Income inequality during the transition: The effects of rent-seeking and capital income

MSc International Economics and Business MA International Economy and Business University of Groningen Corvinus University Budapest

Faculty of Economics and Business Faculty of Economics

Income inequality during the transition: The effects

of rent-seeking and capital income

Supervisor University of Groningen: Dr. R. C. Inklaar

Co-assessor Corvinus University Budapest: Dr. András Székely-Doby

Name student: Hielke Vogelzang

Student ID University of Groningen: s1853767 Student ID Corvinus University Budapest: MBL3M8 E-mail: h.vogelzang@student.rug.nl

2

Abstract

Recently the discussion about income inequality has re-emerged after the publication of Piketty’s ‘Capital in the Twenty-first century’. Piketty discusses inequality in a few developed countries but because of the breakdown thesis the discussion about income inequality is more interesting for the countries under the former Soviet Hegemony. This research will focus on the ambiguous effect of the transition which I predict to reduce inequality by decreasing rent-seeking but at the same time increase inequality by increasing capital incomes. Although I used the new extensive SWIID database the limited amount of data available for my endogenous variables led to non-significant results. The linear and instrumental variable regressions provide some support for the predicted effects of rent-seeking and capital income but these results turnout to be not robust. I also tried to solve the endogeneity problem between the transition and inequality using lagged variables but due to lack of data we did not get significant results here as well.

3

Table of content

1. Introduction ... 5 2. Theory ... 7 2.1. Income inequality ... 7 2.1.1. Gini index ... 8 2.1.2. Previous research ... 8 2.1.3. Welfare state ... 10 2.2. The transition ... 12 2.3. Rent-seeking ... 14

2.4. Capital share in income ... 15

2.5 GDP effects ... 17

3. Data and Methods ... 19

3.1. Method ... 19

3.2. Countries and time period ... 19

3.3. Variables ... 21

3.3.1. Gini index ... 21

3.3.2. Transition indicators ... 22

3.3.3. Rent-seeking ... 23

3.3.4. Share of capital in income ... 25

3.3.5. Control variables ... 25

3.3.6. Limitations & endogeneity problem ... 27

3.4. The model ... 28

4. Results ... 31

4.1. Assumption tests ... 31

4.2. Linear regressions ... 31

4.3. Instrumental variable regressions ... 33

4 4.4.1. Lag structure ... 34 4.4.2. Single-equation regressions ... 35 5. Discussion ... 36 6. Conclusion ... 39 References ... 42 Appendices ... 48

Appendix I: List of abbreviations ... 48

Appendix II: Summary Statistic ... 49

Appendix III: Correlation matrix ... 50

5

1. Introduction

In 2013 Thomas Piketty stirred up the debate about income inequality again by publishing his book ‘Capital in the Twenty-First Century’. In his book Piketty talks about the increase in income inequality while focusing on Western countries. With this Piketty leaves some areas of the discussion about inequality untouched. Firstly, he only focuses on France and Great Britain which are both developed countries so he neglects the non-developed world. Secondly, he mainly focuses on the top incomes and not on the total income distribution while the income distribution is especially relevant for those at the bottom end. And finally, he does not involve a large set of policies in his analysis leaving policy makers with only limited means to change the situation.

To fill these gaps in his research I want to analyse the income inequality in the transition countries in Central and Eastern Europe and Central Asia which are not as developed as France and Great Britain. After the fall of the Soviet Hegemony many of these transition countries shifted from bureaucratic towards market coordination (Kornai, 1992). The quiet radical change in political policy makes the interesting to analyse the effect of certain economic policies on the inequality in these countries. Some scholars found that rent-seeking during the communistic period created inequality and that capital income in the new capitalist system should create more income inequality (Simai, 2006). But over the overall effect of the transition on inequality the literature is unclear. The combination of rent-seeking and capital income leads to an ambiguous effect of the transition on inequality.

6 democracy index in 10 of the Eastern European countries. The majority of former countries under the Soviet Hegemony see a continuing decline in their democracy index during the last decade after their fast transition in the 90s. I want to research into the mechanism how the transition affects the inequality in these countries and to see the effect of the political economy on inequality by looking into two different policies: rent-seeking and capital income. This makes my research not only relevant for scholars but also for policy makers. My study contributes to the literature to the extent that it does an empirical analysis of the income inequality in the transition countries and analysis the effect of the political economy on inequality. Some scholars have already analysed the inequality in these transition countries but their research often lacks thorough empirical evidence (Simai, 2006; Milanovic, 1998) or used datasets only covering a limited set of countries and time periods after the transition (Atkinson and Micklewright, 1992). I want to provide some empirical evidence of the effects of the transition on inequality in the transition countries and find some of the mechanism through which the transition affects the inequality. The empirical backing helps to test several assumptions about the socialistic and capitalist political systems. In doing this research I will use the new Standardized World Income Database which provides extensive an extensive amount Gini indices (Solt, 2014). I will investigate two mechanisms through which the transition influences the inequality in these countries. First, I want to investigate the effect of the transition on reducing rent-seeking which should decrease the inequality. Secondly, I want to investigate the effect of the transition in increasing capital income which should increase the inequality.

7

2. Theory

2.1. Income inequality

In the western world Piketty (2013) sees the return to levels of income inequality equal during the Belle Époque (1971-1914). During this period there existed a class structure existed with a large gap between the rich and the poor. He predicts the return to these levels of inequality because the return to capital is now larger than the growth of the economy as a whole creating a fundamental force of divergence. I won’t go into detail about the critique on this book (therefor see Giles (2014) & Homburg (2015)) but I will treat some of the areas not covered by his research. According to Krugman (2015) we should not use too easy stories about inequality; arguments that inequality is bad for economist growth are invalid. The only proven risks is that inequality will lead to polarization of the political arena which may lead to bad policy because of the insecurity of elections where politicians want to look well in the eyes of the rich.

Piketty looked only at the top incomes because of the available data for the long run in the United Kingdom and France. But we want take a broader perspective on income inequality and want to include the income of the poor. The popular Easterlin paradox showed us that the relation between income and utility is not linear but is flattening so that income is more relevant for the poor (Easterlin 1974). I use the framework of diminishing marginal utility to income were someone at the lower end of the income distribution gains more utility as someone at the higher end of the distribution from the same amount of marginal income (Jones & Klenow, 2010). The scholars disagree about the shape of the decreasing returns to income of the utility curve and with that on the weight with which inequality should decrease social welfare. According to welfare economics it should be all about the social welfare and this social welfare is negative affected by inequality (Jones & Klenow, 2010; Atkinson & Micklewright, 1992).

8

2.1.1. Gini index

Within the utilitarian framework the Lorenz curve and the Gini index are justified measures for inequality making it a popular measure of income inequality (Dasgupta, Sen & Starrett, 1973). The Gini Index ranges from zero to one where zero means everybody has equal income and one means in general that 1 person has everything and the rest has nothing. The Gini index was developed by the Italian statistician and economist Corrado Gini and is based on the Lorenz curve which is a graphical representation of the distribution of income while the Gini index is single number (Milanovic, 2013). With Gini index the index number multiplied by the mean of the distribution equals the expected absolute difference in incomes between two randomly picked persons from the population relative to the mean (Kendall & Stuart, 1969).

The Gini index is the most widely used measure of income distribution and is a single summary measure (Atkinson & Micklewright, 1992). The Gini index is not the most complete measure of inequality because different distributions of income can lead to the same Gini index but being a single summary measure makes it is easy to compare them across time and space. Because it is an accepted measure within in the utilitarian framework which I use and is able to be compared across space and time I use this measure in my research to represent income inequality.

2.1.2. Previous research

9 In the transition countries in 1988 on average 90% of the workers were employed by the state compared to 21.2% for the OECD countries (Milanovic, 1998). Simai (2007) also argues that the social system was not totally egalitarian because of the Nomenklatura which controlled the resources. Direct income distribution was more egalitarian during communism but fringe benefits of the Nomenklatura and various implicit forms of income create more inequality. Milanovic (1998) expected the Gini index to have been between 0.23 to 0.26 slight above the current coefficient of the Nordic countries but the exact numbers are absent for most transition countries. The communist governments didn’t saw the need to collect the required data and their way of measuring makes it hard to compare the data with the western counter parts (Atkinson & Micklewright 1992).

When Pryor compared the Gini indices in 1973 of the communist and western countries he found that the Gini indices of the communist countries where around 6 percent lower than in the Western countries. But he also found that during communism there were large differences in the inequality between the communist countries. The introduction of a market economy is likely to increase inequality because it introduces capital income which is in nature more less equal distributed (Atkinson & Micklewright, 1992). But eventual effect depends on the safety nets introduced after the transition to compensate the losers by state transfers. According to Milanovic (1998) in all transition countries except the Slovak Republic the inequality increased after the transition.

The communist philosophy tries to achieve total equality in society but there is ample proof that the Nomenklatura had multiple fringe benefits creating a class structure of those who were a member of the communist party and who were not. According to some views the inequality under communism was lower as under capitalism while others hold the view that inequality was no lower during that era (Atkinson and Micklewright, 1992). Atkinson & Micklewright (1992) see the change from the communist to the capitalist economic system as trade-off between efficiency and equality. With deeply-rooted egalitarian values in the former communist countries the inequality becomes an even more important political issue because it runs the risk of social unrest (Simai, 2006). The transition towards capitalism means a shift away from the egalitarian communist philosophy towards a capitalist system which advocates the existence of inequality to create the right incentives.

10 differences are seen as ethical in the capitalist system because the gains from the system should eventually trickle down to the poor. Gregory and Stuart (1989) found that in the communist era the distribution of income in the communist countries only had a minor difference with income distribution of the West. During communism the planners were able to replace the capitalist benefits with benefits for the ruling party elite creating a different factor of inequality (Samuelson & Nordhaus, 1989). This source of inequality is actually worse than inequality in capitalism which tries to keep equality of opportunity. In this paper I will only look at income inequality because of the complexity and the many factors that influence equality of opportunity. See the 2013 transition report of EBRD for an analysis of equality of opportunity in the transition countries.

The last decades the Gini indices have increased in general while multiple international organizations were trying to promote shared prosperity (Dollar, Kleineberg, and Kraay, 2014). The major changes of the transition led to high uncertainty in the transition economies which created a major crisis. During a crisis the poor in general suffer the most which can also be observed during the latest financial crisis (EBRD, 2011). During the socialist period poverty and inequality were mitigated by the social policies but the transition created a new distribution of income and wealth raising the inequality substantially in the central Asian countries (Simai, 2006).

2.1.3. Welfare state

Changes in the distribution of income and wealth in the transition countries brought issues of inequality in the national policies (Simai, 2006). Trade liberalization is only good for economic performance if the redistributive effects are not seen as undesirable by the society, or there must be transfer taxes to compensate the burden for those who lose wealth (Rodrik, 2005). The Marxist ideology remained in most of the transition countries and demanded the state to provide job security, health care, education, social services etc. The labour market was the most affected policy area by the reforms. During the Soviet era the countries experienced nearly full employment but after the transition the introduction of the hard budget constraints led to bankruptcies and a sharp increase in unemployment. Simai (2006) also found a sharp drop in the spending of education and health care during the transition.

11 the economies in which large groups lose income and other groups gain income. To keep democratic support for the reforms the countries lead to compensate the losers of the transition by creating a welfare state. The transition countries made different choices in their social policies leading to different welfare states. These welfare states influence the amount of social transfers by the government reducing the income inequality. Cook (2007) found that the impact of the economic pressure on welfare was strongly mediated by domestic political forces. There were pro- and anti-reform forces with a large role of the left overs of the communist regime.

At the start of the transition Offe & Adler (1991) introduced the breakdown thesis where losers would endanger transition because the reforms would be unfavourable for them. They propose that a welfare state should be created to compensate the losers and keep support for the reforms. The problem was a weak market economy cannot compensate the whole population since a welfare state is very costly. Przeworski (1991) elaborated the breakdown thesis with time inconsistency, because the short term welfare effects are negative while the long term welfare effects are positive and voters do not have long term vision. Therefor voters will stop the reforms halfway through the process when only the negative short-term effects of the reforms took place and the economies get stuck in a low level equilibrium. So they needed large investment to create the needed welfare form the start to compensate the losers from the reforms and keep support from the voters through the entire process. Therefor the EBRD and European Investment Bank were introduced and also the EU (transfer money) helped to solve this problem.

Hellman (1998) commented on the breakdown thesis that it’s not the losers who will endanger the economic reforms but the winners. The winners are the rent-seekers who benefit the most from the partial reform equilibrium. More democracy means less rent-seeking groups because voters want to eliminate them. At the same time, rent-seeking groups are afraid of being captured. In a weak democracy, voters will not be able to get rid of them, but in a strong democracy they will. That is why democratization and marketization reinforces each other. But the democratization process needs to pass a certain threshold to prevent winners or losers from blocking the reforms to endanger the transition.

12 reforms. This trade of between laissez faire policy harming the losers and redistribution policy harming the large asset holders and killing entrepreneurship creates tension about the design of the welfare state (Przeworski, 1991). Comisso (1991) analysed the possibilities for welfare states in the transition countries. He found that the countries could only go for gradual privatisation with building up own entrepreneurs or radical privatisation with an inflow of foreign capital. Other forms of creating a welfare state were impossible because of the global landscape would not allow the countries to close their own market for certain sectors as in the French model and the governments did not have the resources to increase the human capital of the Swedish model.

2.2. The transition

The transition in the former Soviet republics existed out of reforms in different areas. Offe & Adler (1991) distinguish 3 different transitions; political, nation building, and economic transition. With the political transition the countries moved from an authoritarian towards a democratic regime. The nation building transition was necessary because during the communist era the Soviet Union tried to create a non-existing Soviet nation and after the transition many of the countries had to (re)build their own nationality of which most already existed prior to the Russian Revolution and the Second World War. Nationalism is very useful to distract the population from inequality problems (Solt, 2011). The economic transition is about the change from bureaucratic to market coordination. Offe and Adler (1991) and Kalemaj (2015) also talked about a fourth transition towards a welfare state which is needed to compensate the losers of the transition and keep the support for the reforms. There is overlap between the areas of the different transitions and they happened around the same time. Especially between political and economic transitions it is sometimes hard to categorize the reforms (Kornai, 1992). My research focuses on the economic transition because of is direct influence on the income distribution.

13 second group of countries liberalisation and privatisation is going slower keeping the soft budget constraints which prevent macro-economic stabilization. The countries in the second group can in general be found in Central Asia.

To decide about their investments the EBRD publishes a set of transition indicators about the progress of the transition on a yearly basis. To analyse the effect of the reforms during the transition have on inequality I use these indicators provided by the EBRD. They use 6 indicators which are: large-scale privatisation, small-scale privatisation, Governance and enterprise restructuring, price liberalisation, trade and foreign exchange system & competition policy. The EBRD grouped these indicators in to two groups where one represent the first 3 indicators, which are called the enterprises indicators and the latest three together are called the markets and trade indicators. Mainly the enterprise reforms help to increase the capital income and increase inequality while the markets and trade reforms help to reduce rent-seeking and inequality. Therefor the total effect of the progress of the transition on inequality is ambiguous in theory. Although we expect the countries with higher scores in the transition to have higher inequality but there is no theoretical ground for this relation so the relation has to work through a certain set of policies. In this research we focus on two of those the first one being rent-seeking and the second being capital income.

Some of the transition countries stayed closer to the communist philosophy than others who reformed more radically which should lead to different results in the income inequality. For more radical reforms there is more need to compensate losers by more redistribution. Simai (2006) saw that the institutional based reforms differed between the Central Asian countries and the Eastern European countries in that the transition towards a market based economy went more difficult in the Central Asian countries.

14 the external force lost a large share of its power and we now see that some of the transition countries are stagnating or sliding backwards in the transition.

2.3. Rent-seeking

The literature distinguishes multiple forms of corruption including; administrative corruption, collusive networks, and clientelism, distortions of the judicial and legal process, deliberate over- or under-regulation in order to elicit monetary and other pay-offs, and rent-seeking of which many took place in the communist system (Karklins, 2002). The Nomenklatura was able to acquire fringe benefits during the communist era which created a form inequality during the communist era (Simai, 2006; Atkinson and Micklewright, 1992). According to Simai (2006) the higher upper class in the transition countries especially in central Asia behaved more like rent-seeking parasites than modern entrepreneurial class. Without major changes in the economic policies in those countries income inequality and the erosion of human capital will get worse.

Poznanski (1999) provides a critical account of the people orchestrating the transition in the transition countries. He argues that the politicians would only have accepted the transition if they could gain something from it. The politicians were therefor very willing to support the initial reforms which led to a partly reformed equilibrium which provided those in power with many opportunities for rent seeking. After the initial reforms they were unwilling to go on with the reforms because it would remove their ability to gain their rents and were blocking the reforms. This cynical view of Poznanski led to a low level equilibrium which the EBRD calls stuck in transition.

15 equilibrium (EBRD, 2013). In this situation the winners and rent-seekers were able to block the further reforms (Hellman, 1998; Kalemaj, 2015).

Corruption is often attributed to authoritarian regimes but Montinola & Jackman (2002) found a non-linear relationship were partial democratic reforms led to more corruption while full democratic reforms led to less corruption than under the authoritarian regime. Karklins (2002) argues that the causes the corruption lies in the communist legacy and the upheavals of the transition. The democratic transition should increase the accountability in the transition countries. According to Tavares (2005) accountability should decrease corruption but the need to raise funds will increase the corruption in the democratic systems. Tavaras (2005) found that countries which democratized within 5 years after the liberalization decreased corruption but in those countries were the democratization process took longer the corruption increased.

Rent-seeking by itself can be considered as negative for society because of harming the morale but its effect on inequality is generally left out of analysis. Rent seeking is the most common manner of corruption during the transition because the scale of the reforms allowed influential people to support partial reforms but block the final reforms to benefit from their position (Poznanski, 1999). The Russian Oligarchs are the most famous example of people who benefitted from the transition in a way that didn’t maximize social welfare. To prevent rent-seeking the transition countries had to implement the markets and trade reforms were especially the market reforms should prevent the rent seeking behaviour. Because the expected effect of rent-seeking on inequality the following hypothesis will tested during this research:

H1: Rent-seeking will increase income inequality 2.4. Capital share in income

16 The existence of a capital class is inherent to capitalist economies which the communist economies lacked because of the limited property rights. So during the transition this capitalist class had to be created. During the transition the countries used several methods to create private ownership in different compositions. Often used methods were mass privatisation, management buy-out and public sales (Bolton & Roland 1992). The different composition of the different methods created a capital class in varying levels of dispersion through the society causing different levels of inequality. For example we can expect mass privatisation by handing out vouchers to the whole population to lead to less inequality as public sales open to international companies.

The economic transition led to liberalized markets which were amongst other things needed to increase innovation. In general the well-educated benefit more from new technologies as the leading to another force of inequality (Aghion, Akcigit, & Howitt, 2013). The well-educated not only benefit from the new technologies but also from the fact that during the communistic period the communist leaders provided relative higher income to blue collar workers compared to white collar workers (Atkinson & Micklewright, 1992).

Capital income is in general more unequally distributed as labour income, neglecting the issue of human capital, because it allows for accumulation. Someone’s labour income is limited to time while this is not the case for capital income. Piketty blames the increase in capital income for the increasing inequality. Simai (2006) argues that the concentration of capital income is larger if the legal framework is less developed which was the case for most of the countries at the start of the transition. The existence of private ownership in the capitalist system allows for capital income. We expect especially the enterprise reforms which are meant to increase private ownership to increase the capital share in income. Because the expected effect of capital share in income on income inequality the following hypothesis will tested during this research as well:

17

Figure 1: Conceptual Model showing the relations between the transition, rent-seeking, capital share in income and income inequality

2.5 GDP effects

In the 80s and 90s the society in the communist countries realized the communist system had failed to deliver the promised welfare. All transition countries experienced a transitional recessions in 90s and substantially recovered by 2002 (Cook, 2007). With the dissolution of the Soviet Union also the trade block around the Soviet Union hegemony was dissolved harming the economy of most of the countries because they dependent on each other’s resources. With the market coordination the budget constraint were hardened which led bankruptcies (Kornai, 1992). The removal of many subsidizes caused by introducing the hard budget constraints led to several bankruptcies which created employment insecurity (Simai, 2006). So both trade reduction and the introduction of hard budget constraints created uncertainty leading to a contraction in the GDP of the transition countries. The expectation was that in the long-run the liberal reforms should create more economic growth because of invisible hand of the market (Smith & Nicholson, 1887) and creative destruction (Aghion et al, 2013). Besides the economic growth caused by market coordination the transition also allowed the former-communist countries to re-integrate into the world economy and catch-up with the West-European countries by finding new and often more profitable trading partners as during the Soviet era (Simai, 2006).

19

3. Data and Methods

3.1. Method

In the empirical research panel data will be used to analyse the effect of the variables over time for the different transition countries. The existence of endogenous and exogenous variables demands an instrumental variable regression. This method allows us to disentangle the importance of the different variables. After the necessary assumption tests I will first perform a normal panel data regressions to analyse the direct effects between the exogenous, endogenous and dependent variables. Secondly, I will perform the instrumental variable regression. Finally, I will perform a two robustness checks; one using lagged variables and in the second I will perform single-equation regressions.

3.2. Countries and time period

To gather the data from the different data sources we first have to specify which countries are considered transition countries and determine the period to cover. Organizations as the EBRD also consider the former Yugoslavian republics as transition countries but we want to focus on the countries that were part of the Soviet Hegemony because these went from communism to capitalism under the same circumstances and in the same period. The EBRD decided to include the former Yugoslavia Republics as transition countries but the regime of Tito was very different from the communist regimes of the Comecon so in this research the Yugoslavia republics are excluded.

The Soviet Hegemony had two different supranational organizations one was the Comecon, which was the trade organization, and the Warsaw Pact which was the military organization. These two organizations nearly cover the same countries as Mongolia was the only country in which these organization different because it was only part of the Comecon. I choose to take the countries of the trade organization the Comecon, so include Mongolia, because we are analysing the economic transition. Warsaw Pact was a military organization so I consider it more related to the political transition.

20 because of the German reunification in 1990 leaving East-Germany out of most the recent databases. Albania is excluded as well because it only participated for short time in the Comecon and Cuba and Vietnam are excluded based on their limited participation and different geographical location and history.

As argued before the EU played an important role in securing progress in the transition for the countries in Central and Eastern Europe. Therefor I created two where the first group exist out of the countries that are currently members of

the European Union (EU-28) and the other group the non-EU members which approximate the current members and associated countries of Commonwealth of Independent States (CIS). Ukraine, Turkmenistan and Georgia are not full members of the CIS but all have their association with the CIS by being former members or being an associate member (CIS STAT, 2015). Just as Connolly (2012) we added Mongolia to the CIS because of its former membership of its geographical location.

To test the relationships in the conceptual model it would have been perfect to do panel data analyses about the period prior and after the transition. But the problem with comparing

socialist countries is that they publish so little relevant data (Lydall, 1979 in Atkinson & Micklewright, 1992) and specifically for the former Comecon members there were so many border changes. The majority of the databases of this research only cover the post transition period so all the research can only cover that era. We already have a hard time finding enough observations for the post transition period including the pre-transition period would have made things even more problematic. I will assume that just after the transition the countries still behave as communist economies to analyse the effect of communist system on inequality. Some of our databases did not cover the last few years so we use our database between 1990

1

Member of the Comecon as member of the Soviet Union

2 _{Member of the Comecon as part of Czechoslovakia}

EU CIS

Bulgaria Armenia1 Czech Republic2 Azerbaijan1 Estonia Belarus1 Hungary Georgia1 Latvia Kazakhstan1 Lithuania Kyrgyzstan1

Poland Republic of Moldova1

Romania Mongolia

Slovakia2 Russian Federation1 Tajikistan1

Turkmenistan1 Ukraine1 Uzbekistan1

21 and 2011. Unfortunately the indicator for rent-seeking covers most of the transition countries since 1999 and the observations for the capital share also lack pretty often. What makes it even worse is that the observations between these two endogenous indicators only have limited overlap. Therefor for some calculations I will use two different datasets for each endogenous variable to not lose too many observations.

3.3. Variables

3.3.1. Gini index

The Gini index is the dependent variable and it measures the inequality within a society. The Gini indices prior to the 1989 are not available because the communist countries did not provide the data to international Western funded organizations like the World Bank. Some survey results exist but these are not in a structural year-to-year base making them not useful for the panel data research. But also after the transition most countries do not publish national Gini index on a year-to-year base. The World Bank provides reliable measures of the Gini index which allow for easy comparison but they have a limited number of observations, especially for the less developed transition countries.

To solve the problem of missing inequality data some other databases are available which have different measurement methods, combine different databases or fill in data gaps in a systematic way. Some for example use household surveys but these surveys were not done in a systematic manner make it hard to use in panel data. Some other databases combine different datasets to fill the gaps like UNU WIDER World Income Inequality Database (WIID) which combines different data sources like household surveys and official Gini indices from national statistic. The most extensive database is the SWIID which has the largest of coverage of Gini indices so far by using multiple imputations to fill up data gaps (Solt, 2014). The SWIID has so far we know of never been used to analyse the inequality in the transition countries specifically.

22 to fill the missing observations, which is very useful for the transition countries were Gini is often lacking but it means we should be careful with interpreting the results.

The SWIID provides the Gini indices for disposable household income measured in two different ways. The SWIID provides both post-tax and post-transfer Gini indices and pre-tax and pre-transfer Gini indices. The post-tax and post transfer Gini indices which Solt calls the net-Gini’s in his database are the most associated with the real welfare of the population and also include the effect of the taxation policies as part of the welfare state. Therefor we will include the net-Gini’s in the research. Figure 2 shows the development of the average Gini indices for the transition countries over time. The graph show an upward trend in the beginning of the 90s after which it stabilizes. We can expect the increase in inequality just after the transition to be caused by the introduction of the capitalist economy with its increased share in capital income.

3.3.2. Transition indicators

The transition indicators are the instrumental variables in this research and they assess the countries progress during the transition. The transition indicators are published by the EBRD which uses them to assess their investments in the region. They were first published in 1994 but the indicators go back to 1990 at the start of the transition. In the research we only use the six indicators for the enterprise reforms and markets and trade indicators although in the 90s also a 3rd category of indicators existed about the financial institution. The EBRD dropped these indicators at the end of the 90s so we are not able to use them if we want to keep the observations in the 00s and not suffer from more data loses. In 2008 the EBRD stopped analysing the transition in the Czech Republic because it nearly completed the transition and they don’t make investment in the country anymore. Therefor EBRD dropped the Czech Republic from their database and I was unable to retrieve the values prior to 1994 so for this country we only have indicators available between 1994 and 2008.

23 The six transition indicators included in this research are; Large-scale privatization, Small-scale privatization, Governance and enterprise restructuring, price liberalisation, trade and foreign exchange system and Competition Policy. For each indicator the EBRD graded each country every year from 1 to 4+. A 1 represents little or no deviation from a rigid centrally planned economy and 4+ represent the standards of an industrialised market economy (EBRD, 2013). For econometric purposes we will grade a 1+ as 1.3 and a 2- as 1.7 and so on.

Figure 3 shows the average value of each

transition indices for each year to see how they developed over time. All indicators show a rapid upward trend right after the transition after which progress slowed down. Most countries started with a 1 for most indicators confirming our prediction that the transition economies would still operate as planned economies just after the transition. The indicators for Governance and enterprise restructuring, Competition Policy and large-scale privatization stay behind the other three indicators which are already close to the maximum score of 4.3.

3.3.3. Rent-seeking

Rent-seeking is the first endogenous variable in the model. The transition should decrease rent-seeking in the economy which should on its turn decrease the income inequality. Rent-seeking can be seen as a form of corruption which is known to be hard to measure. Unethical gains like rent-seeking and corruption do not lent themselves to be measured because it is in itself something hidden and people tend not to report it on their own, some scholar call it even unmeasurable (Galtung, 2006). Because an accepted for measure for rent-seeking does not exist and rent-seeking has many similarities with corruption which both happen under the same circumstance I looked into measures of corruption.

The EBRD and the World also occasionally perform the Business Environment and Enterprise Performance Survey in the transition countries. But these surveys are on a micro level and performed on an occasional basis so therefore not suitable for panel data. Svensson (2005) distinguished three types of measurements of corruption; the first one are the private risk assessment where the International Country Risk is the most is the most popular covering

24 a large amount of time and countries (Knack and Keefer 1995). But this indicator only focuses on the risk for firms and not the broader idea of rent-seeking in society we try to measure. The second group focuses on the perception of corruption and in this way allow to measures a broader amount of types of corruption. The Corruption Perception Index (CPI) is most famous measurement in this second group. A third group focuses on control for corruption and uses a very broad definition of corruption (Kaufmann, Kraay and Mastruzzi 2003).

Svensson (2005) found correlations 0.75-0.97 between the three different groups so choosing which one you take is not really relevant. I consider the first group to be unsuitable because it only measure the risk of corruption and not the amount. Because the second and third group have such a high correlation of 0.97 I take the second group which has a smaller definition as the third group. Within this group we use the popular CPI which also has a large amount of observations in the transition countries as our measure of rent-seeking.

Transparency international started to publish the CPI in 1995 but at that time they did only cover a few countries which did not include most of the transition countries. But they only cover the majority of the transition only after 1999 and even than they have some gaps in the data. The CPI is based on expert opinions leading to critique because it is based on the perception of elites and not on facts (Cobham, 2013). Can even be seen as a self-fulfilling because it is the most important corruption index it creates the perception it is measuring. But by being a highly aggregated index derived from 13 different surveys increasing the coverage and because it is only one number it makes it easy to compare.

Important with interpreting the results using the CPI is that it is measuring the quality of the fight against corruption meaning a higher ranking stands for less corruption and a lower ranking means more corruption. Figure 4 shows the average of the CPI for the transition countries over time and shows a slightly negative trend which seems quiet worrying. This might be caused by the fact that the countries with lower rating in Central Asia were only added to database in a later stage. Analysing the individual developments does not

25 solve this problem because for the individual countries some show an increasing CPI while others see declining scores.

3.3.4. Share of capital in income

Share of capital in income is the second endogenous variables in the model. I expect the progress of the transition to increase the share of capital income and capital income to increase the income inequality. The shares of capital income will be taken form the Penn World Table 8.0 which is a database is developed by scholars from the University of Pennsylvania and the University of Groningen (Feenstra, Inklaar & Timmer, 2013). The database is built to compare components of the living standards across countries. The Penn World provides us with data about the labour share in income where the capital share in income equals 1-‘labour share in income’.

Figure 5 shows that the capital share in income is quiet volatile with a positive trend until the financial crisis in 2008 after which it experienced a small drop. The data on the individual transition countries shows that the just after the transition the capital share was very dispersed and started to converge. Problem is that the PWT does not provide the labour share in income for every year. The PWT does fill up their missing data by using the same value as the previous but because we want to do panel data to see the changes over time this information is not useful. We are only interested in changes in the capital share so we have to drop the value if it is equals the value of the previous year. This led us to drop 310 observations including all observations for Tajikistan, Turkmenistan and Uzbekistan which are all from the CIS group.

3.3.5. Control variables

Because inequality is determined by multiple factors besides rent-seeking and capital income two control variables are used during the research. In the theory we already discussed the interaction between the transition, GDP per capita and inequality and we use it as a control variable. Because we don’t expect the relation to be linear and want to control for heteroskedasticity wherefore we will use the log of GDP per capita. I take the GDP per capita

26 from the Penn World Table by dividing the GDP Price Power Parity (PPP) by the population which is also provided in the database as well. I use PPP because we are interested in real welfare of the population and take the expenditure side because we are interested to compare the living standards. We used ‘cgdpe/population’ for calculating the GDP per capita in our calculations. Figure 6 shows the development of the log of GDP per capita in the selected countries over time confirming a drop in GDP per capita shortly after the transition and a gradual increase in the long run although we see a small dip around the financial crisis of 2008 confirming what we found in the literature.

Population is added as a second control variable because larger countries allow for more inequality by themselves. Multiple scholars already noted that the largest country in our database, Russia, has a larger inequality because of its sheer size (Simai, 2006; Atkinson & Micklewright, 1992). It would have better to have a control variable measuring the land area of the countries as well but these were not included in our databases. For the population we also took the natural logarithm because we don’t expect a perfect linear relationship between the population and inequality and to prevent the effect of heteroskedasticity.

In the research we also wanted to split the database between the EU and CIS group to control for the effect of the authoritarian regimes and natural resources in Central Asia. The EU with its conditionalities seemed to be successful in sticking the Central-East European countries to the reforms while the non-EU countries in generally represent authoritarian regimes which generally also possess more natural resources. These control variables are the most relevant for the measure of rent-seeking because Chang and Golden (2010) found that autocratic regimes and countries that depend on income from natural resources have a tendency for higher corruption. Unfortunately the limited amount of data did not allow us to split the database in an EU and non-EU group to analyse the relations separately.

Within our model we have to choose between random and fixed effects. With random effects the model assumes the observations in our sample to be random selected while in a fixed effects model the individual differences are captured by intercepts (Hill, Griffiths, & Lim,

27 2008). To test which model fits our data we conduct the Hausman test. During the Hausman test we were able to reject the null hypothesis that the random and fixed effects models are the same so we prefer to use the fixed effects model. The fixed effects model removes time invariant characteristics and we also use it to control for characteristic like egalitarian values in the society which influence the inequality.

3.3.6. Limitations & endogeneity problem

With focusing on the transition countries of the former Soviet Hegemony the size of the dataset was already limited. Unfortunately most of the transition countries do not systematically published the countries Gini index prior to the transition to we already had to drop that time period. Combining the 4 different databases and the availability of both endogenous variables put even more constraint on the amount of observations. The control variables were widely available so there was no use in dropping one of them. To prevent the problems of the missing data we performed the instrumental variable regression for both the effect of capital share and rent-seeking separately to reduce the amount of dropped variable.

Variable Source Date downloaded

Dependent variable

Gini indices SWIID 13-04-2015

Endogenous variables

CPI Transparency International 25-03-2015

Capital share of income Penn World tables 8.0 24-03-2015

Instrumental variables

Large-scale privatization EBRD 25-03-2015

Small-scale privatization EBRD 25-03-2015

Governance and enterprise restructuring EBRD 25-03-2015

price liberalisation EBRD 25-03-2015

trade and foreign exchange system EBRD 25-03-2015

Competition Policy EBRD 25-03-2015

Control variables

GDP per capita Penn World Tables 8.0 24-03-2015

Population Penn World Tables 8.0 24-03-2015

28 During our research we also have to deal with an endogeneity problems. The EBRD (2011) argued that those with less inequality had support for more market reforms instead of that the reforms created more equality. Same goes for Solt (2012) who argues that with higher inequality people will see hierarchical structures in society as more natural leading to lower transition index. Jong-Sung and Khagram (2005) have a different argument, leading to the same causality, that inequality gives a few wealthy the ability to participate in rent-seeking and make it hard for the poor to monitor them. They also argue that inequality affects social norms reducing the legitimacy of rules making it easier to accept corruption. Jong-Sung and Khagram (2005) acknowledge that the relation between the institutional quality and inequality also works the other way around creating a vicious circle which the EBRD calls stuck in transition. In this research we focus on the relation from the reforms to inequality because this relation makes it possible for policy makers to influence the situation while with the relation from inequality to reforms there is only little policy makers can do to break the vicious circle. Unfortunately all the previous mentioned articles lacked empirical support for the direction of their causality there in the robust check we try to find empirical support for the direction from the reforms towards inequality by lagging my endogenous and instrumental variables.

3.4. The model

Because we expect the endogenous variables to interact with the instrumental variables by correlating with error term in the relation between the transition progress and inequality we use the instrumental variable regression (Staiger & Stock, 1994). In this model we want to explain inequality so the Gini index is the dependent variable. Because I am interested in the effect of capital income and rent-seeking but expect these to depend on the transition indicators I will treat them as endogenous indicators. I will use the 6 different transition indicators as the instrumental variables.

(1) CPIit   1LSit2SSit3Git4PLit5TFit6CPitit (2) Git   1CPIit2CSit3Xitit

(3) CSit   1LSit 2SSit3Git4PLit5TFit6CPitit Where:

29 - G(it) stands for governance the capital share

- PL(it) stands for price liberalisation indicator

- TF(it) stands for the trade and foreign exchange system indicator - CP(it) stands for the competition policy indicator

- CS(it) stands for capital share in income

- CPI(it) stands for the corruption perception index

- X(it) is a vector of control variables including fixed effects and ε is the error term The research starts with a few assumption tests to see if the variables are suitable for the chosen analysis. After that we will first perform a normal regression analyses between the endogenous and independent variables to analyse the direct effects and will also add the control variables. Then I will assume the endogenous variables to be the independent variables and look for the effect of the instrumental variable on the endogenous variables to see if the instrumental variables are proper variables.

In the second part of the research we will perform instrumental variable regressions. Because of the limited amount of data and especially the lack of observations in which both endogenous variables are included we will perform two separate regression one for each endogenous variable. For an instrument variable regression the number of instrumental variable needs to be at least as large as the number of endogenous variables (r≥K) otherwise we have underidentification. In this research we have 6 instrumental (r) variables and 2 endogenous variables (K) so we no problems with underidentification.

30 (4) CPI_it   ₁LS_{i t, 1} ₂SS_{i t, 1} ₃G_{i t, 1} ₄PL_{i t, 1} ₅TF_{i t, 1} ₆CP_{i t, 1} ₇X_it_it

(5) CS_it   ₁LS_{i t, 1} ₂SS_{i t, 1} ₃G_{i t, 1} ₄PL_{i t, 1} ₅TF_{i t, 1} ₆CP_{i t, 1} ₇X_it_it (6) G_it   ₁CPI_it₂CS_it₃X_it_it

I repeated the calculations using two lags for the instrumental variables (7, 8, & 9) and also lagged the endogenous variables one year to account for a lag between the effect of endogenous variables and income inequality (7, 8, & 10).

(7) CPI_it   ₁LS_{i t,}₂₂SS_{i t,}₂₃G_{i t,}₂₄PL_{i t,}₂₅TF_{i t,}₂₆CP_{i t,}₂₇X_it_it

(8) CS_it   ₁LS_{i t,2}₂SS_{i t,2}₃G_{i t,2}₄PL_{i t,2}₅TF_{i t,2}₆CP_{i t,2}₇X_it_it (9) G_it   ₁CPI_it₂CS_it₃X_it_it

(10) G_it   ₇CPI_{i t, 1} ₈CS_{i t, 1} ₉X_it_it

In the second robustness check the panel data was dropped by using single-equation instrumental variable regression. This way we include the effect of country differences so we also include the effects of the strategies the countries took during the transition. We used the 2sls (11, 12, & 13) method and the Generalize Method of Moments (11, 12, &, 14) method as well because all our variables of interest have fixed dimensions (Imbs, 2004).

(11) CPI_it   ₁LS_t ₂SS_t₃G_t₄PL_t ₅TF_t₆CP_t _it

(12) CS_it   ₁LS_t₂SS_t₃G_t₄PL_t₅TF_t₆CP_t_t (13) G_it   ₇CPI_t₈CS_t ₉X_t _t

31

4. Results

4.1. Assumption tests

Before our analysis we have to perform some assumption tests for which the results are provided in appendix IV. The Woolridge test is used to test for autocorrelation which confirmed the existence of autocorrelation of our variables which will be solved by using clusters. I tested for multicollinearity using the variance inflation factors were all values were below 10 with an average of 4.22 so no need to worry about multicollinearity. The Wald test was used to measure the existence of heteroskedasticity and confirmed that the models are affected by heteroskedasticity which we solved by including robust standard errors. We expect our endogenous variables to be correlated with the instrumental variables. The correlations between the endogenous variables are likely to lead to heteroskedasticity were we are unable to control for but we have to take this into account when interpreting our results knowing that the standard errors will be understated (Imbs, 2004). The control variables GDP and population were tested if they were strong variables and with F-test below 10 we conclude that these are not strong variables so can be included as control variables.

4.2. Linear regressions

32 Table 4 shows the regression results between the instrumental variables and the endogenous variables. Because we want to use the instrumental variable regression the correlation between the instrumental and endogenous variables should be limited. The results confirm that the instrumental variables only have

limited significance in explaining the variation in the endogenous variables. For the CPI only a few of the instrumental variables have significant coefficients while with the capital share none has. The small amount of significant coefficients makes them proper instrumental variables and allows us to perform the instrumental variable regression.

One interesting point about the control variables is the relevance of the population in explaining the CPI. The control variable points out that the larger countries have lower score on the CPI so should have more rent-seeking. Might be explained by the fact that the larger

countries in central Asia in general perform worse in the CPI. But surprisingly the size of the countries did not affect the capital income which would have expected considering the natural resources these large countries in Central Asia possess. To control for this we wanted to perform separate regression for the EU and CIS groups but unfortunately that led to very

variable Model 1 Model 2 Model 3 Model 4

CPI -1.262 -0.411 -0.753* Capital income 251.001*** -13.332 358.559** lgdp -0.445 -2.450 -7.084* lpop -16.608 -27.848 -27.135* Constant -85.940*** 75.284* 112.748* -20.938 N 37 266 124 37 R2 0.610 0.033 0.123 0.873

Table 3 Linear regressions between the endogenous and dependent variables introducing the control variables in model 2-4

NOTE: Model 1 & 4 only include observations having both the CPI and capital share NOTE: Rounded to three decimals

NOTE: *p<0.05, **p<0.01, ***P<0.001

Variable CPI Capital income

Ti_SS -0.016 -0.000 Ti_LS 0.123* -0.004 Ti_G 0.228*** 0.007 Ti_PL -0.161** -0.003 Ti_TF 0.032 0.004 Ti_CP 0.061 -0.002 lgdp 0.058 0.012 lpop -0.286** 0.012 Constant 2.352 0.383 N 281 125 R2 0.128 0.069

Table 4. Regressions between two endogenous and instrumental variables NOTE: Made separated regressions for the CPI and capital share because of limited overlapping data

33 small data groups. The most relevant results from these regressions are the significant and negative relationship between the CPI and inequality and significant and positive relationship between the share of capital income and inequality in model 4 were we control for both GDP per capita and population.

4.3. Instrumental variable regressions

The results of the instrumental variable regression are shown in table 5. In model 5 both the endogenous variables are included and in the models 6 and 7 we performed the calculations using only one of the endogenous variables to save the amount observations. Model 5 confirms both our hypothesises that both rent-seeking capital income increase inequality at great significance but unfortunately

models 6 and 7 do not lead to significant coefficients although the signs are in the right direction. Because model 6 and 7 include more observations we can conclude that the results for model 5 are not robust.

To test if we used the right instrument which do not correlate with endogenous variable we test for over and underidentification (Staiger & Stock,

1994). If one of the test over- or underidentification holds the instruments are not weak and our results are not robust. Because we use cluster robust errors we cannot use the Cragg-Donald Wald F statistic but have to use the Kleibergen-Paap test. For the most important model 5 we were unable to reject the null hypothesis that underidentification exists (7.136). For model 5 we were unable to test for overidentification because of the small amount of data but all other calculations using the Hanson showed that the none of the models came close to overidentification. When we analysed the regressions for the endogenous variables separately the results lost their significant again. But for the CPI the Kleibergen-Paap test now rejects the null hypothesis (11.066) so for the CPI the instrumental variables did not suffer from over- or underidentification meaning the instrumental variables were proper although we do not have significant results. For the capital share the Kleibergen-Paap showed underidentification again and also the results were not significant. Fortunately both endogenous variables keep their expected sign all the time.

Variable Model 5 Model 6 Model 7

CPI -1.366*** -0.667 Capital income 466.714*** 186.779 lgdp -9.319** -0.980 -5.460 lpop -16.027 -11.944* 3.392 N 28 262 118 Kleibergen-Paap 7.136 11.066 6.179 R2 0.167 0.004 0.037

Table 5. Instrumental variable regression

NOTE: Using the six transition indicators as instrumental variable and CPI and capital share as endogenous variables

34

4.4. Robustness tests

To test the robustness of our results we perform two different robustness checks. In the first one I will use lagged variables both to account for delayed effects and to test for the endogeneity problem (Hall et al., 2005). In the second robustness test I will drop the panel data and use single equation instrumental variable regression both using 2sls and GMM to include country differences.

4.4.1. Lag structure

Table 6 show the results for the instrumental variables regressions using different lags. Unfortunately the calculations never let to significant results and the coefficient of the CPI became even positive in the separated calculations for the endogenous variables. When the endogenous variables were combined again the CPI returned to its expected sign. The Kleibergen-Paap statistic shows that the instruments here suffer from underidentification except when we only lag the instrumental variables once for the CPI. We were unable to combine the endogenous variables except for model 12 because with the other lags combining the endogenous variables reducing the amount of observations leading to Fingleton clusters and multicollinearity for the capital share variable. Surprisingly it only worked when lagged the variables the most.

Because lagging the variable is leave use with insignificant results we were unable to prove delayed effects nor to solve the endogeneity problem by Solt (2012), the EBRD (2011) and Jong-Sung and Khagram (2005). We first only lagged the instrumental variable because expect the institutional reforms need time to effect the capital income and corruption. Especially for capital income it will take time to accumulate capital and for reforms to prevent

Model 8 Model 9 Model 10 Model 11 Model 12

Lag instrumental 1 1 2 2 2 Lag endogenous 0 0 0 0 1 CPI 2.686 3.017 -0.638 Capital income 51.475 2.126 3.431 lgdp -1.404 -3.284 -1.508 -2.395 2.979 lpop -7.864 -13.540 -6.946 -19.481 -39.659** N 263 116 264 118 41 Kleibergen-Paap 10.066 6.712 8.710 5.312 2.683 R2 0.171 -0.076 -0.442 0.080 0.318

Table 6. Lagged variables

35 rent-seeking it will take time of citizens to change their behaviour. In the last calculation we also lag the endogenous variables to allow us to combine both endogenous variables in one calculation and as we have seen before some interacting effect exist between the CPI and capital share.

4.4.2. Single-equation regressions

In the second robustness test we perform single-equation instrumental variable regression to prevent the within-country variance to be captured by the control variables and fixed effects (Arcand, Berkes, & Panizza, 2012). Therefore we have to drop the panel data and we use both the 2sls and GMM methods because all the variables have fixed dimensions. The results are shown in table 7 and the they show significant signs for the CPI especially for the GMM method. Both endogenous variables have the expected signs in all the calculations only the capital share is not significant in any of the calculations. Interesting for the GMM calculation is that for the CPI the population turned out to be the significant control variable and for the capital share the GDP per capita is the significant control variable. The relevance GDP per capita can be explained by the effect that higher income countries have more income available to save or invest than low income countries so are also more likely to receive capital income. The relevance of population can be explained by the large size of the countries in Central Asia which in generally score lower on the CPI.

Variable 2sls GMM CPI -2.876* -1.734 -4.195** -2.580*** Capital income 13.417 5.622 14.150 5.622 lgdp 2.973 -0.636 -2.743* 5.507* 0.112 -2.743*** lpop -0.938* -0.742 -0.299 -1.929** -1.445*** -0.298 Constant 11.888 46.557*** 53.518** -4.101 44.504*** 53.518*** N 34 262 119 34 262 119 Kleibergen-Paap 10.560 10.199 8.341 7.735 62.378 25.753 R2 0.985 0.979 0.976 0.981 0.978 0.977

Table 7. Single-equation instrumental-variable regression NOTE: Performed both 2sls and GMM regressions

NOTE: For both regressions we first included both endogenous variables after which we performed separated regression for endogenous variables

36

5. Discussion

From the linear regression model 4 confirms both hypotheses I & II but when we treating the endogenous variables separately this led to non-significant results. This surprised us because we performed the separated the regression to save data hoping that would help us to receive significant results. This can be explained in two ways, one explanation is an interacting effect exist between both endogenous variables which sounds plausible considering the opposite signs they receive. The second explanation lies in the specifics of the countries and time were the overlap between both endogenous variables exists. Because the capital share is totally missing for a quiet a few of the Central Asian countries this sounds plausible as well. The regressions between the instrumental and endogenous variables showed that a significant relation existed between the CPI and a few of the instrumental variables and for the capital share no such relation existed with any of the instrumental variables.

When performing the instrumental variable regression the same happened as with the linear regression that when we combined the endogenous variables this led to significant results and the separate regressions didn’t. But here this was more problematic because model 5 suffered from underidentification. Model 6 did not suffer from the underidentification but that instrumental variable regression did not lead to significant results so overall the instrumental variables regression was unable to provide us with viable results.