge (2002)). Additionally, measurement error is also a potential

concern for rural households, for which it is potentially difficult to derive an accurate report

of the value of produce from subsistence farming. Many studies argue for using household

expenditures instead of household income as the measure of welfare. The reasons for this

are several. First, expenditures are measured more accurately than income (Glewwe, 1991)

as it is easier to recall expenditures on consumption than income. Second, following the

permanent income hypothesis, it may reflect longer term economic status better (Appleton,

2001). Therefore, we use a subset of household expenditures as a proxy for income and

estimate our growth equation in terms of consumption expenditure as well,

gci = x′iαc + βc · c0i + εci (2)

εc ∼ N(0, σc)

which is identical to (1), except that instead of household income, y, we are now modeling

household consumption, c.

3.2 Poverty flows

To confirm our results on income convergence, we estimate a model of poverty transition9.

Since our poverty measure is a binary variable we choose not to work with a growth model

8Alternatively, including the variables in differences would get rid of individual unobserved effects butalso time-invariant variables such as caste, which is one of the main variables we are interested in.

9For a discussion of the use of binary poverty measures versus a continuous variable see Appleton (2001).

92

because its interpretation is not as straightforward. Instead we choose to work with probit

models, that yield more intuitive results, in terms of poverty probabilities10. We start with

a simple probit model that accounts for initial poverty status as an explanatory variable for

final poverty status,

pτi = 1(x′iαp + βp · p0i + εpi > 0) (3)

εp ∼ N(0, 1)

where pτi is our poverty dummy variable, being 1 when poor (below the poverty line)

and 0 when non-poor in the last year of the period, and p0i is the poverty dummy variable

in the initial year. As before, the vector xi includes a constant and all the controls.

We also estimate a recursive biprobit model11, where initial poverty status becomes

endogenous and is determined by the same explanatory variables plus land,

pτi = 1(x′iατp + βτp · p0i + ετpi > 0)

p0i = 1(x′iα0p + δ0p · landi + ε0pi > 0) (4)

E[ετpi ] = E[ε0pi ] = 0

V ar[ετpi ] = V ar[ε0pi ] = 1

Cov[ετpi , ε0pi ] = ρ

where ρ is the correlation coefficient of the unobservables of each of the equations. The

objective is to account for the correlation between the error terms of the two individual

equations and, thus, to infer the importance and direction of the unobservables as drivers

of the two individual equations/poverty statuses in the two periods. In comparison to

the normal probit model, the bivariate probit analysis can yeld useful estimates such as

10Alternatively, Ravallion (2012) analyses poverty reduction as a function of initial poverty levels.11A detailed description of the model can be found in Greene (2008), chapter 23.

93

conditional probabilities (e.g., the probability of being non poor in 2005 given that one

was poor in 1994). For identification purposes, we restrict the variable land to enter only

through the initial poverty variable due to its pre-determinedness. This approach in fact

resembles an IV procedure where land is the instrument for initial poverty. We discuss the

assumption of having land ownership as instrument in Section 4.1.1.

We compare poverty outcomes given equal initial conditions. There is a flow out of

poverty when the probability of ending non-poor is higher than ending poor in 2005 given

that one was poor in 199412:

Pr(pτ =0| p0 =1) > Pr(pτ =1| p0 =1)

4 Results

4.1 Income convergence

Table 4 reports the estimated results from Equations (1) and (2) in columns 1 and 2

respectively. We first analyse the determinants of income/consumption growth and later we

analyse the convergence parameters.

For both income and consumption growth, education is very important. The education

coefficients are higher for income than consumption. All socioreligious groups, compared to

high caste, appear to have both statistically and substantively significant negative effects

on income and consumption growth, except for minorities. Again, the importance of

socioreligious grouping is more negative for income than consumption, and the worst off

are Muslims. In a similar fashion assets seem to have a higher impact on income than

consumption. None of the occupations, in comparison to an occupation in agriculture, are

important for food consumption while all of them are important for income, especially when

the head of the household owns his or her own business. Another difference is the reservation

policy which is statistically significant only for food consumption growth and not for income

12This concept is similar to the convergence concept, in the sense that we are interested in the situationof the initially poor and their probability of catching-up. However we do not call it poverty convergencebecause we are not calculating a speed of poverty reduction as in Ravallion (2012)

94

growth. In contrast, the household’s annual expenditures on education seem to be relevant

only for income growth.

Regarding the beta convergence coefficients, they are very similar, βy = 7.6% and βc

= 7.9%. The fact that both speeds of convergence are so similar can hint at two different

things. First, that measurement error in income is not as bad as expected so that food

consumption is a good proxy for income. Or, second, that there is measurement error both

in initial income and in consumption compared to the final outcomes, which when captured

by the β parameter makes the speeds of convergence similar. Furthermore, measurement

error in the initial variable can lead to high speeds of convergence. For example, when initial

income is below its true value, the growth rate will appear higher than it should be, showing

convergence. Therefore, we now examine the second possibility.

4.1.1 Sensitivity analysis

Initial income may be endogenously determined because the measurement error that it may

contain may be related to the error term in the growth model since initial income enters

also in the RHS. We re-write Equation 1 in levels13:

yτyi = x′iατy + βτy · y0i + ετyi (5)

ετy ∼ N(0, σyτ )

Equation (5) can be estimated instead of Equation (1) without loss of generality, the

only difference is that the parameters are scaled in a different way. We recover the original

growth parameters by dividing by τ , except the y0 parameter, which needs to be subtracted

by one and then divided by τ .

Then we instrument our potentially endogenous variable, initial income, with land

ownership. Land ownership is often inherited and therefore exogenously determined (Glewwe,

13We replace gyi = [yτi − y0i ]/τ into Eq.(1).

95

Table 4: Convergence results

Income growth Consumption growthgy gc

y0 -0.076***[0.002]

c0 -0.079***[0.002]

(omitted: edu illi)edu primary 0.005** 0.004***

[0.002] [0.001]edu middle 0.014*** 0.009***

[0.003] [0.002]edu secondary 0.034*** 0.019***

[0.006] [0.003](omitted: share mid)share old -0.011 0.007

[0.009] [0.008]share children -0.005 0

[0.005] [0.003](omitted: sociorel high)sociorel OBC -0.011*** -0.005***

[0.003] [0.002]sociorel SC -0.018*** -0.009***

[0.006] [0.003]sociorel ST -0.016** -0.013***

[0.006] [0.003]sociorel muslim -0.020*** -0.008***

[0.005] [0.003]sociorel others -0.006 0.001

[0.006] [0.003](omitted: male)female 0.005 0.005**

[0.004] [0.003](omitted: age young)age mid -0.013*** -0.004***

[0.002] [0.001]age old 0.002 0

[0.003] [0.002](omitted: ocu agri)ocu proff 0.009*** 0

[0.002] [0.001]ocu own 0.012*** 0.002

[0.003] [0.002]ocu none 0.007* -0.001

[0.004] [0.002]exp edu 0.060*** 0.008

[0.016] [0.007]assets 0.002*** 0.001***

[0.000] [0.000]policy 0 -0.000**

[0.000] [0.000]split -0.004* 0.001

[0.002] [0.001]Constant 0.688*** 0.667***

[0.015] [0.014]

N 12352 12327R2 0.456 0.44

Note: Standard errors are reported in brackets and allow for intra-village correlation. edu: level of educationof the head, share: share of people in the house according to age, sociorel: socioreligious group, male: genderof the head, age: age of the head, ocu: occupation of the head, exp: expenditure on education. Coefficients forstate dummies are omitted. ***: statistically significant at 1%;**: statistically significant at 5%; *:statisticallysignificant at 10%; ++: statistically significant at 15%; +: statistically significant at 20%.

96

1991). The land market in India is very tight or even almost non-existent due to inheritance

and ownership rules. Morris (2002) points out that the land markets are highly distorted and

inefficient; land records are inaccurate, outdated, and incomprehensible and transaction costs

are high by international standards, all of which have discouraged formal land transactions.

We suggest that it is reasonable to assume that land ownership is exogenous to income

growth, that it is predetermined and cannot, therefore, be simultaneously determined with

income growth. However, it could be related to growth through unobservable variables

such as changes in international food prices. An increase in food prices, while keeping land

constant, increases household income. In our case, a change in international food prices

between our two points in time would be the same for all households. Therefore, we suggest

that this is captured by the constant. Another possible omitted variable could be land

efficiency which we suggest is captured by our asset variable which includes productive

assets. However, of course we realize that ultimately land matters for income. Therefore we

also assume that land is related to initial income rather than to growth, which implies that

land explains growth only via initial income14.

The results are shown in Table 5 and column 3 shows the re-calculated growth parameters.

Regarding the determinants of income growth, we can see that after using our instrument,

the impact of the most important determinants has decreased a bit. However the same

variables are still statistically significant. Surprisingly the βy remains close to the previous

results: βy = 6.9% 15. Therefore, we are confident that our previous results are robust,

especially given that no other instrument is available. It is worth noting that the exogeneity

tests indicate that initial income is indeed exogenous (see Table A.3).

4.2 Poverty flows

Next, we examine the results of our probit and biprobit models (Equation (3) and (4)

respectively). In the first case, initial poverty is restricted to be exogenous and in the second

14It is not an easy task to find other instruments. Assets, for example, are most likely endogenous: thericher the household the more assets it can buy (Glewwe, 1991).

15We estimated the same IV model for food consumption and find βc = 6%, results in Table A.2

97

Table 5: Sensitivity analysis

Income Income - IV Income growth - IVyτ yτiv gyiv

y0 0.161*** 0.241*** -0.069***[0.017] [0.055] [0.055]

(omitted: edu illi)edu primary 0.050** 0.044* 0.004*

[0.025] [0.026] [0.002]edu middle 0.156*** 0.135*** 0.012***

[0.035] [0.037] [0.003]edu secondary 0.378*** 0.332*** 0.030***

[0.069] [0.072] [0.007](omitted: share mid)share old -0.122 -0.125+ -0.011+

[0.096] [0.093] [0.008]share children -0.056 0.024 0.002

[0.058] [0.082] [0.007](omitted: sociorel high)sociorel OBC -0.116*** -0.108*** -0.010***

[0.036] [0.037] [0.003]sociorel SC -0.194*** -0.176*** -0.016***

[0.063] [0.065] [0.006]sociorel ST -0.173** -0.149** -0.014**

[0.068] [0.072] [0.007]sociorel muslim -0.219*** -0.213*** -0.019***

[0.051] [0.051] [0.005]sociorel others -0.067 -0.056 -0.005

[0.063] [0.064] [0.006](omitted: male)female 0.057 0.045 0.004

[0.048] [0.049] [0.004](omitted: age young)age mid -0.143*** -0.137*** -0.012***

[0.024] [0.024] [0.002]age old 0.019 0.024 0.002

[0.031] [0.031] [0.003](omitted: ocu agri)ocu proff 0.099*** 0.115*** 0.010***

[0.026] [0.027] [0.002]ocu own 0.132*** 0.138*** 0.013***

[0.033] [0.033] [0.003]ocu none 0.074* 0.097** 0.009*

[0.041] [0.044] [0.004]exp edu 0.656*** 0.877*** 0.080***

[0.179] [0.231] [0.021]assets 0.018*** 0.015*** 0.0014***

[0.002] [0.003] [0.000]policy -0.002 -0.002 0

[0.004] [0.003] [0.000]split -0.043* -0.028 -0.003

[0.024] [0.027] [0.002]Constant 7.570*** 6.827*** 0.621***

[0.166] [0.516] [0.047]N 12352 12352 12352R2 0.19 0.186

Note: Standard errors are reported in brackets and allow for intra-village correlation. edu: level of educationof the head, share: share of people in the house according to age, sociorel: socioreligious group, male: genderof the head, age: age of the head, ocu: occupation of the head, exp: expenditure on education. Coefficients forstate dummies are omitted. ***: statistically significant at 1%;**: statistically significant at 5%; *:statisticallysignificant at 10%; ++: statistically significant at 15%; +: statistically significant at 20%.

98

case, initial poverty is allowed to be endogenous. In both cases, the original estimations

are presented in the Appendix (Table A.4). It is not intuitive to interpret the coefficients

from Table A.4 directly. Therefore, we calculate the average marginal effects. For the probit

case the average marginal effect of each explanatory variable is evaluated at the outcome

of Pr(poor2005=1) while for the biprobit case, the average marginal effect is evaluated at

the conditional probability of Pr(poor2005 =1| poor1994 =1), which we call ”conditional

probability of poverty”. We choose to work with conditional probabilities instead of joint

probabilities because they resemble regression coefficients by showing the effect of different

explanatory variables of the conditional mean, i.e. slopes of conditional expectations (Greene,

2008). The results are shown in Table 6.

The estimated parameters are very similar for both probit and biprobit models. Actually

the Wald test of ρ, that indicates whether the correlation of unobservables is zero, suggests

that we should accept the null (last line in Table A.4). However, we still continue with the

biprobit analysis in order to obtain the conditional probabilities. In the end, the results are

not wrong, and imply just that we are using a more general model than needed as it takes

into account the potential correlation of unobservables and the potential influence of land

ownership.

Regarding the determinants of the conditional probability of remaining in poverty, we find

that primary and middle education levels reduce such a probability. Regarding occupation,

professional and own business jobs lower such a probability too. Being from OBC, SC, ST

or being Muslim increases the conditional probability of poverty, with the Muslims faring

the worst. It is worth noting that the policy variable is not statistically significant for a

reduction in the probability of poverty. Further, Table 7 shows the joint and conditional

predicted probabilities for the total sample, for each of the socioreligious groups and regions.

We suggest that it is potentially interesting to examine the conditional probability of exit-

ing poverty, Pr(poor2005=0|poor1994=1). The joint probability instead, Pr(poor2005=0,poor1994=1)

is just indicative but not as interesting in our case. To make our point, we can see that the

joint probability of exiting poverty is 20 % and by socioreligious group, it looks like the

99

Table 6: Marginal average effects

Pr(poor05=1) Pr(poor05=1|poor94=1)

(omitted: edu illi)edu primary -0.027* -0.027*

[0.015] [0.015]edu middle -0.065*** -0.065***

[0.022] [0.023]edu secondary -0.111*** -0.114

[0.038] [219.09](omitted: ocu agri)ocu proff -0.053*** -0.053***

[0.016] [0.016]ocu own -0.067*** -0.068**

[0.023] [0.030]ocu none -0.037 -0.036

[0.028] [0.027](omitted: sociorel high)sociorel OBC 0.038* 0.037*

[0.022] [0.022]sociorel SC 0.087** 0.088**

[0.040] [0.040]sociorel ST 0.091** 0.092**

[0.040] [0.040]sociorel muslim 0.126*** 0.123***

[0.031] [0.032]sociorel others -0.028 -0.028

[0.039] [0.038](omitted: male)female -0.076*** -0.074***

[0.029] [0.025](omitted: age young)age mid 0.090*** 0.088***

[0.020] [0.019]age old 0.007 0.006

[0.019] [0.018](omitted: share mid)share old 0.066 0.065

[0.049] [0.048]share children 0.085*** 0.090**

[0.032] [0.034]assets -0.009*** -0.009***

[0.001] [0.001]policy 0.001 0.000

[0.002] [0.002]split 0.023+ 0.024

[0.014] [0.014]poor1994 0.084*** 0.107**

[0.018] [0.053]land 0.00

[0.000]exp edu -0.207* -0.167

[0.115] [0.124]N 12,352 12352

Note: Standard errors are reported in brackets and allow for intra-village correlation. edu: level of educationof the head, share: share of people in the house according to age, sociorel: socioreligious, male: gender of thehead, age: age of the head, ocu: occupation of the head, exp: expenditure on education. Coefficients for statedummies are omitted. ***: statistically significant at 1%;**: statistically significant at 5%; *:statisticallysignificant at 10%; ++: statistically significant at 15%; +: statistically significant at 20%.

100

Table 7: Joint and conditional predicted probabilities (Percent)

Socioreligious Group

Total High caste OBC Dalit Adivasi Muslim Minorities

Joint probabilities

P(poor05=1, poor94=1) 16 7 14 21 24 22 4P(poor05=0, poor94=1) 20 17 19 24 22 22 17P(poor05=1, poor94=0) 23 20 25 23 26 26 12P(poor05=0, poor94=0) 41 55 42 31 28 29 67

Conditional probabilities

P(poor05=1 | poor94=1) 37 25 37 43 48 47 14P(poor05=0 | poor94=1) 63 75 63 57 52 53 86P(poor05=1 | poor94=0) 40 28 40 46 51 50 16P(poor05=0 | poor94=0) 60 72 60 54 49 50 84

Region

Total Northern U. Central L. Central Western Eastern Southern

Joint probabilities

P(poor05=1, poor94=1) 16 8 21 21 12 24 9P(poor05=0, poor94=1) 20 23 21 20 17 27 18P(poor05=1, poor94=0) 23 15 27 27 26 21 20P(poor05=0, poor94=0) 41 53 31 33 44 28 53

Conditional probabilities

P(poor05=1 | poor94=1) 37 22 46 45 36 44 27P(poor05=0 | poor94=1) 63 78 54 55 64 56 73P(poor05=1 | poor94=0) 40 25 49 49 39 47 30P(poor05=0 | poor94=0) 60 75 51 51 61 53 70

Note: Results are calculated from estimated coefficients from Table A.4

most disadvantaged groups (Dalit, Adivasi and Muslim) have a higher probability of exiting

poverty (24%, 22% and 22% respectively). However, if we consider instead the conditional

probability of exiting poverty, we see that for the total sample, it is 63%. By examining the

estimated probabilities by socioreligious group, we see that the High caste and minorities

have the highest conditional probabilities of exiting poverty (75% and 86% respectively).

Thus the conditional probability is taking into account the previous state of poverty, which,

we argue, is the correct thing to do when allowing for the existence of poverty traps.

Considering regional variation, the conditional probability of exiting poverty is highest

for the Northern (78%) and for the Southern region (73%). By contrast it is lowest for

the Upper Central (54%) region. Similarly the conditional probability of staying non-poor,

Pr(poor2005=0 | poor1994 =0), is the highest for the Northern (75%) and Southern (70%).

Regarding our poverty results, when conditioned on initial poverty, the probability of

ending non-poor is higher than ending poor:

101

Pr(pτ =0| p0 =1)=63% > Pr(pτ =1| p0 =1)=37%

We find the same relation within all socioreligious groups, namely that the probability

of exiting is higher than the probability of staying in poverty. However for the Adivasis and

Muslims the difference is not as high. In general, we find no evidence of poverty traps and

instead we find that poverty is declining - and, thus, the poor will not stay poor.

5 Discussion

5.1 Comparison with previous research

The main difference between our analysis and that in Krishna and Shariff (2011) is that

even though they separate their sample by the household’s poverty situation in 1994, it

is not the same as conditioning on the initial poverty status(as we do). They essentially

re-sample their analysis. First, they choose those households that were poor in 1994 and

non-poor in 2005 and then they choose households that were not poor in 1994 and poor in

2005. Instead we work with the whole sample and with conditional outcomes. Regarding

the explanatory variables, they include community characteristics, and we do not, out of

a concern for endogeneity due to location choice. Plus, they include variables from 2005,

which we try to avoid due to potential simultaneity.

Although the methodology in Jha et al. (2012) is very different, we try to compare

our findings with some of their results. To begin with, they measure the vulnerability of

being poor as the probability of entering poverty and staying poor. Later, they include

their vulnerability measure in their initial year in a multinomial logit to obtain outcomes of

different combinations. From all those results combined we think the result that is most

comparable to our conditional probability would be the poor-poor result. They find that

education is very important just as we do. However, the other variables are defined differently

and hence difficult to compare.

102

5.2 Migration

In the introduction we mentioned that although rural India has not gained from the skill-

biased pattern of development since the 1990s, studies show that the rural sector has

indirectly benefited from rural-urban migration, in particular, from remittances. Data from

the NSSO 64th round (2007-08) show that the average yearly consumption expenditures

for rural households receiving remittances was Rs. 41,000 as compared to Rs. 38,000 for

rural households not receiving remittances, and that 36.5% of rural households received such

remittances (Parida and Madheswaran, 2010). Remittances, therefore, are an important

source of income and consumption-smoothing for certain rural households and this may have

had consequences for rural incomes and poverty in recent years.

Unfortunately our sample of households only includes those families who have actually

stayed in the rural areas in the period under study. This means that our sample is restricted

to the households that have not migrated. However, one or more family members may have

migrated. We do not have information on whether or not a member has migrated in 1994

but our total household income measure includes remittances in any case.

In the literature migrants are generally found to be positively self selected in terms of

both observables and unobservables. However, their economic performance after migration

could vary depending on job availability in urban areas and many could actually be worse-off

after migrating if they do not possess the skills demanded in urban labour markets in India

which have experienced skill-biased technological change in recent times. Therefore the

potential selection bias could go in either direction.

A drawback of our data is that we cannot separate remittances from the total income in

the survey in 1994 but only in 2005; furthermore we do not have detailed information on

remittances. Therefore we exploit only the information on household total remittances in

2005 to analyse its effects on convergence.

Only 6% of our sample received remittances in 2005 (not as many as claimed by Parida

and Madheswaran (2010)). We augment the convergence and poverty analysis by now

103

including a dummy for those households that received remittances in 2005 16. The results

show no significant differences from the ones discussed previously.

6 Conclusions

This paper explores the dynamics of income in rural Indian households over the period

from 1994 to 2005, when India underwent several liberalization reforms, by examining the

time-conditional inequality measures of income convergence and flows out of poverty. The

identification strategy explicitly addresses issues pertaining to the potential measurement

error in income and the potential measurement error in initial income and poverty.

Despite the fact that the raw data shows increasing inequality in income and increased

poverty rates over this period, we find that there is evidence of income and poverty conver-

gence which indicates that poverty traps are not inescapable and that poor households are

indeed catching up to the rich households.

Among the most important results we find that education and asset ownership are crucial

for income accumulation and also for poverty reduction. One policy recommendation would

be to provide access to productive assets to families and at the same time increase public

expenditures on education in rural areas, perhaps emphasizing training of usage of such

assets.

Another robust finding running through income to poverty convergence is that the lower

castes (Dalits and Adivasis as well as OBCs) as well as Muslims, show less income growth

and slower poverty convergence than the High castes and Minorities. This is the case despite

decades of reservation policies for Dalits and Adivasis, suggesting that these groups require

further attention in future inequality reducing measures and so do OBCs and Muslims.

Extensions could consider incorporating the effects of migration and searching for

alternative instruments (within the confines of the previously mentioned issues regarding

these issues). In the future the survey will be conducted again which will allow us to exploit

one more wave of data to confirm our currents results on income convergence and poverty

16We are aware that this may cause problems of simultaneity.

104

transition for households in rural India.

105

Tab

leA

.1:

Defi

nit

ion

ofva

riab

les

and

des

crip

tive

stat

isti

cs

Sh

ort

Use

dN

am

eD

escr

ipti

on

Ob

sM

ean

Std

.D

ev.

Min

Max

y0

lyo94

05p

c1994

an

nu

al

hou

seh

old

inco

me

per

cap

ita

in2005

pri

ces

(base

don

rura

ld

eflato

rsby

Sta

te)

.In

logari

thm

s.

12352

8.6

641

0.8

065

5.1

928

12.6

02

yτ

lyo05p

c2005

an

nu

al

hou

seh

old

inco

me

per

cap

ita

inco

nst

ant

pri

ces.

Inlo

gari

thm

s12352

8.6

465

0.8

204

6.9

078

13.1

1

yo05p

c2005

an

nu

al

hou

seh

old

inco

me

per

cap

ita

inco

nst

ant

pri

ces

12352

8433.1

12641

1000

494000

y94

m0

Dem

ean

edin

com

eb

ase

don

lyo94

05p

cy05

m0

Dem

ean

edin

com

eb

ase

don

lyo05p

cgy

g1p

An

nu

al

gro

wth

rate

of

inco

me

per

cap

ita

inp

rice

sof

2005.

12352

-0.0

016

0.0

877

-0.4

502

0.4

897

def

05

Sta

te-l

evel

rura

ld

eflato

rsfo

r1993/1994

yea

r.B

ase

yea

r:2004/2005

12352

0.5

719

0.0

188

0.5

381

0.5

987

c 0lc

94

05p

1994

an

nu

al

hou

seh

old

food

con

sum

pti

on

per

cap

ita

in2005

pri

ces(

base

don

rura

ld

e-fl

ato

rsby

Sta

te)

.In

logari

thm

s.T

he

pro

d-

uct

sare

:ce

reals

an

dp

uls

es,

sugar,

milk,

edib

leoil,

mea

tan

dfi

sh,

eggs,

veg

etab

les,

fru

its

an

doth

ers.

12351

8.0

971

0.4

616

5.5

685

10.5

87

c τlc

05p

2005

an

nu

al

hou

seh

old

food

con

sum

pti

on

per

cap

ita

inco

nst

ant

pri

ces.

Inlo

ga-

rith

ms.

12328

8.0

65

0.5

103

3.1

781

10.9

47

c94

m0

Dem

ean

edco

nsu

mp

tion

base

don

lc94

05p

c05

m0

Dem

ean

edin

com

eb

ase

don

lc05p

gc

gcp

cA

nnu

al

gro

wth

rate

of

food

con

sum

pti

on

per

cap

ita

inp

rice

sof

2005.

12327

-0.0

029

0.0

501

-0.4

188

0.2

455

p0

poor9

4=

1P

oor,

ifcu

rren

tin

com

ep

erca

pit

ain

1994

isb

elow

the

pover

tylin

ein

1994.

12352

36%

0.4

793

01

poor9

4=

0N

on

poor,

ifcu

rren

tin

com

ep

erca

pit

ain

1994

isab

ove

the

pover

tylin

ein

1994.

12352

64%

0.4

793

01

pτ

poor0

5=

1P

oor,

ifcu

rren

tin

com

ep

erca

pit

ain

2005

isb

elow

the

pover

tylin

eof

2005.

12352

39%

0.4

884

01

poor0

5=

0N

on

poor

ifcu

rren

tin

com

ep

erca

pit

ain

2005

isab

ove

the

pover

tylin

eof2

005.

12352

61%

0.4

884

01

exp

edu

rexe9

4S

hare

of

exp

ense

son

edu

cati

on

of

tota

lan

-nu

al

hou

seh

old

inco

me.

12352

0.0

339

0.0

908

04.2

739

ass

ets

ass

94

Su

mof

wei

ghte

dp

rod

uct

ive

an

du

np

rod

uc-

tive

ass

etin

dex

.P

rod

uct

ive

ass

ets:

trac-

tor,

win

now

er,

etc.

Un

pro

du

ctiv

eass

ets:

car,

bic

ycl

e,te

levis

ion

,et

c.

12352

4.8

82

6.8

805

063

106

Sh

ort

Nam

eU

sed

Nam

eD

escr

ipti

on

Ob

sM

ean

Std

.D

ev.

Min

Max

share

ad

sna

Sh

are

ofnu

mb

erofad

ult

s(1

5-5

9yrs

ofage)

wit

hre

spec

tto

size

of

hou

seh

old

.12352

59%

0.2

127

01

share

old

sno

Sh

are

of

nu

mb

erof

old

peo

ple

(m

ore

than

59

yrs

of

age)

wit

hre

spec

tto

size

of

hou

se-

hold

.

12352

7%

0.1

392

01

share

child

snch

Sh

are

of

nu

mb

erof

child

ren

(6-1

4yrs

of

age)

wit

hre

spec

tto

size

of

hou

seh

old

.12352

34%

0.2

097

01

poli

cyre

sP

oliti

cal

rese

rvati

on

for

dia

dvanta

ged

mi-

nori

tygro

up

s.S

hare

of

seats

inst

ate

as-

sem

bly

rese

rved

for

SC

or

ST

inea

chst

ate

.

12352

4.3

82

7.7

691

024.7

86

split

1if

the

ori

gin

al

hou

hse

hold

in1994

was

split

in2005

12352

0.3

015

0.4

589

01

vill

Villa

ge

12352

386.3

6227.1

71

789

edu

illi

1.e

du

hH

ead

of

hou

seh

old

’sed

uca

tion

.Il

lite

rate

12352

51%

0.4

999

01

edu

pri

m2.e

du

hP

rim

ary

edu

cati

on

12352

28%

0.4

499

01

edu

mid

3.e

du

hM

idd

leed

uca

tion

an

dM

atr

icu

lati

on

12352

17%

0.3

741

01

edu

sec

4.e

du

hS

econ

dary

edu

cati

on

an

dab

ove

12352

4%

0.1

94

01

soci

ore

lh

igh

1.c

ast

rel0

5H

ead

of

hou

seh

old

’sso

ciore

ligio

us

gro

up

.B

rah

min

+F

orw

ard

cast

e12352

21%

0.4

073

01

soci

ore

lO

BC

2.c

ast

rel0

5O

ther

Back

ward

Cla

sses

(OB

C)

12352

34%

0.4

727

01

soci

ore

lS

C3.c

ast

rel0

5D

alit

or

Sch

edu

led

Cast

es(S

C)

12352

25%

0.4

331

01

soci

ore

lS

T4.c

ast

rel0

5A

div

asi

or

Sch

edu

led

Tri

bes

(ST

)12352

9%

0.2

932

01

soci

ore

lm

us

5.c

ast

rel0

5M

usl

im12352

8%

0.2

695

01

soci

ore

loth

6.c

ast

rel0

5S

ikh

,Jain

an

dC

hri

stia

n12352

3%

0.1

678

01

male

1.s

exh

Hea

dof

hou

seh

old

’sgen

der

.M

ale

12352

95%

0.2

081

01

fem

ale

2.s

exh

Fem

ale

12352

5%

0.2

081

01

you

ng

age

1.a

geh

Hea

dof

hou

seh

old

’sage.<

30

yea

rs12352

60%

0.4

894

01

ad

age

2.a

geh

31-5

5yea

rs12352

15%

0.3

549

01

old

age

3.a

geh

>55

yea

rs12352

25%

0.4

329

01

ocu

agri

1.o

cuh

Hea

dof

hou

seh

old

’socc

up

ati

on

.C

ult

iva-

tion

+allie

dagri

cult

ure

+ca

ttle

12352

48%

0.4

996

01

ocu

pro

f2.o

cuh

Wages

,sa

lari

esan

dp

rofe

ssio

nal

12352

33%

0.4

716

01

ocu

ow

n3.o

cuh

Ow

nb

usi

nes

s12352

11%

0.3

096

01

ocu

non

e4.o

cuh

Oth

er(n

ow

ork

or

work

at

hom

eand

not

spec

ified

.12352

8%

0.2

686

01

lan

dla

nd

94

Hou

seh

old

ow

ns

lan

din

acr

es.

Note

:M

in.

an

dM

ax

valu

eis

inon

ed

ecim

al

poin

tse.

g.

Max.

valu

eis

3600

mea

ns

360

acr

esof

lan

d.

12352

37.3

72

98.0

16

03600

2.s

tate

2.s

tate

Sate

.H

imach

al

Pra

des

h12352

6%

0.2

338

01

3.s

tate

3.s

tate

Pu

nja

b12352

6%

0.2

322

01

5.s

tate

5.s

tate

Utt

ara

nch

al

12352

1%

0.1

162

01

6.s

tate

6.s

tate

Hary

an

a12352

7%

0.2

551

01

8.s

tate

8.s

tate

Ra

jast

han

12352

9%

0.2

82

01

107

Sh

ort

Nam

eU

sed

Nam

eD

escr

ipti

on

Ob

sM

ean

Std

.D

ev.

Min

Max

9.s

tate

9.s

tate

Utt

ar

Pra

des

h12352

4%

0.2

054

01

10.s

tate

10.s

tate

Bih

ar

12352

5%

0.2

135

01

18.s

tate

18.s

tate

Ass

am

12352

0%

0.0

27

01

19.s

tate

19.s

tate

Wes

tB

engal

12352

7%

0.2

62

01

20.s

tate

20.s

tate

Jh

ark

hand

12352

2%

0.1

403

01

21.s

tate

21.s

tate

Ori

ssa

12352

7%

0.2

597

01

22.s

tate

22.s

tate

Ch

hati

sgarh

12352

5%

0.2

184

01

23.s

tate

23.s

tate

Mad

hya

Pra

des

h12352

11%

0.3

076

01

24.s

tate

24.s

tate

Gu

jara

t12352

6%

0.2

289

01

27.s

tate

27.s

tate

Mah

ara

shtr

a12352

11%

0.3

147

01

28.s

tate

28.s

tate

An

dh

raP

rad

esh

12352

6%

0.2

415

01

32.s

tate

32.s

tate

Ker

ala

12352

2%

0.1

512

01

33.s

tate

33.s

tate

Tam

ilN

ad

u12352

5%

0.2

093

01

1.r

egio

n1.r

egio

nR

egio

n.

Nort

her

n12352

19%

0.3

884

01

2.r

egio

n2.r

egio

nU

pp

erC

entr

al

12352

13%

0.3

316

01

3.r

egio

n3.r

egio

nL

ow

erC

entr

al

12352

32%

0.4

649

01

4.r

egio

n4.r

egio

nW

este

rn12352

17%

0.3

729

01

5.r

egio

n5.r

egio

nE

ast

ern

12352

7%

0.2

632

01

6.r

egio

n6.r

egio

nS

ou

ther

n12352

13%

0.3

379

01

wei

ght

swe0

5w

eight

from

2005

12352

5664

7292

543

308216

sum

of

wei

ghts

=su

mof

pop

ula

tion

esti

-m

ate

sin

mil

lon

sof

per

son

s70

Note

:S

am

ple

rest

rict

edto

yo05p

c>=

1000.

Most

of

the

vari

ab

les

are

from

1994

surv

ey,

un

less

oth

erw

ise

state

d.

108

Table A.2: Sensitivity analysis for consumption

Consumption Consumption - IV Consumption growth - IVcτ cτiv gciv

c0 0.133*** 0.353++ -0.059***[0.018] [0.242] [0.022]

(omitted: edu illi)edu primary 0.047*** 0.039** 0.004**

[0.014] [0.017] [0.002]edu middle 0.102*** 0.081*** 0.007***

[0.018] [0.031] [0.003]edu secondary 0.214*** 0.170*** 0.015***

[0.036] [0.058] [0.005](omitted: share mid)share old 0.08 0.071 0.006

[0.086] [0.079] [0.007]share children -0.004 0.125 0.011

[0.031] [0.150] [0.014](omitted: cast high)cast OBC -0.054*** -0.046** -0.004**

[0.020] [0.023] [0.002]cast SC -0.100*** -0.074* -0.007*

[0.028] [0.040] [0.004]cast ST -0.142*** -0.105** -0.010**

[0.033] [0.053] [0.005]cast muslim -0.093*** -0.083** -0.008**

[0.030] [0.033] [0.003]cast others 0.008 0.009 0.001

[0.038] [0.039] [0.004](omitted: male)female 0.057** 0.050++ 0.005++

[0.029] [0.031] [0.003](omitted: age young)age mid -0.042*** -0.040** -0.004**

[0.016] [0.017] [0.002]age old 0.003 0.01 0.001

[0.019] [0.019] [0.002](omitted: ocu agri)ocu proff 0.004 0.018 0.002

[0.014] [0.021] [0.002]ocu own 0.017 0.02 0.002

[0.018] [0.019] [0.002]ocu none -0.009 0.009 -0.001

[0.024] [0.032] [0.003]exp edu 0.085 0.095 0.009***

[0.079] [0.078] [0.007]

assets 0.008*** 0.005* 0.000*[0.001] [0.003] [0.000]

policy -0.003** -0.004** -0.000**[0.002] [0.002] [0.000]

split 0.006 0.028 0.003[0.014] [0.030] [0.003]

Constant 7.342*** 5.423** 0.493**[0.159] [2.118] [0.193]

N 12327 12327 12327R2 0.26 0.231

Note: Standard errors are reported in brackets and allow for intra-village correlation. edu: level of educationof the head, share: share of people in the house according to age, cast: caste, male: gender of the head, age:age of the head, ocu: occupation of the head, exp: expenditure on education. Coefficients for state dummiesare omitted. ***: statistically significant at 1%;**: statistically significant at 5%; *:statistically significant at10%; ++: statistically significant at 15%; +: statistically significant at 20%.

109

Table A.3: Convergence parameters

Income Consumptiongy gyiv gc gciv

βy -7.6% -6.9%βc -7.9% -5.9%Test of endogeneity

Ho: variable is exogenousF-test 2.24 1.14p-value 0.130 0.285Explanatory power of IV

R2 0.390 0.319Adjusted R2 0.387 0.317Partial R2 0.044 0.004F-test 9.74 7.87prob F 0.002 0.005

110

Table A.4: Probit and Biprobit models

Probit Biprobit

(omitted: edu illi)edu primary -0.075* -0.073* -0.113**

[0.042] [0.044] [0.050]edu middle -0.182*** -0.174** -0.404***

[0.063] [0.069] [0.074]edu secondary -0.319*** -0.301** -0.960***

[0.113] [0.118] [0.135](omitted: share mid)share old 0.185 0.185 0.1

[0.137] [0.136] [0.203]share children 0.238*** 0.200* 1.756***

[0.091] [0.114] [0.104](omitted: cast high)cast OBC 0.107* 0.105* 0.134*

[0.062] [0.063] [0.073]cast SC 0.244** 0.239** 0.386***

[0.111] [0.112] [0.124]cast ST 0.254** 0.248** 0.508***

[0.112] [0.115] [0.127]cast muslim 0.349*** 0.346*** 0.096

[0.086] [0.087] [0.103]cast others -0.084 -0.084 -0.074

[0.118] [0.118] [0.170](omitted: male)female -0.218** -0.214** -0.193**

[0.086] [0.087] [0.093](omitted: age young)age mid 0.247*** 0.246*** 0.06

[0.054] [0.054] [0.057]age old 0.019 0.016 0.089

[0.052] [0.053] [0.076](omitted: ocu agri)ocu proff -0.148*** -0.157*** 0.120*

[0.045] [0.047] [0.062]ocu own -0.189*** -0.188*** -0.281***

[0.065] [0.065] [0.086]ocu none -0.104 -0.111 0.206**

[0.079] [0.080] [0.086]exp edu -0.580* -0.640* 4.758***

[0.324] [0.375] [0.425]assets -0.026*** -0.025*** -0.044***

[0.004] [0.004] [0.007]policy 0.002 0.002 -0.013**

[0.006] [0.006] [0.007]split 0.064+ 0.057 0.378***

[0.039] [0.043] [0.057]land -0.008***

[0.001]poor94 0.236*** 0.308*

[0.050] [0.165]Constant -0.776*** -0.783*** -1.211***

[0.119] [0.119] [0.151]

N 12,352 12,352 12,352Wald test of rho=0

p-value 0.62

Note: Standard errors are reported in brackets and allow for intra-village correlation. edu: level of educationof the head, share: share of people in the house according to age, cast: caste, male: gender of the head, age:age of the head, ocu: occupation of the head, exp: expenditure on education. Coefficients for state dummiesare omitted. ***: statistically significant at 1%;**: statistically significant at 5%; *:statistically significant at10%; ++: statistically significant at 15%; +: statistically significant at 20%.

111

References

Appleton, S. (2001). The rich are just like us, only richer: Poverty functions or consumption

functions? Journal of African Economies 10 (4), 433–469.

Azam, M. and V. Bhatt (2012, May). Like father, like son? Intergenerational education

mobility in India. Discussion Paper 6549, IZA.

Bardhan, P. (2007). Poverty and inequality in China and India: Elusive link with globalisation.

Economic and Political Weekly 42 (38), pp. 3849–3852.

Cain, J. S., R. Hasan, R. Magsombol, and A. Tandon (2010, March). Accounting for

inequality in India: Evidence from household expenditures. World Development 38 (3),

282–297.

Chin, A. and N. Prakash (2011, November). The redistributive effects of political reservation

for minorities: Evidence from India. Journal of Development Economics 96 (2), 265–277.

Deaton, A. and J. Dreze (2002, September). Poverty and inequality in India: A re-

examination. Working Papers 184, Princeton University, Woodrow Wilson School of

Public and International Affairs, Research Program in Development Studies.

Deaton, A. and V. Kozel (2005). The great Indian poverty debate. Macmillan.

Desai, S. and A. Dubey (2011). Caste in 21st century India: Competing narratives. Economic

and Political Weekly 46 (11), 40–49.

Desai, S., A. Dubey, B. L. Joshi, M. Sen, A. Sharif, and R. Vanneman (2010, March). Human

Development in India: Challenges for a Society in Transition. Number 9780198065128 in

OUP Catalogue. Oxford University Press.

Desai, S., A. Dubey, B. L. Joshi, M. Sen, A. Shariff, and R. Vanneman (2008). Indian human

development survey: Design and data quality. Indian Human Development Survey .

112

Desai, S. and R. Vanneman (2008). India human development survey (IHDS), 2005. Working

papers, Inter-university Consortium for Political and Social Research.

Glewwe, P. (1991, April). Investigating the determinants of household welfare in Cote

d’Ivoire. Journal of Development Economics 35 (2), 307–337.

Greene, W. H. (2008). Econometric Analysis. Prentice Hall.

Hnatkovska, V., A. Lahiri, and S. B. Paul (2013). Breaking the caste barrier intergenerational

mobility in india. Journal of Human Resources 48 (2), 435–473.

Jha, R., W. Kang, H. K. Nagarajan, and K. C. Pradhan (2012). Vulnerability as expected

poverty in rural India. Technical report.

Jolliffe, D. (2002, January). Whose education matters in the determination of household

income? evidence from a developing country. Economic Development and Cultural

Change 50 (2), 287–312.

Kotwal, A., B. Ramaswami, and W. Wadhwa (2011). Economic liberalization and Indian

economic growth: What’s the evidence? Journal of Economic Literature 49 (4), 1152–1199.

Krishna, A. and A. Shariff (2011, April). The irrelevance of national strategies? rural

poverty dynamics in states and regions of India, 1993-2005. World Development 39 (4),

533–549.

Lenagala, C. and R. Ram (2010). Growth elasticity of poverty: estimates from new data.

International Journal of Social Economics 37 (12), 923–932.

Morris, D. (2002, January). Land markets in India: Distortions and issues. Economic

Development and Cultural Change 50 (2), 287–312.

Oldiges, C. (2012, February). Cereal consumption and per capita income in India. Economic

and Political Weekly 50 (2), 287–312.

113

Parida, J. K. and S. Madheswaran (2010). Determinants of migration and remittance in

India: Empirical evidence. In 52nd Annual Conference of the Indian Society of Labour

Economics, pp. 17–19.

Ravallion, M. (2012). Why don’t we see poverty convergence? American Economic

Review 102 (1), 504–23.

Shariff, A. (April/June 2009). Rural income and employment diversity in India during 1994

and 2005. Journal of Developing Societies 25 (2), 165–208.

Singh, A. (2010, July). Does returns to farming depend on caste? new evidence from India.

MPRA Paper 26526, University Library of Munich, Germany.

Wooldridge, J. M. (2002). Econometric Analysis Cross Section Panel. MIT press.

114

DEPARTMENT OF ECONOMICS AND BUSINESS AARHUS UNIVERSITY

SCHOOL OF BUSINESS AND SOCIAL SCIENCES www.econ.au.dk

PhD Theses since 1 July 2011 2011-4 Anders Bredahl Kock: Forecasting and Oracle Efficient Econometrics 2011-5 Christian Bach: The Game of Risk 2011-6 Stefan Holst Bache: Quantile Regression: Three Econometric Studies 2011:12 Bisheng Du: Essays on Advance Demand Information, Prioritization and Real Options

in Inventory Management 2011:13 Christian Gormsen Schmidt: Exploring the Barriers to Globalization 2011:16 Dewi Fitriasari: Analyses of Social and Environmental Reporting as a Practice of

Accountability to Stakeholders 2011:22 Sanne Hiller: Essays on International Trade and Migration: Firm Behavior, Networks

and Barriers to Trade 2012-1 Johannes Tang Kristensen: From Determinants of Low Birthweight to Factor-Based

Macroeconomic Forecasting 2012-2 Karina Hjortshøj Kjeldsen: Routing and Scheduling in Liner Shipping 2012-3 Soheil Abginehchi: Essays on Inventory Control in Presence of Multiple Sourcing 2012-4 Zhenjiang Qin: Essays on Heterogeneous Beliefs, Public Information, and Asset

Pricing 2012-5 Lasse Frisgaard Gunnersen: Income Redistribution Policies 2012-6 Miriam Wüst: Essays on early investments in child health 2012-7 Yukai Yang: Modelling Nonlinear Vector Economic Time Series 2012-8 Lene Kjærsgaard: Empirical Essays of Active Labor Market Policy on Employment 2012-9 Henrik Nørholm: Structured Retail Products and Return Predictability 2012-10 Signe Frederiksen: Empirical Essays on Placements in Outside Home Care

2012-11 Mateusz P. Dziubinski: Essays on Financial Econometrics and Derivatives Pricing 2012-12 Jens Riis Andersen: Option Games under Incomplete Information 2012-13 Margit Malmmose: The Role of Management Accounting in New Public Management Reforms: Implications in a Socio-Political Health Care Context 2012-14 Laurent Callot: Large Panels and High-dimensional VAR 2012-15 Christian Rix-Nielsen: Strategic Investment 2013-1 Kenneth Lykke Sørensen: Essays on Wage Determination 2013-2 Tue Rauff Lind Christensen: Network Design Problems with Piecewise Linear Cost

Functions

2013-3 Dominyka Sakalauskaite: A Challenge for Experts: Auditors, Forensic Specialists and the Detection of Fraud 2013-4 Rune Bysted: Essays on Innovative Work Behavior 2013-5 Mikkel Nørlem Hermansen: Longer Human Lifespan and the Retirement Decision 2013-6 Jannie H.G. Kristoffersen: Empirical Essays on Economics of Education 2013-7 Mark Strøm Kristoffersen: Essays on Economic Policies over the Business Cycle 2013-8 Philipp Meinen: Essays on Firms in International Trade 2013-9 Cédric Gorinas: Essays on Marginalization and Integration of Immigrants and Young Criminals – A Labour Economics Perspective 2013-10 Ina Charlotte Jäkel: Product Quality, Trade Policy, and Voter Preferences: Essays on

International Trade 2013-11 Anna Gerstrøm: World Disruption - How Bankers Reconstruct the Financial Crisis: Essays on Interpretation 2013-12 Paola Andrea Barrientos Quiroga: Essays on Development Economics