University of Southern California Dissertations and Theses

Income inequality and economic growth: A theoretical and empirical analysis

(USC Thesis Other)

Income inequality and economic growth: A theoretical and empirical analysis

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content NOTE TO USERS This reproduction is the best copy available. ® UMI Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. INCOME INEQUALITY AND ECONOMIC GROWTH: A THEORETICAL AND EMPIRICAL ANALYSIS Copyright 2004 by Grigor Martin Sukiassyan A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY FO SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (ECONOMICS) August 2004 Grigor Martin Sukiassyan Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UMI Number: 3145297 INFORMATION TO USERS The quality of this reproduction is dependent upon the quality of the copy submitted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleed-through, substandard margins, and improper alignment can adversely affect reproduction. In the unlikely event that the author did not send a com plete manuscript and there are m issing p ages, th ese will be noted. Also, if unauthorized copyright material had to be rem oved, a note will indicate the deletion. ® UMI UMI Microform 3145297 Copyright 2004 by ProQ uest Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United S tates Code. ProQ uest Information and Learning Company 300 North Z eeb Road P.O. Box 1346 Ann Arbor, Ml 48106-1346 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. DEDICATION To my mother and father who inspired me to pursue for my Ph.D. degree. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. iii ACKNOWLEDGEMENTS I would like to acknowledge my thankfulness and gratitude to God for this opportunity. I would like to thank my dissertation advisor and committee chair, Dr. Jeffrey Nugent for all his advisement, guidance, inspiration, knowledge, encouragement, and support throughout the whole period of writing my dissertation and during the entire time while being a student at the USC Economics Department. He was able to develop interest in me towards the theory and empirical analysis of economic development, growth, and inequality. He was always able to find time in his busy schedule to read and provide very useful comments and suggestions on the drafts and papers prepared by me that have been subsequently developed into constituent essays of my dissertation. It was a real pleasure to have professor Nugent as my academic advisor and I especially would like to mention his good personality, helpfulness, availability, and professionalism. I would also like to thank Dr. Krishna Kumar and Dr. Cheng Hsiao for their advisement as members of my dissertation committee. I especially thank professor Kumar for his many helpful and insightful comments and suggestions, which were very valuable in the direction of developing my theoretical work. He was always friendly, encouraging, and constructive. I found also very helpful his support throughout my employment endeavors, given especially the very competitiveness of the current job market. I would also like to thank professor Hsiao for the extremely useful empirical material I have learned that I have subsequently made an extensive Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. use while writing my dissertation. It was a pleasure also being his teaching assistant for the course of Applied Econometrics that gave me an opportunity to deepen my understanding in the subject. Many thanks go to all the professors who instructed me in the Economics and with whom I have worked in the capacity of being Teaching Assistant. This was a great learning experience for me. I would also like to thank Ms. Young Miller and Ms. Sheila Williams for their incredible help with many administrative issues. I would also like to thank the Graduate School for granting me the Final Year Dissertation Fellowship. I am grateful to my father, who when he was alive, always encouraged me to pursue my dreams. I owe all my thanks to my mother and sister for their love, encouragement, and support throughout the whole period of my studies. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. V TABLE OF CONTENTS Dedication ii Acknowledgements iii List of Tables vii List of Figures ix Abstract x 1. Introduction 1 2. Inequality And Growth: What Does The Transition Economy Data Say? 9 2.1 Introduction To Essay One 10 2.2 Data Selection And Description 16 2.2.1 Sources Of The Data 16 2.2.2 Socioeconomic Evaluation 17 2.2.3 Regional Description 21 2.3 The Model And The Estimation Method 25 2.4 Estimation Results And Interpretation 30 2.4.1 Basic Results 30 2.4.2 Reduced Form Results 37 2.4.3 Results For Changes In Inequality 41 2.4.4 Results For Other Estimations 43 2.4.5 Sub-Period Estimations: Speculating About Future Sign Reversal? 47 2.5 Concluding Remarks 50 Chapter 2 Bibliography 53 3. Human Capital, Fertility, And Enterprise: Implications For Income Distribution And Growth 55 3.1 Introduction To Essay Two 56 3.1.1 Brief Literature Review And Discussion 61 3.2 The Household Sector 64 3.2.1 The Household Utility Maximization Problem 64 3.2.2 The Child Quantity -Quality Tradeoff 70 3.3 The Occupational Choice And The Credit Allocation Mechanism 73 3.3.1 The Market Investment Structure And The Optimal Contract 73 3.3.2 Credit Rationing And The Choice Of Lending Or Borrowing 79 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. vi 3.3.2.1 Both Entrepreneurs And Workers Share The Operating, Monitoring Costs 81 3.3.2.2 Only Entrepreneurs Are Responsible For Paying The Operating, Monitoring Costs 84 3.3.3 Occupational Choice 89 3.4 Static Equilibrium Conditions 90 3.4.1 Labor Market Equilibrium 90 3.4.2 Capital Market Equilibrium 92 3.4.3 Competitive Equilibrium 92 3.5 The Dynamic Behavior And The Stationary State 94 3.5.1 Individual Dynamics (This Subsection Is Related To Case (A), 3.3.2.1, When Both Entrepreneurs And Workers Share The Operating And Monitoring Costs) 94 3.5.2 Distributional Dynamics 97 3.5.3 The Stationary State 99 3.6 The Case Of The Social Planner’s Problem 99 3.7 Computational Experiments 102 3.7.1 Calibration 102 3.7.2 Quantitative Examples 103 3.7.2.1 Example 1. (The Degenerate Human Capital Distribution Case) 103 3.7.2.2 Example 2. (The Bivariate Uniform Distribution Case) 104 3.7.2.3 Example 3. (The Bivariate Lognormal Distribution Case) 110 3.8 Conclusion 115 Chapter 3 Bibliography 117 4. Education And Self Selection: The Role Of Household And Individual Characteristics In Wage Earnings Distributions In East Germany After Reunification 120 4.1 Introduction To Essay Three 121 4.2 Data Selection And Estimation 127 4.3 Wage Inequality Dynamics And Segregation By Skill 141 4.4 The Decomposition Method 147 4.5 The Results Of The Decompositions And Their Interpretation 152 4.6 A Self-Selection Model 159 4.7 Concluding Remarks 170 Chapter 4 Bibliography 173 5. Conclusion 176 Comprehensive Bibliography 186 Appendix 192 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. vii LIST OF TABLES Table 2.1 Summary review of the empirical literature 14 Table 2.2 Initial values of per capita real GDPs and inequality indices for the 26 countries 18 Table 2.3 Summary statistics of variables used in estimations. Both the whole period as well as sub-period data are provided 27 Table 2.4 Fixed and Random Effects Panel Estimations 31 Table 2.5 Random Effects and MLE Estimations 32 Table 2.6 Random effects panel IV estimation with only linear inequality explanatory term involved 33 Table 2.7 Random effects panel IV estimation with both linear and quadratic inequality explanatory terms involved 34 Table 2.8 Random effects panel IV estimation with the difference Terms of the Gini coefficient included 35 Table 2.9 3SLS estimation with selected educational and fertility variables in controls 36 Table 2.10 3SLS estimation with the difference terms of Gini Coefficient included 37 Table 2.11 Reduced form results, Random effects panel IV estimation 38 Table 2.12 Reduced form results, three stage least squares estimation 39 Table 2.13 Random effects panel IV estimation with the difference terms in inequality and lagged inequality as dependent variables 41 Table 2.14 Three stage least squares estimation with the difference terms in inequality and lagged inequality as dependent variables 42 Table 2.15 Panel IV estimation using Anderson and Hsiao method 45 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 2.16 Panel estimation using Baltagi and Chang random effects estimation method 46 Table 2.17 Three stage least squares simultaneous estimation of GDP growth rates involving the three sub-sample periods of 1988-92, 1993-97, and 1998-2001 48 Table 3.1 Partial equilibrium at several fixed wages (uniform distrib.) <j)=0.078, y=0.3, x=0.2, x, = 1, b, = 1, R=5% 105 Table 3.2 <|>=0.078, y=0.4, x=0.2 106 Table 3.3 <|>=0.15, y=0.3, x=0.2 106 Table 3.4 Partial equilibrium population growth at some fixed wages (lognormal distrib.) <|)=0.078, y=0.3, x=0.2, x t = 1, 6, = 1, a=0.5, R=5% 110 Table 3.5 Population growth at some wages and c t ’ s (lognormal) 111 Table 4.1 Averages of selected variables by year. East German Sample 128 Table 4.2 Instrumental variable estimation of log monthly net income 140 Table 4.3 Wage decomposition by year, weighted Oaxaca method 154 Table 4.4 Wage decomposition by year, cellular method 155 Table 4.5 Wage decomposition, comparison of weighted Oaxaca and cellular methods 157 Table 4.6 Wage decomposition, weighted Oaxaca and cellular methods comparison 161 Table 4.7 Logit estimation of the indicator function I(i), Employed (both part- and full- time) 167 Table 4.8 Logit estimation of the indicator function I(i), Full-time sample 168 Table 4.9 Logit estimation of the indicator function I(i), Part-time sample 169 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ix LIST OF FIGURES Figure 2.1 Cross-country distributional dynamics of earnings inequality, RGDP growth rate, investment ratio, and total fertility rate 19 Figure 2.2 The Regional dynamics 20 Figure 3.1 The timing of the events 66 Figure 3.2 The classification of the agents 76 Figure 3.3 Monitoring costs 82 Figure 3.4 Classification of agents, case 1 84 Figure 3.5 Classification of agents, case 2 88 Figure 3.6 The dynamics when agents have equal wealth 95 Figure 3.7 The dynamics when agents have equal human capital 97 A Chart 3.1 The bt (x) and bt (x) lines (uniform distribution) 107 Chart 3.2 PDF functions, uniform distribution 108 A Chart 3.3 The bt (x) and b, (x) lines (lognormal distribution) 112 Chart 3.4 PDF functions, lognormal distribution 113 Figure 4.1 Kernel density estimates of Treiman occupational prestige scale, East German workers 130 Figure 4.2 Returns to education and experience, east-west comparison of the coefficients from the Mincer regressions 132 Figure 4.3 Kernel density estimates for wage inequality. Log monthly wages are used 138 Figure 4.4 Inter-industry coefficients for education, experience, and log wages, east-west comparisons 145 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ABSTRACT This dissertation consists of three interrelated essays and focuses on the effects of wealth and human capital distributions on choices of investment and employment, income inequality, and economic growth. The first essay is devoted to the theme of inequality and economic growth for the economies in transition from Central and Eastern Europe and the Commonwealth of Independent States (CIS). Despite their initial similarities, subsequently they have experienced significantly different growth rates and changes in income inequality. The empirical analysis reveals that the effect of inequality on growth is negative and strong. This result is robust to the use of the different specifications and estimation methods that in different previous studies had led, even when sharing the same data, to very different results. The second essay discusses the distributional effects of human capital and wealth on small business startups, credit rationing, child quality-quantity tradeoff, and income inequality. The following results based on an OLG model are received: Given the initial distribution in wealth and human capital, the smaller is the gap between wages and entrepreneurs’ returns, the more the child quality-quantity tradeoff is biased towards the former. Lower wages increase the number of both potential and credit rationed entrepreneurs. Higher wages increase the human capital threshold to be eligible for credit. The higher is the correlation between wealth and human capital, the fewer are the potential entrepreneurs who are credit rationed. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The third essay examines wage inequality dynamics in East Germany from its reunification with West Germany until 2001. While under the centrally planned system even vocational education was sufficient to secure employment, after reunification East German workers had to either receive re-training and apply for jobs in western industries offering higher wages, or remain in eastern, less competitive industries offering lower salaries and little prospect for future growth. The results show that individual characteristics played a significant role in labor adjustment process. While wages were set by industries, the inequality dynamics was largely influenced by the workers’ education, employment experience, and related background. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1 1. Introduction My dissertation examines the relationship between income inequality and economic growth both theoretically and empirically. Its purpose is to analyze the effects of wealth and human capital inequality on choices of investment and employment, income distribution, wealth accumulation, and economic growth. The empirical analysis is based on both cross-country macro and single country household survey data from countries in transition. The dissertation consists of several interrelated essays. The first is devoted to the theme of economic growth and inequality at the cross-country level, focusing on why some countries are developed and others less so? What’s the impact of differences in income inequality, savings, and fertility on subsequent per capita GDP growth rates? The neoclassical convergence theory asserts that assuming earnings inequalities, savings, and fertility rates are the same across countries, per capita incomes in poor countries should grow faster and eventually, living standards in all countries must converge. However, persistent cross-country differences suggest that there are differences in attitudes, preferences and cultures across societies that change only slightly over time. Differences in policies are also relevant. For instance, different countries have different policies towards encouraging savings and investment, control of population growth, alleviating bureaucracy and corruption, etc. The essay focuses on a specific group of countries namely, the economies in transition from Central and Eastern Europe and the Commonwealth of Independent Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2 States (CIS). One attractive feature of this group of countries is that their starting points were quite similar. They initiated their transitions simultaneously, with inherited low values of income inequality, similar per-capita GDP and GDP growth rates, many common policies and objectives. Despite these similarities, subsequently significant changes in growth rates and income inequality have taken place both within and across these countries. My empirical estimates indicate that the effect of inequality on growth is negative and strong. This result is rather robust to the use of the different specifications and estimation methods that had yielded conflicting results in different studies, despite having used the same data sources. The policy implication of this study is that the quality of government activity and decision-making is indispensable for rapid recovery and sustainable growth. Well- performing governments safeguard the integrity of financial markets and investment mechanisms, effectively privatize the industrial and agricultural sectors, and protect the property rights, that are necessary in order to compete effectively in a globalized market environment and a smooth and rapid transition. The speed of the realization of these policies was quite different across these countries, yielding substantial differences in growth rates and income distributions. My results, based on transition economy data, indicate a negative relationship between initial inequality and subsequent growth. It is also shown that there is a negative relationship between the change in inequality and initial income and a positive relationship between the squared change in inequality and initial income. The results do not show any significant lagged effect of inequality on growth. In Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3 contrast to conflicting results found in the literature these empirical findings for transition economies indicate a strong, contemporaneous growth-inequality relationship in the short to medium run for similar estimations and specification methods carried out by various authors. These findings imply that the results are sensitive to the specific choice of sample countries. My results also support Kuznets’ inverted - U hypothesis. From these findings I project that inequality will decrease over time as these countries continue to grow and as the transitional effects attenuate and the trend effects of economic growth become dominant. The second essay of the dissertation focuses on market imperfections pervasive in the developing countries. In the absence of complete financial markets, discontinuities in production technology and uncertain production lead to moral hazard effects in lending and borrowing and thereby to credit rationing. In such situations initial inequality may result in poverty traps and hence perpetual poverty for some groups. The essay specifically focuses on the effects of human capital and wealth distributions on small business startups, credit rationing, and small business income and wage earnings inequality. A critical assumption and one that distinguishes this essay from others in the literature is that human capital (measured by years of schooling), or entrepreneurial skills can substitute for wealth as collateral in borrowing. The literature emphasizes that when agents are heterogeneous in their wealth and entrepreneurial ability, or human capital, both income distribution and economic growth in the long run depend on the initial distribution of agents’ inherent Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 4 entrepreneurial skills. Little attention has been paid however to how these entrepreneurial skills evolve over time. Namely, we endogenize human capital and its evolution over time. In our analysis, human capital investment becomes a choice variable in that individuals decide how much education their children would receive as opposed to having more children. In aggregate terms the paper attempts to address the issue of how a certain path of human capital accumulation affects the level of entrepreneurial activity, investment, earnings distribution and wealth accumulation. With setup costs involved in the acquisition of certain occupations or skills, including entrepreneurial skills, and the borrowing constraints for poor agents, the initial distribution of wealth and human capital affects the aggregate skill composition of the economy, entrepreneurship, small business activity, investment, and output. Poor families don’t find much incentive to invest in the education of their children, locking their descendants into a poverty trap. High initial inequality thus may perpetuate itself. Countries with historically high poverty rates may have persistently low per capita incomes. The dynamic implications of the OLG model employed in this study are reflected in fertility dynamics resulting from the child quality-quantity tradeoff mechanism. Distributional dynamics are also considered. The results are as follows: The effect on fertility: Given the initial distributions in wealth and human capital, the higher are the wages relative to the entrepreneurs’ returns, the lower is fertility in the long run. This is a replication of a standard and well-documented result. My results also indicate that if the initial distributions are highly unequal, an increase in Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 5 earnings inequality raises fertility in subsequent periods. If, however, the initial distributions reflect low inequality, the increase in earnings inequality causes fertility to decline in subsequent periods. Occupational choice effect: The higher are the wages relative to the entrepreneurial return, the higher is the threshold human capital level that allows one to qualify for credit among potential borrower investors with the same wealth level. Credit rationing effect: With lower wages the number of potential investors and of those who are credit rationed increases. Also, the higher is the correlation between other forms of wealth and human capital, the fewer are those who would be credit rationed. These results reveal that by simply introducing entrepreneurial skills in the production function and considering them as collateral for purpose of borrowing I obtain qualitatively different results than those recorded in literature, where this consideration has been absent. The third essay investigates how and to what extent wage inequality has changed during post-reunification period of 1990 to 2001 in East Germany. Under the central planning system most East German workers received vocational education, leading to considerable specialization beginning at an early age. Moreover, the socialist system guaranteed job security and as a result workers had little incentive to acquire skills to cope with unexpected events, such as loss of employment. After reunification, with exposure to outside competition, demands for certain skills increased, rendering other types of training obsolete. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 6 The education acquired under central planning became less needed under the western, market system, as the old education system was biased toward the hard sciences and engineering, neglecting the social sciences, law, business management, and public policy. Under these new conditions, East German workers had to solve the dilemma to either needing training to become qualified for applying to jobs in industries offering higher returns to skills and wage incentives compatible with western standards, or remaining in the eastern, older industrial system, offering modest salaries with little future prospects. The German Socioeconomic Panel (GSOEP) household survey data for the period from 1990 to 2001 is used for the analysis. After reunification there was a substantial shift in labor demand from traditional manufacturing to trade, services, and finance as West German firms’ investments in East Germany were predominantly concentrated in these industries. To find out which industries were exactly the ones having the most of the influx of western investments and hence the ones offering the highest incentives to East German workers in terms of higher returns to skills, education and experience, I have conducted wage decompositions with industry dummy variables as the regressors. Conventionally, decompositions are mostly conducted for the purpose of tracking the effects of workers’ education and experience on the dynamics of wage inequality. In this respect my approach of decomposing wages by industry is innovative. It allows me to identify the industries that have contributed to the increase in inequality as well as those offering the highest incentives for skilled workers. The use of Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 7 sectoral decomposition reveals also how and by how much education, experience, and wages influence inter-sectoral skilled labor movements from the sectors offering the least incentives to the ones offering the most. Two distinct generalized wage decompositions are applied, namely, Blinder- Oaxaca and “cellular” methods. The results of these two methods are generally quite similar. Following these decompositions results, a self-selection model in which workers decide either (1) to invest time and resources to get re-trained and become qualified for the ‘western’ jobs offering higher returns to skills or (2) to continue working in the conventional, lower-pay occupations and poor prospects for improvement is applied. An indicator function, outlining those industries where there were significant labor outflows and those ones experiencing significant labor inflows is constructed based on the wage decompositions results. This indicator function is estimated using logistic procedure on general, vocational, and higher education, work experience, number of children, and other individual background variables to find out the effects of these individual labor aspects on the changes in labor supply schedule and wage inequality dynamics over the period. The analysis shows that individual and household characteristics play a significant role in the adjustment of labor to the new, western market. While wages are largely set by industries, determined by the industry specific needs for qualified labor, wage inequality and its dynamics are largely influenced by individuals’ education, work experience, household, and other related background. Specifically, the results show a significant, negative effect of vocational education on inter Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 8 industry labor movements. While, the effect of secondary general education is insignificant, the effects of higher education and monthly salaries are positive and significant. Surprisingly, while the effect of experience is positive, it is not statistically significant. Surprisingly also, the results show that women have moved to ‘western’ industries to a greater extent than men. The coefficient of number of children is negative and significant. Hence, the results indicate that the adjustment of East German workers to the inter-industry shifts in labor demand was largely based on and affected by their individual characteristics. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 9 2. Inequality and Growth: What Does the Transition Economy Data Say? Despite an extensive literature on inequality and growth, there remains considerable disagreement on the effect of inequality on subsequent growth. This paper attempts to empirically re-evaluate the relationship between inequality and growth with data from the transition economies of Central and Eastern Europe and the Commonwealth of Independent States (CIS). One attractive feature of this group of countries is that their starting points were remarkably similar. They initiated their transitions almost simultaneously, with inherited low values of income inequality, similar levels of per capita GDP and GDP growth rates, many common policies, and similar objectives. Despite their similar initial conditions, subsequently enormous changes in growth rates and income inequality have taken place both within and across these countries. Hence, this data set offers considerable promise for investigating the relationship between growth and inequality. The econometric estimations indicate that the effect of inequality on growth is negative and strong. This result is rather robust to the use of the different specifications and estimation methods that have been applied in the relevant literature. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 10 2.1. Introduction to Essay One The vast existing literature on the effects of initial inequality on subsequent growth has resulted in unresolved disagreements. Indeed, different cross-country empirical studies arrive at different conclusions on the subject, even when they share common sources of data. This essay attempts to re-evaluate empirically this relationship for a quite different set of countries, namely, the relatively sizeable group of countries from Central and Eastern Europe and the Commonwealth of Independent States (CIS) that have been undergoing transitions to market economies. There are several reasons for focusing on the transition economies. First, the countries began their transitions at virtually the same time and shared many initial characteristics. On account of the former political regime, all these countries inherited relatively low levels of income inequality (reflected in low Gini coefficients), similar levels of per capita GDP and GDP growth rates (in fact the newly formed CIS member countries and those from the Baltic region were constituent parts of the same country). The countries also shared common policies towards education, social security, employment, family planning, healthcare, and political structure. Second, as changes in the political system were taking place, these countries shared similar transition objectives, such as creating new financial markets, investment mechanisms, privatizing the industrial and agricultural sectors, creating market mechanisms for intra and inter-industry trade, and some of them creating Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. their own central banks. Yet, the speed of the implementation of these policies was sufficiently diverse across the countries, yielding substantial differences in growth rates. Given that the initial inequality was more or less the same across the countries, these policy differences provide a thus far unexplored opportunity to identify the ones that are the most effective in promoting growth. Furthermore, the analysis identifies the country characteristics that are both positively and negatively related to the growth. Third, since most existing papers on growth and inequality have been international cross-section studies, they suffer from two econometric problems: measurement error and omitted variable bias. The use of various data sources allows to obtain an extensive list of variables characterizing macro, financial, social, and political features of the sample transition economies and thereby to mitigate the omitted variable bias problem. Similarly, measurement error, which may arise because countries have different definitions of key variables, particularly of inequality, and/or varying standards and degrees of accuracy in data collection, is mitigated for the transition economies because these countries share similar measurement standards and rules. Fourth, the existing literature has largely omitted these countries from cross country growth-inequality analyses, primarily due to the lack of coverage of the transition economies in the well-known and widely used institutional data sources. In particular, the major source of data on inequality for existing studies is Deininger and Squire’s (1996) comprehensive cross-country inequality data. Since Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 12 such data has observations over time as well as across countries, it affords an opportunity to use a variety of estimation techniques, and since these different observations are given quality ratings it also affords an opportunity to average the sample over different sub-periods and hence different numbers of observations and of course different functional forms. Yet, the only transition economies included in the data set are Bulgaria, Hungary, and Poland, and even in those cases observations are available for at most two to three years. Most of the recent cross-country studies have drawn their country samples from that data set. For example, Barro’s (2000) sample consists of 84 countries from that data set, while Forbes’s (2000) sample contains only the 45 countries whose income inequality data was deemed to be of “high quality”. Banerjee and Duflo (2003) use both the 45 country high quality sample of Deininger and Squire (1996), as Forbes does, and a 50 country sample from Barro. Other studies, such as Alesina and Perotti (1996), Perotti (1996), Persson and Tabellini (1994), etc., use various earlier sources of inequality data than that provided by Deininger and Squire (1996) and as a rule there are no transition countries in their samples. Thus, due to the lack of earlier sources of reliable data on these transition countries1 they have largely been omitted from the growth-inequality literature. 1 One reason of the absence o f sufficiently reliable data in the mainstream literature could be the existing “Nomenklatura”, or in other words the Iron Curtain during the Soviet era and until before the 1990’s when the system was over. After the fall of the system, the ensuing transition didn’t take up quite smoothly in most o f the countries. The markets and financial institutions were in the stage o f formation. Interest rates were unrealistically high while inflation was skyrocketing. Moreover, the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 13 Even when using relatively similar data sets with minimal or no inclusion of transition economies, the various authors of such studies obtain quite different results. In large part this seems to be the result of different estimation techniques and empirical specifications. Table 2.1 lists the data sources used, the estimation methods, and the results obtained in these works. From the table, Alesina and Perotti (1996), Perotti (1996), and Persson and Tabellini (1994), though not correcting for time-invariant omitted variable bias, find significant and negative effects of income inequality on subsequent economic growth. Forbes (2000) and Li and Zou (1998), on the other hand, use panel analysis and find a positive relationship. Barro (2000) uses 3SLS, claiming that the use of fixed effects eliminates the main (cross-sectional) source of variation in the data. With random effects, no significant relationship between inequality and growth is found for the whole sample. Yet, when the sample is divided into sub-samples of poor and rich countries, it turns out that the inequality-growth relationship is negative in the sample of poor countries but positive in the sample of rich countries. These results suggest that the inequality-growth relationship is likely to vary across samples. Banerjee and Duflo (2003) emphasize that the existing relationships between variables might be far from being linear. Their results on inequality and growth are inconclusive for different samples and different specifications (using both the Perotti and Barro specifications) and estimation methods. Once again, these results new governments were merely taking over and there was a “vacuum” o f information until after the stabilization process took over in mid 90’s. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 14 Table 2.1. Summary review of the empirical literature Author Data source Estimation Specification Inequality effect Coverage Accounti method on growth of transition ng for econom ies nonlineari ties Persson Summers and OLS Based on their Negative None No and Heston political Tabellini (1988), economy (1994) Maddison (1982), World Bank (1984), other sources model Perotti Barro and OLS Perotti Negative None No (1996) Wolf (1989), Barro and Lee (1993), other sources Alesina Barro and 2SLS, Bivariate Negative None No and Perotti Wolf (1989), 3SLS simultaneous (1996) Barro and Lee (1993), other sources equation based on their sociopolitical stability model Lee and Deininger and Panel, Based on their Positive None No Zou Squire (1996), fixed model (1998) Barro and Lee effects (1996), other sources Barro Deininger and 3SLS Barro Whole sample: Three 2 Yes (2000) Squire (1996), Barro and Lee (1996), other not significant; Poor: negative; Rich-positive countries sources Three1 4 Forbes Deininger and Panel, Perotti (lag 1) Positive No (2000) Squire (1996), fixed Barro and Lee effects (1996), other countries sources Three1 4 Banerjee Deininger and Panel, Perotti, Inconclusive: Yes and Duflo Squire (1996), fixed Barro Positive for countries (2003) Barro and Lee effects, (1996), other random sources effects; First difference Perotti, mixed for Barro specifications 2 Those are Bulgaria, Hungary, and Poland. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 15 depend on the specific choice of the group of countries, the selected specifications and the estimation methods, and the underlying assumptions about the existing relationships. However, they also find that (1) changes in inequality (in any direction) are associated with lower future growth rates, (2) there exists a strong, negative relationship between changes in inequality and past inequality, and (3) there is a negative relationship between growth rates and inequality lagged one period among countries where the level of inequality was not very high to start with. On account of (3) it is important to re-examine this relationship given the initially low inequality levels of transition economies. Our econometric estimates for the transition economy sample show that the effect of inequality on growth is negative and strong. Surprisingly, the result is rather robust to the use of the different specifications and estimation methods applied in the relevant literature. Our results extend existing knowledge on the effect of inequality on growth by investigating the relationship for an important set of countries that has been largely omitted from the literature. The following section describes the data and compares the cross-regional change of several key variables over the period of transition. Section 3 focuses on the model and the estimation methods. Section 4 examines the results. Section 5 concludes. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 16 2.2. Data Selection and Description 2.2.1. Sources of the Data Our primary data source is the TransMonee 2002 Database3, provided by UNICEF. This database provides information on the 26 countries from the Central and Eastern Europe and the Commonwealth of Independent States (CIS) listed in Table 2.2. The rest of the data is from the Penn World Tables, International Financial Statistics (IFS), and International Country Risk Guide (ICRG). Our data set covers the period from 1988 to 2002. Most of the inequality in earnings data, measured by Gini coefficient, is taken from TransMONEE database4. Some of the pre-1992 values for this index are taken from Atkinson and Micklewright (1992), the rest interpolated from grouped data from household budget surveys reported to the MONEE project. As indicated in the table, the sociopolitical variables, such as political stability, investment risk, the level of corruption, bureaucracy, law and order, democratic accountability are obtained from ICRG, the financial macro-indicators, such as lending rates, inflation rates, government consumption expenditures, and trade 3 The MONEE project is a part o f the research activities carried by the UNICEF Innoceti Research Center. The project analyses social conditions and public policies affecting children and their families in Central and Eastern Europe (I denote CE and EE respectively) and the Commonwealth of Independent States (CIS). 4 Other variables used from the TransMONEE database include the total fertility rate (TFR) (births per woman), the crude birth rate (live births per thousand population) (CBR), the educational enrollments, which include secondary general, secondary vocational, and higher educations (gross rates, percentages o f the relevant age population), the average age o f mothers at first birth, and the ratio of public expenditures on education to GDP. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 17 balances are from IFS dataset, and the remaining macroeconomic variables, such as per capita GDP (as a percent of US real GDP), the GDP growth rate (as a percent of the base year, 1995, growth rate), openness, the investment to GDP ratio, the price level of investment, measured as the PPP of investment/exchange rate relative to the US (PPPI), are all taken from Penn World Tables. The above-mentioned variables as controls are chosen in the econometric analysis of inequality to growth relationship based on the specifications used in the literature that is discussed in Section 3. 2.2.2. Socioeconomic Evaluation A brief comparison of the socioeconomic characteristics of this group of countries relative to the rest of the world is provided in Table 2.2. As the table reports in 1989 the sample transition countries had roughly equivalent levels of per capita GDPs and belonged to the upper-to-middle5 income group according to the World Bank’s classification. The 26 countries of the sample can be aggregated into seven subgroups based on geographic, social, and economic criteria of the countries as follows: Central Europe (CE), the Balkans (BALK), East Europe (EE), the Baltic (BALT), West CIS (WCIS), the Trans-Caucasus (TCA), and Central Asia (CASIA). 5 According to the World Bank 1999/2000 World Development Report, p.291, countries are ranked by their PPP adjusted per capita GNPs as follows (1998 level): low income, $760 or less; low-to-middle income $761-$3030; upper-to middle income, $3031-$9630; and high income, $9630 and more. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 18 Table 2.2. Initial values of per capita real GDPs and inequality indices for the 26 countries Distribution 1 of earnings: i Gini i coefficient < Distribution I D f income: i Gini < coefficient RGDP Real GDP per capita, constant price, growth Laspeyres, adjusted for terms of trade (index, changes (som e of the data in 1990 is 1989=100) from the first available year after 1990) Year 1989 1989 2001 1990 1996 2000 Albania 109.7 1819.9 3070.8 3402.1 Armenia 0.258 0.251 63.1 2828.4 2393.3 2830.6 Azerbaijan 0.275 0.308 56.2 2350.2 2046.8 3292.5 Belarus 0.234 0.229 87.3 8790.0 5667.2 8090.9 Bulgaria 0.233 73.7 7567.6 5893.3 5868.2 Croatia 0.36 83.5 7427.9 8839.9 Czech Rep 0.204 0.198 101.2 14010.2 13457.7 13959.8 Estonia 0.253 0.28 86.7 7041.9 7509.6 10047.5 FR Yugosl. 49.7 FYR 74.4 3988.0 4571.2 5131.4 Macedonia Georgia 0.301 0.28 35.4 4478.8 4888.5 Hungary 0.268 0.225 108.9 9566.3 8707.8 10386.1 Kazakhstan 0.276 0.281 76.1 6005.1 5882.2 7770.4 Kyrgyzstan 0.26 0.27 69.2 3439.4 2596.0 2951.8 Latvia 0.244 0.26 68.3 11059.8 6194.1 7671.4 Lithuania 0.26 0.263 67.5 6867.8 6495.3 7793.0 Moldova 0.25 0.251 34.2 2264.2 2031.4 Poland 0.207 0.275 129.1 6743.9 7712.9 9061.1 Romania 0.155 0.237 80 4997.9 4983.8 4702.3 Russia 0.271 0.265 66.5 9432.2 7106.8 9285.7 Slovakia 0.2 105.7 12929.0 9992.9 11771.5 Slovenia 0.219 116.8 12746.1 13151.5 15860.5 Tajikistan 0.276 0.281 50 985.7 1287.1 Turkmenist. 0.255 0.279 82.9 Ukraine 0.244 0.228 41.4 9884.8 4402.5 4764.9 Uzbekistan 0.257 0.28 101.1 2418.9 2803.8 Sources: TransMONEE 2002 Database (UNICEF) and Penn World Tables Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. As Figure 2.1.1 displays in 1989 the Gini coefficients of all transition countries were rater tightly clustered around 0.26 but thereafter became more diverse and generally much higher as 1992, 1996, and 2000 density charts reveal. After the initial slowdown in Figure 2.1. Cross-country distributional dynamics of earnings inequality, real GDP growth rate, investment ratio (percent of real GDP), and total fertility rate. 1.1, Gini, Earnings Distribution 1.2 Real Per Capita GDP Distribution 5 0.10 ® 0.08 0.08 a) 0.06 0.02 0.00 o o o o o o o o o o <2 O ) C O 8 Ul ro c r > -o 00 C O ro 4k _k _k k) to k> C O C O 4 k 8 * v | 00 1 0 * n | o > o 00 o> • * > 00 Ol 00 • V l I 4 k O 4k 00 to O ) C J 1 C O 2 o> C O 1989 Gini values 1992 — 1996 - 2( P er capita GDP, m easured in USD ■1989 — » -1 9 9 2 i 1996 ~s<—2000 1.3, Investment Rate Distribution 1.4, TFR Distribution -1989 ) -k ) w o > ro ro O - J k . IO P ° C O C O C J l C O C O O ) 0 0 % GDP bi ro b) 1992 •2000 0.30 0.25 0.20 & tn c 0.15 a > a 0.10 0.05 0.00 O -k CO co o at — k N ) t o CO CO ^ 00 t o o — ^ - £k ^ ( O ro o o CO •1989 TFR values •1992 1996 “ 2 0 I Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 20 growth due to the transitional shock, some of these countries began to recover rapidly and Figure 2.2. Regional Dynamics o o < o 2.1, Lending Rate 250 200 0 ) 2 150 O ) c T J c 0 ) _ l 1 0 0 CO — k . CO CO CO CO CO CO G J CO CD CO CO CO BALK « BALT * CAS IA • TCA a CE x E E -■+ WCIS 2.2, Investment Rate ..- H h s a s t ..X x x | J L to < p < o W £ C J l 1200 1000 0 ) g 800 | 600 C O £ 400 200 0 CO CO C O C O — 1 . C O C O CD CO C O C O C O C O CO O) C D -N| C O 2.3, Inflation Rate f J I j!\ ■ */ \ \ i I W i \ J \ v \ V f - J i \ \ | A [■.■'A- * * r t r i r t i i . . . 2.4, GDP Growth as a % of Base Year 110 100 * « * * y 00 00 C D - X " > * CO CO CO CO CO CO CO CO CO CO C O O O C O C D C O C O C O C O C O C O C O C O Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 21 ended up with growth rates well above their pre-transitional levels in recent years. Others, however, are still experiencing only sluggish growth if any. As Table 2.2.1 and Figure 2.2.4 indicate, only the countries from central Europe (CE) were able to recover rapidly and attain their pre-transitional growth rates by the late transitional period. To control for social, political, and institutional factors from the ICRG dataset the following variables are selected: law and order, government involvement in politics, democracy, corruption, and bureaucratic quality. At the beginning of the period the different countries of this transition sample had rather similar values for each of these factors6. This analysis suggests that the transitional countries belong to the middle-income group and have average level democracy and governance indicators compared to the rest of the world. 2.2.3. Regional Description Figures 2.2.1-2.2.3 show that from 1990 to 1995 all the regions except CE and EE experienced an explosion in inflation and lending rates and a commensurate decline in investment rates (investment to GDP ratio), meaning that the initial fall in investment was more than that of real GDP. Low savings due to falling incomes and 6 According to Aggregate Governance Indicators Dataset (AGGIND) by Kaufimann, Kraay and Zoido-Lobaton (1999b) the democracy and political stability indicators for these transition countries were between 25th and 50th percentiles relative to the rest o f the world in the initial stage of transition. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 22 high inflation rates7 , high debt default risks due to the absence of credit markets and enforcement legislation were the main causes for the high lending rates8 and low credit supply in the initial transitional period. Tight monetary policies, such as pegging domestic currencies to the US dollar and other major currencies, although successful in bringing down the inflation rates substantially, (even to negative rates in late transitional period for certain countries), also caused credit crunches, especially in those countries where the foreign investment wasn’t substantial. Those regions, which have started the transition earlier (CE, EE, Baltic), have experienced smoother transition with relatively lower peaks of lending and inflation rates. This is because they were able to quite rapidly develop and implement adequate institutional frameworks that ensured proper functioning of credit markets and received significant Western investment inflows9 that enhanced the supply of credit. After 1992 the CE countries have experienced a gradual increase in investment rates, especially as the prospect of their admission to 7 For example, in 1993 the inflation rates were 2.2*1014% in FR Yugoslavia, 1517% in Croatia, 4735% in Ukraine, 3731% in Armenia, 1662% in Kazakhstan, 1190% in Belarus, 410% in Lithuania, 256% in Romania. 8 The lending rates for selected countries in 1993 are 71.6% in Belarus, 33.6% in Estonia, 48.6% in Slovenia, 184.2% in Ukraine, 1443.6% in Croatia, 58.3% in Bulgaria, 28.4% in Hungary, 91.8% in Lithuania, etc. Source: IFS data. 9 Campos and Coricelli use data from the United Nations Economic Commission for Europe’s Economic Survey (2001, no. 1) to calculate cumulative FDI inflows during 1988-99 by region. The following by each region is received (billion $): Baltic 2.1, Balkan 2.07, Visegrad (includes CE, Slovakia and Slovenia) 14.41, Asia (includes TCA and central Asia using own terms) 1.73, and WCIS 8.17. In per capita terms these inflows are respectively ($/person): 923.67, 277.5, 1122.8, 183.00, and 91.00. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 23 the EU began to boost Western investment1 0 . The others, however, were still in a declining mode. Figure 2.1.1 depicts the distributional dynamics of earnings inequality, per capita GDP, investment rate, and total fertility rate, based on nonparametric kernel estimations1 1 results. Figure 2.1.1 for earnings inequality shows that as mentioned earlier, during the transition there was a sharp increase in inequality both within and across countries. The support of the density gradually moves to the right and substantially flattens over time while the support range increases. In numerical terms, from Table 2, the initial inequality ranged from 0.16 to 0.25, whereas in the late transitional phase it has ranged from 0.56 to 0.76. Figure 2.1.2 for real per capita GDP shows an opposite movement of the density while, once again, the support range increases, meaning a decrease in per capita GDP income in all countries and an increase in cross country income inequality. Only the last, 2000-year density shows a tendency of increase in per capita income for certain countries, as the right tail of the density extends further than those of the previous years’. Comparing Figures 2.1.1 and 2.1.2, it appears to be that there is a negative relationship between inequality and per capita income. 1 0 A comprehensive discussion o f growth impacts o f FDI in transition economies is provided in Campos and Kinoshita (2001). nEpanechnikovkernel function with quadratic weights is used, = j 0.75(1- z 2), - 1< z < 1; with the [ 0, otherwise optimal bandwidth from Silverman (1986), h* = 1.06min (a, 0.75IQR)N'1 /5 . IQR is the inter quartile range of the difference between 75th and 25t h percentiles. The estimated density is: f lr l rl Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 24 With a similar note Figure 2.1.3 shows an initial decline in investment rate, while the late period dynamics is somewhat more ambiguous. The 1989-year density support moves to the left and becomes flatter as 1992-year chart shows. A subsequent rebounding follows according to charts corresponding to 1996 and 2000. The latter clearly delineates two peaks with two separate investment groups of countries, where the left-peaked group has much lower investment rates than is the peak investment rate for 1989-year chart while the right -peaked group has much higher investment rates. Figure 2.2.4 depicts the distributional dynamics of total fertility rate. Again, as noted before, there is a fertility decline across the countries as the 1999 density support moves to the left. The year-2000 chart peaks at 1.23 births per woman, with an increased variation across the countries, however, as there is a slight density increase around the value of 2.5 births per woman. This, again, along with the declining per capita income supports the Malthusian theory1 2 , meaning that the income effect is dominating in childbearing decisions. From these regional charts and the results discussed above three separate phases of transition can be observed that the countries have undergone. The initial shock period between 1989 and 1992 was characterized by high and rising inflation and interest rates, and declining investment rates, per capita GDP levels, and real wages. This was followed by stabilization and leveling-off phase in GDP and interest rates 1 2 Micevska and Zak (2002) explain the fertility decline in the transition economies by the temporary fall in income below subsistence consumption level, while uncertainties increased substantially during the transition. They provide some empirical, cross-country evidence to support their arguments. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 25 between 1992 and 1996, when inflation and interest rates fell dramatically but not as far as their pre-transition period levels. After 1996, in the late transition phase (1996- 2002), inflation and interest rates declined further to their pre-transition levels, while per capita GDPs and investment rates for the first time rebounded from their lows. 2.3. The Model and the Estimation Method As mentioned in Barro (2000), an attractive feature of multicountry panels is that they capture large variations in government policies and other country specific effects over time. The transitional period of the transition economy sample was characterized by important institutional changes, such as the creation and development of private financial institutions and credit markets, privatization of industrial and agricultural sectors, decline in government expenditures, especially in subsidies to education, health and food programs. While as noted above the countries shared similar initial levels of per capita income, and similar values of social, welfare, growth, and inequality indicators, the early phases of transition brought about increased inequality in these indicators both within and across the countries. Therefore, cross-country panel estimation should provide a reasonably accurate assessment of my prior conjectures on inequality and growth relationship for countries in transition, after controlling for government policies, investment, education, fertility, degree of openness, etc. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 26 The summary statistics of the variables used in the estimations are provided in Table 2.3, for both the whole data (1988 to 2002) and for each of the three aforementioned sub-periods: 1988 to 1992, 1993 to 1997, and 1998 to 2002. This division into three sub-period groups is designed to avoid the serial correlation from business cycles, that would be apparent if annual data were utilized. To estimate the relationship between GDP growth rate and inequality the following equation from Banerjee and Duflo (2003) is used: (ytt+ a - yu)/a = ay it + Xufi + ygu +v; + sit, (2.3.1) where y it is the log RGDP in the beginning of the period with a years of length (a=5 in the analysis). The left hand side of the equation is the RGDP growth rate for a years. X u is a set of variables to control for possible sources of spurious correlations averaged over the period, and git is country i’s Gini coefficient in the beginning of • 1 • • period t . y it is included among the control variables to capture the convergence effects, v, is the country-specific, time-invariant component of the error term, and sit is the time-varying component of the error term. 1 3 In my data set the Gini coefficients for earnings inequality and the logs o f per capita GDPs are the starting year or the closest year to the starting year values for each sub-period, as some early observations are missing. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 27 Table 2.3. Summary statistics of variables used in estimations. Period 1988 - 2002 Variable Description Obs Mean Std. Dev. Min 1 Max Inequality Distribution of earnings: Gini coefficient 75 0.32 0.08 0.16 0.55 Income RGDP chain rule relative to the US (US=100) 72 23.53 12.44 3.83 50.87 PPPI Price level o f investment (ppp/exch. rate current prices) 72 40.36 20.8 0.72 83.93 Secondary General secondary education (enrollment, % Education relevant age) 78 26.47 8.14 7.85 43.8 Vocational Vocational/technical secondary education Education (enrollment, % relevant age) 78 36.98 18.94 5 71.4 Higher Higher education (enrollment, % relevant age) Education 78 21.47 9.66 4.15 54.03 Gov. Spend. Government share of RGDP (%, const prices) 72 26.51 9.49 7.02 48.63 Investment Investment share o f RGDP (%, const prices) 72 14.24 6.34 1.91 28.56 CBR Crude birth rate (number of births per 1000 pop.) 78 15.05 7.05 8 37.28 TFR Total fertility rate (births per woman) 78 1.97 0.82 1.1 4.83 First birth Average age o f mothers at first birth 75 23.18 0.94 21.3 26.17 Bureaucracy Bureaucracy quality (index) 60 2.34 1.74 0.55 9.24 Democracy Democratic accountability (index) 60 4.22 1.07 2 6 Law law and order (index) 58 4.23 0.94 1.52 5.73 Sub-period 1988 - 1992 Variable Description Obs Mean Std. Dev. Min 1 Max Inequality Distribution o f earnings: Gini coefficient 25 0.25 0.04 0.16 0.36 Income RGDP chain rule relative to the US (US=100) 24 28.11 13.3 8.53 50.87 PPPI Price level o f investment (ppp/exch.rate) 24 26.04 22.91 0.72 73.55 Secondajy General secondary education (enrollment, % Education relevant age) 26 25.35 9.51 7.85 38.75 Vocational Vocational/technical secondary education Education (enrollment, % relevant age) 26 41.57 15.62 16.5 67.8 Higher Higher education (enrollment, % relevant age) Education 26 17.91 5.84 8.63 32.95 Gov. Spend. Government share of RGDP (%, const prices) 24 25.83 10.62 7.02 48.63 Investment Investment share o f RGDP (%, const prices) 24 14.59 6.89 1.91 28.56 CBR Crude birth rate (number of births per 1000 pop.) 26 18.2 7.76 10.93 37.28 TFR Total fertility rate (births per woman) 26 2.37 0.86 1.44 4.83 First birth Average age o f mothers at first birth 25 22.93 0.77 21.3 24.35 Bureaucracy Bureaucracy quality (index) 20 2.26 1.78 0.55 9 Democracy Democratic accountability (index) 20 3.93 1 2 5 Law law and order (index) 18 4.23 1.01 1.52 5.66 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 28 Table 2.3. Summary statistics of variables (Continued from the previous page) Sub-period 1992 - 1997 Variable Description Obs Mean Std. Dev. Min Max Inequality Distribution o f earnings: Gini coefficient 25 0.33 0.07 0.24 0.46 Income RGDP chain rule relative to the US (US=100) 24 22.96 10.8 5.56 43.75 PPPI Price level o f investment (ppp/exch. rate current prices) 24 47.73 15.76 21.34 83.93 Secondary General secondary education (enrollment, % Education relevant age) 26 25.41 6.89 12.58 43.58 Vocational Vocational/technical secondary education Education (enrollment, % relevant age) 26 35.17 19.63 8.68 66.84 Higher Higher education (enrollment, % relevant age) Education 26 19.67 7.21 6.6 33.58 Gov. Spend. Government share of RGDP (%, const prices) 24 27.39 9.43 7.42 46.48 Investment Investment share of RGDP (%, const prices) 24 13.71 5.42 2.96 22.99 CBR Crude birth rate (number of births per 1000 pop.) 26 14.57 6.79 8.78 32.48 TFR Total fertility rate (births per woman) 26 1.9 0.77 1.28 4.15 First birth Average age of mothers at first birth 25 23.02 0.9 21.8 25.1 Bureaucracy Bureaucracy quality (index) 20 2.4 1.82 1 9.24 Democracy Democratic accountability (index) 20 4.23 0.97 2 5.32 Law law and order (index) 20 4.42 0.92 2.13 5.73 Sub-period 1997-2002 Variable Description Obs Mean Std. Dev. Min 1 Max Inequality Distribution o f earnings: Gini coefficient 25 0.37 0.08 0.21 0.55 Income RGDP chain rule relative to the US (US=100) 24 19.53 12.04 3.83 46.75 PPPI Price level o f investment (ppp/exch. Rate) 24 47.31 15.52 22.21 76.17 Secondary General secondary education (enrollment, % Education relevant age) 26 28.63 7.66 14.03 43.8 Vocational Vocational/technical secondary education Education (enrollment, % relevant age) 26 34.21 21.04 5 71.4 Higher Higher education (enrollment, % relevant age) Education 26 26.82 12.44 4.15 54.03 Gov. Spend. Government share o f RGDP (%, const prices) 24 26.32 8.66 9.4 41.98 Investment Investment share o f RGDP (%, const prices) 24 14.41 6.84 4.74 28.43 CBR Crude birth rate (number of births per 1000 pop.) 26 12.39 5.38 8 29.37 TFR Total fertility rate (births per woman) 26 1.64 0.65 1.1 3.85 First birth Average age o f mothers at first birth 25 23.6 1.02 21.7 26.17 Bureaucracy Bureaucracy quality (index) 20 2.36 1.7 1 8.45 Democracy Democratic accountability (index) 20 4.48 1.2 2.06 6 Law law and order (index) 20 4.04 0.89 2 5.08 Sources: TransMONEE 2002 Database (UNICEF), International Financial Statistics (IFS), Penn World Tables, and International Country Risk Guide (ICRG).___________________________________________ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 29 Numerous versions on estimating (2.3.1) exist in the literature. For example, Barro (2000) uses a similar convergence framework, derived from an extended version of the neoclassical growth model Dy = F(y, y ) , where y is the long-run target level of per capita output, and Dy is the growth rate of per capita output. Forbes (2000) uses a model similar to the model used in Perotti (1996), equivalent to (2.3.1), with the right hand side variables taken at t-1, instead of t. While using equations with similar forms, the choices of control variables and methods of estimations have varied considerably from one study to another. For example, Barro chooses random effects over fixed effects as they account for both within country across time, as well as cross-country inequality variations. He runs three stage least squares estimations, where the current values of the regressor are instrumented with its lagged values. Also, he includes an extensive list of control variables, including the investment share of GDP (lagged one period), government spending (lagged one period), fertility, and education as plausible channels through which inequality could affect growth. He criticizes the related literature on the choice of control variables and shows that the results are sensitive to having fertility in the regressions. Forbes, on the other hand, conducts fixed effects panel estimations. She assumes that the main source of variation comes from the within country, time effects, rather than from the cross-country variation. She adheres to Perotti’s specification for control variables, i.e. only male and female education, and the purchasing power parity of investment goods (a measure of distortion). In contrast to Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 30 Perotti, however, she uses one period lag of these variables rather than their contemporaneous ones. I follow Banerjee and Duflo (2003) in using alternative sets of control variables used in Perotti (1996), Forbes (2000), and Barro (2000). 2.4. Estimation Results and Interpretation 2.4.1. Basic Results First, Banerjee and Duflo (2003) estimation procedures are applied to the transition economy sample. First, Table 2.4 reports the results of (2.3.1) for Perotti fixed and random effects and for Forbes random panel specifications. The estimations show a significant, negative inequality-growth relationship. The Hausman specification tests do not reject the null hypothesis of possible correlation between country-specific effects and the other explanatory variables; suggesting the choice of random effects over fixed effects. Table 2.5 presents the results based on estimations techniques and specifications similar to those in Table 2.4, but with a term of Gini squared added. The results reported are for panel, random effects and maximum likelihood, random effects estimations. The latter method assumes normally distributed error terms. While the coefficient of Gini Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 31 Table 2.4. Fixed and Random Effects Panel Estimations (Corr. B&D Table 1) Estimation method Fixed Random Randc> m Specification Perotti X(t) Perotti X(t) Forbes X(t-a) Dependent variable: RGDP growth Coef. t Coef. z Coef. z Distribution o f earnings: Gini coeff. -93.34 -3.16 -135.33 -5.45 -110.08 -5.27 log RGDP (US=100) 4.71 0.86 1.95 0.48 -3.20 -1.01 Price level of investment -0.35 -3.66 -0.24 -2.81 0.29 3.78 Secondary education 0.50 1.33 0.04 0.12 -0.19 -0.74 Higher education enrollment 0.63 2.21 0.81 3.39 0.18 0.78 Constant 76.15 3.11 102.17 5.21 109.04 8.37 R-sq: within 0.68 0.64 0.57 between 0.07 0.34 0.58 overall 0.28 0.46 0.54 Number of obs 69 69 70 Hausman specification test chi2( 5)= 8.92 8.98 Prob>chi2 0.11 0.11 squared is insignificant for Perotti specification, it is significant and negative for Forbes specification in both estimations, implying a possibility of existing non- linearities in the effect of initial inequality on growth. It must be noted that in Banerjee and Duflo (2003) the estimations are carried out using panel, random effects methods. However, as mentioned in Forbes (2000), a problem with both fixed effects and random effects is that equation (2.3.1) contains a lagged endogenous term of log income among the explanatory variables. Indeed, since (GDP growth rate)t+i = (GDP index)t+i - (GDP index)t, (2.3.1) can be rearranged into the following form: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 32 yit+i = © yit + ygit + XitP + V i + sh, (2.4.1) where 0 = 1 + a. It is apparent that yjt and S it-i are correlated. Table 2.5. Random Effects and MLE Estimations (Corr. B&D Table 1) Estimation method Random effects Maximum likelihood Specification Perotti X(t) Forbes X(t-a) Perotti X(t) Forbes X(t-a) Dependent variable: RGDP growth Coef. z Coef. z Coef. z Coef. z Distribution o f earnings: Gini coeff. -134.60 -0.89 165.94 1.20 -143.82 -0.98 168.18 1.29 Distribution o f earnings: Gini coeff. 0.08 0.00 -397.22 -2.00 17.86 0.08 -398.59 -2.14 squared log RGDP (US=100) 2.03 0.47 -4.53 -1.43 2.38 0.54 -4.49 -1.50 Price level of investment -0.24 -2.82 0.26 3.59 -0.25 -2.94 0.26 3.81 Secondary education 0.04 0.13 -0.27 -1.08 0.07 0.22 -0.26 -1.13 Higher education enrollment 0.80 3.36 0.30 1.36 0.80 3.55 0.31 1.46 Constant 101.61 4.03 68.18 2.91 101.00 4.24 67.26 2.97 R-sq: within 0.64 0.62 between 0.34 0.58 overall 0.45 0.55 Number of obs 69 70 69 70 Hausman specification test chi2( 6)= 9 Prob>chi2 0.165 To correct for this endogeneity problem, dynamic panel estimation techniques are used. The results for panel, instrumental variable random effects are presented in Tables 2.6 to 2.8. Note, that regardless of model and specification choice, the coefficient of Gini is always negative and significant in those regressions wherever the quadratic term Gini is not included1 4 . This allows concluding that the impact of 1 4 While in Table 4, Forbes specification, the coefficient of gini is insignificant and for its squared term is negative and significant, in Tables 5 and 8, presenting the dynamic panel results, I don’t observe significance in quadratic gini coefficients, or joint significance o f gini and its squared term, analogous to the results reported in Banerjee and Duflo. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 33 inequality on growth for the dataset of countries in transition is negative and significant. These results are in sharp contrast with Forbes’ findings, although they support Barro’s empirical results. Tables 2.5, 2.7 and 2.10 present estimation results where both Table 2.6. Random effects panel IV estimation with only linear inequality explanatory term involved (Corr. B&D Table 1) Estimation method Panel, IV, Random effects Specification Perotti X(t) Perotti X(t) Forbes X(t-a) Dependent variable: RGDP growth Coef. z Coef. z Coef. z log RGDP (US=100) -14.53 -0.89 -11.64 -0.65 -5.45 -0.54 Distribution of earnings: Gini -190.88 -3.03 -165.30 -2.31 -133.29 -3.78 Price level o f investment at t -0.24 -2.47 -0.19 -1.90 Lagged price level o f investm. (at t- 0.16 1.63 0.29 3.49 A ) Secondary education -0.15 -0.38 -0.07 -0.21 -0.07 -0.30 Higher education 1.03 2.90 0.75 1.87 0.13 0.51 Constant 170.63 2.42 148.92 1.87 120.77 3.04 R-sq: within 0.54 0.60 0.54 between 0.24 0.34 0.50 overall 0.36 0.44 0.48 Number o f obs 67 67 67 Note to Table 6: Instrumented variable: log RGDP(t), instruments are: log RGDP(t-a), log RGDP(t-2a) Gini and squared Gini terms are included. Tables 2.7, 2.8, and 2.10 report also the estimations results with selected specifications where the changes in Gini and its squared term, (g(t)-g(t-a)) and (g(t)-g(t-a)) , are included. From Table 2.7 it can be Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 34 Table 2.7. Random effects panel IV estimation with both linear and quadratic inequality explanatory terms involved (Corr. B&D Table 3) Specification Perotti X(t) Forbes X(t-a) Perotti X(t) Forbes X(t-a) Dependent variable: RGDP growth Coef. z Coef. z Coef. Z Coef. z log RGDP (US=100) -13.55 -0.81 -10.14 -0.84 2.77 0.28 -2.38 -0.20 Distribution o f earnings: Gini coeff. 20.60 0.10 193.50 1.07 Distribution of earnings: Gini -308.78 -0.85 -497.75 -1.70 coeff. squared First difference in Gini g(t)-g(t-a) -68.70 -3.74 -56.46 -2.95 Squared first difference: (g(t)-g(t- a))2 Price level o f investment -0.23 -2.35 0.27 3.02 -202.44 -0.24 -1.64 -2.21 -155.32 -1.23 0.26 2.70 Secondary education -0.33 -0.67 -0.12 -0.48 0.14 0.39 -0.09 -0.31 Higher education enrollment 1.01 2.79 0.27 0.95 0.44 1.54 0.31 0.97 Constant 138.46 2.65 84.20 2.27 64.22 1.94 67.88 1.98 R-sq: within 0.54 0.57 0.58 0.48 between 0.31 0.41 0.02 0.22 overall 0.41 0.45 0.21 0.29 Number o f obs 67 67 66 66 Note to Table 7: Instrumented variable: log RGDP(t), instruments are: RGDP(t-2a) log RGDP(t-a), log seen that the coefficient of the linear change in Gini is negative and significant, while the quadratic change is insignificant for both Forbes and Perotti specifications. When the level of Gini coefficient is also included, as in Table 2.8, then the coefficients of both terms of change in Gini become insignificant, while the coefficient for Gini itself is negative and significant. These results do not change when Barro’s specification is applied, as Table 2.10 shows. My results are inconsistent with those Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 35 Table 2.8. Random effects panel IV estimation with the difference terms of the Gini coefficient included (Corr. B&D Table 3) Specification Perotti X(t) Forbes X(t-a) Dependent variable: RGDP growth Coef. z Coef. z log RGDP (US=100) -25.01 -1.09 -15.76 -0.89 Distribution o f earnings: Gini coeff. -275.04 -2.12 -181.79 -2.18 First difference in Gini g(t)-g(t-a) 47.00 0.91 23.67 0.60 Squared first difference: (g(t)-g(t-a))2 197.46 0.77 5.17 0.03 Price level o f investment -0.33 -2.47 0.29 3.06 Secondary education -0.19 -0.42 0.07 0.26 Higher education enrollment 1.35 2.32 0.24 0.79 Constant 225.49 2.10 161.69 2.18 R-sq: within 0.47 0.46 between 0.17 0.29 overall 0.28 0.33 Number o f obs 66 66 Note to Table 8: Instrumented variable: log RGDP(t), instruments are: log RGDP(t a), log RGDP (t-2a) found in Banerjee and Duflo. Their results show a significant, negative value for the coefficient of the quadratic change in inequality, while the coefficient of the linear change term is insignificant, regardless of whether or not the Gini coefficient is included. Since the nonlinear change term has insignificant coefficients regardless of the specification in the regressions, it can’t be supported Banerjee and Duflo’s conclusion that changes in inequality in both directions reduce growth1 5 . 1 5 In Banerjee and Duflo (2000) this result is essential to explain a political economy “hold-up” model, where there exists a certain output sharing rule between two classes, running as competing political parties. In each period the ruling party has a bargaining power in terms o f holding up the economy from its potential growth and re-negotiates with the opposition party about the possible amendment o f the output-sharing rule on its behalf. The redistribution is costly and hinders the potential growth rate. Since the ruling party can be either rich or poor, the model predicts that redistribution in either way reduces the output growth rate. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 36 Table 2.9. 3SLS estimation with selected educational and fertility variables in controls (Corr. B&D Table 1) Specification_________________ Barro X(t) Dependent variable: RGDP growth Coef. z Coef. z Coef. z Coef. z Distribution o f earnings: Gini -74.08 -2.47 -83.77 -2.66 -73.34 -2.40 -93.98 -3.55 coeff log RGDPC lagged (t-a) -100.89 -2.85 -108.00 -3.08 -98.83 -2.79 -55.52 -1.65 Square o f log RGDPC (t-a) 15.34 2.59 16.78 2.88 15.01 2.54 7.61 1.33 Lagged government share of -0.53 -2.28 -0.57 -2.44 -0.52 -2.23 -0.53 -2.60 RGDP Secondary education lag (t-a) -0.34 -0.93 -0.40 -1.10 Secondary education 0.12 0.48 -0.66 -1.88 Higher education, lag (t-a) 0.95 3.05 0.76 2.87 0.99 3.24 Higher education 1.12 3.37 Avg age o f mothers at first 1.41 0.62 birth Total fertility rate -4.03 -0.81 31.03 1.75 Crude birth rate -0.56 -1.01 -0.81 -1.57 -2.69 -1.35 Law and order 3.78 1.56 4.81 2.08 3.70 1.53 1.12 0.55 Democratic accountability -29.16 -2.54 -28.38 -2.46 -28.28 -2.47 -18.35 -1.55 Square of democratic 4.05 2.80 3.88 2.67 3.95 2.73 2.61 1.76 accountability Lagged investment share o f 0.50 1.44 0.54 1.53 0.49 1.39 0.62 1.89 RGDP Constant 290.45 5.00 291.66 4.99 285.82 4.94 169.45 1.89 "R-sq" 0.68 0.67 0.68 0J3 Number o f observations 51 51 51 51 Note to Table 9: Instrumented variable: log RGDP(t), instruments are: log RGDP(t a), log RGDP (t-2a)_________________________________________________________________________ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 37 Table 2.10. 3SLS estimation with the difference terms of Gini coefficient included (Corr. B&D Table 3) Specification_____________________________________________ Barro X(t) Dependent variable: RGDP growth Coef. z Coef. z Coef. z Distribution o f earnings: Gini coeff -170.91 -0.94 -66.28 -2.09 Distribution o f earnings: Gini coeff. squared First difference in Gini g(t)-g(t-a) 102.14 0.37 -71.12 -3.47 -38.59 -1.54 Squared first difference: (g(t)-g(t-a))2 -17.52 -0.15 23.29 0.20 log RGDPC lagged (t-a) -52.88 -1.66 -51.79 -1.60 -50.78 -1.63 Square o f log RGDP lagged (t-a) 7.39 1.37 7.42 1.37 7.30 1.40 Lagged government share of RGDP -0.46 -2.21 -0.70 -3.52 -0.55 -2.70 Secondary education -0.75 -2.06 -0.71 -1.95 -0.60 -1.69 Higher education 1.21 4.02 1.16 3.64 1.12 3.65 Crude birth rate 0.67 1.04 0.87 1.24 0.63 0.93 Law and order 1.15 0.52 2.05 0.97 1.45 0.71 Democratic accountability -18.07 -1.58 -6.99 -0.59 -13.14 -1.11 Square o f democratic accountability 2.57 1.77 1.38 0.91 2.03 1.36 Lagged investment share of RGDP 0.52 1.56 0.69 2.02 0.65 1.98 Constant 218.78 3.49 145.43 2.53 177.77 3.11 "R-sq" 0.71 0.71 0.74 Number o f observations 51 50 50 Note to Table 10: Instrumented variable: log RGDP(t), instruments are: log RGDP(t a), log RGDP (t-2a) The results show that there is only a strong, negative relationship between the Gini coefficient is included. Since the nonlinear change term has insignificant linear change term and growth, meaning that an increase in inequality reduces growth, but a reduction in inequality raises it. 2.4.2. Reduced Form Results Tables 2.11 and 2.12 report the “reduced form” results based on Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 38 (yu+a -y u )/a = ay it + X ufi + ygit.a +v, + sit, (2.4.2) where the growth rate is regressed on the level of inequality lagged one period, g(t- a). As in Banerjee and Duflo (2003) this reduced form is derived from two structural equations. In addition to the effect of inequality on growth, according to their “hold up” Table 2.11. Reduced form results, Random effects panel IV estimation (Corr. B&D Table 4) Specification Perotti X(t) Forbes X(t-a) Dependent variable: RGDP growth Coef. z Coef. z Coef. z log RGDP -23.62 -0.36 -15.35 -0.37 55.71 1.08 Distribution o f earnings: Gini -13.07 -0.21 -62.92 -0.27 54.84 0.89 coefficient lagged (t-a) Gini square lagged (t-a) 80.36 0.24 Price level of investment -0.60 -1.29 -0.53 -1.70 0.44 2.39 Secondary education 0.16 0.23 0.02 0.03 -0.39 -0.85 Higher education 0.37 0.89 0.50 1.24 -0.56 -0.51 Constant 162.74 0.66 142.27 0.85 -110.78 -0.73 R-sq: within 0.09 0.16 0.46 Between 0.24 0.26 0.18 Overall 0.06 0.04 0.19 Number o f observations 66 66 66 Note to Table 11: Instrumented: log RGDP(t), instruments are: price level of investment, secondary education, higher education, log RGDP relative to US(t-a), log RGDP relative to US(t- 2a) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 39 Table 2.12. Reduced form results, three stage least squares estimation (Corr. B&D Table 4) Specification___________________________Barro X(t)_______ Barro X(t) Barro X(t-a) Dependent variable: RGDP growth Coef. z Coef. z Coef. z Distribution o f earnings: Gini coeff 23.74 0.89 -104.59 -0.57 0.09 0 lagged (t-a) Gini square lagged (t-a) 194.38 0.71 log RGDP lag (t-a) -63.22 -1.76 -59.35 -1.64 -135.09 -2.98 Square log RGDP lag (t-a) 8.32 1.38 7.81 1.29 20.71 2.72 Lagged government share of RGDP -0.86 -3.87 -0.80 -3.38 -0.78 -2.74 Secondary education -1.23 -3.32 -1.22 -3.32 -0.60 -1.26 Higher education 1.59 4.80 1.60 4.87 1.33 3.78 Crude birth rate 1.89 2.68 1.90 2.71 -0.28 -0.44 Law and order 2.78 1.18 2.24 0.91 1.25 0.41 Democratic accountability 2.28 0.18 0.12 0.01 -25.96 -1.65 Square o f democratic accountability 0.19 0.12 0.50 0.30 3.72 1.87 Lagged investment share of RGDP 0.51 1.34 0.56 1.46 0.24 0.57 Constant 143.41 2.18 158.65 2.30 328.33 3.87 "R-sq" 0.64 0.65 0.58 Number o f observations 50 50 47 Note to Table 12: Instrumented: log RGDP relative to US and the price level of investment, instruments are the corresponding first and second order lags o f the instrumented variables. model, the changes in growth rates are related to distributional conflicts, or inequality changes. They set up a two-equation structural form to estimate the model. In the first equation the growth rate is regressed on the change in inequality, g(t+a)- g(t), while, since the change in inequality is not exogenous, the second equation estimates this change regressing it on the one period lag of the inequality coefficient and the other control variables. Combining these two equations gives the (2.4.2) reduced form. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 40 Banerjee and Duflo found: (1) that for their whole sample they don’t find any significant relationship between the lagged Gini and the growth rate, and (2) that, where inequality is not too high, inequality, lagged one period is negatively correlated with growth. For the transition economy sample, the coefficient of the lagged Gini is statistically insignificant for all specifications in both Tables 2.11 and 2.12. This result doesn’t change either when the lagged value of the squared Gini coefficient is included. The results are thus at odds with Baneijee and Duflo especially, since the transition countries had very low levels of inequality in the beginning of their transitions. Since, their results show the effect of contemporaneous inequality on growth is insignificant, while I find strong, negative contemporaneous relationship, at a very minimum my results show that the Banerjee and Duflo reduced form model may not be generally applicable and be misspecified at least for transition economies. Nor does my results support their wealth effect interpretation of the model, based on the lagged inequality effect, according to which the wealth effect might raise investment in human capital, affecting growth with a lag. Raising human capital through investment might not be a reasonable explanation for a lagged wealth effect in transition economies since, thanks to substantial educational subsidies, they had sufficiently high human capital levels and investments prior to transition. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 41 2.4.3. Results for Changes in Inequality Tables 2.13 and 2.14 present the estimation results of the first equation of the structural form namely, the relationship between the changes in inequality and past inequality. The results are consistent with those of Banerjee and Duflo, which is one of their main findings. In all specifications, the coefficient of inequality lagged one period is significant Table 2.13. Random effects panel IV estimation with the difference terms in inequality and lagged inequality as dependent variables (Corr. B&D Table 4) Specification Forbes X(t) Perotti X(t-a) Dependent variable: g(t)-g(t-a)* (g(t)-g(t-a))2 * * g(t)-g(t-a)* (g(t)-g(t-a))2 * * Coef. z Coef. z Coef. z Coef. z log RGDP -0.22 -3.25 0.04 2.04 -0.16 -1.98 0.02 1.85 Distr of earnings: Gini coeff lagged -1.36 ■ ■11.21 0.25 4.02 -1 .3 2 -8 .8 2 0.21 4.53 (t-a) Price level of investment 0.00 -1.49 0.00 -1.08 0.00 -0.05 0.00 2.11 Secondary education 0.00 2.30 0.00 0.48 0.00 -1.48 0.00 0.80 Higher education 0.00 0.17 0.00 -1.60 0.00 2.45 0.00 -3.03 Constant 1.06 5.67 -0.18 -2.58 0.91 2.93 -0.11 -2.40 R-sq: within 0.88 0.47 0.88 0.53 Between 0.13 0.09 0.06 0.15 Overall 0.40 0.25 0.53 0.40 Number o f observations 65 65 65 65 Note to Table 13: Instrumented: log RGDP relative to US and the Gini coefficient lagged one period, instruments are: log real GDP lagged one period and log real GDP lagged two periods. * The first difference in Gini coefficient ** The first difference in Gini coefficient squared_________________________________________ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 42 and is negative in those equations where the dependent variable is the change in Gini and is positive in those equations where the dependent term is the square of the change in Gini. In light of these results, it can be examined whether or not any Kuznets curve type growth-inequality relationship is observed in this group of countries. In Table 2.13 the coefficients of log GDP are negative and significant for linear change in inequality as dependent variable equations for Forbes and Perotti specifications, while in the quadratic change equations they are positive and Table 2.14. Three stage least squares estimation with the difference terms in inequality and lagged inequality as dependent variables (Corr. B&D Table 4) Specification Barro Dependent variable: (g(t)-g(t-a))* (g(t)-g(t-a))2 * * Coef. z Coef. z Distr o f earnings: Gini coefficient lagged (t-a) -0.84 -6.51 0.07 2.60 log RGDP lagged (t-a) 0.11 0.65 -0.02 -0.62 Square log RGDP lagged (t-a) -0.01 -0.32 0.00 0.21 Lagged government share o f RGDP 0.00 2.69 0.00 -0.70 Secondary education 0.01 3.31 0.00 -1.32 Higher education 0.00 -2.80 0.00 -0.30 Crude birth rate -0.01 -3.48 0.00 0.79 Law and order -0.01 -1.09 0.00 0.42 Democratic accountability -0.14 -2.29 0.03 1.90 Square o f democratic accountability 0.02 2.15 0.00 -1.88 Lagged investment share o f RGDP 0.00 0.77 0.00 0.44 Constant 0.33 1.04 0.01 0.13 "R-sq" 0.75 0.54 Number of observations 50 50 Note to Table 14: Instrumented: the Gini coefficient lagged one period, instruments are: log real GDP lagged one period and log real GDP lagged two periods. * The first difference in Gini coefficient ** The first difference in Gini coefficient squared__________________________________________ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 43 significant. From table 2.14, however, all of these coefficients are insignificant when Barro’s specification is used. The Kuznets curve: the hypothesis that income inequality first increases and then decreases with development, deals with the question of how the level of income affects income distribution. From the kernel density charts of the Gini coefficient in section 2 it can be noticed that the countries have experienced an increase in inequality over the whole period of the sample. This supports the Kuznets’ hypothesis and I would anticipate for inequality to decrease in the late transitional-developmental phases of these transition economies. For this reason it would be expected the coefficients of log per capita income to be negative in linear inequality change equations and to be positive in quadratic inequality change equations. In this regard, my results are consistent with Kuznets curve in Perotti and Forbes specifications, but not when Barro’s specification is used. 2.4.4. Results for Other Estimations Next to determine the robustness of these findings, other specifications than those of Perotti-Forbes and Barro that have been used above are further applied. This would seem especially important given Barro (2000) claim that the estimations results are sensitive to the choices of control variables. Therefore, the Barro’s specification of controls, which is more extensive than that used by other authors has been used. It includes such variables as primary, secondary, and higher schooling, a Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 44 rule of law index, a democracy index, the total fertility rate, and openness. Since preliminary estimates showed that the primary schooling variable had virtually no explanatory power, it has been dropped from the list of control variables. This is explained by the historically high and relatively constant levels of primary enrollments throughout the sample period, attributable to the high priority assigned to primary schooling in centrally planned economies and to the enforcement of mandatory attendance rules. But, on the other hand, two more fertility indicators are included, following Barro’s argument that the results are sensitive to the inclusion of fertility variables in the estimations. Tables 2.15 and 2.16 report the results using this specification. Table 2.15 presents the Anderson and Hsiao (1981) instrumental panel, first-differenced estimation results. According to the first difference method, the one period lag of (2.4.1) is subtracted from the contemporaneous equation to eliminate the v; country specific, fixed effects. This yields: yit+i-yu = 0 (yit - yu-i) + y(gu- git-0 + P(Xit -X it.i) + sit - eit.i, (2.4.3) Anderson and Hsiao use either yin, or (y^-i -yu-2) as instruments to estimate 0. The results using this estimation method still show strong negative coefficients for the inequality term, where the dependent variable is once again the real GDP growth rate1 6 . 1 6 I have also conducted Arrelano-Bond GMM linear dynamic panel estimations. Since the method requires at least three periods, my averaged sample with three periods is insufficient to generate results. Despite the obvious limitations, I instead used the yearly, as an alternative to the period averaged data. The results still show negative and significant coefficient for the Gini. As expected, with my yearly sample, however, the autocorrelation tests for the residuals show first-order serial Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 45 Table 2.15. Panel IV estimation using Anderson and Hsiao method Estimation 1 2 Dependent variable: RGDP growth rate Coef. z P>|z| Coef. z P>|z| Gini coefficient (earnings) -133.52 -2.86 0.00 • -140.12 -3.05 0.00 Log real GDP relative to US -28.29 -1.72 0.09 -25.56 -1.63 0.10 Government share o f RGDP -0.57 -1.79 0.07 -0.63 -2.08 0.04 Government share o f RGDP (-1) 0.20 0.53 0.60 0.28 0.76 0.45 Investment share o f RGDP 1.42 2.33 0.02 1.34 2.33 0.02 Investment share o f RGDP (-1) 0.54 0.92 0.36 0.54 0.94 0.35 Higher education 0.90 2.22 0.03 0.87 2.26 0.02 Higher education (-1) 0.49 0.92 0.36 Secondary education (-1) -0.43 -0.59 0.56 0.04 0.07 0.94 Vocational secondary (-1) 0.19 0.76 0.45 0.27 1.17 0.24 Avg. first birth yr. per woman 1.58 0.40 0.69 2.22 0.58 0.56 Crude birth rate -3.48 -0.97 0.33 -4.21 -1.20 0.23 Total fertility rate 47.10 1.37 0.17 51.16 1.52 0.13 Investment profile 2.16 1.14 0.26 2.03 1.12 0.26 Government involve. In politics -1.24 -0.76 0.45 -1.29 -0.81 0.42 Law and order 1.85 0.42 0.67 1.86 0.45 0.66 Constant 74.42 87.14 0.85 53.12 0.66 0.51 R-sq: within 0.40 0.40 Between 0.79 0.80 Overall 0.64 0.65 Number of obs. 50 50 Note to Table 15: Instrumented variable: RGDP growth rate lagged one period, instrument: Log real GDP relative to US lagged one period Estimation 1 in Table 2.16 presents the results obtained when the Baltagi (1995) error components two stage least squares (EC2SLS) estimation method is used, while estimation 2 displays the estimation results when the Baltagi-Chang (2000) consistent correlation, although there is no second-order serial correlation. Sargan’s test of overidentitying restrictions is not met either. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 46 Table 2.16. Panel estimation using Baltagi and Chang random effects estimation method Estimation 1 2 Dependent variable: RGDP growth rate Coef. z P>|z| Coef. z p>|z| Gini coefficient (earnings) -96.00 -3.14 0.00 -95.16 -3.15 0.00 Log real GDP relative to US -2.16 -0.41 0.68 -2.00 -0.40 0.69 Log real GDP relative to US (-1) -12.51 -2.59 0.01 -11.01 -2.21 0.03 Government share o f RGDP -0.15 -0.64 0.52 -0.09 -0.39 0.69 Investment share o f RGDP 1.06 1.99 0.05 1.24 2.42 0.02 Investment share o f RGDP (-1) 0.01 0.03 0.97 -0.14 -0.39 0.70 Openness 0.00 -0.09 0.93 -0.02 -0.30 0.76 Avg. first birth yr. per woman 2.53 0.74 0.46 1.69 0.60 0.55 Higher education 0.36 1.47 0.14 0.34 1.51 0.13 Higher education (-1) 0.29 1.21 0.23 0.30 1.20 0.23 Secondary education (-1) 0.18 0.52 0.60 0.20 0.61 0.54 Vocational secondary (-1) 0.36 2.20 0.03 0.38 2.35 0.02 Trade balance 0.00 0.32 0.75 0.00 0.17 0.87 Constant 47.75 0.67 0.51 59.63 1.01 0.31 R-sq: within 0.65 0.61 between 0.72 0.75 overall 0.73 0.73 Number o f obs. 63 63 Note to Table 16: Instrumented variables: Log RGDP and Investment to RGDP ratio. estimators of variance components method is instead used. The EC2SLS method accounts for the complications related to the use of relatively small panel datasets and missing observations. It also accounts for random error components structure of the disturbances. I choose to verify once again the estimations results using this method, because the panel has a relatively small size and it involves the problem of omitted data. The results, shown in Table 2.16, again confirm the strong, negative effect of the inequality term on the real GDP growth rate. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 47 2.4.5. Sub-period Estimations: Speculating About Future Sign Reversal? In a final note to this section three stage least squares (3SLS) and seemingly unrelated regressions (SUR) are conducted, when real GDP growth rate is regressed in a system of three equations corresponding to the initial, leveling off, and recovery sub-periods of transition. The interest here is the transitional dynamics of inequality- growth relationship and specifically, whether the coefficient of inequality is significant in each sub-period and furthermore, whether there is a sign change from one sub-period to another. I again apply (2.3.1) in each sub-period. The estimation results in Table 2.17 show that the Gini coefficient is negative and insignificant in early-transition period, negative and significant in mid-transition period, but, quite unexpectedly, it turns to be positive and significant in SUR results and insignificant in 3SLS results in late-transition period, when economic recovery is underway. This last, recovery period reversal in the sign of Gini coefficient implies that the counties with higher earnings inequality experience faster growth than those with lower inequality. These results are thus in disagreement with those of Banerjee and Duflo, showing an insignificant effect of contemporaneous inequality on growth. They also claim that the relationship between growth and lagged level of inequality is U shaped. Specifically, high lagged inequality causes lower growth, but once the Gini coefficient is higher than 0.4, its further increases bring about increase in Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 48 Table 2.17. Three stage least squares simultaneous estimation of GDP growth rates involving the three sub-sample periods of 1988-92,1993-97, and 1998-2001 Estimation method___________________________Seemingly unrelated reg Three stage least squares Dependent variable: RGDP growth rate Sub-period 88-92 Coef. Z P>|z| Coef. z P>|z| Gini coefficient (earnings) 89 -13.90 -0.42 0.68 -19.39 -0.53 0.60 Log real GDP relative to US -5.36 -2.15 0.03 -3.95 -0.77 0.44 Government share o f RGDP -0.04 -0.29 0.78 -0.03 -0.16 0.87 Investment share of RGDP 0.38 1.78 0.08 0.36 1.69 0.09 Constant 101.36 8.70 0.00 97.85 5.68 0.00 Sub-period 93-97 Coef. Z P>|z| Coef. z P>|z| Gini coefficient (earnings) 93 -133.63 -4.83 0.00 -122.62 -3.53 0.00 Real GDP growth rate 88-92 0.98 3.95 0.00 1.61 2.47 0.01 Log real GDP relative to US 1.93 0.58 0.56 6.42 0.85 0.40 Government share of RGDP -0.85 -2.72 0.01 -0.88 -2.44 0.02 Government share o f RGDP (-1) 0.54 2.09 0.04 0.55 2.02 0.04 Investment share o f RGDP 1.40 3.24 0.00 1.18 2.00 0.05 Investment share o f RGDP (-1) -0.50 -1.52 0.13 -0.66 -1.77 0.08 Constant 18.68 0.71 0.48 -46.46 -0.71 0.48 Sub-period 98-2001 Coef. Z P>|z| Coef. z P>|zj Gini coefficient (earnings) 98 64.66 2.29 0.02 84.72 1.27 0.21 Real GDP growth rate 93-97 1.25 11.09 0.00 1.31 4.29 0.00 Log real GDP relative to US 0.18 0.07 0.94 5.22 0.68 0.50 Government share of RGDP 0.75 2.18 0.03 0.67 1.47 0.14 Government share o f RGDP (-1) -0.40 -1.10 0.27 -0.24 -0.56 0.58 Investment share o f RGDP 1.32 3.49 0.00 1.17 2.36 0.02 Investment share o f RGDP (-1) -0.68 -1.72 0.09 -0.79 -1.48 0.14 Constant -52.28 -2.76 0.01 -76.67 -1.35 0.18 Obs R-sq Obs R-sq 23 0.30 23 0.29 23 0.87 23 0.83 23 0.96 23 0.95 Equation Real GDP growth rate 1992 Real GDP growth rate 1997 Real GDP growth rate 2002 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 49 growth rates. In particular, they replicate Barro’s structural estimate and use the same sample of countries to investigate the issue of non-linearity by breaking the sample into high (Gini above 0.4-0.45, which mostly includes countries from Latin America) and low inequality countries. Estimating separately these high and low inequality sub-samples, they find that high levels of inequality are linked to higher subsequent growth in Latin America, while they are associated with lower subsequent growth for the rest of the sample. As Figure 2.1.1 shows, the Gini earnings density support has moved to the right between 1989 and 2000. The 2000-year density is observably much flatter and stretched out compared with 1989-year density, showing a substantial inequality increase for both within and across countries. The summary statistics presented in Table 2.2 shows that the mean inequality has risen throughout the whole period, rising to an average of 0.37 and a maximum sample value of 0.55 for year 2002. The mean value is very close to 0.4, the reversal point of inequality-growth relationship reported in Banerjee and Duflo. Thus, the sign reversal in the results is consistent with the Banerjee and Duflo finding of a U-shaped relationship, but only for the contemporaneous effect of inequality on growth. This kind of sign reversal is widely reflected in the literature. Galor and Zeira (1993) for example, pointed out that non-convexities in the investment technology, such as the existence of a minimum scale for investment, have the effect of reversing this inequality-growth relationship over some ranges. Other articles on the same theme include Banerjee and Newman (1993), Aghion and Bolton, (1997), and Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 50 Bernhardt and Lloyd-Ellis (2000), again, explaining the cause of these non-linearities by market imperfections, financial constraints, occupational choice, distribution of entrepreneurial talent, etc. Such reasoning is very applicable to transitional economies. Furthermore, besides the above-mentioned ‘usual’ imperfections, it could be added other, transition-specific causes of non-linearities, such as incomplete markets, institutional rearrangements, etc. The transition economy data contains observations only for 15 years and the results reflect negative inequality effect on growth for the short to medium run. It’s still too early to know whether the sign reversal in inequality coefficient in late transition phase will be sustained in a longer period of time or not. Therefore, for the long run these results cannot be considered as definitive. Future analysis will be able to shed light on whether the negative inequality to growth relationship will continue to be observed in the long run, steady state, or a sign reversal is possible. 2.5. Concluding Remarks In this essay it has been reexamined various dimensions of the growth-inequality debate in the specific context of countries undergoing transition from centrally planned to market systems. While at the beginning of the transition these countries shared many similar characteristics and specifically, had low levels of income inequality, over time they diverged considerably. Hence, they provide a potentially rich experience for examining the relation between income inequality and growth. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Several different specifications and estimation methods that in the literature have been employed to arrive at very different conclusions on the relation between inequality and growth even with the same datasets are applied. In contrast to these conflicting results for samples, which include no, or at most three transition economies, my empirical findings for transition countries indicate a strong, negative contemporaneous growth-inequality relationship for all of these specifications and estimation methods in the short to medium run. Therefore, this result is most consistent with Barro (2000), which showed that for low and medium income countries this relationship is negative. Also, a negative relationship between the change in inequality and initial income, and a positive relationship between the squared change in inequality and initial income as in Banerjee and Duflo (2003) are found. Nevertheless, the results do not show a significant relationship between lagged inequality and growth as Banerjee and Duflo (2003) found. The policy implications of the results are that good government performance is indispensable for rapid recovery and sustainable growth and that government implements and safeguards the integrity of the financial markets, investment mechanisms, effectively privatizes the industrial and agricultural sectors, and protects property rights. The implementation speed of those policies was quite different across the transition countries, yielding substantial differences in growth and income inequality. The results suggest that the transition governments need to Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 52 carefully nurture and protect these institutions and develop essential mechanisms for competitive performance in a market environment. In other respects, these results support Kuznets’ inverted - U hypothesis and I would anticipate decreasing inequality over time as the initial transitional effects attenuate and the trend effects of economic growth become more apparent. It can be concluded that, in light of the present controversies in literature on the growth-inequality theme, these findings are consistent with the view that the empirical results are sensitive to the specific choice of sample of countries. In this study, however, the results are not sensitive to the specifications and estimation methods chosen. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 53 CHAPTER 2 BIBLIOGRAPHY [1] P. Aghion and P. Bolton (1997), “A Trickle-down Theory of Growth and Development With Debt Overhang,” Review of Economic Studies 64(2), 151-72. [2] A. Alesina and R. Perotti (1996), “Income Distribution, Political Instability, and Investment,” European Economic Review, 40,1203 - 1228. [3] T. Anderson and C. Hsiao (1981), “Estimation of Dynamic Models with Error Components,” Journal of the American Statistical Association, 76, 598- 606. [4] A. Atkinson and J. Micklewright (1992), “Economic Transformation in Eastern Europe and the Distribution of Income,” Cambridge: Cambridge University Press. [5] B. Baltagi (1995), “Econometric Analysis of Panel Data,” Chichester: Wiley. [6] B. Baltagi and Y. Chang (2000), “Simultaneous Equations with Incomplete Panels,” Econometric Theory, 16,269-279. [7] A. Banerjee and E. Duflo (2003), “Inequality and Growth: What Can the Data Say?,” NBER Working Paper No. 7793. [8] A. Banerjee and A. Newman (1993), “Occupational Choice and the Process of Development,” Journal of Political Economy 101(2), 274-298 [9] R. Barro (2000), “Inequality and Growth in a Panel of Countries,” Journal of Economic Growth, 5, 5 - 32. [10] D. Bernhardt and H. Lloyd-Ellis (2000), “Enterprise, Inequality, and Economic Development,” Review of Economic Studies, 67,147-168. [11] N. Campos and F. Coricelli (2002), “Growth in Transition: What we Know, What we Don’t, and What we Should,” Journal of Economic Literature, Vol. XL, 793-836. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 54 [12] N Campos and Y. Kinoshita (2002), “Foreign Direct Investment as Technology Transferred: Some Panel Evidence from Transition Economies,” The Manchester School, 70(3), 398-419. [13] K. Forbes (2000), “A Reassessment of the Relationship Between Inequality and Growth,” The American Economic Review, 90(4), 869 - 886.0. [14] Galor and J. Zeira (1993), “Income Distribution and Macroeconomics,” Review of Macroeconomic Studies 60, 35-52. [15] D. Kaufinann, A. Kraay, and P. Zoido-Lobaton (1999b), "Governance Matters". World Bank Policy Research Department Working Paper No. 2196. [16] S. Kuznets (1955), “Economic Growth and Income Inequality,” The American Economic Review, 45(1), 1-28. [17] H. Li and H. Zou (1998), “Income Inequality is not Harmful for Growth: Theory and Evidence,” Review of Development Economics 2(3), 318 - 334. [18] M. Micevska and P. Zak (2002), ‘What Accounts for the Emergence of Malthusian Fertility in Transition Economies?,” Mimeo. [19] R. Perotti (1996), “Growth, Income Distribution, and Democracy,” Journal of Economic Growth 1,149-187. [20] T. Persson and G. Tabellini (1994), “Is Inequality Harmful for Growth?,” The American Economic Review, 84(3), 600-621. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 55 3. Human Capital, Fertility, and Enterprise: Implications For Income Distribution and Growth This essay studies how the initial wealth and human capital inequality affects the fertility through the child quality-quantity tradeoff, the population growth, the investment in physical capital through the market credit allocation mechanism, the distribution of wealth, and the growth rate in an incomplete credit market environment. An overlapping generations model is developed including the household, production, and credit market sectors in which human capital, or entrepreneurial talent attained in the process of schooling, can be used as collateral for borrowing and be a constituent part in a firm’s production function. Agents choose to become either workers or entrepreneurs, based on their wealth and educational credentials. I receive quite different results compared to those, recorded in the relevant literature, namely from those discussed in Aghion and Bolton (1996) by simply extending their model to allow for an extra dimension in terms of human capital. Also, the partial equilibrium distributional dynamics of the model is examined using numerical simulations. The computational results support the view that a larger middle class consensus is better for economic development and growth. These results show that there is a role for policies towards subsidizing education and favoring high per-child human capital option against high fertility to ensure higher in investment and growth rate. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 56 3.1. Introduction to Essay Two There is a vast endogenous growth literature examining the interaction of fertility, demographic transition, and human capital accumulation with the distribution of wage income and its effects on growth1 7 . Little attention however, has been paid to the potentially important effects of human capital accumulation on entrepreneurial skills, investment, and income distribution. This essay analyzes the extent to which the initial distribution of wealth and human capital affects fertility through the child quality-quantity tradeoff, and the investment in physical capital through the market credit allocation mechanism, and thereby the distribution of wealth, and the output growth rate in an incomplete credit market environment. The effect of learned entrepreneurial skills through schooling is particularly important due to the empirical significance of financial constraints in firm start-ups. Potential entrepreneurs are generally unable to self-finance all of their financial needs for start up or expansion and hence, require external financing. Insufficiency of external finance can constitute an important constraint on the choice to become an entrepreneur and start a business. Wealthier investors are less dependent on the availability of external funding and more likely to make investments even without it. As a result, financial constraints create links between the initial wealth distribution 1 7 For example, Becker, Murphy, and Tamura (1990), Morand (1999), Galor and Weil (1996), etc. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 57 and the emergence of entrepreneurship. Skilled entrepreneurs may also need relatively less capital and spend less on monitoring and operating expenses to operate a firm and thus are more likely to obtain loans for an equivalent project than a less skilled entrepreneur would. Generally speaking, credit contracts reflect both the borrower’s (firm’s) and the lender’s characteristics, as well as the characteristics of the market in allocating the size of the loan and the terms and conditions of the contract to ensure its proper implementation and successful completion. Our analysis is especially applicable to less developed countries with imperfect and incompletely institutionalized credit markets and educational systems. This is because the dominant forms of enterprise in both rural and industrial areas of these countries are sole proprietorships and partnerships, whose investment activities are mainly financed by family, friends, informal moneylenders, and commercial bank loans. In those countries small agricultural firms and individual businesses account for a significant portion of employment and gross domestic product. The recent research has highlighted the importance of income gains in agriculture and small individual businesses as the source of growth for these labor-intensive sectors, and furthermore, resolving the massive employment problems and alleviating poverty in developing countries. Even for the industrialized countries, bank loans constitute 60 to 70 percent of the external financing of the businesses1 8 . Both moneylenders and banks usually make their lending decisions based on certain information obtained from borrowers 1 8 Colin Mayer, “Financial Systems, Corporate Finance, and Economic Development,” in Asymmetric Information, Corporate Finance, and Investment, (1990) ed. R. Glenn Hubbard. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 58 concerning their initial capital, collateral, educational, and entrepreneurial credentials, as well as family, dependents, and social relations to ensure borrower solvency. In order to focus on the entrepreneurial talent, or human capital effects1 9 the technology is deliberately excluded from the production function. A standard neoclassical model with both a household sector and a production sector is used, the latter involving a two-factor production function in terms of labor and human capital (or entrepreneurial talent), rather than physical capital20. This allows to disaggregate the conventional aggregate production function to the firm level and to relate heterogeneity in firm output to that in their entrepreneurial skills. In addition, a credit sector with market imperfections is considered, such that moral hazard can affect agents’ occupational choices between becoming investor-entrepreneurs and workers, thereby affecting the investment activity and hence aggregate output. Otherwise, with only a conventional aggregate production function, such effects would not be possible. In modeling the credit market environment an Aghion and Bolton (1997) framework is adopted, but with extending it to allow for labor participation in the production function (so that individuals make occupational choices between 1 9 Galor and Tsiddon (1997) discuss about the effects of technological innovation on income and human capital distribution, and economic growth. 2 0 Lucas (1998) shows that even a linear human capital accumulation function alone (without taking into account the capital mobility and the technological spillover effects) is sufficient for sustaining long-term physical capital and output growth. Furthermore, he also shows that such growth is an increasing function o f the amount invested in the human capital, differences o f which can explain the substantial differences in growth rates across countries. My results support this argument. But I also show that the distributional effects on the child quality-quantity tradeoff can be linked to the speed of human capital and wealth accumulation. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 59 becoming entrepreneurs and wage workers) and to allow for accumulated human capital (or entrepreneurial skills) to affect lending and borrowing. These extensions are made to answer the following questions: (1) Is there a role for human capital (entrepreneurial skills) in successful entrepreneurship and thus in investment and growth? And (2) does this extension bring about new, qualitatively different results? An overlapping generations model is used to formulate the child quality-quantity tradeoff as in Becker and Barro (1988) and to capture the nature of the distributional dynamics. Yet, again, rather than using an aggregate production function, as in most neoclassical models, my disaggregated production approach introduces heterogeneity and allows for capital market imperfections to affect individual investment decisions. Due to the non-convexities involved in the credit and production sectors, it is not feasible to track the dynamic behavior of the model or to identify the general equilibrium conditions ensuring ergodicity in the stationary state. Instead, computational experiments and simulations are conducted assuming partial equilibrium. Note that qualitatively the partial equilibrium approximations to wage and interest rate equilibrations, which clear the labor and credit markets, should not be affected by the results, because both the wage-entrepreneurial earnings ratio and the real interest rate usually remain within a narrow range over an extended period of time. Based on these simulations, the results I receive are as follows: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 60 (1) Given the initial human capital and wealth distributions, the lower is the wage worker-entrepreneur income inequality gap, the lower is fertility2 1 and slower is the population growth. Consequently, the higher is the wage-entrepreneurial income inequality, the higher is the fertility and population growth rate, given everything else remains constant. This result is standard and is well documented in the literature. (2) i) Wage effect: The lower the wage worker - entrepreneur income inequality gap, the fewer borrower-investors and the fewer are those rationed out of the credit market. The higher are wages, the higher is the opportunity cost of becoming an entrepreneur, ii) Human capital effect: Regardless of the initial distribution, with the increase in wage rate per unit of human capital, the threshold human capital to become an entrepreneur increases among the agents with the same wealth level. (3) There are relatively more credit rationed potential investors if the wealth and human capital are quite equally distributed. There are relatively less credit rationed potential investors if the wealth and human capital are unequally distributed. The larger is the middle class of agents the more are those willing to borrow and invest and the more are those credit rationed, although there are more investments made than in the former case. (4) If the initial wealth and human capital are rather unequally distributed, an increase in income inequality (due to a decrease in wage rates) will raise fertility and population growth. On the other hand, if the initial wealth and human capital are 2 1 Azariadis (1996) discusses about wages as an opportunity cost o f parenting. If wages are too low a demographic trap may result. Dahan and Tsiddon (1998) and Morand (1999) make similar conclusions. They also show that both fertility and income distribution follow an inverted 17-shaped dynamics along the path to demographic transition. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. more equally distributed, an increase in income inequality (decrease in wage rates), that is large enough for wages to fall below the subsistence income level, will reduce both fertility and population growth. This last effect is consistent with the Malthusian theory of a positive income to fertility relationship. (5) The higher is the correlation between human capital and wealth in the initial distribution (wherein the wealthy have a better chance to receive education than the poor), the lower is the number of agents subjected to credit rationing and the less significant is the credit rationing effect. This result is also valid for the dynamic behavior of the model. The results in (1) -(3) show that the larger is the middle class in the population the better are the economy’s prospects for development. These findings are in accord consistent with many contributions to the literature2 2 but for quite different reasons. The next subsection discusses some of the papers in the literature on the theme of entrepreneurial ability, investment, inequality, and growth. It also presents some empirical evidence on the effect of human capital and education on investment. 3.1.1. Brief Literature Review and Discussion As mentioned above, Aghion and Bolton (1996) consider growth and income inequality in the presence of imperfect capital markets. They show that, if the accumulation of capital is sufficiently rapid across generations, the economy 2 2 For example, Alesina and Perotti (1996), Persson and Tabellini (1994), Benhabib and Rustichini (1998), Easterly (2001). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 62 converges to a unique and invariant wealth distribution. For this to take a place, they also stress the importance of the equality in wealth distribution, because it causes greater equality of opportunity and it accelerates the trickle-down effect of capital accumulation from the rich to the poor. However, neither human capital, nor its accumulation is included in their analysis. Likewise, Lloyd-Ellis and Bernhardt (2000) emphasize the role of the development of financial institutions for efficient credit allocation in the economy. Assuming that agents are heterogeneous in their inherited wealth and entrepreneurial ability (human capital), both income distribution and economic growth in the long run depend on the initial distribution of agents’ inherent entrepreneurial skills. Once again, there is no role for education and no human capital dynamics is discussed. It must be noted that due to the discontinuities in the earnings distribution, generated by credit market imperfections, most of the research work is based on static models, or a Markov transformation process is adopted to track the wealth accumulation mechanism. For example, in both of the above papers, as well as in related on the theme papers by Banerjee and Newman (1993) and Piketty (1993), the wealth accumulation is governed by nonlinear Markov transition processes with multiple invariant distribution outcomes. Conditioned on a sufficiently rapid wealth accumulation scheme the Markov transformation process is monotone and consequently, according to the Monotone Mixing Condition in Hopenhayn and Prescott (1992), the distribution of wealth and income will converge to a unique invariant distribution. Seldom, however, does this condition appear to be met. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Lucas (1978) examines size-distribution of firms given the distribution of persons by managerial talent. The most talented become managers while others with talents less than a certain cutoff level become workers. The model elaborates this cutoff level to be dependent on wage rate and rental price of capital. Given the firm’s output is a function of entrepreneurial skill or talent, so called managerial technology, the model shows that the average firm size is an increasing function of per capita wealth and entrepreneurial talent. Our model considers similar relationship, but in an overlapping generations context and studies the effects of intergenerational human capital and wealth accumulation and the distributional dynamics. A recent empirical case based on micro-enterprise survey data in Mexico is studied in McKenzie and Woodruff (2002). The results show a positive relationship between the returns to capital investment and the level of the entrepreneurs’ education. Also, there is a positive relationship between entrepreneurial ability and the level of investment. Another recent empirical study on top shares of income and wages in the US over 1913 to 1998 by Piketti and Saez (2001) shows top entrepreneurs wage shares increase in recent decades compared with their capital income. This is another empirical evidence, emphasizing the increasing importance of the human capital in entrepreneurship. That the existence of a larger middle class, or a middle class consensus is essential as a critical determinant in development, which follows essentially from the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 64 computational results in (l)-(3), is broadly discussed in Easterly (2001). It is noted that a strong middle class consensus, a situation of high level of relative equality and ethnic homogeneity, ensures high levels of income and growth. Particularly, the middle class consensus secures a higher level of human capital, which accounts for higher levels of education and health in the society. On the other hand, societies lacking a middle class consensus see the economic, political, ethnic elite under invest in human and infrastructure capital because of their fear empowering the opposition. The rest of the essay is organized as follows. The following section provides the household utility maximization model. Section 3 focuses on capital market investment aspects and occupational choice. Section 4 provides the labor market, credit market, and general equilibrium conditions for the model. Section 5 examines the dynamics and the stationary state behaviors of the model. Section 6 reconsiders the model dynamics for a centralized economy. Some parameter calibration and simulation results are given in Section 7. Section 8 concludes. 3.2. The Household Sector 3.2.1. The Household Utility Maximization Problem An overlapping generations model is developed, in which agents live for two periods: childhood and adulthood. During childhood they receive education. The amount of education a child would attain is decided and financed by parents, as well Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 65 as parents provide for the other needs of children. As children enter adulthood they have human capital ht based on the amount of the education they have attained, receive inheritance from their parents bt, and decide on their consumption ct, the number of children to have nt, and the education et for each of them to receive, i.e. their human capital ht+i. Individuals in the economy are provided with equal amount of time endowments in each period, normalized to one, which they supply inelastically at no disutility cost for either employment, or entrepreneurial activities. They also can devote a certain portion of their time to raising their children. Figure 3.1 describes the timing of events. By the end of the adulthood agents leave bequests to their children and die. Based on their inherited wealth and attained human capital, agents make their occupational choices and investment plans. After resolving the occupational issues, they decide on the number of children to have and on the amount of household consumption. There are two types of assets to invest: 1) A fixed asset, or an investment firm, with face value k (it is assumed that all investment firms have the same physical size ) that requires entrepreneurial skills and involves a production activity yielding an idiosyncratic return rt which can be either higher than the market rate in case of a successful investment and zero in case of project failure, and 2) invest in a mutual 2 3 Here, an assumption is made that firms are homogeneous in their size to rule out the issues related with the scale effects from the physical size o f the firm and rather, to focus on the ‘pure entrepreneurial’ effect o f human capital on firm’s return. I implicitly assume therefore, that the dominant forms o f enterprise are small partnerships and sole proprietorships, where entrepreneurial skills apparently play more significant role than those in larger enterprises, where the risks to default are substantially less. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 66 | Childhood__________|________ Adulthood________ |_ 1. Bom 1. Receive inheritance 1. Leave bequests 2. Receive 2. Decide on: 2. Die education > Consumption > Children > Occupation Figure 3.1. The timing of the events fund yielding the sure market return Rt. The potential investors with lesser wealth than k need to borrow from the market at market rate Rt to be able to invest, while wealthier agents lend their excess wealth to the mutual fund at the ongoing market rate. The occupational choices are as follows: 1) if an agent is an investor then he becomes an entrepreneur and he is the manager of the firm. He invests his wealth in a risky project by making an initial down payment k . Investing in a firm is however, riskier as it may lead to losing the initial investment k in case of project failure, 2) If the representative agent isn’t able to invest then he becomes a worker for an entrepreneur and receives wages. Wages are paid at the end of the period after the revenues are realized. Agents are risk-neutral and prefer entrepreneurship to working if the expected return of an investment is higher than the lifetime wage income. Conditional on a firm’s production schedule and how much human capital workers and entrepreneurs have, the probability of successful outcome p of the investment project depends largely on how much workers and entrepreneurs are willing to spend Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 67 to install monitoring equipment that will enhance the reliability of the production process and on operating expenses to improve working conditions in terms of providing better lighting, workspace heating and ventilating, etc. These costs are increasing and convex in the probability of successful outcome, but are decreasing in 2 the entrepreneur’s and workers’ human capital, This cost setup ensures quite 2x, realistically, that the monitoring and overhead costs are increasing in the number of workers, as it becomes more difficult to manage them and more complex to monitor the production. Given this problem setup, households maximize their utilities with respect to their consumption ch number of children nt, child relative human capital xt+i=ht+i/ht+i, and child bequests bt+ i: MaxU, = lnct + y(lnxt+int + lnbt+int ) , (3.2.1) where ht (bt ) = j htdFt(ht,bt) is the average human capital for households 0 owning bt wealth2 4 at t, Ft is the joint distribution of human capital and wealth (or physical capital, assumed households inherit ownership rights on physical capital). There is perfect foresight such that xt+i is known to each parent, as shown later on, 2 4 Throughout this essay I use the term wealth for the wealth that a household has inherited at the beginning of adulthood. I separate the inherited wealth from the other income sources received by the end o f adulthood, namely the labor income and the returns from investing in the mutual fund and from the risky entrepreneurial project. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 68 ht+ i is known from equilibrium conditions. In (2.1) y> 0 captures parental altruism effect. An adult’s budget constraint with human capital ht and bequest bt depends on the occupation and the investment options chosen, as described above, and is given by: Wage worker: ct +etntwt + ntbt+ i = wt xt(l -(jm t ) + Rtbt„ (3.2.2a) Entrepreneur: ct +etntwt + ntbt+ i = E(kJ(1 -< frit), (3.2.2b) where wt is the wage per unit of human capital and nt is the net profit form entrepreneur’s investment. Parents spend etw, wealth on each child for them to receive e, teaching time of education. It is assumed that a teachers’ human capital equals to the population’s average human capital and hence xt = 1 for teachers, ensuring that the cost of education does not depend on parental wages. Additionally, raising each child takes a fixed proportion of parental time $ so that the time available for work, or entrepreneurship is l-(jmt. Accordingly, the smaller is nh the greater is the occupational time and the higher are the income earnings. From right hand side of (3.2.2a) and (3.2.2b) parents with higher human capital and thus higher income have higher opportunity cost to rising children and therefore, substitute child quality for child quantity by deciding to have fewer children with more education. From (3.2.2a), workers’ incomes consist of wages and returns on investing their wealth in the mutual fund. The last term in (3.2.2a) does not bear any marginal decisive role in utility maximization setup. From (3.2.2b), entrepreneurs’ incomes Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. consist of the returns of the entrepreneurship and are functions of time devoted to the latter. Since entrepreneurial returns are still uncertain in this period, it is the expected net returns that are used in the budget constraint. The relative human capital of children xt+ j depends on their parents relative human capital xh and education et: where Gt(xti,bt ) is the joint distribution of wealth and relative human capital. And the law of motion for the distribution of wealth and relative human capital is: e, = 0, they still have some positive human capital by learning from others. This allows generating a threshold human capital level, such that if the majority o f population initially falls short o f this threshold human capital a poverty trap might occur, while if most of the population has above threshold initial human capital then there is a rapid growth perspective with a subsequent decrease in inequality. I set by default the smallest human capital at zero and show however, that in contrast to their paper, the credit market imperfections alone can generate alternative growth paths, depending on households’ fertility-occupational choices. r 1- t %t+l ~ x l e t , (3.2.3) where z e (0, 1), child human capital is convex in education25. The population evolves over time according to: (3.2.4) 2 5 Croix and Doepke (2001) have a similar model setup, but introduce a parameter capturing the general level o f education and knowledge spillover effect. Even if children do not receive education, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 70 p,+ iG < + i (b',x') = P, J J «,7(6/+ 1 < b 'xt+ l < x')dG ,{xt,b, ) , (3.2.5) where the indicator function I(b'x) 0 otherwise is defined. In the next two subsections, abstracting from integer constraints associated with the family size and childbearing, and ignoring complications related to child mortality, twins, and the like, the (3.2.1) to (3.2.5) maximization problem is solved, giving the household optimal decision rules for each occupation. Additional inequality constraints, such as nt>0, xt>0, bt >0, or bt <0 (borrow from children) are observed. 3.2.2. The Child Quantity -Quality Tradeoff: a) The Worker's Case A worker’s utility maximization gives the optimal number of children, child education, optimal bequests, and consumption: (3.2.6) r (3.2.7) T n yr w (x t + R ,bt 1 + 2 y w ,< f> x, (3.2.8) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. c , = _ W'X, + R tb, 1 + 2 y 71 (3.2.9) The second order conditions for a maximum are satisfied, given the concavity of the utility function in choice variables. It follows from (3.2.6) - (3.2.9) that skilled workers invest more in per child education and bequests, than in their number. These results are intuitive and reflect well-documented facts . From (3.2.3) and (3.2.6)-(3.2.8) it follows, v bl+ l = ( 1 -r) *,+ l = (1 - r ) f (3.2.10) (3.2.11) Equation (3.2.10) provides the optimal rule for investing in a child’s human and physical capital. Equations (3.2.7), (3.2.8), and (3.2.11) give the dynamic equations for the intergenerational wealth, stock of descendants, and human capital. The asymptotic behavior of nh which gives the lowest possible fertility rate as xt approaches to infinity according to (3.2.8) is given by lim n. = Y — (3.2.12) ' (1 + 2y)< /> 26 For example, Galor and Weil (1996) receive similar results. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 72 b) The Entrepreneur’ s Case In a similar way an entrepreneur’s utility maximization problem yields: 0 _ (1 7 ^ ) p 2 13) ' WtT bt+ l = (3.2.14) n, = Y - (3.2.15) #(1 + 2 r) c, = x‘E (n > } (3.2.16) 1 + 2y Like workers, according to (3.2.13) - (3.2.16), entrepreneurs prefer investing more in per child education and bequest, rather than in their stock. The optimal bequest-human capital allocation rule is: */+ i = \ * J ( 1 - r )w r 1 + 1 ■ " a W tT X ,, (3.2.17) (3.2.18) Equations (3.2.14), (3.2.15), and (3.2.18) give the dynamic equations for the intergenerational wealth, stock of descendants, and human capital. It must be noted Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 73 that while there is no uncertainty related with equations (3.2.10) and (3.2.11) and both bt+ x and xl+ l can be expressed in terms of xt, it is not possible to have similar dynamics for the entrepreneur’s case as from (3.2.17) and (3.2.18) the dynamic equations aren’t deterministic. From (3.2.12) and (3.2.15) the model stipulates that entrepreneurs’ fertility behavior corresponds to the asymptotic case of workers. 3.3. The Occupational Choice and the Credit Allocation Mechanism 3.3.1. The Market Investment Structure and the Optimal Contract Agents decide on occupation based on their type (bt , Xf) and, assuming perfect foresight, on their information of the equilibrium unit human capital wages and 00 expected potential investment profits. Let b, = J6(c/G,(jc,,6,) be the average wealth 0 inherited in cohort bom at t. As described in section 2, the production technology — 7 7 — — requires a fixed initial capital outlay k . It is assumed at least initially k> bt , a part of potential investors need to borrow if they are willing to invest in high yield projects and become entrepreneurs. Since agents are risk-neutral, investing in risky projects is preferred to investing in the riskless mutual fund assets and earning 2 7 In Piketty (1996) a concave production function is used yielding a stochastic output, which depends on the level o f monitoring cost. For any given interest rate r, there is a profit maximizing capital input k(r). I explore three state variables, fertility, bequests, and human capital, to examine the fertility-human capital tradeoff and would like to maintain the simplest possible production function with respect to the physical capital. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 74 wages. A typical investor uses his wealth to purchase a firm paying k , hires labor at existing wages wt2 8, realizes profits by the end of the period, sells it, once again at k , to an entrepreneur from the younger generation, and makes a loan repayment for the amount borrowed from the market. The return on investment is given by: where p e (o,l) is the probability of successful outcome2 9 , I is the “quality” labor force employed, i.e. the number of workers hired multiplied by their human capital. scale with respect to labor selection (the k firm size is still fixed). Although the probability of success is increasing in monitoring and operating expenses, these costs Given the G(xt,bt) distribution, the investment market classifies the agents according to their individual (xt , bt ) human capital and wealth. Specifically, agents with the same xt are classified according to their initial wealth as follows: (1) The very wealthy investors with bt > k who are able to invest in their projects as well as 281 would like to make a note here that wages are paid at the end of the period, while the initial outlay k is raised from the inherited wealth plus any additional funds borrowed to cover the shortfall. 2 9 The likelihood o f success, p, is increasing in monitoring and operating expenditures, both being costly in terms o f “resources”, for example these expenses may include purchasing and installing monitoring equipment, improving work environment, etc. The resource cost is increasing and convex (quadratic) in the likelihood o f success the entrepreneur has to choose. with probability p with probability 1 - p (3.3.1) Since the firms are small-to-medium sized, a e (0,l) to ensure decreasing returns to are increasing and convex, namely, c(p) =— • Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. in other investors’ projects by lending their excess wealth to the credit market (since the initial capital outlay requirement project is fixed at k , additional wealth investment in own project doesn’t produce any extra yield); (2) the middle class borrowers with bt (xt)< k , who are willing to invest in own high yield projects, but need to borrow k - bt(x,). The potential lenders either obtain credit and invest, or are credit rationed and work unwillingly; (3) the poor investors, who are better off by lending their wealth to the market at rate Rt and working. Figure 3.2 depicts this classification of agents based on their (xh bt) characteristics. This diagram is quite different from the one provided in Aghion and Bolton (1996), when agents are classified solely on the basis of their initial wealth. For instance, as the figure xt xt Relative human capital Borrower V entrepreneur Worker ^ ,> x and f Credit rationed worker lender choice -'4- Worker and lendei by choice bt ( x ,) < L > fi 2 a. i § Wealth bt Figure 3.2. The classification of agents Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. shows, those potential investors whose wealth falls within the range between some as workers by self choice, workers due to credit rationing, or borrower entrepreneurs, depending on how much human capital they have, whereas according to Aghion and Bolton (1996) they are classified as credit rationed workers only. An optimal credit contract specifies a repayment schedule between borrowers and lenders based on the condition that the repayment cannot exceed a borrower’s end of period wealth: where p is the unit loan repayment rate arranged by the mutual fund for each individual borrower and based on the borrower’s credentials. Given the optimal contract, a middle-class borrower who contemplates becoming an entrepreneur maximizes his net expected return on investment by choosing the optimal labor demand and the operating costs30: cost function and allowing for labor force participation. The production function has essentially a Cobb-Duglas form, where instead of “traditional” production factors, capital and labor, I use entrepreneurial monitoring and the quality labor. b,(xl) and bt(x,), where it is assumed bt(xt)<b,(xt), then they could be classified (k - b , ) p , if the outcome is successful 0, if failure (3.3.2) maMikpXh < £ (* )> (3.3.3) s.t. E(jtt)- ( w > t xt + Rtbt ) > 0, 3 0 I extend the production function in Aghion and Bolton (1996) by adding the human capital in the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 77 /, >0, wt>0, 0<pt<l. The first order conditions yield the firm’s profit maximizing labor demand as a function of the manager’s individual human capital and wealth: I, = = 1 - a w, , £ \ l~a — k p ,= \ — kp, a ) (3.3.4) If bt<£then the second best level of monitoring expenditures is: P s b = x t (k£, - p ( k - b ty)n . (3.3.5) Similarly, for wealthy investors who are able to invest without borrowing, bt > k, the maximization problem yields the first best level p F B : P Fb =xtkst 32 where e t = s{w,) = a r \ - c t N 1 -a d st dw. < 0 . (3.3.6) (3.3.7) From (3.5), p x>0 and pb>0, capturing the existing moral hazard with respect to human capital and wealth and for this reason p F B > p SB. Obviously, investors with lower x t and bt combination exhibit higher risk of project default and hence would spend less on monitoring. 311 restrict the parameters so that p is between 0 and 1, as it is the probability of successful outcome. 3 2 Again, 0<p<l must be satisfied. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 78 Lemma 3.1. 1) Assuming perfectly competitive labor, production, and credit markets, the firm’s size is a function of the borrower investor’s human capital and wealth. Thus, the ratio of the initial capital investment over the labor employed may not be fixed33. From (3.3.4), (3.3.6), and (3.3.7) while, given xt, this ratio is indeed constant for non-borrower investors, from (3.3.4), (3.3.5), and (3.3.7) this property no longer holds for the borrower-investors due to the moral hazard effect. 2) The labor demand is increasing in the amount of the initial capital requirement k meaning that firms are larger in capital-intensive industries. 3) The firm size is inversely related with the unit loan repayment rate p. Consequently, it is increasing in the supply of loanable funds. The proof directly follows from (3.3.4) -(3.3.7). The expected profit maximizing investment revenues net of labor costs are therefore given by: E(yl) = (\/oc)kple(wt) £(H, ) = (£/>,)“/,'““ ~w,l, = (1 / a)kPls(w,) - ( ( ! - « ) / a)kp,s(w,) = kpt£(wt) = ccE(yt) The expected return on investment: (1- « ) £ ( * ) + * 3 3 This result is in contrast with the widely known stylized fact, recorded for example in Lucas (1978), that the capital labor ratio remains constant over time. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 79 The credit market sets the mutual fund rate of return Rt balancing the aggregate supply and the aggregate demand of funds. It is assumed that the loan market is perfectly competitive, so that the lenders’ expected return on funds lent equals to the market rate of return: Rt -p(bt, Xf)p. (3.3.8) Using (3.4) and (3.5), the unit repayment rate is assigned individually: 1 - 4Ri-( £ ~ b‘l . (3.3.9) x,(kst)2 J It follows that p> x< 0 and pt<0, implying that the unit repayment rate is lower, the higher is the human capital and the higher is the initial wealth. These once again emphasize the moral hazard that borrowers do not carry any responsibility for debt repayment in case of project default. 3.3.2. Credit Rationing and the Choice of Lending or Borrowing ks, p = —r = ------ 2 {k-b t) From (3.3.9) it follows there is no real unit repayment rate if the expression under the square root is negative. This happens if 0<b, = (3.3.10) 4 K, Investors with xt human capital whose wealth is less than that specified by (3.3.9) and wishing to borrow for making investments are credit rationed. (3.3.7) and (3.3.9) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 80 imply that the expected value of their unit repayment rate p(xt ) falls short of the risk free rate of return Rt. From (3.3.10), it follows bR > 0, bw > 0, bx < 0. Lemma 3.2. Type {xt,bt) investor’s credit rationing threshold wealth level is increasing in Rt and wt. Moreover, the unit loan repayment rate is increasing in R, and decreasing in wt. Higher wages raise production costs and the default risk. Also, the credit rationing wealth level is a decreasing function in the human capital. Using (3.3.8) in (3.3.4) and rearranging the following expression is received: xtAk£, P,(x„b,)= „ if b, <bt < k (3.3.11) xt(Aket) and p t(x,,b,) = x,Aks,, if b, > k Here the manager’s monitoring cost is expressed as a function of his wealth and skills, where p R < 0 , p w < 0 , p x > 0, and p b >0. The higher is R, the less borrowing is desirable, such that at some level b,(xt) investors feel indifferent between borrowing, or lending at the market rate. In order to provide the incentive constrain the following assumptions are made: a) The workers as well as the entrepreneurs share the monitoring or the 2 operating costs in (3.3.3) and hence, each of them pays an amount of i — . This 2x, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. option is motivated with the reasoning that the workers are concerned about the successful outcome of the project. b) Only the entrepreneurs are responsible for paying the operating, or monitoring costs — . 2xt 3.3.2.1.Both Entrepreneurs and Workers Share the Operating, Monitoring Costs If assumption a) holds, then the incentive constraint for the potential entrepreneur will be expressed as: C kp,)al)~a - w tl, - p tp { k - b , ) - ^ - = R,bl + w , x , (3.3.12.1) 2xt 2xt or aE(y,) - Rt (k - bt) = Rtbt + wtx, The left hand side is the expected net return on risky investment, while the right hand side is the return from choosing the alternative to work and invest in the mutual fund. Solving for the threshold costs of monitoring level from (3.3.12.1) gives: * * ’ ' * ' • (3.3.13.1) KE( where p R > 0 , p w > 0 , and p x > 0. Even if bt is greater than bt (x,), implying that the investor isn’t credit rationed, he might choose not to invest if pt >pt and instead become a worker. Figure 3.3 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. graphs p,(xt), and p,(b„xt) as functions of bt . pt(xt) is a horizontal line, as it doesn’t depend on b, . p t, is an increasing function of individual wealth due to the moral hazard effect. p (x ,b ) A Monitoring costs x A k e P xAks 2 Wealth k b Figure 3.3. Monitoring costs Lemma 3.1 Individuals’ choice to become entrepreneur or work doesn’t depend on their wealth. The threshold monitoring cost level is a function of individual human capital and aggregate prices, wage and interest rates. Proof. The proof follows from (3.3.13.1), as only those individuals withxt > 0 are willing to exert real and positive managerial monitoring cost. From (3.3.12.1) and (3.3.13.1) the indifference wealth level is determined to be or (3.3.14.1) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 83 (ke,(w)) R, w. (k£,(w)y - i + ■ x,s b,(xt) is convex in x, and reaches its minimum when x,=x,= R,k (wt(w , -(kS'iw))2) 1/2 A A b, (x,) is monotone increasing in w,. Also, given xh b, (x,) is convex in R, and reaches its minimum when Rt =R, =xtet{w ) f r wl/2 w -1 The results depicted in this figure classify agents according to their human capital and wealth in a very similar way that was assigned apriori in Figure 3.2. Credit Rationing. Given the investor’s type xh there is credit rationing if &,(*/) e ( & , ( * „ 6,(x,,.K,,w,))and b{xt,Rt,wt) > bt{xt,Rt,wt) . Proposition 3.1.1 With the decrease in wages the proportion of investors who borrow decreases, reducing the number of credit rationed. If wages fall below w, < ^ e1 then there is no credit rationing. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 84 Lender entrepreneur k Borrower Credit rationed Work b i Subsidized non-worker Work Relative >> human capital Figure 3.4 Classification of agents, case 1 Proof. The proof follows from (3.3.10) and (3.3.14.1) as in this case b(xt,Rt,wt) and bt(xt ,Rt,wt) do not intercept. 3.3.2.2.0nly Entrepreneurs are Responsible for Paying the Operating, Monitoring Costs If assumption b) is applied then the threshold wealth level bt (xt), when the potential investor feels indifferent between borrowing, or lending at the market rate will be given by Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The left hand side is the expected net return on risky investment, while the right hand side is the return from choosing the alternative to work and invest in the mutual fund. Solving (3.3.12.2) for the break-even level of the monitoring cost is received: will consider becoming entrepreneurs. This result is similar to the result presented in Lemma 3.1. Proof. The proof follows from (3.3.13.2), as only those individuals withxt > 0 are willing to exert real and positive managerial monitoring cost. Equation (3.3.13.1) still remains applicable to this case, while Figure 3.3 graphs pt(xt), and ptQ > t,xt) which are derived from (3.3.12.2) and (3.3.13.1) respectively. (3.3.13.2) where p R >0, p w > 0, and p x < 0. Where x, = is denoted the threshold level of human capital above which p is real and positive. Lemma 3.2 Only those individuals with relative human capital J c t > 0 or greater Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 86 Even if b, is greater than bt (x,), implying that the investor is able to borrow, he might decide not to invest if pt >pt and instead become a worker. From (3.3.12.2) and (3.3.13.2) his break-even wealth level is determined to be ( i M = ii + M 'P . - M f e Q O ) _ P,Ake(v , x , R,x, S, This expression is equivalent to the one received in (3.14.1) for case (a). Proposition 3.1.2 There is credit rationing in equilibrium for type (xb bj investors, whenever k(Aet)2(w,)xt < | 4 ' k Proof. The proof is provided in Appendix 3.1. Lemma 3.4. Suppose the market equilibrium interest rate is within the credit rationing range given by (3.3.15.2) for the set of (xt, bt), b, < k type households. The credit rationing range for Rt is increasing in xt, given bt, and decreasing in wt. 3 (Aks,(w,)) 8 w, (3.15.2) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 87 Proof: The proof follows from (3.3.15.2), assuming there exists a positive - k(As Y x credit rationing range for (x6 bj, b, <k type households. Both —— ‘ A— l and ( 3 (Ate,)2 ^ -w, V 8 x. are increasing in xt and decreasing in wt, since sw < 0. Corollary. With the increase in the market rate R, the range of human capital increases for those who have the same wealth level and are credit rationed. The higher is the market interest rate the higher is the credit rationing range with respect to human capital. Following Proposition 3.3.1 and Lemma 3.3.2, Figure 3.3.5 diagrammatically depicts the breakdown by occupational choice according to individual (bt, xt) characteristics. The individuals with less than x, human capital find themselves better in lending their wealth and working, even if their wealth is more than k , because from (3.3.10), expected returns from investing fall short of the working and lending alternative. This self-choice wage-working alternative extends to the range of (x, ,x,) for the individuals having wealth levels below b,(xt) . Given the equilibrium interest rate Rh from (3.3.15), the individuals with relative human capital falling within the range Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 88 r— <x, < _ — ■ . and with less than b,(xt) wealth are subjected to credit 3(Aksly -S w l k(A£,y 8 R k rationing and have to choose to become workers, where x. = ---- =— -------- 3{Aks,y -8w, Wealth bt . k bt =bt Lender entrepreneur 5 l v \ &,(*,) Borrower entrepreneur Credit fationed Worker x. xt xt xt Figure 3.5. Classification of agents, case 2 Relative ^ human capital and x, = = respectively. In contrast, individuals with more than x, human k{As,y capital do not experience any difficulty with borrowing, regardless of their initial wealth. The potential entrepreneurs falling in the segment bordered with the lines of bt(xt) and 6,(x,)ffom below and with b, =k from above are the borrowers, while those falling in the segment bordered by the line b, =k from below and x, = x, from the left are the wealthy, lender entrepreneurs. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3.3.3. Occupational Choice 89 The further analysis in this essay is conducted for case (b) only, when only the entrepreneurs have the sole responsibility for the success in their investment. In this case, individuals occupational choice on whether to invest in a firm and become an entrepreneur or to invest in the mutual fund and become wage worker is defined from equations (3.3.10) and (3.3.14.2), subsection (3.3.2.2). Individuals with inherited wealth bt(x,) > max(&,(xt,R,,wt);bt(xt,R,,wt)) choose to become entrepreneurs, while others become workers. Given b,(xt) > b t > 0, credit rationing holds for some potential investors, from (3.3.12.2) and (3.3.14.2) the following partial derivatives are unambiguously 7*' ~ a , a , defined: bw > 0, b ^ > 0 , bR > 0, and bx < 0. The following lemma follows immediately from these results. Lemma 3.5. A , a) From bw > 0 it follows that with the increase in wages less individual adults choose to invest and more choose to become wage workers. b) The higher is Rt, the less type xt investors would be willing to invest, as bR > 0 . A, c) bx < 0 implies that even though agents might have the same bt amount of wealth inherited, they might choose to be wage workers or become managers. The Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. higher is the individual human capital, given the same wealth, the lower is the A threshold wealth bt to becoming an investor. 3.4. Static Equilibrium Conditions 3.4.1. Labor Market Equilibrium Equation (3.2.4) gives the population evolvement. If Et is denoted as the total number of adult firm owners at time t and Lt the aggregate number of workers, then As at any time t Pt is the total of adult population, it follows Pt= E,+ L,. The labor demand a a firm’s level, l(wbRt,xt,b(), is given by the equations (3.3.4)- (3.3.7). Partially deriving lt with respect to its arguments yields lR < 0 , lw < 0 , lx > 0, and lb > 0. It is assumed that schools are regular businesses owned and run by their investor- managers and employ their teacher wage workers, who by assumption made in Section 2 have the population average human capital, such that the teachers’ relative human capital is xt=l. Using Figure 3.4 and denoting g,(xt,bt) the distribution function of Gt(x,,bt) the aggregate labor supply at any t is given as (3.4.1) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 91 Ls ,{wt) = Pt x b(x) x & (x) j j g ,( x l,bl)dbtdx,+ J jg'ix^b^db'dx, + J Jg,(x,,6,)c$,rft, 0 0 X 0 X 0 (3.4.2) The aggregate labor demand is then derived to be: CO X oO £ ? ( w , ) = J \l(wt,x„bl)gi(xi,bt)dbtdxt + ^ jl(wl,xl,bl)g,(x,,bl)dbldxl Xb(x) x b(x) 0 0 00 J Jl(wt,x,,bl)gl(x,,bt)dbtdxl = 1 - a a I « co f PsBg,(xt,btyibt + jp F B g,(x,,b,)db, b(x) dx, + (3.4.3) ] K 0 0 \pS B gXxt,bt)dbt + fPng'fa^ydb, *« k 00 *« + J ft 00 dx, Using (3.4.2) and (3.4.3), wages wt~w(Ls,LD ) are endogenously derived from the labor market clearing condition L](w,,R,) = LD t (w,,R,). (3.4.4) 3.4.2. Capital Market Equilibrium Following Figure 4 the aggregate supply of loanable funds is derived to be: Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 92 K?(Rt)= jjb'g'ix^b'yib'dx, + JJ(Z), -k)g,(x„bt)dbtdxt + 0 0 x k x b(x) j \b,g(xt,bt)dbtdxt + J Jbtg{xt,bt)db,dx (3.4.5) x 0 The aggregate demand of funds is accordingly given by The market rate of return Rt=R(Ks,KD ) is endogenously determined from the 3.4.3. Competitive Equilibrium Definition 1. Given the initial distribution F o ( h o ,b o ) , and the initial population size P o , a competitive equilibrium consists of a sequence of prices {wt, Rt}, aggregate quantities {ht, bt, Pt+i), distributions Ft+i(ht+ i,bt+i), and decision rules {ct, bt+ i, rt, nt, et, ht+i} such that: 1. Households make decisions on the optimal quantities of ct, bt+ i, nt, et, ht+i, subject to the constraints (3.2.2) and (3.2.4) to maximize their utility. 2. Individuals maximize their income earned during adulthood by selecting from either one of the two occupational options: a firm investor-manager or a credit market equilibrium condition K? (w, ,R,) = K f (w, ,R,). (3.4.7) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 93 worker, according to the conditions (3.3.10) and (3.3.11), subject to their school- learned skills and wealth. 3. Entrepreneur-managers choose from (3.3.4)-(3.3.7) the amount of the managerial monitoring cost to exercise and the amount of labor to hire in order to maximize their expected net investment return, subject to their financial constraints and credit allocation contract conditions (3.3.9) imposed by the market. 4. Wages per unit of human capital wt clear the labor market according to the labor market clearing condition (3.4.4). 5. The market rate Rt clears the capital market according to the credit market clearing condition (3.4.7) 6. The wealth and human capital evolves according to (3.2.7) and (3.2.11) for a wage worker and (3.2.14) and (3.2.18) for an firm-owner-entrepreneur. 7. The aggregate variable hi is derived in each period according to ht(b^) = 00 | htdFt (h, ,bt), while Pt is given by (3.2.4). There is a perfect foresight, as there are 0 no uncertainties involved at the aggregate level. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3.5. The Dynamic Behavior and the Stationary State 94 3.5.1. Individual Dynamics (this subsection is related to case (a), 3.3.2.1, when both entrepreneurs and workers share the operating and monitoring costs) A Given (3.3.10) and (3.3.14.1) define explicitly bt(x() and bt(x,) break-even and credit rationing wealth levels for type xt investors, when the assumption in section 3.3.2 (a) is in effect, that is, the entrepreneur shares the operating and monitoring costs with the workers then it is possible to track the individual dynamics. (3.2.7) and (3.2.10) essentially present the individual wealth dynamics for a worker, while for an entrepreneur it is given by the equations (3.2.14) and (3.2.17). Using these wealth transition equations and figure 3.4, figures 3.6 and 3.7 diagrammatically represent the one period individual dynamics. Case 1) Assume agents have equal wealth, but they are different in their human capital. The dynamics for the agents with bx =b2 =b and x, < x2 is graphically depicted. From (3.2.7) a type (b,,xt) worker’s next generation offspring receives w d x x bl+ l = ‘ wealth. Figure 3.6 illustrates this dynamics with two horizontal lines T corresponding to levels and — . Similarly, from (2.14) borrower T T investors with the same wealth Leave W ‘^ P ,S‘— wealth to their T children. This equation is used to draw the positive sloping lines in Figure 3.6 for two agents with the same wealth^ =b2 = b , but with x, < x2 different human Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 95 b,(x,) Borrower , Lender cr\ >»eur- neur | Credit ratjioning barrier Figure 3.6. The dynamics when agents have equal wealth capitals. In this case, given that both have the same wealth, bt , the slopes are the same. As shown, there is a credit-rationing gap between the borrower entrepreneur’s next generation wealth, who is just qualified to receive credit and that of the worker, whose human capital isn’t just enough to be qualified for loan. For wealthy investors with k <bt the transition schedule line is steeper than for those who borrow, as they don’t make loan repayments34. 3 4 McMillan and Woodruff (1999) discuss empirical evidence from Vietnam about firm contracting when the legal framework is underdeveloped. Under these circumstances contracting in parts rests on the risk o f future loss o f business. To ensure agreements are kept, firms rely on other informal devices Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 96 Case 2) Agents are heterogeneous in their initial wealth, but they are identical in their human capital (hence X ; = x2 = xt = 1). This case is corresponding to the one considered in Aghion and Bolton (1996) when there is no human capital involvement. The dynamics for the agents with 6, < b2 is graphically depicted. From (3.2.7) type (b,,x,) workers’ next generation receives equal amount of bl+ l = w< ^ x‘ T wealth. Figure 3.7 shows the dynamics for this particular case. Again using (3.2.14) borrower investors with the same human capital leave w< ^P ist— —^)) weaith r to the next generation offspring. This equation is used to draw the positive sloping lines in Figure 3.7 for the agents having the same human capital but different wealth levels. In this case again, the wealthiest investors’, those with initial wealth k <bt and who lend the excess capital at the market rate, transition schedule line is steeper than that for borrower investors’, as they don’t make loan repayments. It is interesting to compare these results with those found in Aghion and Bolton (1996). First, as Figure 3.7 displays, the wealth transition dynamics from parent to a child is a function of parental human capital. The higher is the human capital the more wealth is transferred regardless of the occupation of the parent. In this regard there is a role of redistribution policies, for instance, subsidizing education that will increase the average household’s human capital. While the redistribution of wealth to maintain repeated-game incentives, such as community sanctions, or more elaborate governance structures. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 97 Lender Borrower entrepre-. _i. _ eatapr.e; neur i Figure 3.7. The dynamics when agents have equal human capital from rich to poor is found to be increasing the aggregate welfare in Aghion and Bolton (1996) the role of the human capital is omitted. 3.5.2. Distributional Dynamics (3.2.4) and (3.2.5) give the dynamics equations of the population evolvement and the law of motion of the distribution function G,(xt, b). Using the optimal policy Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 98 functions (3.2.8) and (3.2.15) for both wage worker and entrepreneur fertility in the population evolvement and distributional dynamics equations the next period population level and distribution functions can be received. After several simplifying transformations the following expressions are received: ^ i = P jT * H x ) p L * £ (* ) p L f ( ^ L dG{xt, bt) + ] \ - L ^ x bJ + T { - L + d G i x M + l Q Q V V i f 0 wtx, ,(3.5.1) P ,+ lG l+ x (x,b)-- P,yx #1+2y) ff— A*,+ 1 *x,bM <b)dG(x„b,)+ \ ] W J(X i+ i <x> ( + ] <b)dG(xl,bl) + oow/xt i 0 W tX t xb(x) R ,b, X t ( X) 00 0 0 J } - £ - L/ ( * < + ! £ * , bl+ 1 < b )dG(xt, h ,) + J J / ( x ,+ 1 < x , bM < b’ )dG(xt, b,) x 0 W IX I 0 0 , (3.5.2) where in the first three integrals bl+ l and xl+ l are determined in the I indicator function from (3.2.7) and (3.2.11) wage workers’ wealth and relative human capital transfer dynamic equations from parents to children and in the fourth integral again, (3.2.7) and (3.2.11) are used for (xt+ u bt+j) corresponding to wage workers, while (3.2.14) and (3.2.18) are used for (xt+ i, bt+i) corresponding to investors. From (3.5.2) Gt-i(b \h ) the next period distribution function is obtained. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 99 3.5.3. The Stationary State Suppose < X > , = ®(vf,,Rt,P,,G{xt,bt)) completely determines the aggregate state of the economy. Accordingly, agents’ utility maximization defines the optimal decision rules for fertility, child human capital, and bequest allocation: nlt =n(hll,bind)l), h « + \ , bM =b(hjl,bjl,njl,< t> t). Definition 3.2. A stationary state is a competitive equilibrium which satisfies ®,+ y =®, =®‘ for j=l, 2, ... such that P t+ J Gt+ j(b*,x*)=PtGt given the agents maximize their utility choosing the optimal rules for fertility, human capital, and bequest allocation. Due to production sector non-convexities involving credit constraint conditions, rather considering the transitional dynamics of the model a heuristic approach is taken by assuming that the economy has reached a stationary state and examine the behavior of the model in equilibrium. Because of the non-convexities arising from the credit and production sectors, it is difficult to derive explicitly the stationary equilibrium conditions from the theoretical results. Therefore, rather a quantitative approach to analyze the model behavior is taken. 3.6. The Case of the Social Planner’s Problem In this section the social planner’s case is considered, whose objective is to maximize the social welfare. As an efficient allocation in the economy is the one, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 100 which maximizes output (per capita), the social planner maximizes the aggregate output and wealth subject to labor and capital market clearing conditions and given that the resource allocation is feasible. 1 -a Y rr. - . . r e — ( \ - C C ^ « ~= \\{Akptyi]-adGt{xn bt)= \\Akp, I <E> <E> dGt{x,,bt), (3.6.1) where E denotes the set of entrepreneurs. The labor and capital markets in equilibrium must satisfy (3.4.4) and (3.4.7) conditions. An allocation is feasible if it does not utilize more than the available resources in any given period of time. Hence, Pt >Zf +Et =Zf +Et, (3.6.2) where Zf =Lf in equilibrium and Et =Pt -Zf =Pt -Z f is the number of firms or entrepreneurs. Given the bequests and the number of children for each household the aggregate wealth evolves over time as follows: W mPm =P, (*,,&,) (3.6.3) <P> i> # + 2 y ) T <P/E>l + 2 r T 1 ! + l l+ l l+2y \\x,R,btdGt(xt,bl)+ +Rlb,)dGl(xl,bl) \< E > <P/E> Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1 0 1 where <P/E> is the population at a given time not including the entrepreneurs. The relative human capital evolves according to (3.2.11) and (3.2.18): =p, J. (1 - * ¥ n1 - t S ' n l V Rfb, < E> \ " V K w i J x,dGt{x„bt)+ JJI-1 T — xtdGt{xt,bt) s l - r <P/E>\ 1 J (3.6.4) (1-r)^ 1 - r J| — x,dG, (x, ,b,)+ J|x,dGt (x,, b,) \ 1 ~ T x w < E > \ t J <PIE> The aggregate capital stock in the economy in any period is given by the number of entrepreneurs times the firm’s capital, Etk. Since the transfer of wealth from the older to the younger generation takes in the form of capital stock bequests, a richer economy is the one that has more entrepreneurs, or firms. Hence, PtWt = Etk (3.6.5) Pt+J^t+i = Et+ \k P t+ 1 — Lt+ \ +Et= i =L+ U +Et+ l (3.6.6) Pt+ ,Wt+x=(Pt+l- L s t+ l)k Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 102 3.7. Computational Experiments 3.7.1. Calibration The values to the model parameters are assigned in a way to satisfy the assumptions imposed by the model and are reasonable in terms of being actually observed in reality, (j) is set equal to 0.078, meaning that each child requires 15% of parental time and parental attention for half of their childhood. As r determines the correlation between parents and children’s human capital, it is set equal to 0.2, small enough to avoid human capital divergence over time. The parental altruism parameter y, which determines the population growth rate at a stationary state, is set at 0.3 ensuring that it maintains a slightly above-zero value to the population growth. Since the overall size of the population is a scale parameter, the population change over generations is considered in per thousand of population terms, k is normalized at 1.2. The initial broader market interest rate is set at 5% meaning Rt=1.05. a=l/3 to ensure that labor share in production is 2/3, which is a widely accepted number in the literature. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 103 3.7.2. Quantitative Examples 3.7.2.1.Example 1. (The degenerate human capital distribution case) The simplest case is to consider that the human capital has a degenerate distribution so that the problem is simplified to the Aghion and Bolton (1996) setup, when the agents are heterogeneous with respect to their wealth. This assumption implies xt =Xt = 1. The uniform wealth distribution is chosen o <b <—k to ensure 3 that g l(x,,b,) = It follows bt =}b,gt(x„bt)dbt = -k< k, not all of the agents - k o 3 3 can afford purchasing high yield projects at k. Obviously, if x t = 1 < xt, or x, < xt there is no steady state equilibrium. Our calculations show that equilibrium is not obtained if xt < xt < x t either. There is excess aggregate supply of funds over the aggregate demand of funds for a reasonable range of market interest rates and wages. If x, < X t < xt, from the labor and capital markets equilibrium conditions it follows that a stationary equilibrium exists if wages per unit human capital are set at 0.46 (approximately a third of initial capital outlay requirement). This simple computational experiment reveals that given the wealth distribution in the population, the existence of the equilibrium in the model largely depends on what is the average human capital level in the economy. If the average human capital Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. is too low, there are too few borrower entrepreneurs who promise a successful investment outcome. Another simple case to examine is the degenerate distribution with respect to wealth (perfect equality). This case clearly doesn’t generate an equilibrium, because there must be at least some spread in wealth among the agents, some of them having above k wealth to lend to the borrower investors. To examine the dynamic behavior, the model is simulated for a given set of wages and market interest rates and when in the initial sage the wealth and the human capital are jointly uniformly distributed. The averages of b and x are scale parameters and normalized to be approximately equal to one, ensuring that not all the agents are sufficiently wealthy to be able to make investment without borrowing. The model is calibrated using the above-listed parameter values. The experiments are conducted using following uniform distribution function: evenly selected from all the relevant areas of interest depicted in Figure 3.4. The results of the computational experiments are provided in the appendix. 3.7.2.2.Example 2. (The bivariate uniform distribution case) 3 J b . This initial distribution ensures that the agents are Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 105 Table 3.1 displays the partial equilibrium results of population growth for selected wages per unit human capital. These results demonstrate that the population growth rate is very sensitive to the wage level, as a slight wage increase brings about a significant Table 3.1. Partial equilibrium at several fixed wages (uniform distrib.) (|)=0.078, y=0.3, t=0.2, x t = 1, bt = 1, R=5% Coh./ W 0.15 0.19 0.21 0.23 0.27 0.3 0.31 0.32 0.36 0.38 1 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 2 61821 13770 10527 9029 2661 2478 1892 1358 1046 970 3 4* 10* 190054 110983 81665 7156 6141 3627 1843 1095 940 4 24*106 3*10° 106 738675 19240 15221 6952 2503 1146 912 5 15 * 10y 36*106 12*106 7* 106 51735 37725 13325 3399 1199 884 fertility and population growth reduction. It can be concluded that a decreased income gap between workers and entrepreneurs lowers fertility and the population growth. The simulation results for population growth are even more sensitive to changes in the model parameters, such as the parental altruism coefficient, or the cost of raising children. Tables 3.2 and 3.3 show that the higher are these parameter values the higher is the population growth rate. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 106 Table 3.2. <(>=0.078, y=0.4, t=0.2 Table 3.3. (j)=0.15, y=0.3, x=0.2 Cohort w=0.27 w=0.34 1 1000.0 1000.0 2 3382.8 1239.2 3 11465.7 1536.3 4 38861.7 1904.7 5 131718.2 2361.3 Cohort w=0.27 1 1000.0 2 10260.0 3 105305.0 4 1080808.9 5 11092991.0 Chart 3.1 shows the dynamic behavior of b, (x) and bt(x) for the case of uniform distribution. With the increase in per unit human capital wage rate, the area under bt (x) , corresponding to the set of potential entrepreneurs subjected to credit rationing decreases and becomes zero after a certain wage rate. This implies that as the wages become more appealing there is an increase in the opportunity cost to becoming an entrepreneur and more agents choose to work instead. Chart 3.1 also displays the (xt,bt) pairs of a generated joint uniform distribution sorted in ascending order. This will correspond to a perfect positive correlation case (the correlation coefficient is one) of wealth and human capital. As the diagonal distributional line remains outside the credit rationing area, there is no credit rationing in this case. With the decrease in the correlation coefficient, the distribution will be more scattered around this diagonal line and there would be more (xt,bj pairs falling in the credit rationing area. It can be concluded therefore, that the higher is the correlation between The charts show that with an increase in the wage rate per unit human capital, the area under bt (x), corresponding to the set of agents who are credit rationed decreases and eventually becomes zero. This can be explained by the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 107 increase in the opportunity cost of employment as an alternative to entrepreneurship, which eventually forces all Chart 3.1. the bt (x) and bt (jc) lines (uniform distribution) w = 0.27 t D O J C V I C M C O i - l O t D d d d d d d o d relative human capital d d d o d relative hum an capital 1.80 1.60 1.40 1 .2 0 I 1 0 0 | 0.80 0.60 0.40 0.20 0.00 < o r ^ i - a > c o T - o - j T - C N I C O l O t O O O O C M ' s f d d d d d ^ ^ - ^ relative hum an capital * b » bhat - bbar 1.80 1.60 1.40 1.20 I 1 0 0 | 0.80 0.60 0.40 0.20 0.00 cocnoir^oiCTOOt-i^ M i o f ' - a > T - c o i n < o c o c » d d d O T - T — r*- r - T — t— relative human capital w = 0.32 w = 0.3 the agents interested in participating in entrepreneurship to willingly become wage workers. The more unequal the initial human capital and wealth are distributed, the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chart 3.2. PDF functions, uniform distribution r=[1.13 1.12 ] w=[0.25 0.27 ] r=[1.13 1.12 ] w=[0.27 0.29 ] 1.20 1.20 1.00 1.00 0.80 0.80 0.60 0.60 0.40 0.40 0.20 0.20 0.00 0.00 CM CO CM ^ If) CO CO CO O CO o> o Tf CO O) CO CO (SI ^ O CO Series 1 Series2 S eries 1 S eries2 r=[1.13 1.12 ] w=[0.29 0.31 ] r=[1.13 1.12 ] w=[0.32 0.35 ] 1 .2 0 1.0 0 0.80 0.60 0.40 0.20 0.00 n ' t m <o ■ q - C D oo o CM CM oo o> o ■cf CO 05 CM S eries 1 S eries2 1 .2 0 1 .0 0 0.80 0.60 0.40 0.20 0.00 CO CM 'J- CD 00 S eries 1 S eries2 r=[1.13 1.12] w=[0.35 0.37] 1 .2 0 1 .0 0 0.80 0.60 0.40 0.20 0.0 0 o > S eries 1 ■ Series2 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 109 lower is the number of agents subjected to credit rationing and the less significant is the credit rationing effect. The result would also remain valid for the dynamic behavior of the model. This result is quite different from the results in Aghion and Bolton (1996), as it is shown that credit rationing is a function of human capital and depends on the correlation between the level of human capital and wealth. The model is simulated for two periods to examine the distributional dynamics depends on the parameters of the model. Chart 3.2 depicts in the two-dimensional plane the two consecutive period distributions. From these results two inferences follow. First, there is an increased inequality in the second period as the second period distribution function is stochastically dominated by the first period distribution function. Second, the simulation results show that with the increase in wages the second period distribution function becomes flatter, meaning more inequality in income distribution. This can be explained again by the increase in the opportunity cost to becoming an entrepreneur with the increase in second period wages. 3.7.2.3. Example 3. (The bivariate lognormal distribution case) In this example the simulations are carried out using the lognormal bivariate initial distribution with given variances and for specified wages, market interest rates and other fixed model parameters. The averages of b and x are again normalized to be approximately equal to one. While the uniform distribution in the previous Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 110 example assumes an equal spread of agents across wealth and human capital, the lognormal distribution assumes the average wealth and human capital levels are more densely populated. Consequently, the latter provides more equal distribution than the former. This allows comparing these two simulations results in terms of the differences in distributions. Table 3.4 displays the partial equilibrium results for selected wages per unit human capital. The results are somewhat different from those in Table3.1. Particularly, when wages fall below a certain threshold, the fertility and population growth rates decline. Table 3.4. Partial equilibrium population growth at some fixed wages (lognormal distrib.) <(>=0.078, y=0.3, x=0.2, x t = 1, bt = 1, cr=0.5, R=5% Coh./W 0.22 0.25 0.27 0.29 0.33 0.37 0.41 0.43 0.45 0.47 1 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 2 489 751 1279 2459 3201 6726 3596 1533 1196 978 These results demonstrate that if the initial population wealth and human capital inequality is relatively low, with the drop in wage incomes decreases fertility and population growth rate, corresponding to the Malthusian view of positive relationship between these two variables. Table 3.5 presents the results for different a values and for different wages. As a increases there is more inequality in joint human capital and wealth distribution. The Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. I l l Table 3,5. Population growth at some wages and ct’ s (lognormal distribution, same parameters) Cohort c t /W [0.27 0.291 [0.33 0.351 [0.41 0.431 1 1000 1000 1000 2 0.3 613.25 1602.23 1292.82 0.5 1279.66 3201.99 1533.49 0.8 2185.2 6960.65 2051.46 1.2 5873.95 9237.54 3947.26 1.6 10275.86 18345.81 5216.85 results indicate that with the increase in inequality in both respects there is an increase in population growth. bt (x) and bt (x) exert similar dynamic behavior as in the previous exercise and this is clearly seen comparing Chart 3.1 and Chart 3.3. The diagram shifts further to the right with the increase in per unit human capital wage rate, while the area under b, (x), corresponding to the set of potential entrepreneurs subjected to credit rationing decreases and becomes zero after a certain wage rate. Again, as the wages become more appealing there is an increase in the opportunity cost to entrepreneurship and more agents choose to work, while those who choose to be entrepreneurs, given the same wealth, start from a higher level of relative human capital than if wages were lower. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. A Chart 3.3. the bt (x) and b, (x) lines (lognormal distribution) 112 w=0.27, 0.29 and sigma=0.8 « 0.8 | 0.6 1 0.4 0.2 0.0 T - c o ' s j - m t o o o o c v i i n c o Relative human capital b bhbar w =0.27,0.29 and sigma=0.8 1.4 1 .2 2 q 1 .0 s y. | £ > 0.8 r a ■ = ■ i « 0.6 0.4 0.2 0.0 T- C O Tf If) 10 00 0 0 0 0 0 0 O C O N- C O C O Relative hum an capital — ■ — bhbar — cdf cohort 2 w=0.33; 0.35 and sigm a=0.8 1.4 1 .2 1 .0 0 .8 2 0 .6 0.4 0 .2 0 .0 C O o b o to C O N O ) C M 0 0 0 0 ^ “ Relative hum an capital b —i — bhbar w=0.33; 0.35 and sigm a=0.8 1.4 0.8 m 0 .6 I 0.4 0.2 r|rt^iONCDOT|'CO b o o b d d t1 ^ y - 1 Relative hum an capital —« — b h b a r cdf cohort 2 The two period partial equilibrium distributional dynamics is provided in Chart 3.4 for different wages and a. Again an increase in inequality in the second period is Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. observed for most of the cases, although it is less obvious compared to the uniform distribution case. Chart 3.4. PDF functions, lognormal distribution w=[0.27 0.29] cr=0.3 g = 0.5 a=0.8 s a=1.2 c t =1.6 w=[0.33 0.35] <r=0.3 c t =0.5 c t =0.8 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 114 Chart 3.4. PDF lognormal distribution (Continued from the previous page) o=1.2 <7=1.6 w=[0.41 0.43] <7=0.3 <7=0.5 < 7= 0.8 <7=1.2 <7=1.6 * * * * * 0 . 8 0 . 7 { 0 6 O S 0 . 4 ' 0 . 3 0 . 2 0 . 1 0 5 1 0 I S 2 0 2 5 3 0 3 S 4 0 From these results one can infer that the lower are the wages per unit human capital and hence, the higher is the income inequality, it is more likely that the second Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 115 period wealth and human capital joint distribution is more unequal. This effect is to the contrary to what is observed when the distribution is uniform. 3.8. Conclusion This essay studies the effect of the initial wealth and human capital distribution on investment and growth through the channels of household utility maximization. In particular, agents’ decisions on child quality-quantity tradeoff, i.e. the parental choice of investing in per-child human capital and wealth versus number of children and their occupational choices are functions of inequality and income distribution and these choices play an important role in human capital and wealth transition mechanism. This is an alternative approach to those adopted in the literature, which focuses on stochastic transition processes as a wealth transition channel. It is shown that the lower is the initial inequality and wage worker-entrepreneur income inequality gap the lower is fertility and population growth rate and the higher is the parental investment in per child human capital and wealth. Therefore, the results underline the importance of education and/or training in increasing population human capital as an effective policy towards increasing investment and the number of credible loan allocations. If the initial wealth and human capital are more unequally distributed, the lower is the wage worker - entrepreneur income inequality gap, the less is the number of borrower investors and the less is the number of those credit rationed. If however, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the initial wealth and human capital are more equally distributed, the higher are the wages, the less is the number of borrower investors, but there are relatively more of credit rationed potential investors. The higher are the wages, the higher is the opportunity cost to becoming an entrepreneur. If the initial wealth and human capital are more unequally distributed, with higher wages more potential borrower-investors are willing to become workers. With this regard, this effect somehow attenuates the credit rationing mechanism described in the literature. The dynamic effect of this mechanism is even less evident, as the computational experiments reveal. A further extension of the essay would be to consider the policy implications and general equilibrium analysis. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 117 CHAPTER 3 BIBLIOGRAPHY [1] A. Alesina (1994), “Political Models of Macroeconomic Policy and Fiscal Reforms,” in S. Haggard, and S. Webb, eds. Voting for reform: Democracy, Political Liberalization, and Economic Adjustment, (New York, NY: Oxford Univ. Press, 1994). [2] A. Alesina and R. Perotti (1996), “Income Distribution, Political instability, and Investment,” European Economic Review, 40,1203 - 1228. [3] P. Aghion and P. Bolton (1996), “A Theory of Trickle-Down Growth and Development,” Review of Economic Studies, 64, 151 - 172. [4] C. Azariadis (1996), “The Economics of Poverty Traps Part One: Complete Markets,” Journal of Economic Growth, 1,449-486. [5] G. Becker and R. Barro (1988), “A Reformulation of the Economic Theory of Fertility,” The Quarterly Journal of Economics, 103(1), 1 - 25. [6] G. Becker, K. Murphy, and R. Tamura (1990), “Human capital, Fertility, and Economic Growth,” Journal of Political Economy, 98(5), S12 - S37. [7] J. Benhabib and A Rustichini (1998) “Social Conflict and Growth,” Journal of Economic Growth, 1(1), 143-158. [8] D. Bernhardt and H. Lloyd-Ellis (2000), “Enterprise, Inequality, and Economic Development,” Review of Economic Studies, 67, 147 - 168. [9] D. Croix and M. Doepke (2001), “Inequality and Growth, Why Differential Fertility Matters,” Working Paper, UCLA. [10] M. Dahan and D. Tsiddon (1998), “Demographic Transition, Income Distribution, and Economic growth,” Journal of Economic Growth, 3, 29 - 52. [11] J. Greenwood and B. Jovanovic (1990), “Financial development, Growth, and the Distribution of Income,” Journal of Political Economy, 98, 1076- 1107. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 118 [12] W. Easterly (2001), “The middle Class Consensus and Economic Development,” Journal of Economic Growth, 6, 317 - 336. [13] D. Evans and B. Jovanovic (1989), “An Estimated Model of Entrepreneurial Choice under Liquidity Constraints,” Journal of Political Economy, 97(4), 808 - 827. [14] O. Galor and D. Weil (1996), “The Gender Gap, Fertility, and Growth,” The American Economic Review, 86(3), 374 - 387. [15] O. Galor and D. Tsiddon (1997), “Technological Progress, Mobility, and Economic Growth,” The American Economic Review, 87(3), 363 - 382. [16] P. Krusell and A. Smith (1998), “Income and Wealth Heterogeneity in the Macroeconomy,” The Journal of Political Economy, 106(5), 867-896. [17] S. Kuznets (1955), “Economic Growth and Income Inequality,” The American Economic Review, 45(1), 1-28. [18] R. Lucas (1978), “On the Size Distribution of Business Firms, ’’The Bell Journal of Economics, 9(2), 508 - 523. [19] R. Lucas (1988), “On the Mechanics of Economic Development.” Journal of Monetary Economics, 22, 3 -42. [20] T. Malthus (1952), “An Essay on Population.” London: J.M. Dent & Sons Ltd; New York: E.P. Dutton & Co. Inc. [21] D. McKenzie and C. Woodruff (2002), “Is There an Empirical Basis for Poverty Traps in Developing Countries?” Mimeo, Stanford University. [22] J. McMillan and C. Woodruff (1999), “Dispute prevention without courts in Vietnam,” The Journal of Law, Economics, and Organization, 15(3), 637- 658. [23] O. Morand (1999), “Endogenous Fertility, Income Distribution, and growth,” Journal of Economic Growth, 331 - 349. [24] T. Persson and G. Tabellini (1994), “Is Inequality Harmful for Growth?,” American Economic Review, 84, 600-621. [25] T. Piketti and E. Saez (2001), “Income Inequality in the United States, 1913- 1998,” NBER Working Paper No. 8467. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 119 [26] T. Piketti (1996), “The Dynamics of the Wealth Distribution and the Interest Rate with Credit Rationing.” Review of Economic Studies, 64, 173 - 189. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 120 4. Education and Self Selection: The Role of Household and Individual Characteristics in Wage Earnings Distributions in East Germany After Reunification After German reunification workers from Eastern Germany had to adapt swiftly to the skill and other requirements needed to compete in the labor markets of unified Germany. This adjustment process was of varying difficulty for different types of workers. The reunification forced industries in Eastern Germany to face increased competition from their West German counterparts, causing some industries to lose their markets and lay off their employees, or to offer lower salaries compared with western standards. This has caused wage discrepancy between the “western” and “eastern” workers. The East German workers thus faced the dilemma of either to continue working in their current “conventional” occupations or to receive retraining and get hired by an advanced “western” firm. Our goal in this essay is to analyze the dynamics of labor earnings inequality in Eastern Germany in light of inter industry and intra industry labor movements with a focus on the extent with which individual and household characteristics have contributed to the workforce occupational choice. I use sectoral decomposition procedure to reveal how and to what extent these effects have influenced the earnings inequality. I find that indeed these characteristics have played significant hole in occupational choice process and thus on the labor earnings’ distributional dynamics after the reunification. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 4 .1 .Introduction to Essay Three 1 21 After the reunification workers from Eastern Germany had to adapt swiftly to the skill and other requirements needed to compete in western market environment. This adjustment process was almost effortless for some workers, while for others it was much more difficult. As the neoclassical theory postulates, wages are determined in the labor market from labor supply and demand equilibrium. The reunification caused industries in Eastern Germany to face increased competition from their West German and other western counterparts, forcing some industries using obsolete production technologies to lose their markets formally assigned by the central planning system and to lay off their employees, or to offer salaries lower than those paid by their western competitors. This has caused wage discrepancies between “western” and “eastern” workers. Thus, East German workers faced the dilemma of either continuing to work in their current “conventional” occupations or receiving training and applying to the advanced, “western” industry jobs for higher compensation. Some of the “eastern” jobs became redundant causing layoffs3 2, but many new occupations, in management, trade, services, while unnecessary under the central planning rule became increasingly needed. Many workers have found themselves in a position of exploring opportunities in these “new” areas. 3 2 An anecdotal evidence is that the professors o f political economy or Marxism philosophy in Eastern universities had to look for new, less prestigious and low paid occupations despite their education and experience in the academe as these subjects weren’t in demand any more. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 122 The purpose of this essay is to examine the dynamics of labor earnings inequality in Eastern Germany in light of these inter industry and intra industry labor movements with focusing on the extent with which individual and household characteristics have contributed to the occupational choice and transfer from older sector to the technologically advanced one. In this regard, sectoral decomposition method is used to reveal how and to what extent these effects have influenced the earnings inequality. There is a vast literature on this theme based extensively on US labor data. For instance, Daly and Valletta (2000) analyze the household inequality and poverty dynamics in the US during the years 1969-1998. They explain the rise in labor earnings inequality over this period through the widening dispersion in individual labor earnings, particularly, those between men and women, and through such behavioral changes, as the increase in female labor supply, changes in family composition, and living arrangements. Our argument nevertheless, is that the above normal rate of the earnings inequality growth is largely due to the contribution of the technological change, rather than the “regular” interpretations about the increased participation of women in the labor force, or other changes in living arrangements over the very short post-reunification period. East German workers’ “sudden” and significant exposure to western industries and labor markets could be compared with a “big” technology shock. Acemoglu (2000) shows that technical change has been skill biased during the past century. He also discusses that the past few decades have witnessed an acceleration of skill bias Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 123 and an increase in returns to skills, in terms of both education and experience, causing increase in wage inequality. He further argues that this skill-biased technical change was in response to the acceleration in the supply of skills, as firms find developing skill-biased techniques is more profitable. Our approach on the other hand, is to focus largely on the labor supply side and particularly, on how workers’ background and individual characteristics contribute the skill distributional dynamics, given the technology shock. With this regard, the German reunification allows analyzing this effect, namely, how well were workers able to adjust to the “large” technology shock exposure, given the individual and household characteristics. It is shown that this adjustment was largely achieved via substantial increase in inter-industry skilled labor movement as a response to the exposure to western industries. In a similar note Galor and Tsiddon (1997) explore the relationship between technological progress and wage inequality. They show that in the initial phase of the technology shock there is an increase in inequality in the return to skills, as there are relatively fewer workers to work in the new sector. Over time, as the technology becomes more accessible, this inequality decreases. The essay presents the effect of mobility, the allocation of talents across occupations, as a technological link between wage inequality and economic growth. Under the centrally planned regime most of the workers had received vocational education, leading to narrow specializations at an early age. Moreover, employment was guaranteed by law, leaving workers with little incentive to acquire multiple Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. skills to cope with unexpected events and uncertainties, such as loss of employment or closure of the industry. The education and experience attained under the old regime turned out not very applicable under the new, market conditions, as the old and new occupations were hardly substitutable. The old educational system was biased towards hard sciences and engineering, neglecting such skill areas as social sciences, law, business, public relations, and public policy, highly in demand under the market conditions. The reunification has brought increased exposure to outside competition, leaving workers to solve the dilemma whether to stay in the old, conventional industry, or receive adequate training and become qualified for highly demanded, high pay occupations. As right after the unification East German workers were still at levels of education and experience obtained during the former regime, one would expect that their post-unification levels of “returns” to experience and education was much lower than those of their western counterparts. To this end I conduct Mincer (1974) regressions and compare the results for East and West German sub samples. The German Socioeconomic Panel Studies (GSOEP) household survey database is used over the period from 1990 to 2001, which essentially includes the most intensive adjustment period to market. First, the results show that West German workers generally had higher rates of return to education and experience than their East German counterparts had during the most of the period and especially right after the reunification. Second, there was a sharp decline in East German workers’ returns to schooling and experience, especially between 1992 and 1994 with slight rebounding Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 125 after 1996. The plunge in return estimates shows that the entry and participation to the much more competitive West German labor market was a costly and difficult process for East German workers. Our results confirm theoretical predictions in Galor and Tsiddon (1997), where they conclude that if older technologies require less education than new ones, this may generate a downward bias in the cross-section estimates of the returns to education. They further argue that in cohort analysis this lock-in effect will tend to decrease the returns to education in older generations relative to younger ones. To further examine how East German workers have responded to this initial plunge in their returns to skills especially, whether there was a significant labor movement from less competitive industries to more competitive western, or high pay industry, to compensate for the initial plunge in their returns to skills the inter industry segregation coefficients for education, experience, and log wages are calculated. Similar analysis is conducted in Kremer and Maskin (1996) using the data for US, Britain, and France based on their model of segregation indices. Their results suggest that the past decades’ growth in wage inequality has been accompanied with an increase in segregation by skill. They also show that these increases in segregation and inequality were due to the technological change, or due to observed changes in skill-distribution. Our results for the German data show that the coefficients are indeed increasing throughout the period. It is also found that this increase in segregation indices is much higher in East Germany, than in West Germany, US, Britain, and France. Our conclusion is that the inter-industry labor Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 126 movement was especially high in East Germany due to the sudden exposure to the significant western “technology shock”. Further, cross section and panel estimations are conducted to analyze the effects of various personal and household features on the change in wage earnings inequality over the period. Also, decomposition of wages is carried out, given worker personal abilities and individual and household circumstances, such as educational background, past work experience, marital status, number of children, etc, using generalized Blinder-Oaxaca (Blinder (1973), Oaxaca (1973)) and dummy variable, or “cellular” decomposition methods. Further, a self-selection model is considered when workers decide either to invest time and resources to get re-trained and become qualified for the newer jobs, or to continue working in the conventional, lower-pay industry. Section 4.2 discusses the selection of the data and provides panel and instrumental variable estimations results for monthly wage earnings during 1992 to 2001. Section 4.3 focuses on skill segregation coefficient and the results obtained. Section 4.4 describes the decompositions. Section 4.5 provides and compares the results of the decompositions. Section 4.6 discusses about the self-selection model and provides the empirical results when the East German panel data is applied to the model. Section 4.7 concludes. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 127 4.2.Data Selection and Estimation Our analysis is based on a panel from GSOEP dataset over the period 1990 to 2001 that includes both East and West German samples. Table 4.1 lists average values of some of the variables from the Eastern, pooled men and women sample and for men and women separately. The table shows that there was an increase in individual labor earnings over the period for both men and women, although average wages for women remain well below those of men throughout the period. The table also reveals that there was an increase in the labor earnings spread over the period. The coefficient of variation of labor income has increased from 16885 German Marks to 30158 for men and women, from 6143 to 31286 for women, but it has decreased from 30230 to 27885 in men’s sub sample. While this decrease in variation is plausible for the latter, its increase is remarkable for the former. Certainly, this increase in the total coefficient of variation is explained by the increase in competition and uncertainty that has caused a decline in importance of such wage equating factors within an industry or a firm as compensating differentials, efficiency wages, or pressure for equity. Table 1 shows also that there was an increase in household asset income and in household public transfers, both considered to be household non-labor income. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 4.1. Average values of selected variables by year. East German sample 128 Year Individual Coeffici- Annual Employm HH labor yr ent of employm level asset earnings Variation hours (FT/PT) income HH Treiman : public occup ' transfers prest i Satisf Satisf with life life in now 5years Total 1990 21661 16885 468 12595 35.17 6.67 7.45 1992 18199 10711 1292 1.91 364 3006 1994 23303 16551 1247 1.95 616 5047 23.97 6.26 6.45 1996 24240 22260 1179 2.01 812 6192 22.84 6.40 6.27 1998 23863 25870 1056 6305 23.21 6.38 6.07 2000 24180 29112 1126 2.07 1278 6271 22.79 6.37 6.16 2001 23979 30158 1085 2.10 1473 6385 20.78 6.40 6.11 Male 1990 26506 30230 891 16413 37.82 6.62 7.51 1992 22704 8662 1599 1.68 364 3118 1994 28778 13108 1546 1.74 617 5332 27.45 6.26 6.47 1996 29370 18676 1469 1.80 850 6415 26.04 6.36 6.22 1998 28693 22627 1107 6497 25.40 6.37 6.06 2000 28903 26393 1388 1.88 1296 6412 24.91 6.36 6.10 2001 28607 27885 1317 1.93 1516 6478 23.19 6.38 6.10 Female 1990 18578 6143 10165 32.89 6.71 7.39 1992 14234 11240 1023 2.10 363 2908 1994 18486 18604 986 2.14 614 4795 20.92 6.27 6.44 1996 19727 24773 924 2.20 779 5995 20.04 6.44 6.32 1998 19612 28105 1012 6136 21.28 6.39 6.07 2000 20025 30755 897 2.23 1262 6147 20.92 6.38 6.20 2001 19906 31286 881 2.24 1434 6303 18.66 6.41 6.12 ■JO Note that most of the variables exhibit a gradual increase in their values . Particularly, the increase in the employment level reflects an increase in the number of workers in full-time jobs and a decrease in part-time employee numbers, 3 3 The other variables of interest, not listed in the table are education and number o f children. Both remain relatively constant throughout the period at about average values o f 12.1 and 2.05 correspondingly. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 129 representing an increase in labor force participation. However, there was a decline in the annual working hours. Men work longer hours per annum than women. Surprisingly enough, while workers became more satisfied with present life between 1992 and 2001, the expectations about the future and levels of satisfaction with life in five years were steadily in a declining mode. This might have related with the fact that workers gradually have been repositioned to less prestigious occupations than they were originally holding, according to the Treiman occupational prestige scale34. Comparing the indices for men and women, men on average hold more prestigious positions than those women hold. Nonetheless, over the period of time this difference between the indices for men and women has remained more or less unaffected. Figure 4.1 depicts the distributional dynamics of the Treiman occupational prestige code where the estimation is conducted by non-parametric, Kernel approach. The results show that indeed, there was a decline in occupational prestige for both men and women. It is interesting to notice that the sample distributions are double peaked, meaning a sharp polarization between high-skilled and low-skilled labor 3 4 Treiman’s Standard International Occupational Prestige Scale (SIOPS) is an occupational status scale. As the division of labor is an essential part o f social inequality, occupation therefore is one of the main dimensions o f social stratification. The SIOPS classifies occupations according to their “occupational status”, covering prestige, socioeconomic status, and class measures, based on the 1988 International Standard Classification of Occupations (ISCO-88). The 4 digit ISCO-88 code classifies the most prestigious occupations with the lowest code numbers and assigns higher code numbers to the occupations with lesser prestige. For instance, the code assigns numbers starting with 1000 to Legislators, Senior Officials & Managers, 1200 to Corporate Managers, 3000 stands for Technicians and Associate Professionals, 4000 for Clerks, 5000 for Service Workers and Shop and Market Sales Workers, 6000 for Skilled Agricultural and Fishery Workers, 7000 for Craft and Related Trades Workers, 8000 for Plant and Machine Operators and Assemblers, 9000 for Elementary Occupations. The Treiman prestige code has been extensively developed and modified in Ganzeboom and Treiman (1996), where three cross-nationally standardized measures of occupational status are proposed for the occupations listed by ISCO-88. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 130 Figure 4.1. Kernel density estimates of Treiman Occupational Prestige Scale, East German Workers. Kernel Estimate, Occupational Prestige Code o 0.040 2 0.035 0.025 0.020 0.015 0.010 0.005 0.000 co<qa><qoioa>ooT-;'*j;r'-o<'Ocoeo-^-'l; cm in m co -Total 90 -Total 93 Treiman scale -Total 01 Kernel Estim ate, Occupational P restige Code, Men o 0.040 s 0.005 0.000 ■ Men 90 ■ ■ Men 93 -Men 01 Treim an scale Kernel Estimate, Occupational Prestige Code, Women 0 0.040 1 0.035 0.030 0.005 0.000 3 o in a i co t - Tf iv, o co p eo ■ > - . MCMnn^'fininioio -W om en 90 -W om en 93 -W om en 01 Tmiman scale Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 131 force. Over the same period the peak corresponding to the more prestigious occupations range has decreased, while the peak corresponding to the less prestigious range has increased. Similar dynamics is observed separately for men and women. It can be concluded therefore, that the decision to attain training and relocate to a competitive, “western” industry offering to a greater extent higher compensations wasn’t that easy for most of the Eastern workers in terms of attaining less prestigious occupations than they had previously held. To further track worker skill returns, I conduct Mincer (1974) regressions for each year over the period and compare the results obtained from East and West German sub samples. The Mincer regressions for yearly cross sections are ln(w, ) = A0 +&xSt + A2 exp eri + X3 exp erf + £., (4.2.1) where i stands for an individual, W j is the individual wage, s, is the years of schooling, exper = age - s - 6 is a proxy for obtaining work experience, and s is the error term. Xi is the Mincerian return to education and + 2^expert is the Mincerian return to experience. Figure 4.2 compares the estimated returns to education and experience for East and West German sub samples for selected age and gender groups. The first chart for both men and women shows that the return to education starts lower for East German workers, but increases steadily at a higher rate than that of West German workers, which Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 132 Figure 4.2. Returns to education and experience, east-west comparison of the estimated coefficients from the Mincer regressions. Return to Education, Age 17 to 65 0.100 0.080 0.060 0.040 0.020 0.000 co < 0 O < D CO rs> c o CO CO o > -* N > co o co o 00 o ■East -W est Return to Education, Age 17 to 65, Women 0.100 0.080 0.060 0.020 0.000 co CO o - » ■ fsj CO CO o CO CO o cn co o •E a st H i -W est Return to Education, Age 26 to 35 Men 0.020 0.000 ■East -W est Return to Education, Age 17 to 65, Men 0.080 0.060 0.040 0.020 0.0 0 0 & ■East -W est Return to Education, Age 26 to 35 0.160 0.140 0.120 0.100 0.080 0.020 0.000 Return to Education, Age 36 to 45 0.100 0.090 0.080 0.070 0.060 0.050 0.040 0.030 0.020 0.010 0.000 -E ast -W est Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 133 Figure 4.2. (Continued) Return to Education, Age 36 to 45 Men 0.100 0.060 0.040 0.000 Return to Education, Age 36 to 45, Women 0.140 0.120 0.080 0.060 0.020 0.0 0 0 CO CO CO (0 (0(0 o rs) ^ CO CO o (0(0 0 o>________00 o - E a s t M W est Return to Education, Age 46 to 55 0.120 0.100 0.080 0.060 0.040 0.020 0.000 K ) CD CD CD CD IS) CD CD CD CD CD C O 0 0 O E ast W est Return to Education, Age 46 to 55, Men 0.120 0.100 0.080 0.060 0.040 0.020 0.000 C O CD O C D C D ro CD £ CD CD O ) N ) CO O CO O 00 O -E ast W est Return to Education, Age 46 to 55, Women 0.120 0.100 JKr-m 0.080 0.060 0.040 0.020 0.000 h o CD C D O CD CD N > CD CD C D C O O ) C D CD 0 0 O East ■West Return to Education, Age 46 to 55, Women 0.120 0.100 0.080 0.060 0.040 0.000 — k -a. — * . N) CD CD CD O CD CD CD O ■ U O ) 0 0 O Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 134 Figure 4.2. (Continued) Return to Experience, Age 17 to 65 0.016 0.014 0.012 0.010 0.008 0.006 0.004 0.002 0.000 0.012 0.010 0.008 0.006 0.004 0.002 0.000 Return to Experience, Age 17 to 65, Women C O CO o CO CO N ) CO c o CO CO o > CO CO c o o o o ■East W est Return to Experience, Age 17 to 65, Men 0.018 0.016 0.014 0.012 0.010 0.008 0.006 0.004 0.002 0.000 & -E ast -W est Return to Experience, Age 17 to 26 0.350 0.200 0.150 0.050 0.0 0 0 -E a st —■ — •W est Return to Experience, Age 26 to 35 0.050 0.045 0.040 0.035 0.030 0.025 0.020 0.015 0.010 0.005 0.000 W est Return to Experience, Age 26 to 35, Men 0.060 0.050 0.040 0.030 0.020 0.010 0.000 c o CO c o c o c o o > c o c o o o E ast —■ — W est remains essentially flat throughout the period, reaching and staying above it after 1998. A similar dynamics is recorded for East males, while that of East females Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 135 Figure 4.2. (Continued) Return to Experience, Age 36 to 45 0.018 0.016 0.014 0.012 0.010 0.006 0.004 0.000 Return to Experience, Age 36 to 45, Men 0.025 0.020 0.015 0.005 0.005 0.015 to < o o CO CO to C O s <0(0 0 (o C D o c r > c o o -E ast W est Return to Experience, Age 36 to 45, Women 0.040 0.030 0.020 0.010 0.000 - 0.010 - 0.020 ro o < o C O < 0 < 0 < 0 < 0 0 > C O C O O o -E ast —■ — W est Return to Experience, Age 46 to 55 0.015 0.010 0 .0 0 0 0.010 c o c o o c o co ro c o s. N ) < 0 C O O CO CO o 0 ) 0 0 o -E ast -W est remains lower throughout the period, although it again, rises steadily. The age and gender subgroup charts further show that this difference in returns to education was more obvious for older age subgroups, especially, for those of women, while for younger men’s subgroups between ages of 26 to 35 this difference is hardly noticeable. From the experience return charts it can be observed that although the returns to experience for the East German sub sample again exhibits a gradual increase, it remains lower than that of the West German sub sample for the whole period of Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 136 time. This result remains unchanged for men and women estimated separately. From the sub group charts by age and gender it can be concluded that, again, the gap is the largest for the oldest groups of workers especially, for the age group 46 to 55. Comparing these education and experience returns it can be concluded that again, the entry and participation to the much more competitive West German labor market was a costly and difficult process for East German workers. First, there was a sharp initial plunge in workers’ returns to schooling and experience, especially between 1990 and 1992 with subsequently slight rebounding for the rest of the period. Second, the results show that West German workers generally had higher rates of returns to education and experience than their East German counterparts had over most of the period and especially shortly after the reunification. Third, this difference in returns is particularly significant for the older age groups and especially for women (less than half of those of West German workers) and these groups were the most affected from the initial shock. This last result is particularly explainable in terms of the loss in experience that the workers undergo. Jovanovic and Nyarko (1996) show that a switch of technologies temporarily reduces expertise and the bigger is the technological gap the bigger is the loss. If ability is a complement to sector-specific on-the-job training, high-ability individuals may lock themselves into older technologies. The prospect of productivity drop may prevent older age workers, who have the most sector-specific 3 5 Although not listed, the estimations for age groups 55 and older show that this difference is even larger. Due to the lack of sufficient observations and due to the space constraints these results aren’t posted in the essay, although they are available upon request. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 137 experience, from changing industries. In contrast, younger age workers are the less experienced, less skilled, and therefore, less productive in the conventional sector and they may switch industries with much more ease. These results confirm also the theoretical predictions in Galor and Tsiddon (1997), where they conclude that if older technologies require less education than new ones, this may generate a downward bias, or a lock-in effect, in the cross-section estimates of the returns to education. They argue also that in cohort analysis this effect will tend to decrease the returns to education in older generations relative to younger ones. To track and compare the distributional dynamics of labor earnings again kernel density estimation method is applied for East and West German sub samples. From these 1992 and 2000 estimations results, depicted in Figure 4.3, it can be observed that there was a slight change in the spread for East German women, while it is hardly noticeable any spread change for East German men. Both distributions however shifted rightwards, indicating an increase in labor earnings over the period. Table 4.1 also records this increase in average wages (annual labor earnings in German Marks are listed). For men it has increased from 22704 to 28607 between 1992 and 2001, or by 2.88% annually. The average wages for working West German men have increased from 54238 to 72130 during the same period (number of observations is 2793), or by 3.6% annually. It can be concluded therefore that Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 138 Figure 4.3. Kernel density estimates for wage inequality. Log monthly wages are used. Eastern Germany Western Germany Men, 1992 Log Individual Labor Earnings 11.6 5.89 Log Individual Labor Earnings ^ 3 2 5 4.02 Men, 2000 0.857 L0037 Log Individual Labor Earnings 6.28 12.03 0.86 4.92 Log Individual Labor Earnings 13.39 Women, 1992 J.737 Log Individual Labor Earnings 11.75 4.45 0.613 ' 4.07 Log Individual Labor Earnings 13.07 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 139 Figure 4.3. (Continued from the previous page) Women, 2000 0.56 0.0031 i Log Individual Labor Earnings 6.05 12.3 0.51 Density 0 3 gg Log Individual Labor Earnings 13.1 although the labor earnings have increased in Eastern Germany they were growing at a lower rate than those of the West German men. Similar wage dynamics is recorded for East and West German women. Table 4.1 reveals that East German working women’s average wages have increased from 14234 to 19906 from 1992 to 2001, or by 4.4% annually36, while West German women’s average wages have increased from 30608 to 42105 during the same period, or by 4.17% annually. One difference that appears to be significant for West German women is the minimum wage effect, as the density curve has a second peak at lower wages. Table 4.2 displays instrumental variable estimations results for log individual monthly labor income3 7 over the period of 1990 to 2001, using separately East and West German sub samples. Since the number of children 3 6 Similar wage dynamics women against men is observed in the US data. For example, DiNardo, Fortin, and Lemieux (1996) show that over the period 1973-1992 women’s wages grow faster than those o f men. Juhn and Murphy (1997), focus on employment changes for married couples and hence, on the male-female distributional dynamics between 1969-1989. They have found that the increased labor supply by married women is not motivated largely to compensate for the decline in men’s wage growth rates. This period is especially characterized with the increase in return to skill, and de- unionization in US. 3 7 In Germany as in most o f the European countries the salaries are on a monthly base, rather than on an hourly or annual base prevalent in the US. For this reason monthly salaries are considered and not the hourly wages. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 140 cannot be treated as an independent variable, it has been instrumented, where the instruments are one and two period lagged log individual labor earnings, marital status, years of education, work experience, and annual work hours. Table 4.2. Instrumental variable estimation of log net income last month Sub sample East W est Explanatory variables Coef. t 1 P>|tl Coef. t P>|t| Number of children -1.1416 -1.99 0.046 -0.5339 -4.03 0 Experience 0.1274 2.75 0.006 0.0931 7.83 0 Square of experience -0.3013 -2.48 0.013 -0.1880 -6.78 0 Gender -0.2103 -4.96 0 -0.3392 -19.3 0 Marital status -0.0910 -1.78 0.076 -0.0734 -3.05 0.002 Years of education 0.0370 5.21 0 0.0525 12.05 0 Annual work hours 0.0003 10.92 0 0.0006 26.11 0 Worried economic develop 0.0405 1.18 0.238 Worried finances 0.0584 1.38 0.167 Worried about job secur -0.0575 -2.25 0.024 Worried about child care 0.2196 2.04 0.041 Life satisfaction today 0.0248 1.63 0.103 Life satisfaction in 5yr -0.0147 -0.97 0.331 Treiman occ. Prest. Code 0.0081 6.92 0 Constant 9.3853 19.68 0 8.9672 63.5 0 Number of obs 4085 21181 R-squared 0.1567 0.4524 Adj R-squared 0.1551 0.4522 Instrumented: Instruments: Number of children in household Experience, square of experience, log wages lag 1, log wages lag 2, marital status, years of education, annual work hours The results show significant positive influence of work experience, years of schooling, the level of occupational prestige, and annual work hours on wages. In Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 141 contrast, the coefficients for the number of children, experience square, and gender3 8 all take strong, negative signs. While the coefficient for marital status3 9 is negative in both estimations, it is insignificant when using East German sub sample but it is significant when using West German sub sample. This negative relationship implies that married and single individuals have higher wage income than those separated, divorced, or widowed. The results support the theory of skill biased technological advancement discussed in Section 4.1, as jobs opened in Eastern Germany by “western” companies were uncommon during the central planning system, but were requiring skills, education, and experience, specific to the technologically advanced, western market economy and had to offer special training courses, higher skill premiums, and higher wages than the conventional sector was offering due to the insufficient supply of qualified East German labor supply to those industries. 4.3.Wage Inequality Dynamics and Segregation by Skill This section focuses on the effect of wage gap between high- and low-skill workers and its effect on wage inequality dynamics. The reunification and introduction of western-type firms has caused a shift of skill requirements, as some of the occupations became redundant, while many new types of occupations and skill requirements became increasingly in demand. This process has caused a rise in skill 3 8 Gender is a dummy variable. It takes the value 1 in case o f men and 2 in case o f women. 3 9 Marital status takes 1 if married, 2 if single, 3 if widowed, 4 if divorced, 5 if separated. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. dispersion. At the industry level, given that workers of different skill levels are imperfect substitutes and output is more sensitive to skill in some industries than in others, firms that once used to be hiring both high- and low-skill employees, the increased competition after the unification forced them to specialize in one level or the other, thereby fostering segregation by skill. Once skill distribution becomes sufficiently dispersed, a further increase in mean skill-level may raise the wages of high -skilled workers but cause those of poorly skilled workers to decline, increasing inequality as a result. This process results in simultaneous increases in inequality and skill segregation40. The increase in skill segregation particularly would mean that the “technology shock” has increased inequality not only due to the shift in employers’ labor demand schedule, but also, due to the workers self-selection process, when they choose to undergo retraining and the costs of industry-change become qualified for the advanced sector. Of course, these decisions are largely dependent upon workers’ background and individual characteristics. The generalized segregation index proposed in Kremer and Maskin (1996) is used to measure segregation by skill across industries during the transitional period after the reunification. The proposed index is a measure of correlation that can be applied to variables taking many values, rather than being a measure of a dichotomous state, such as race or sex. It is defined as follows: 4 0 A related literature focuses rather on workforce composition and labor productivity, for example Haltiwanger, Lane, and Spletzer (1999). This way o f thinking is motivated by matching and sorting models, which suggests that firms with different production process will employ workforces with different skill structures. My approach is more self-selection literature oriented due to the existing large technological gap between East and West German firms. As the reunification has caused an immense amount of “technology shock”, East German workers then had to overcome this technology barrier to be qualified working in western type firms. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 143 - ? ) ] £ ( ? * -<1)1*1 p - — le Z j- k Zh___________ , (4.3.1) XX^-*)2 j - 1 AeZy where workers are indexed by i and k. qt is the measure of skill level of worker i, q is the mean skill level in the sample of N workers. Zj is the number of workers in j- th firm, where firms are indexed by j=l,...,J. If p = 0, a correlation of zero, then all firms have the same skill-mix of workers. If on the other hand p = 1, then there is a complete segregation, meaning that all workers within a firm have the same q, but the firms have completely different workers in terms of their skills. It must be noted that the index is invariant to the affine transformations of the units in which skill is measured. Equation (4.3.1) can be represented in terms of the variance of q within and between firms as follows: where s] and si represent the variance of q between and within firms, respectively, and s2 r represents the total variance of q in the economy. Since p depends only on the variance of q in the population and on the variance of mean q between firms, it can be calculated using separate data sources for workers and firms. This is especially useful if data linking employees and their firms are unavailable, for Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. example in the case of household survey data. To correct (4.3.2) for the differences in the number of employees in each firm the following normalization can be performed: ! s l W - J ) s 2t / ( N - 1) (4.3.3) The following confidence interval is applied at 5% significance level in the calculations: where bf is an F-distributed random variable with (N-J, J-l) degrees of freedom. Figure 4.4 presents the inter-industry segregation coefficients for education, experience, and log wages for East and West German sub samples over the period 1990 to 2001. From the education coefficients chart it can be seen that while the East and West coefficients were approximately equal in 1990, there was a subsequent gradual increase in both. The chart however shows, that the trend was much higher in East Germany (the confidence intervals do not intercede). This increase indicates that, first, there was a highly educated labor movement from one industry to another both in Eastern and Western Germany such, that some industries increasingly became overflown with highly educated labor force, probably due to a better pay, while others have 1 1 (4.3.4) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 145 Figure 4.4. Inter-industry coefficients. The confidence intervals are provided at five percent significance level (the estimates are adjusted for variations in firm size). Coefficient of S egregation, Education 0.100 0.090 0.080 0.070 0.060 0.050 0.040 0.030 0.020 0.010 0.000 _ k — X _ i. _ x N) N> CO CO CO CO CO CO CO CO CO CO o o CO CO CO CO % CO CO CO CO CO o o o CO tn O ) •Nj 03 CO o Coefficient of S egregation, Experience Coeff. East Coeff. W est L ow er bound W Upper bound W L ow er bound E Upper bound E Coeff. East - - a- - - Coeff. W est — — — L ow er bound E — Upper bound E — - — L ow er bound W — Upper bound W _ x _ x N ) N > C O C O C O C O C O C O C O C O C O C O o o C O C O C O C O C O C O C O C O C O C O o o o N 3 0 3 c* 0 3 - N l 0 0 C O o U .4U U 0.440 0.390 0.340 0.290 0.240 0.190 0.140 r\ n a n Coefficient of S egregation, Log Wage 0.320 0.300 0.280 0.260 0.240 0.220 0.200 0.180 0.160 0.140 0.120 _ i. _ x . - k . N5 N3 CO CO CO CO CO CO CO CO CO CO o o CO CO CO CO CO CO CO CO CO CO o o o ro C O A Ul O) *vj 0 0 CO o Coeff. East Coeff. W est Low er bound E Upper bound E Low er bound W Upper bound W Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 146 witnessed its depletion, probably due to becoming less competitive and unable to maintain competitive wages. The corresponding coefficient values were 0.00587 and 0.0067 (adjusted for differences in firm size ) for East and West Germany in 1990 and were 0.0818 and 0.0688 in 2001. Similar dynamics is observed for the segregation coefficients by experience and log monthly wages. The values for experience in East and West Germany were correspondingly 0.098 0.218 in 1990 and were 0.45519 and 0.3474 in 2001. This increase is especially remarkable when comparing the East and West German results. Neither such a trend is recorded in the other western countries. For instance, the coefficient has increased in France from 0.11 in 1986 to 0.16 in 1992 (source: Kramarz, Lollivier, and Pele (1996)). For the segregation index of log wages, from Figure 4 it appears that the coefficient remains essentially unchanged for West German industries, while for East German industries it gradually increases and the slope especially becomes steeper after 1999. The numeric values in East and West Germany were correspondingly 0.213 and 0.1579 in 1990 and were 0.303 and 0.179 in 2001. Again, the trend of the East German index is remarkably higher than that for West Germany. Referring to other western data, the coefficient has increased in France from 0.36 in 1986 to 0.44 in 1992 (the same data source as above). Davis and Haltiwanger (1991) list the index values for wages in US manufacturing production sector. The coefficients were 0.76 in 1975 and 0.80 in 1986, a very low trend. The coefficients show no trend in Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 147 segregation among non-production workers either. The segregation index was 0.447 in 1975 and 0.443 in 1986. Although wages, experience, and education are all imperfect indicators of skill, together they demonstrate a consistent increase in segregation. It therefore can be concluded that such remarkable increase in East German inter-industry segregation indices wasn’t entirely driven by the increase in skilled labor demand, sufficiently qualified to work in western type industries. Hence, the “technology shock” wasn’t skill driven, rather the labor had to adjust in response to the shock by moving from the industries that became less competitive to the ones that were offering better skill return. This adjustment wasn’t costless however in terms of the training time and the attainment of less prestigious occupations in western industries as Figure 4.1 shows. 4.4.The Decomposition Method I further conduct decomposition of East German workers wage variance to find out the extent to which the distribution of personal and household characteristics, such as education, age, experience, sex, number of children, marital status, etc. and their corresponding estimated coefficients, as a proxy for labor supply, and the sectoral and occupational distribution of labor as a proxy for labor demand schedule influence the dynamics of wage inequality during the post unification period. There are many studies that propose different approaches and procedures for estimating changes in overall wage inequality, but it is difficult to compare all these methods or Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. their results because of their computational and methodological differences. For this reason I exploit two quite different, but generalized approaches, namely the unified, or weighted Oaxaca approach, and the cellular, or dummy variable approach, developed in Lemieux (2002). The results are compared against each other in terms of how well they interpret the changes in the wage distribution. This kind of approach enables examining the robustness of the alternative methods, thereby increasing the confidence in results attained. The choice of these two particular methods is based on the fact that first, both are unified approaches as they incorporate decomposition with re-weighting of the characteristics distributions, second they are quite different in terms of their methodology, namely, while the former method is based on decomposing via variances and covariances of the regressors, the latter is essentially based on decomposing through mean values of the cells, third, both methods are applied on the results of the same regression, where the explanatory variables are the industry and occupational dummies, allowing to track the inter- and intra-industry dynamics of inequality and occupational mobility41. The first, developed in Yun (2001), is essentially a weighted Blinder-Oaxaca (Blinder (1973), Oaxaca (1973)) approach that unifies the decomposition in Juhn, Murphy, and Pierce (1993) and weighting proposed in Fields (2001). Another modification is that rather than proceeding along the lines of the regular Blinder- 4 1 Another approach that could be o f interest to apply in this essay, where essentially self selection is considered, is the propensity score estimation method. For example, Hirano, Imbens, and Ridder (2000) propose a weighted estimator to estimate the impact of binary treatments, such as, training programs. The use o f household data, however, doesn’t allow to identify the treatment group at industry level. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 149 Oaxaca procedure, observing the difference in the wage levels, it examines the change variance between two periods. According to Juhn, Murphy, and Pierce 91993), given the regression equation K -i y« = P0 , + X Pk,xk u + £u > (4.4.i ) k=\ changes in wage inequality for year t come from three sources: changes in the distribution of individual characteristics (i.e., changes in the distribution of the Xi’s ), changes in the prices of the observable skills (i.e., changes in the P’s), and changes in the distribution of the residuals e *, where yu =ln(wit) are log individual monthly wages in a particular year. Rewriting (2) in variance and covariance terms gives o '2O '/) = Z cr(Pkxk J i) + <r(e,y,)> (4-4.2) k = \ where if equation (4.4.1) is estimated using OLS method, then a(ey) = (r2(s). Each individual characteristic in (4.4.2) is assigned a relative inequality weight, defined in Fields (2001)42. The weights are computed using the estimated values in (4.4.1) according to the following equation 4 2 In Fields (2001) this relative weights are proposed and applied to analyze the wage inequality change over time. However, no decomposition is proposed. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 5 = o(<Pkxkiy) _ Pk°(xk)p(xk>y) k <r(y) 150 (4.4.3) where cr(xk) is the standard deviation of X k and p(xk,y) = E^Xk,y} is the 0(xk)cr(y) correlation between X k and y. The first unified decomposition method with relative inequality weights therefore takes the following form: o -2 (y, ) - o -2 (yj) = Z [ p ^ k , )pi.xk i » y M y * ) - Pk p {x k ] )p(xk J > y, M t , )] * = i K - \ - + Z \PHffi.xk l)pi.xtl >y,)<r(y,) - (4AA) k = 1 + k 2( ^ ) - o - 2(^ )], K - 1 where A T is the index for the error term, y = /?o y + ^ f i jtxk i +e, are the counterfactual k=\ wages, and i and j are the corresponding cross sections. For the purpose of this study i is for year 2000 and j for 1992. In (4.4.4) the first, second and last terms are correspondingly the characteristics effects, coefficients effects, and residuals effects to changes in wage variance. The second, cellular, or dummy variable approach, proposed in Lemieux (2002), is again a unified approach and is also based on and can be viewed as a generalization of Oaxaca-Blinder method. Rather observing the difference in the wage levels, the change in wage variance between two periods is decomposed in Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 151 terms of differences in means across and within the cells43. It is shown that the set of OLS estimates bt (with no intercept term) from equation (2), where x^u k=l,..,K is an exhaustive set of dummy variables, are just the sample means of yit in each cell k. The sample proportions, equal to the xk t average values in each cell, completely describe the changes in the distributions of x’s. The variance of wages is thus decomposed into between- and within-group components. The total decomposition of variance is described by the following equations: The first term on the right hand side is the contribution of changes in bt to changes in variance: the second term is the contribution of changes in the distribution of x’s (i.e. sample proportions O ut) to both between and within group variances 4 3 The proposed procedure unifies elements from Juhn, Murpy, and Pierce (1993) and DiNardo, Fortin, and Lemieux (1996). (4.4.5) (4.4.6) k v; - v; = X 0b iy b - y, ) 2 - z O k * O '* - y b , f + Z ( * * - ^ K > ( 4 -4 -7 ) k k k Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 152 and the last term represents the effect of changes in the residual variance within each cell (4.4.8) k The main difference between (4.4.4) and (4.4.5) is that while the former method is based on decomposing via variances and covariances, the latter is based on decomposing via means of the cells. 4.5. The Results of the Decompositions and their Interpretation This section proceeds with applying the unified wage inequality decomposition methods described in the previous section using data on East German employed men to preserve maximum continuity in observations44. First, OLS regressions are run for 1992 and 200045. Using the estimated coefficients, as a second step, wage variances are decomposed for each year by both, weighted Oaxaca and cellular (T. Lemieux) methods. Tables 4.3 and 4.4 report these results. 4 4 Estimations using data on men versus women are more accurate as women generally have more discontinuous employment records due to childbirth, child rising, and other household related circumstances. Since my focus is on the effect o f the “western” technology shock, I only use data on men. 4 5 The dependent variable is the log monthly wage rather than the log hourly wage, as most of the workers’ compensations are in terms of monthly salaries in Germany. The (dummy) explanatory variables are age group 16 to 35, age group 46 to 55, education less than high school, high school education, education more than high school, agriculture, energy industry, mining, manufacturing, construction, trade, transport, bank insurance, services (industry), professional technical (occupation), managerial, sales, services (occupation), operational- clerical, transportation, labor, craft, farming, private household, married, no children, two children, working hours (20th percentile), and working hours (99th percentile). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. From Table 4.3 it can be noted that there was a substantial increase in wage inequality over the period. The variance of log wage has increased from 0.345 in 1992 to 0.696 in 2000, or by 101.7%. The variance of the residuals has increased as well, from 0.266 in 1992 to 0.526 in 2000, or by 97.7%. Hence, the increase in the variance of residuals, or in the within group, intra-industry inequality, explains 74.1% in the increase in total variance of log wages46. The remaining portion of the rise in inequality is due to the increase in between group, or inter-industry inequality. As it was discussed in section 4.3, the latter is explained by the increase in inter industry skill and wage segregation. This high percentage increase in within group variance supports the results in Galor and Tsiddon (1997), who show that a faster rate of technological progress may change the observed wage profile within a sector since it changes the quality mix of those who remain behind. This effect is particularly apparent in this empirical analysis, where there is a “sudden” and “significant” technology shock. Our results support also the findings in Galor and Moav (2000), where they demonstrate that ability-biased, which certainly is the case of German reunification, as opposed to education-biased technological change enables a simultaneous account of a rise in between and within groups inequality. 4 6 This is perhaps not surprising since standard regressors such as experience and education account for a relatively small fraction the variance of the wages (R-square is typically in the 0.2-0.3 range). Similar results are obtained for US data. For example, Juhn, Murphy, and Pierce (1993) first documented that residual, or within-group inequality accounts for most o f the growth in wage inequality. In other words, dispersion in the residuals in a standard Mincer wage regression model is shown to grow much more than that of the explainable fraction o f wages. Lemieux (2003) arrives at a similar conclusion using US data over the period between 1973 and 2002. Acemoglu (2000) also argues that the postwar US economy data shows that increases in within group (residual) inequality accounts for much o f the inequality rise, especially starting from 1970s. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 154 Table 4.3 shows that while high school and less than high school education were contributing to the earnings inequality negatively in 1992 and the latter by -11.5%, Table 4.3. Wage decomposition by year, weighted Oaxaca method Variable 1992% Total 2000% Total Age group 16 to 35 0.004 5.6 0.004 2.6 Age group 46 to 55 0.001 0.8 0.000 0.3 Education less than high school -0.009 -11.5 0.004 2.9 High school education 0.000 -0.5 0.004 3.0 Education more than high school 0.010 13.1 0.020 14.1 Agriculture -0.001 -0.9 -0.001 -0.8 Energy industry 0.002 2.8 0.004 3.2 Mining 0.000 0.3 0.004 2.6 Manufacturing 0.004 5.3 0.020 14.3 Construction 0.019 25.7 0.012 8.7 Trade 0.001 1.7 0.005 3.2 Transport 0.002 2.4 -0.001 -0.5 Bank insurance 0.002 2.9 0.002 1.5 Services 0.007 9.7 0.027 19.4 Professional technical occupation 0.000 0.5 0.004 2.6 Managerial occupation -0.002 -3.1 -0.004 -3.0 Sales occupation 0.001 0.7 0.002 1.3 Services occupation 0.000 0.6 0.000 -0.1 Operational clerical 0.000 0.5 0.001 0.5 Transportation, labor, craft 0.000 0.6 0.000 -0.2 Farming private household 0.000 0.0 0.000 0.0 Married -0.002 -2.8 0.005 3.2 No children 0.000 0.3 0.004 3.0 Two children 0.003 4.1 0.001 0.5 Working hours (20th percentile) 0.027 36.6 0.023 16.2 Working hours (99th percentile) 0.003 4.5 0.003 1.9 Sum of variables 0.075 23.0 0.142 24.5 Residual 0.266 77.0 0.526 75.5 Log monthly wages 0.345 100.0 0.696 100.0 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 155 their effect on wage inequality was positive in 2000. At the same time more than high school education has had significant and positive effect on wage inequality in both years. There were significant changes in other industry and occupation related variables contributions to the earnings inequality. For instance, while construction Table 4.4. Wage decomposition by year, cellular method Variable 1992% Total 2000 % Total Age group 16 to 35 0.338 11.0 0.180 5.5 Age group 46 to 55 0.160 5.2 0.117 3.5 Education less than high school 0.014 0.5 0.086 2.6 High school education 0.284 9.3 0.450 13.6 Education more than high school 0.047 1.5 0.194 5.9 Agriculture 0.013 0.4 0.008 0.3 Energy industry 0.002 0.1 0.002 0.1 Mining 0.018 0.6 0.002 0.0 Manufacturing 0.074 2.4 0.056 1.7 Construction 0.034 1.1 0.026 0.8 Trade 0.024 0.8 0.021 0.6 Transport 0.015 0.5 0.044 1.3 Bank insurance 0.001 0.0 0.004 0.1 Services 0.069 2.3 0.141 4.3 Professional technical occupation 0.116 3.8 0.108 3.3 Managerial occupation 0.165 5.4 0.292 8.8 Sales occupation 0.017 0.6 0.014 0.4 Services occupation 0.022 0.7 0.042 1.3 Operational clerical 0.027 0.9 0.015 0.4 Transportation, labor, craft 0.278 9.1 0.226 6.9 Farming private household 0.010 0.3 0.007 0.2 Married 0.540 17.7 0.483 14.6 No children 0.283 9.3 0.520 15.7 Two children 0.122 4.0 0.066 2.0 Working hours (20th percentile) 0.342 11.2 0.149 4.5 Working hours (99th percentile) 0.042 1.4 0.051 1.5 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 156 was the major contributor to wage inequality in 1990 at 25.6%, its effect was much less, only 8.7% in 2000. Similarly, manufacturing and services were the biggest contributors to the earnings inequality in 2000. The effect of working hours, 20th percentile has declined from 36.6% to 16.2% over the period. Table 4.4 however, provides different results than those listed in Table 3. In this case high school education variable has the strongest effect among the other two educational groups. While the role of construction is much less, 2.4% and 1.7% in 1992 and 2000 respectively, professional-technical and managerial occupations have much stronger effect, the former has grown from 5.4% to 8.8% while the latter has declined from 3.8% to 3.3% over the period. Both tables however, show that among the younger age group, 16 to 35 wage was less equally distributed than among the mid age group, 46 to 55. Among the married workers inequality has seen an increase. Those with no children had less equal wage distribution than those with two children in 2000. Table 4.5 lists together and compares the results attained from each method, including the total, coefficient, characteristic, and residual effects. While some of the effects have similar signs for the same explanatory variable, others show different signs and magnitudes when using the two different methods. Hence, there is inconsistency in these methods, as these different decompositions generate quite different interpretations for characteristic and coefficient effects. Particularly, from Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 157 Table 4.5. Wage decomposition, comparison of weighted Oaxaca and cellular methods Decomposed: Log Monthly wages Lem Oax Lem Oax Lem Oax Lem Oax List of Variables Total % Coeff % Caract% Resid % Age group 16 to 35 -63.8 -0.7 -6.4 1.7 <o ■ M ; C \l : I -10.2 0.5 Age group 46 to 55 -17.2 -0.3 -7.1 0.0 5.5 -1.2 1.9 Education less than high school 29.1 19.0 1.7 0.8 1 6 4 64.3 -0.1 High school education 67.3 6.8 3.6 17.9 -18 8 -57.6 17.9 Education more than high school 59.3 15.2 -1 4 7.0 34 3 22.3 4.8 Agriculture -1.8 -0.8 -0.3 0.4 -0 5 -4.8 0.0 Energy industry -0.2 3.6 0.1 3.8 -01 -4.6 -0.1 Mining -6.5 5.2 0.0 4.4 -0.4 -1.3 -1.4 Manufacturing -7.5 24.4 -2.3 20.3 4.3 -5.6 -0.6 Construction -3.0 -10.3 -0.1 -9.8 1.8 8.0 -1.2 Trade -1.2 5.0 -1.4 6.1 3.7 -10.1 0.1 Transport 11.8 -3.8 -0.4 0.1 -5.3 -14.2 4.8 Bank insurance 1.1 -0.2 0.0 0.0 -0.1 -0.5 0.3 Services 28.9 30.1 -3.0 29.5 14.2 -27.4 5.6 Professional technical occupation -3.1 4.9 -7.1 4.0 13.2 -0.5 2.8 Managerial occupation 51.2 -2.8 -5.5 -5.2 5.6 13.6 16 2 Sales occupation -1.5 1.8 -1.2 1.8 3.0 -1.9 0 0 Services occupation 8.2 -0.9 -0.8 -0.4 1.9 -1.7 2.2 Operational clerical -4.8 0.4 -2.4 0.0 4.5 1.2 0.0 Transportation, labor, craft -20.8 -1.2 -9.9 -0.7 1 5 7 -1.4 0.9 Farming private household -1.2 0.0 -0.7 -0.1 1.2 0.5 0.1 Married -22 9 9.9 -34.6 5.8 24 8 9.1 24.1 No children 95.8 6.1 -13.6 5.0 68 3 -0.8 15.8 Two children -22.6 -3.6 -2.8 -0.9 -9.1 -8.9 0.5 Working hours (20th percentile) -78.1 -6.7 -1.8 8.2 -67.1 -61.9 4.5 Working hours (99th percentile) 3.7 -1.0 -2.6 0.2 10 7 -4.4 0.3 the table, although the total effect of age groups, 16 to 35, and 46 to 554 7 is negative, its magnitude is much higher in cellular case. Mining, manufacturing, trade, professional and technical occupations, sales, farming, and married dummy variables have opposing signs for the total effect. Similar sign discrepancy is observed when 471 purposely omit age 36-45 group to avoid the “cohort” effect, when one generation becomes older over time and enters into the next age subgroup. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 158 comparing selected variables for the coefficients and characteristics effects on wage inequality dynamics. For instance, while the total effect is positive for the education related variables, the coefficient effect is positive when weighted Oaxaca method is applied, but negative when the cellular method is applied for more than high school education variable, meaning that according to the former the wage range has increased in this subgroup while according to the latter it has actually decreased. Another example is manufacturing industry. According to weighted Oaxaca method the characteristic effect is negative at -5.6% while according to the cellular model it is positive at 4.3%. This means that according to the former there was an outflow of workers from this industry, while according to the latter there was actually an inflow. I explain these differences in decomposition results through the differences in the decomposition methods and the way that the weights have been applied in these decompositions. First, as mentioned earlier, while the weighted Oaxaca approach is based on decomposing by means of variances and covariances, the cellular approach is based on decomposing by means of average values in cells. Second, the way that the relative weights are accounted for characteristics and coefficients effects calculations is quite different in these methods. Namely, the second term of the sum in (4.4.4) and (4.4.5) contributes to the price effect, or the effect of the change of coefficient values on wage distribution, given the distribution of labor population is unchanged at its 2000 level. While in weighted Oaxaca method the weight effect is captured by the change in covariances between the variables and the log wage, in the cellular method the weight effects are captured solely by the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 159 change in variance of average wages (the sum of the squared difference between average wages in each cell and the average wage of the whole sample). Third, in a similar note, the first term of the sum in (4.4.4) and (4.4.6) is accounted for the characteristics effect, when the coefficients are held unchanged. Again, there is a difference in the use of weights in these methods. In weighted Oaxaca method there is no re-weighting for the distribution of prices and rather the 1992 prices, or the estimated coefficients are used. In the cellular method, however, as a first step, the between and the within, or residual effects are separated, and second, while no weight is assigned for the latter, the former is weighted by the between cell variances of actual and counterfactual wages 1992, where again, for the counterfactual case the variance is over the wages in 1992, while the distribution is that of 2000. Finally, as in each case the total effect is the sum of the characteristics, the coefficients, and the residuals effects for each variable, and these effects are different, the total effect for each variable is also different. 4.6. A Self-Selection Model In this section a self-selection model is considered, where East German workers decide whether they would like to receive re-training and become qualified to apply for jobs in the advanced, western industries, or firms that offer higher employment Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 160 compensation on average, or to stay with the conventional industry offering only moderate compensation schedule. In order to identify the industries that have experienced inflow of qualified workforce due to offering higher benefits, and the ones, that have experienced labor outflow due to becoming less competitive and being unable to offer competitive wages I suggest the following approach. The first step is to carry wage decomposition, where the explanatory variables are the industry and occupation dummies of interest. Second, from the decomposition results select the industries, or the occupations having positive total, or characteristic effects and those having negative effects on wage inequality into two subgroups. If the industry has negative characteristic effect, then there was an outflow of qualified labor to other industries, thus reducing the wage inequality in the sector. If, on the other hand, the effect was positive, then there was an inflow of qualified labor from less competitive, moderate pay industries. Third, use these labor inflow or outflow effects to construct an industry indicator binary function, where it takes value one if the industry has experienced an inflow of labor, and it takes a value of zero if the industry has experienced an outflow of labor. Forth, regress the indicator function on selected individual and household background characteristics using logit procedure. Based on the regression results one can explain how these individual characteristics have affected the decision to move to an advanced industry. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission o f the copyright ow ner. Further reproduction prohibited without p erm ission. Industry, occupation Lem Select Oax Lem Select Oax Lem Select Oax Lem Total % Coeff % Caract % Agriculture S !i:E ! ;§ f l l i i l fill® i l i i i i i + - i l l l i l l ■ llffi + Energy industry 11 WMk ++ + + ++ - i l l l l i l U llB I Mining -- Il f ! B ill ++ f l'iil + ++ - i i i B t l p lliii - Manufacturing -- + ++ -- + ++ ++ iilllm li l lllii - Construction - iiy illlilt; -- - i l j i i l If -- ++ + ++ - : Trade i i l i l l ! + ++ - t i l l ! ! ! ++ ++ + ■ ■ ill + Transport ++ + - M ill § 1f llg lS + -- f i l i a l ! i i i i i ++ Bank insurance + + - + l l i l l l l l : - - : i i i i i i i l l l l f i + Services ++ + ++ - a i i i s i ++ ++ + - ++ Professional technical occupation -- l l l l l l l i ++ -- l i l i u ++ ++ + H l i l ++ Managerial occupation ++ + -- -- i i i i i i ' i t - ++ + ++ ++ Sales occupation - + + - fjiiiiiiiiii1 + ++ + — + Services occupation ++ + - - llfM llll - ++ + |S li; l ++ Operational clerical -- - + - + ++ + + - Transportation, labor, craft — - - — - + l i l t ! + Farming private household 1 1 :-: : ; ■ _ l l E l - - - + + + + Table 4.6. W age decomposition, weighted Oaxaca a n d cellular methods comparison 162 Table 4.6 lists the industry and occupation variables and compares the signs of the decomposition effects attained by weighted Oaxaca and or the cellular approaches. If the signs do not coincide, then the sign corresponding to the highest absolute value is chosen. I construct the industry indicator function based on the selection results listed in the table4 8. The following self-selection model is used to choose the regression variables for step four. According to the model, workers decide either to receive re-training to become qualified applying to the jobs in a competitive industry offering higher wages, or to stay in the conventional, less competitive and lower pay sector. The work in the advanced sector promises higher returns to education and experience, whereas working in the conventional sector is not rewarding for education and experience. Receiving re-training is costly and requires financial and time commitment. The choice is then solely based on workers’ educational background, work experience, individual and household characteristics. It is assumed that the market is heterogeneous and the population is diverse, x, is denoted as the observable fraction of individual r’s human capital including educational background, work experience, and other characteristics, r, is a random variable reflecting individual f s unobserved fraction of earnings capabilities. It is 4 8 According to the results listed in Table 6, the following lists o f variables are obtained: Total effect- negative: agriculture, energy, mining, construction, professional-technical, operational-cleric, transport (occupation), laborer, craftsman; positive: manufacturing, trade, transport (industry), banking-insurance, managerial, sales, services (occupation). Characteristics effect- negative: agriculture, energy, mining, manufacturing, transportation (industry), banking-insurance; positive: construction, trade, services (industry), professional-technical, managerial, sales, service (occupation), operational-clerical. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 163 assumed that infinitely lived agents’ gross lifetime utility is the present value of wage earnings49. There are two sectors in the economy, conventional and advanced, indexed accordingly as j=l,2. Individual lifetime earnings of workers from each sector are denoted by VJ t = V(xJt,r ,) , xJ t = (y7,x,), where y } j=l,2, corresponds to labor market conditions. In the initial period agents decide on the sector they would like to be employed. While there is no cost involved joining the conventional, lower-pay sector, there is a cost associated with joining the advanced sector, Cl2 i = C (z,,^l2i) , where zi is the set of individual specific variables affecting the cost (the most capable workers adjust easily, while for less capable workers this adjustment process is much costlier), and (pni is the unobserved cost of changing jobs. Lets assume it takes st periods to change jobs. Consequently, the net expected future earnings from applying to the advanced sector are V2 l - Cw . Workers choose to work in the advanced sector, if V2 i - C1 2 ; > Vu . It is assumed that expected earnings grow in both sectors exponentially over time according to the following rule: wji ( 0 = Wjt e x P(SjiO > ( 4 .6 .1 ) where W y , are the starting wages when t=0, and g ji,j= l,2 are the wage growth rates in each industry. 4 9 Given the increase in uncertainty with the entry of East German workers into the market system it is reasonable to assume that agents face infinite horizon. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 164 It is further assumed that for each person i there is a constant discount rate , where rt > max(glpg 2(). Also, the costs of changing jobs enter into the utility function exponentially. Under these assumptions, the net utility of changing jobs from conventional to the advanced sector will be: u 1 2 i = v2 i - C1 2 ( = JW 2 < exp(g2 , (t - st)) exp(-r,t)dt (4.6.2) w. -exp (-r,s,). rt ~ S 2 , The net utility of staying in the conventional sector is: c o u u i = vu = Jw1 ( exp(g1 ( f) exp(-rtt)dt (4.6.3) Define /, = In Su - r, = In ( u ^ [ u niJ = In / ri -Sli ^1/ ri - S i i exp (rrtst) (4.6.4) where if Uw » UUl, then /, « 0 otherwise, I { > 0. Taking Taylor series approximation around the population mean values of (g\,g2,r) , yields (see the Appendix for the computational details): Ii=aQ+ In w2 l - In wu + a xg u + a 2g2 i + a 2 rt . (4.6.5) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 165 Assume base wages and the wage growth rates are linear in labor market conditions and individual capabilities (human capital): \n.Wji=ylJxXJi+iiji, j=l,2 (4.6.6) gji = r2,*2i/+ V (4-6.7) As workers’ individual discount rate is a function of individual characteristics and family circumstances, it can be expressed accordingly, (4.6.8) Equations (4.6.5) - (4.6.8) then constitute the structural form of the econometric analysis. Substituting (4.6.6) - (4.6.8) into (4.6.5) then yields the reduced form that can be estimated using logit procedure: h = A > + £ i • (4.6.9) The proposed model is consistent with the theoretical argument that technological advancement increases income inequality, applied to the case of German re-unification. However, rather concentrating on the skill based technology advancement argument, the interest is on the workers decision whether to join the advanced sector or not, based on their individual and household background. More specifically, the analysis focuses on general as opposed to vocational education and the goal is to estimate whether there is a significant labor movement from older to advanced, western industries based on this educational background. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 166 Such a result for instance, is obtained for US data in Bartel and Sicherman (1999). They match a variety of industry-level measures of technological change to a panel of workers observed between 1979 and 1993 and examine the role of observed and unobserved worker heterogeneity, such as education and experience. Their results show that the wage premium associated with technological change is primarily due to the sorting of more able workers into those industries. They explain this increase in premium in terms of increased demand for innate ability or other unobserved characteristics of more educated workers. In a similar note, Krueger and Kumar (2004) develop a theoretical model of technology adoption where it is shown that the skill-specific, or vocational versus general education policies largely explain US-Europe growth gap increase in 1980s and 1990s when new technologies emerged at a more rapid pace than in the earlier decades of 1960s and 1970s. Our empirical analysis therefore tests this hypothesis using the East German household database by examining whether there is a link between workers choice of either old-conventional, or western industries to work, based on their educational backgrounds, general or higher education as opposed to vocational education. Table 4.7 provides the logit estimation results for equation (4.6.9), where the dependent variable is the industry and occupation indicator function as discussed earlier and the explanatory variables are years of education, experience, square of experience, log net income last month, employment level (part time or full time employment), number of children, gender, annual work hours, and dummy variables Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 167 for higher (includes, college, university, technical higher, etc.), secondary vocational5 0, and secondary general educations. Quite interestingly, the results show that indeed individual and household characteristics play a significant role in workers occupational decisions on whether to move to an advanced industry. Particularly, the coefficients for years of education and higher education are positive and significant when either the industry indicator Table 4.7. Logit estimation of the indicator function I(i), Employed (both part- and full- time) I Total effects Characteristics effects Explanatory variables Coef. z P>|z| Coef. z P> z| Years of education 0.1375 10.94 0.0000 0.1361 10.87 0.0000 Experience 0.0104 0.91 0.3610 0.0160 1.43 0.1510 Square of experience 0.0109 0.45 0.6490 -0.0260 -1.11 0.2650 Log net income last month 0.7123 14.24 0.0000 0.8326 16.61 0.0000 Employment level of individ. 0.0848 0.91 0.3640 0.4221 4.56 0.0000 Number of children -0.1579 -5.16 0.0000 -0.1891 -6.26 0.0000 Gender 1.1497 20.82 0.0000 0.9501 17.99 0.0000 Annual work hours 0.0000 -0.23 0.8210 0.0001 1.63 0.1040 Higher education 1.1599 5.13 0.0000 1.1583 5.17 0.0000 Secondary vocational educ. -0.5698 -4.18 0.0000 -0.3880 -2.82 0.0050 Secondary general educ. 0.2492 0.43 0.6710 -0.2017 -0.39 0.6940 Constant -9.8429 -17.74 0.0000 -11.3917 -20.40 0.0000 Number of observations 9586 9586 Pseudo R-square 0.0989 0.0941 Employed (both part and full time) Gender: Male-1, Female-2 Employment level: Full time-1, Part time-2 5 0 Under this category fall vocational training, practical training, vocational without apprenticeship, apprenticeship, specialized vocational school, specialized technical school, school o f health care, civil servant training, other vocational education. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 168 I(i) is constructed using the total effects or the characteristics effects results of the wage decomposition.. The coefficients are negative and significant for number of children and secondary vocational education. This last result is interesting as it shows that while those workers who have received higher education have a positive impact on labor migration to western industries, those with vocational background have largely stayed in older industries, thus having a significant and negative impact on it. Similar regressions are conducted for full-time and part-time subgroups. The results are provided in Tables 4.8 and 4.9. Table 4.8 shows the results don’t alter much if only full-time employees sub-sample is estimated. Table 4.8. Logit estimation of the indicator function I(i), Full-time employed Total effects Characteristics effects Explanatory variables Coef. z P>|z| Coef. z P>|z| Years of education 0.14225 9.78 0.0000 0.1365 9.40 0.0000 Experience -0.00035 -0.03 0.9790 0.0100 0.76 0.4450 Square of experience 0.02316 0.83 0.4050 -0.0237 -0.87 0.3850 Log net income last month 0.99377 14.77 0.0000 1.1803 17.35 0.0000 Number of children -0.14511 -4.13 0.0000 -0.1905 -5.47 0.0000 Gender 1.14057 17.83 0.0000 0.9845 16.11 0.0000 Annual work hours 0.00012 1.84 0.0660 0.0003 4.79 0.0000 Higher education 0.90356 3.48 0.0010 0.8446 3.32 0.0010 Secondary vocational educ. -0.39544 -2.49 0.0130 -0.2600 -1.62 0.1040 Secondary general educ. 0.08506 0.14 0.8870 -0.3609 -0.68 0.4990 Constant -12.89721 -18.33 0.0000 -15.0586 -21.06 0.0000 Number of observations 7619 7619 Pseudo R-square 0.1042 0.1081 Gender: Male-1, Female-2 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 169 role in deciding on moving to an advanced industry. It is interesting to note that while for the whole sample as Table 4.7 shows the coefficient of annual work hours is insignificant, for full-time workers’ case it is positive and significant Table 9 shows similar results when only part-time employees sub-sample is considered, but in this case the coefficient for annual work hours is negative and significant. Therefore, it can be concluded that those who prefer working full-time are interested in moving to the advanced sectors, while those who prefer working part-time are interested in staying in the conventional sector. Table 4.9. Logit estimation of the indicator function I(i), Part-time employed Total effects Characteristics effects Explanatory variables Coef. z P>|z| Coef. z P>|z| Years of education 0.0689 2.53 0.0120 0.0633 2.35 0.0190 Experience 0.0218 0.94 0.3490 0.0115 0.51 0.6110 Square of experience 0.0026 0.05 0.9580 -0.0065 -0.14 0.8910 Log net income last month 0.4693 5.30 0.0000 0.5902 6.74 0.0000 Number of children -0.2330 -3.58 0.0000 -0.2288 -3.59 0.0000 Gender 1.3771 12.00 0.0000 1.1226 10.02 0.0000 Annual work hours -0.0003 -2.00 0.0450 -0.0004 -2.99 0.0030 Higher education 2.0685 3.85 0.0000 2.2943 3.78 0.0000 Secondary vocational educ. -1.2843 -4.38 0.0000 -0.9575 -3.33 0.0010 Secondary general educ. Constant -6.6930 -8.01 0.0000 -7.0266 -8.52 0.0000 Number of observations 1964 1964 Pseudo R-square 0.1171 0.0936 Gender: Male-1, Female-2 Note: General educ is dropped due to insufficient amount of general education in part time data Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 170 4.7. Concluding Remarks The purpose of this essay is to analyze on the empirical basis whether there is a relationship between individual and household characteristics and the dynamics of the labor income inequality and specifically, the extent with which individual and household characteristics have contributed to the workforce occupational choice and movements from one industry to another in East Germany after the reunification. My focus is largely on analyzing the labor supply effects. The German Socioeconomic Panel (GSOEP) household survey data for the period from 1990 to 2001 is used for the analysis. After reunification there was a substantial shift in labor demand from traditional manufacturing to trade, services, and finance as West German firms’ investments in East Germany were predominantly concentrated in these industries. To find out which industries were exactly the ones having the most of the influx of western investments and hence the ones offering the highest incentives to East German workers in terms of higher returns to skills, education and experience, I have conducted wage decompositions with industry dummy variables as the regressors. Conventionally, decompositions are mostly conducted for the purpose of tracking the effects of workers’ education and experience on the dynamics of wage inequality. In this respect my approach of decomposing wages by industry is innovative. It allows me to identify the industries that have contributed to the increase in inequality as well as those offering the highest incentives for skilled workers. The use of Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 171 sectoral decomposition reveals also how and by how much education, experience, and wages influence inter-sectoral skilled labor movements from the sectors offering the least incentives to the ones offering the most. Two distinct generalized wage decompositions are applied, namely, Blinder- Oaxaca and “cellular” methods. The results of these two methods are generally quite similar. Following these decompositions results, a self-selection model in which workers decide either (1) to invest time and resources to get re-trained and become qualified for the ‘western’ jobs offering higher returns to skills or (2) to continue working in the conventional, lower-pay occupations and poor prospects for improvement is applied. An indicator function, outlining those industries where there were significant labor outflows and those ones experiencing significant labor inflows is constructed based on the wage decompositions results. This indicator function is estimated using logistic procedure on general, vocational, and higher education, work experience, number of children, and other individual background variables to find out the effects of these individual labor aspects on the changes in labor supply schedule and wage inequality dynamics over the period. The analysis shows that individual and household characteristics play a significant role in the adjustment of labor to the new, western market. While wages are largely set by industries, determined by the industry specific needs for qualified labor, wage inequality and its dynamics are largely influenced by individuals’ education, work experience, household, and other related background. Specifically, the results show a significant, negative effect of vocational education on inter Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. industry labor movements. While, the effect of secondary general education is insignificant, the effects of higher education and monthly salaries are positive and significant. Surprisingly, while the effect of experience is positive, it is not statistically significant. Surprisingly also, the results show that women have moved to ‘western’ industries to a greater extent than men. The coefficient of number of children is negative and significant. Hence, the results indicate that the adjustment of East German workers to the inter-industry shifts in labor demand was largely based on and affected by their individual characteristics. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 173 CHAPTER 4 BIBLIOGRAPHY [1] D. Acemoglu (2000), “Technical Change, Inequality, and the Labor market,” NBER Working Paper No. 7800. [2] A. Bartel and N. Sicherman (1999), “Technological Change and Wages: An Interindustry Analysis,” Journal of Political Economy 107 (2), 285-324. [3] M. Blis and P. Klenow (1998), “Does Schooling Cause Growth or the Other Way Around?,” NBER Working Paper No. 6393. [4] A. Blinder (1973), “Wage Discrimination: Reduced Form and Structural Estimates,” Journal of Human resources 8,436-455. [5] M. Daly and R. Valletta (2000), “Inequality and Poverty in the United States: The Effects of Changing Family Behavior and Rising Wage Dispersion,” Mimeo, Federal Reserve bank of San Francisco. [6] S. Davis and J. Haltiwanger (1991), “Wage Dispersion Between and Within US Manufacturing Plants, 1963-86,” Brookings Papers on Economic Activity: Microeconomics, 115-180. [7] J. DiNardo, N. Fortin, T. Lemieux (1996), “Labor Market Insititutions and the Distribution of Wages, 1973-1992: A Semiparametric Approach,” Econometrica 64 (5), 1001 - 1044. [8] G. Fields (2002), “Accounting for Income Inequality and its Change: A New Method With Application to the Distribution of Earnings in the US,” Mimeo, Cornell University (Forthcoming: Research in Labor Economics). [9] O. Galor and O. Moav (2000), “Ability-Biased Technological Transition, Wage Inequality, and Economic Growth,” Quarterly Journal of Economics 115(2), 469-497 [10] O. Galor and D. Tsiddon (1997), “Technological Progress, Mobility, and Economic Growth,” American Economic Review 87 (3), 363-382. [11] H. Ganzeboom and D. Treiman (1996), “Internationally Comparable Measures of Occupational Status for the 1988 International Standard Classification of Occupations,” Social Science Research 25, 231-239. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 174 [12] J. Haltiwanger, J. Lane, and J. Spletzer (1999), “Productivity Differences across Employers: The Roles of Employer Size, Age, and Human Capital,” The American Economic Review 89 (2), Papers and Proceedings, 94 -98. [13] K. Hirano, G. Imbens, G. Ridder (2000), “Efficient Estimation of Average Treatment Effects using the Estimated Propensity Score,” NBER Technical Working Paper No. 251. [14] B. Jovanovic and Y. Nyarko (1996), “Learning by Doing and the Choice of Technology,” Econometrica 64,1299-1310. [15] C. Juhn, K. Murphy, and B. Pierce (1993), “Wage Inequality and the Rise in returns ot Skill,” Journal of Political Economy 101 (3), 410 -442. [16] C. Juhn and K. Murphy (1997), “Wage Inequality and Family Labor Supply,” Journal of Labor Economics 15 (1, Part 1), 72-97. [17] F. Kramarz, S. Lollivier, and L. Pele (1996), “Wage Inequalities and Firm- Specific Compensation Policies in France,” Annales D ’ Economie et de Statistique, Institute National de la Statistique et des Etudes Economiques, No. 41/42, 369-386. [18] M. Kremer and E. Maskin (1996), “Wage Inequality and Segregation by Skill,” NBER Working Paper No. 5718. [19] D. Krueger and K. Kumar (2004), “Skill-Specific rather than General Education: A Reason for US-Europe Growth Differences?” Journal of Economic Growth 9,167-207. [20] T. Lemieux (2002), “Decomposing Changes in Wage Distributions: A Unified Approach,” Mimeo, University of British Columbia. [21] T. Lemieux (2003), “Composition Effects, Wage Measurement, and Growth in Within-Group Wage inequality,” Mimeo, University of British Columbia. [22] J. Mincer (1974), “Schooling, Experience and Earnings,” New York, Columbia University Press. [23] R. Oaxaca (1973), “Male-Female Wage Differentials in Urban Labor Markets,” International Economic Review 14 (3), 693 - 709. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 175 [24] R. Willis and S. Rosen (1979), “Education and Self-Selection,” The Journal of Political Economy 87 (5), Part 2: Education and Income Distribution, S7 - S36. [25] M. Yun (2002), Earnings Inequality in USA, 1961 - 1999: Comparing Inequality Using Earnings Equations,” Working Paper, University of Western Ontario. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 176 5. CONCLUSION This dissertation analyzes both theoretically and empirically the effects of inequality on growth through individuals choices of their investments in human and physical capital, employment, occupation, fertility, as well as by means of income and credit allocation mechanisms of markets. The dissertation consists of three interrelated essays. The first is devoted to broader conceptual theme in inequality, development, and economy-wide perspectives of growth with focusing on a specific group of countries namely, the economies in transition from Central and Eastern Europe and the Commonwealth of Independent States (CIS). Several very essential dimensions of the growth-inequality debate have been re examined in the specific context of this group of countries undergoing transition from centrally planned to market systems. In the beginning of the transition these countries shared many similar characteristics and specifically, had inherited low values of income inequality, similar levels of per-capita GDP and GDP growth rates, they initiated their transitions almost simultaneously, had many common policies and similar transitional objectives. Despite these apparent similarities, significant changes in growth rates and income inequality have taken place both within and across these countries over the subsequent phases of the transition. Hence, they provide a potentially rich experience for examining the relation between income inequality and growth. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. I l l My results, based on this transitional data, indicate a negative relationship between initial inequality and subsequent growth. This finding is rather robust to the use of the different specifications and estimation methods that in the literature have been employed to arrive at very different conclusions on the relation between inequality and growth even when the same datasets are applied. In contrast to these conflicting results for samples, which include no, or at most three transition economies, my empirical findings for transition countries indicate a strong, negative contemporaneous growth-inequality relationship for all of these specifications and estimation methods in the short to medium run. This result is most consistent with Barro (2000), which showed that for low and medium income countries this relationship is negative. Also, my results show a negative relationship between the change in inequality and initial income, and a positive relationship between the squared change in inequality and initial income as it is found in Banerjee and Duflo (2003). Nevertheless, the results do not show a significant relationship between lagged inequality and growth as Banerjee and Duflo (2003) found. In contrast to conflicting results found in the literature on the theme these empirical findings for transition economies indicate a strong, contemporaneous growth-inequality relationship in the short to medium run for similar estimations and specification methods carried out by various authors. In other respects, these results support Kuznets’ inverted - U hypothesis. I would anticipate that the inequality would decrease over time as these countries continue to Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 178 grow and as the initial transitional effects attenuate and the trend effects of economic growth become more dominant. It can be concluded that, in light of the present controversies in literature on the growth-inequality theme, these findings are consistent with the view that the empirical results are sensitive to the specific choice of sample of countries The policy implications of the results are that adept government performance is indispensable for rapid recovery and sustainable growth and that government implements and safeguards the integrity of the financial markets, investment mechanisms, effectively privatizes the industrial and agricultural sectors, and protects property rights, necessary for competitive performance in a market environment in order to maintain a smooth and rapid transition. The implementation speed of those policies was quite different across the transition countries, yielding substantial differences in growth and income inequality. The results suggest that the transition governments need to carefully nurture and protect these institutions and develop essential mechanisms for competitive performance in a market environment. A further extension of the essay would be to analyze the effects of the specific policies conducted by the governments of these countries on the speeds of recovery and output growth rates. The second essay of the dissertation focuses on market imperfections pervasive in the development world and informal institutions that arise to fill in the resulting shortcomings. The shortcomings to perfect market arise in the paper when the assumptions of continuous and convex production technology and of complete credit Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 179 market are dropped, the latter causing moral hazard effects in lending and borrowing. The disparities arise from the importance of asset ownership in access to credit, tenancy of credit markets, e.g., limited access of the poor to credit owing to lack of collateral, and inadequacy in human capital. A critical assumption that distinguishes the essay is that human capital in years of schooling, or entrepreneurial skills can be supplemented for wealth as a collateral in borrowing and for physical capital in production function. This assumption allows analyzing two forms of self-reinforcing equilibria widely debated in the existing literature. The first is the so-called capital versus skill complementarity, which put in terms of the essay’s terminology means human capital in terms of entrepreneurial skills complementarity to inherited wealth in terms of physical capital. The existing theoretical literature on the theme predicts multiple equilibra (associated with differing levels of development starting from poverty traps to rapid economic growth and prosperity) depending on the economic agents decisions and actions. The second form of self-reinforcement arises from the possibility of historically given initial conditions, which within the framework of our paper corresponds to the initial distributions of wealth and human capital among the economic agents. Although this approach is not necessarily based on the phenomenon of the multiple equilibria, given the stochastic nature of the problem setup, the model doesn’t guarantee that the subsequent evolution of the economy may be uniquely defined. Given the presence of these two self-reinforcement forms, the model dynamics is Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 180 considered either at the individual level, or at aggregate level with conducting partial equilibrium analysis. Since little attention has been devoted in the literature to the effects of the accumulation of human capital on entrepreneurial skills, investment, income distribution, and the speed of wealth accumulation, this essay attempts to address these issues, with specifically focusing on the extent to which human capital accumulation affects the level of entrepreneurial activity, investment, earnings distribution and wealth accumulation. The main forms of enterprises are assumed to be sole proprietorships and partnerships, the small to medium sized firms, constituting a significant part of economic activity in developing countries. This essay studies the effect of the initial wealth and human capital distributions on small business startups, credit rationing, and small business income versus wage earnings inequality. In particular, agents’ decisions on child quality-quantity tradeoff (OLG model is considered), i.e. the parental choice of investing in per-child human capital and wealth versus number of children through the channels of household utility maximization and their occupational choices are functions of inequality and income distribution and these choices play an important role in human capital and wealth transition mechanism. This is an alternative approach to those adopted in the literature, which focuses on stochastic transition processes as a wealth transition channel. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 181 The model predicts that with setup costs involved in the acquisition of the entrepreneurial skills, and borrowing constraints for poor agents, the initial distributions of wealth and human capital affect the aggregate skill composition of the economy, entrepreneurship, small business activity, investment, and output. In this respect, poor families might find little incentives in investing in their children’s education, thus locking their descendants into a poverty trap. High initial inequality distribution thus may perpetuate itself, or in other words countries with historically high poverty rates may have persistently low per capita incomes. The numerical calculations generate the following results: Given the initial distributions in wealth and human capital, the smaller is the income inequality gap between labor wages and entrepreneurs’ returns the lower is fertility in the long run. This result is standard and well documented in the literature. Wage effect: (a) The lower are the wages, the higher is the number of potential investors and of those who are credit rationed, (b) The higher are the wages relative to the entrepreneurial return, the higher is the opportunity cost and the threshold human capital level to be qualified for credit among potential borrower investors with the same wealth level. If the initial wealth and human capital are more equally distributed, the higher are the wages, the less is the number of borrower investors, but there are relatively more of credit rationed potential investors. If the initial wealth and human capital are rather unequally distributed, then the lower is the wage worker - entrepreneur income inequality gap, the less is the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 182 number of borrower investors and the less is the number of those credit rationed. Also, in this case, as wages increase, more potential borrower-investors are willing to become workers. With this regard, this effect somehow attenuates the credit rationing mechanism described in the literature. The dynamic effect of this mechanism is even less evident, as the computational experiments reveal. If the initial wealth and human capital distributions are highly unequal, an increase in income inequality raises fertility in subsequent periods. If however, the initial distributions are highly equal, the increase in earnings inequality causes decline in fertility in subsequent periods. The higher is the correlation between wealth and human capital (one may assume rich have more opportunities to receive education) the less is the number of credit rationed. Therefore, these results underline the importance of education and/or training in increasing population human capital as an effective policy towards increasing investment and the number of credible loan allocations. The results also reveal that simply introducing entrepreneurial skills in the production function and considering it as collateral to wealth in borrowing generates qualitatively different results compared to those recorded in literature, failing to consider its effect. My third essay, addresses the issue on how and to what extent has wage inequality changed in East Germany during the post German reunification period from 1990 to 2001. The analysis is based on decomposition allowing to examine whether there is a relationship between individual and household characteristics and Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 183 the dynamics of the labor income inequality and specifically, the extent with which individual and household characteristics have contributed to the workforce occupational choice and movements from one industry to another in East Germany after the reunification. My focus here is largely to analyze labor supply effects. Under the central planning system East German workers were mostly receiving vocational education, leading to an excessive specialization at an early age. Moreover, the socialist system guaranteed job security and workers had no incentives in acquiring skills to cope with unexpected events, such as loss of employment. After the reunification, with the exposure to outside competition, demands for certain skills increased, rendering other types of training obsolete. The education acquired under central planning became less needed under the western, market system, as the old education system was biased toward the hard sciences and engineering, neglecting the social sciences, such as law, business, management, public policy, etc. Under these new conditions East German workers had to solve the dilemma of either receiving training and being qualified for jobs in western industries and firms and get paid wages compatible with the western standards, or remain in the eastern, older industries and firms, offering lower salaries and having lost their perspectives for growth in future. I use data on East German workers based on the GSOEP household survey database over the post-unification period. A sectoral decomposition procedure is conducted to analyze labor movement from eastern, less competitive to western, competitive sectors. After reunification there was a substantial shift of labor demand Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 184 from traditional manufacturing to trade, services, and finance sectors. The use of sectoral decomposition reveals particularly how and to what extent education, experience, and wages have influenced these inter-sectoral labor movements and earnings inequality. Further, wage decompositions are conducted using generalized Blinder-Oaxaca and dummy variable, or “cellular” approaches to further examine how individual and household characteristics, such as education, work experience, number of children, marital status, expectations about future, etc. of East German workers have affected their wage inequality dynamics over the transitional period. The results also allow comparing these two methods. Following these decompositions results, a self-selection model when workers decide either to invest time and resources to get re-trained and become qualified for the newer jobs, or to continue working in the conventional, lower-pay industry is considered. The analysis shows that individual and household characteristics play a significant role in the adjustment of labor to the new, western market. While wages are largely set by industries, determined by the industry specific needs for qualified labor, wage inequality and its dynamics are largely influenced by individuals’ education, work experience, household, and other related background. Specifically, the results show a significant, negative effect of vocational education on inter industry labor movements. While, the effect of secondary general education is insignificant, the effects of higher education and monthly salaries are positive and significant. Surprisingly, while the effect of experience is positive, it is not statistically significant. Surprisingly also, the results show that women have moved Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 185 to ‘western’ industries to a greater extent than men. The coefficient of number of children is negative and significant. Hence, the results indicate that the adjustment of East German workers to the inter-industry shifts in labor demand was largely based on and affected by their individual characteristics. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. COMPREHENSIVE BIBLIOGRAPHY 186 [1] D. Acemoglu (2000), “Technical Change, Inequality, and the Labor market,” NBER Working Paper No. 7800. [2] P. Aghion and P. Bolton (1996), “A Theory of Trickle-Down Growth and Development,” Review of Economic Studies, 64,151 - 172. [3] P. Aghion and P. Bolton (1997), “A Trickle-down Theory of Growth and Development With Debt Overhang,” Review of Economic Studies 64(2), 151-72. [4] Alesina (1994), “Political Models of Macroeconomic Policy and Fiscal Reforms,” in S. Haggard, and S. Webb, eds. Voting for reform: Democracy, Political Liberalization, and Economic Adjustment, (New York, NY: Oxford Univ. Press, 1994). [5] Alesina and R. Perotti (1996), “Income Distribution, Political instability, and Investment,” European Economic Review, 40,1203 - 1228. [6] T. Anderson and C. Hsiao (1981), “Estimation of Dynamic Models with Error Components,” Journal of the American Statistical Association, 76, 598- 606. [7] A. Atkinson and J. Micklewright (1992), “Economic Transformation in Eastern Europe and the Distribution of Income,” Cambridge: Cambridge University Press. [8] C. Azariadis (1996), “The Economics of Poverty Traps Part One: Complete Markets,” Journal of Economic Growth, 1,449-486. [9] B. Baltagi (1995), “Econometric Analysis of Panel Data,” Chichester: Wiley. [10] B. Baltagi and Y. Chang (2000), “Simultaneous Equations with Incomplete Panels,” Econometric Theory, 16, 269-279. [11] A. Banerjee and E. Duflo (2003), “Inequality and Growth: What Can the Data Say?,” NBER Working Paper No. 7793. [12] A. Banerjee and A. Newman (1993), “Occupational Choice and the Process of Development,” Journal of Political Economy 101(2), 274-298 [13] R. Barro (2000), “Inequality and Growth in a Panel of Countries,” Journal of Economic Growth, 5, 5 - 32. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 187 [14] A. Bartel and N. Sicherman (1999), “Technological Change and Wages: An Interindustry Analysis,” Journal of Political Economy 107 (2), 285-324. [15] G. Becker and R. Barro (1988), “A Reformulation of the Economic Theory of Fertility,” The Quarterly Journal of Economics, 103(1), 1 - 25. [16] G. Becker, K. Murphy, and R. Tamura (1990), “Human capital, Fertility, and Economic Growth,” Journal of Political Economy, 98(5), S12 - S37. [17] J. Benhabib and A Rustichini (1998) “Social Conflict and Growth,” Journal of Economic Growth, 1(1), 143-158. [18] D. Bernhardt and H. Lloyd-Ellis (2000), “Enterprise, Inequality, and Economic Development,” Review of Economic Studies, 67,147 - 168. [19] A. Blinder (1973), “Wage Discrimination: Reduced Form and Structural Estimates,” Journal of Human resources 8,436-455. [20] M. Blis and P. Klenow (1998), “Does Schooling Cause Growth or the Other Way Around?,” NBER Working Paper No. 6393. [21] N. Campos and F. Coricelli (2002), “Growth in Transition: What we Know, What we Don’t, and What we Should,” Journal of Economic Literature, Vol. XL, 793-836. [22] N Campos and Y. Kinoshita (2002), “Foreign Direct Investment as Technology Transferred: Some Panel Evidence from Transition Economies,” The Manchester School, 70(3), 398-419. [23] D. Croix and M. Doepke (2001), “Inequality and Growth, Why Differential Fertility Matters,” Working Paper, UCLA. [24] M. Dahan and D. Tsiddon (1998), “Demographic Transition, Income Distribution, and Economic growth,” Journal of Economic Growth, 3, 29 - 52. [25] M. Daly and R. Valletta (2000), “Inequality and Poverty in the United States: The Effects of Changing Family Behavior and Rising Wage Dispersion,” Mimeo, Federal Reserve bank of San Francisco. [26] S. Davis and J. Haltiwanger (1991), “Wage Dispersion Between and Within US Manufacturing Plants, 1963-86,” Brookings Papers on Economic Activity: Microeconomics, 115-180. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 188 [27] J. DiNardo, N. Fortin, T. Lemieux (1996), “Labor Market Insititutions and the Distribution of Wages, 1973-1992: A Semiparametric Approach,” Econometrica 64 (5), 1001 - 1044. [28] W. Easterly (2001), “The middle Class Consensus and Economic Development,” Journal of Economic Growth, 6, 317 - 336. [29] D. Evans and B. Jovanovic (1989), “An Estimated Model of Entrepreneurial Choice under Liquidity Constraints,” Journal of Political Economy, 97(4), 808 - 827. [30] G. Fields (2002), “Accounting for Income Inequality and its Change: A New Method With Application to the Distribution of Earnings in the US,” Mimeo, Cornell University (Forthcoming: Research in Labor Economics). [31] K. Forbes (2000), “A Reassessment of the Relationship Between Inequality and Growth,” The American Economic Review, 90(4), 869 - 886.0. [32] O. Galor and O. Moav (2000), “Ability-Biased Technological Transition, Wage Inequality, and Economic Growth,” Quarterly Journal of Economics 115(2), 469-497 [33] O. Galor and D. Tsiddon (1997), “Technological Progress, Mobility, and Economic Growth,” American Economic Review 87 (3), 363-382. [34] O. Galor and D. Weil (1996), “The Gender Gap, Fertility, and Growth,” The American Economic Review, 86(3), 374 - 387. [35] Galor and J. Zeira (1993), “Income Distribution and Macroeconomics,” Review of Macroeconomic Studies 60, 35-52. [36] H. Ganzeboom and D. Treiman (1996), “Internationally Comparable Measures of Occupational Status for the 1988 International Standard Classification of Occupations,” Social Science Research 25, 231-239. [37] J. Greenwood and B. Jovanovic (1990), “Financial development, Growth, and the Distribution of Income,” Journal of Political Economy, 98, 1076- 1107. [38] J. Haltiwanger, J. Lane, and J. Spletzer (1999), “Productivity Differences across Employers: The Roles of Employer Size, Age, and Human Capital,” The American Economic Review 89 (2), Papers and Proceedings, 94 -98. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 189 [39] K. Hirano, G. Imbens, G. Ridder (2000), “Efficient Estimation of Average Treatment Effects using the Estimated Propensity Score,” NBER Technical Working Paper No. 251. [40] B. Jovanovic and Y. Nyarko (1996), “Learning by Doing and the Choice of Technology,” Econometrica 64,1299-1310. [41] C. Juhn, K. Murphy, and B. Pierce (1993), “Wage Inequality and the Rise in returns ot Skill,” Journal of Political Economy 101 (3), 410 -442. [42] C. Juhn and K. Murphy (1997), “Wage Inequality and Family Labor Supply,” Journal of Labor Economics 15 (1, Part 1), 72-97. [43] D. Kaufmann, A. Kraay, and P. Zoido-Lobaton (1999b), "Governance Matters". World Bank Policy Research Department Working Paper No. 2196. [44] F. Kramarz, S. Lollivier, and L. Pele (1996), “Wage Inequalities and Firm- Specific Compensation Policies in France,” Annales D'Economie et de Statistique, Institute National de la Statistique et des Etudes Economiques, No. 41/42, 369-386. [45] M. Kremer and E. Maskin (1996), “Wage Inequality and Segregation by Skill,” NBER Working Paper No. 5718. [46] D. Krueger and K. Kumar (2004), “Skill-Specific rather than General Education: A Reason for US-Europe Growth Differences?” Journal of Economic Growth 9, 167-207. [47] P. Krusell and A. Smith (1998), “Income and Wealth Heterogeneity in the Macroeconomy,” The Journal of Political Economy, 106(5), 867-896. [48] S. Kuznets (1955), “Economic Growth and Income Inequality,” The American Economic Review, 45(1), 1-28. [49] T. Lemieux (2002), “Decomposing Changes in Wage Distributions: A Unified Approach,” Mimeo, University of British Columbia. [50] T. Lemieux (2003), “Composition Effects, Wage Measurement, and Growth in Within-Group Wage inequality,” Mimeo, University of British Columbia. [51] H. Li and H. Zou (1998), “Income Inequality is not Harmful for Growth: Theory and Evidence,” Review of Development Economics 2(3), 318 - 334. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 190 [52] R. Lucas (1978), “On the Size Distribution of Business Firms, ’’The Bell Journal of Economics, 9(2), 508 - 523. [53] R. Lucas (1988), “On the Mechanics of Economic Development.” Journal of Monetary Economics, 22, 3 -42. [54] T. Malthus (1952), “An Essay on Population.” London: J.M. Dent & Sons Ltd; New York: E.P. Dutton & Co. Inc. [55] D. McKenzie and C. Woodruff (2002), “Is There an Empirical Basis for Poverty Traps in Developing Countries?” Mimeo, Stanford University. [56] J. McMillan and C. Woodruff (1999), “Dispute prevention without courts in Vietnam,” The Journal of Law, Economics, and Organization, 15(3), 637- 658. [57] M. Micevska and P. Zak (2002), ‘What Accounts for the Emergence of Malthusian Fertility in Transition Economies?,” Mimeo. [58] J. Mincer (1974), “Schooling, Experience and Earnings,” New York, Columbia University Press. [59] O. Morand (1999), “Endogenous Fertility, Income Distribution, and growth,” Journal of Economic Growth, 331 - 349. [60] R. Oaxaca (1973), “Male-Female Wage Differentials in Urban Labor Markets,” International Economic Review 14 (3), 693 - 709. [61] R. Perotti (1996), “Growth, Income Distribution, and Democracy,” Journal of Economic Growth 1,149-187. [62] T. Persson and G. Tabellini (1994), “Is Inequality Harmful for Growth?,” The American Economic Review, 84(3), 600-621. [63] T. Piketti (1996), “The Dynamics of the Wealth Distribution and the Interest Rate with Credit Rationing.” Review of Economic Studies, 64, 173 - 189. [64] T. Piketti and E. Saez (2001), “Income Inequality in the United States, 1913- 1998,” NBER Working Paper No. 8467. [65] R. Willis and S. Rosen (1979), “Education and Self-Selection,” The Journal of Political Economy 87 (5), Part 2: Education and Income Distribution, S7 - S36. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 191 [66] M. Yun (2002), Earnings Inequality in USA, 1961 - 1999: Comparing Inequality Using Earnings Equations,” Working Paper, University of Western Ontario. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 192 APPENDIX 1. The Proof of Proposition 3.1.2 on Page 86 The proof follows from a similar proposition in Aghion and Bolton (1997). _ 2 Denote 0(p) = spkA - — , which is increasing and concave in p. Using (3.10) in 2x o p p rkrA (3.14) gives p(b) = — — and p(b) = ------ = --------. It follows then that xksA p(b) 2 «(/» = — V ks k e ----- 4 3 _ —x(ksA)2 doesn’t depend on R. 8 At b , 0(p(b)) = Rk + w x , which is a function of R. Finding R such that a3 0(p(b(R' )) = 0(p(b(R~)) gives \x{keA )2 =R*k + w x, or R* = = —x(ksA)2 - w v8 8 k Since p is increasing in the amount invested b and 0(p) is increasing in p, it follows that if R< R*, then 0(p(b(R*)) > 0(p(b(R*)) and£(i?*) > b(R*). Likewise, if R< R* A * * then b(R ) < b(R ) and those type x investors, whose wealth falls within the range [£(/?*), 6(Z ?*)j are credit rationed.. Finally, from (3.10) b(R) < 0 if > 1, or R < Hence, it is x Jtis ^4)^ x ^3 — ^ shown that for —-------- < Rt —x, (ks,A)2 - wt there exists credit rationing. 8 / Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 193 2. The derivation of the indicator function (4.6.5) on page 164 Linearizing equation (12) by applying Taylor series approximation in equation. Taking Taylor series approximation in (12) around the population mean values of ( I 7 !»Si»r) , yields: & ^=^(a»a»r)+C a-ahr- S i ) ~ ~ (a*/) + ( ^ ) | +<(&>&/) ( s - a /) In w2 i - In wX i + ln-z—zr~ ~ rst + [(r, - r ) - (gu - g x ) ]r r^ r r r ~ g 2 r ~ g\ ■ k - r ) - ( g 2 i - g 2)]=-^=— (r, - r)s, r ~ S 2 = In w2 i - In wu + In-?— ^ - r. 8 J 8i _ rs. r ~ g 2 r ~g\ r - g 2 (r-giXr-gi) = o c0 + In w2 j - In wu + a x g x i + a 2g 2 i + a 3 rt, where r ~ 8 x 1 1 a 0 = In— — , a x = - — — , a 2 = — r - g 2 r - g i r - g 2 and a3=- 8 2 - 8 1 ( r - g x)(r-g2) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Controlling for biases from measurement errors in health outcomes research: A structural equation modeling approach

PDF

An econometric evaluation of IMF program effectiveness

PDF

Immiserizing effects of aid-induced price distortions

PDF

Dual labor markets, public debt management, and exchange rate movements

PDF

Buscando la prosperidad: Migration as long -term investment in El Salvador

PDF

Insurance mechanisms, forest clearance, and the effect of government policies in rural economies

PDF

Health interventions, health status, and labor market outcomes in Indonesia: Three essays

PDF

Essays on agrarian institutions

PDF

Economic analysis of ground lease-based land use system

PDF

Cost analysis of three pharmacy counseling programs for diabetics in a health maintenance organization

PDF

Design and performance of space -time codes

PDF

Dependence on foreign labor, quality of education and unemployment in the GCC countries: In search of solutions

PDF

Growth, human capital, and technological change in a centrally planned economy: Evidence from Cuba

PDF

Essays on technology innovation and new product development strategy

PDF

Asymmetry and risk aversion in first -price sealed bid auctions: Identification, estimation, and applications

PDF

Countertrade: A theoretical and empirical investigation

PDF

Irrigation institutions and technical change: A case study from South India

PDF

Individual heterogeneity and program evaluation

PDF

Complex economic growth

PDF

An analysis of SME export assistance needs

Asset Metadata

Creator Sukiassyan, Grigor Martin (author)

Core Title Income inequality and economic growth: A theoretical and empirical analysis

School Graduate School

Degree Doctor of Philosophy

Degree Program Economics

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag economics, general,OAI-PMH Harvest

Language English

Contributor Digitized by ProQuest (provenance)

Advisor Nugent, Jeffrey B. (committee chair), Hsiao, Cheng (committee member), Kumar, Krishna (committee member)

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c16-559468

Unique identifier UC11340857

Identifier 3145297.pdf (filename),usctheses-c16-559468 (legacy record id)

Legacy Identifier 3145297.pdf

Dmrecord 559468

Document Type Dissertation

Rights Sukiassyan, Grigor Martin

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the au...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA