Financial Development and Income Inequality: Differentiating between Income Definitions

Transcription

1 Financial Development and Income Inequality: Differentiating between Income Definitions Author: Martijn Zielschot Master thesis Abstract. This thesis is the first to test the relationship between financial development and income inequality differentiating between gross- and net income inequality in a cross-country setting, using a new standardized database of income inequality. Financial development does negatively affect income inequality measured using the definition of net income, but does not affect gross income inequality. This result is striking and gives reason to question the quality of datasets used in existing empirical work testing this relationship. Next to this the effect of financial development is analyzed on the growth of income share of all quintiles. The poorest quintiles benefit from financial development while the richest quintile loses income share. The analysis throughout this thesis controls for inflation, educational attainment, trade openness, GDP per capita growth, corruption and ethnical fractionalization. Date: 27 th of June, 2013 Supervisor: J. Grazell I

2 1. Introduction Income inequality is one of the most discussed phenomena in society. Income inequalities can be huge. For instance, in Argentina in 2006 the richest 10% are about 31 times richer than the poorest 10% (2006). This ratio is about 4 for Norway. Income inequality is at the heart of politics and even political ideology, since redistribution is a vital function of policy. Income inequality also has its repercussions on growth and stability. Studying the determinants and the effects of income inequality is therefore very valuable from a societal point of view. In this master thesis I examine the relationship between financial development and income inequality. If financial development is able to decrease inequalities, it could be seen as a substitute and even a complement to more progressive policies. Analysis takes place in a crosscountry setting, covering 72 countries between 1960 and Theoretical models give us two ways to think about this relationship: an inverted U- shaped relationship and a negative linear relationship. Using the Standardized World Income Inequality Dataset (2011), I examine this relationship differentiating between -coefficients based on net income and gross income. Existing literature does not account for different income concepts. Financial development does seem to have a negative effect on net income inequality but not on gross income inequality. The results question whether existing empirical literature correctly accounts for different income concepts. This thesis also examines the effect of financial development on the growth of the income share of the quintiles in a country. Since the sum of all growth rates of all quintiles should be zero by definition, it implies that one or more quintiles pay this relative price. It will turn out that quintile 1 till 4 all benefit from financial development at a decreasing rate, and the richest quintile will lose income share. 1

3 2. Literature review This thesis adds to the literature examining the relationship between financial development and growth, which has been researched intensively. Summarized in Levine (2005), the main finding of this literature is that financial development boosts growth. The reason for this is that financial development improves the allocation of capital and reduces capital market imperfections. This literature does not answer the question which part of society actually benefits from this growth. On the one hand the poor may benefit since growth can increase employment. On the other hand it can benefit the wealthier due to increased returns and profit margins. The finance inequality literature, and thus this thesis tries to answer this question. Theoretical papers regarding the finance-inequality relationship are not numerous. There are two distinct theoretical hypotheses: an inverted U-shaped relationship and a negative linear relationship between financial development and inequality. Greenwood and Javanovic (1990) predict an inverted U-shaped relationship. This means that they expect that at early stages of financial development income inequality will increase and later on will decrease. The benefits of financial development in early stages will mainly flow towards the wealthier, since they are able to pay the fixed cost required being able to use financial intermediation. The reason for the increase in inequality is due to the assumption that access to financial intermediation increases returns and simultaneously reduces idiosyncratic risk. Since intermediation alleviates financial market imperfections and improves the selection of projects, aggregate growth will increase. Growth implies that more people will ultimately afford it to join financial intermediaries, increasing their returns. As a result income inequality will reduce and stabilize. Galor and Zeira (1993) and Banerjee and Newman (1993) predict a negative relationship between financial development and income inequality. Galor and Zeira (1993) model income inequality in an economy with indivisible investments. Agents live two periods and can chose to work unskilled for two periods or invest in their education and work 2

4 as a skilled worker in the second period. Credit market imperfections prohibit the poor to invest in human capital. Generations are linked through bequests. Only those with access to external credit or with a large inheritance (rich parents) will be able to invest in human capital. The development of financial markets and thus the alleviation of credit constraints will decrease income inequality, since the poor will have improved access to credit and thus invest in higher future income. An implication of this is that when a country does not develop financially, income inequality is likely to be persisting. The model of Banerjee and Newman (1993) finds similar predictions. Before proceeding with the rest of this thesis it is necessary to understand the Lorenz Curve and the -coefficient. These concepts are quickly reviewed in Appendix A. The theoretical papers above describe a negative linear relationship and an inverted U- shaped relationship between financial development and income inequality. These hypotheses have been tested empirically a few times. Li, Squire and Zou (1998) were the first that made a significant contribution in examining the determinants of inequality. They use two categories of determinants. The first one is called the political economy argument and describes the link between political policy and inequality. The rationale behind this link is that the rich are assumed to have enough influence in the political arena through lobbying, in order to establish policies that are beneficial to them. This influence may be harmful to other parts of the population and to growth. They measure this political economy argument using a measure of political freedom and the level of secondary schooling. The second determinant is capital market imperfections (as described in theoretical papers above). They are measured using the distribution of land (which is a proxy for collateral) and a measure of financial development (M2/GDP). In a sample of 49 countries between 1947 and 1994 they find evidence that better schooling, more civil liberties, deeper financial markets and a more equal distribution of land decrease inequality. They find these results both using Ordinary Least Squares and using an instrumental variable method. In the sensitivity analysis they control for several other variables; among others: Initial GDP per capita (real), urbanization ratio and black market premium. They 3

5 only find a negative significant effect for initial GDP per capita on inequality. Implementing these controls does not change the signs of the coefficients or the significance of the explanatory variables. Clarke, Xu and Zou (2006) conduct a cross country analysis of 83 countries between 1960 and They find that financial development decreases inequality and improves growth. The dependent variable in their regressions is the level of inequality, measured by the coefficient. They control for initial GDP, inflation, government consumption, ethnic fractionalization, the protection of property rights and the sectorial structure of the economy. According to their empirical results a 1% increase in financial development (measured by Private Credit to GDP) is expected to result in a lower -coefficient with about 0.31%. They do find support for an inverted U-shaped relationship, but this result is not robust in all of their specifications. They also find a positive effect of risk of expropriation on income inequality, a positive effect of ethno linguistic fractionalization on income inequality and a negative effect on inequality when an economy is more modern (i.e. less agricultural). The cross-country study of Beck et al. (2007) contains a sample of 72 countries between 1960 and It finds evidence for three important conclusions. First of all, financial development reduces inequality. Secondly, financial development alleviates absolute poverty. Finally, financial development results in a disproportionately growth of income of the poorest quintile. Their dataset is publicly available. This gives me the chance to review their work in a meaningful way. I will discuss their econometric methods and the construction of their data thoroughly. It is the latter that will give the most valuable insights for conducting my own empirical analysis. Beck et al. (2007) use ordinary least squares and an instrumental variable approach (Generalized Methods of Moments). For their OLS regression they employ the following specification: [ ]. Where [ ] represents the growth rate of the dependent variables -coefficient, income share of the poor or poverty (measured as headcount) for country i over the period [t-1;t]. represents the logarithm of 4

6 the -coefficient, share of lowest income quintile or headcount in the first year of observation for country i. Using these lagged values in the equation they allow for the possibility that past values influence growth rates. is private credit to GDP, which is a proxy for financial development. is a vector of control variables for country i measured at time t. The construction of the dataset took place as follows. All variables, dependent and independent, need to have matching observations over time. Only countries with at least ten years between the first and last observation are allowed. Countries that do not meet this requirement are dropped, resulting in a sample of 72 countries. There are three main specifications with respect to the dependent variables. As mentioned above these variables are the growth rates of the -Coefficient, income share of the poor and headcount. The growth rate of the -coefficient is calculated as the difference between the natural logarithm of divided by the number of years between those two observations. This variable can be fully replicated using UNU/WIDER World Income Inequality Database (WIID). The Growth rate of the lowest income share is calculated in a similar way. It is calculated as the difference of the natural logarithms of the first and last observation divided by the number of years in between. The sources of this variable are WIID and Dollar and Kraay (2002). When I tried to replicate this variable I found 26 not coinciding growth rates in the dataset. Beck et al. (2007) simulated this data assuming a lognormal income distribution function, as suggested by Dollar and Kraay (2002). The standard deviation is calculated as. Where denotes the cumulative normal distribution function and G is the -coefficient between 0 and 100. The income share of the 20 th percentile is calculated as. After I recalculated all these observations I can retrieve 21 out of 26 observations. Four observations are most likely not replicable due to data revisions. One observation, Switzerland, could not be replicated at all, the number I find when calculating the poorest quintile income share is far off. The usage of simulated data is my main point of criticism towards this paper. Beck et al. (2007) did not explicitly report that they simulated the income share of the poorest quintile for many countries. In my opinion empirical analysis 5

7 should not use assumptions regarding the likely distribution of a variable. It should focus on information, not on simulation. I will come back to this issue in my own empirical analysis. Next to the point of criticism mentioned above (simulating the income share of the poorest quintile using a lognormal distribution) there is another important point I find worth highlighting. Beck et al. (2007) do not explicitly mention the potential danger for poor data quality. -coefficients and quintiles income shares are calculated using different concepts of income. In some surveys it is even unsure which income concepts were used. Replicating their data I found that the first observation of the time span comes directly from Dollar and Kraay (2002). For most of the country observations it is possible to track these observations in WIID2c. WIID2c documents several characteristics, like the income definition, unit of measurement and the source. Looking at the usage of data stemming from several income definitions, comparability across countries cannot be assured. There even is a lack of comparability within countries. For instance, Bangladesh and Spain use the gross income definition in their first observation and consumption in their last observation. According to Deininger and Squire (1996), expenditure based coefficients should be increased by 6.6 points (where has a value between 0 and 100) in order to make them comparable to income based coefficients. This corresponds to the average difference. In my own empirical analysis I will overcome this problem and differentiate between gross- and net income s. Next to examining the effect of inequality, which is relative by nature, it is also important to look at absolute effects. It is important to highlight that income inequality does not say anything about poverty. For instance, income can be perfectly equally distributed, but at the same time everybody can live in poverty. On the other hand inequality can be huge in a country, while at the same time no-body is living in poverty. Absolute poverty is measured as the percentage of the population living below a certain threshold, which is usually $1 or $2, this is called headcount. In Beck et al. (2007) Growth of headcount is calculated as the annualized growth rate of headcount for the longest available time span on headcount. Data 6

8 on headcount is only available since 1980 s, so the requirement of a minimum of 10 years between observations is dropped. The explanatory variable is in all specifications Financial Development. Financial development is proxied by private credit issued by deposit money banks and other financial intermediaries. The source of this data is IMF s International Financial Statistics (IFS) and own calculations. The variable is calculated as the natural logarithm of the average of private credit of the first and last observation. The control variables used when Beck et al. (2007) regress Growth of or Growth of Income share of the poorest quintile on Financial development are GDP per capita growth, educational attainment, inflation, and trade openness. The reason they control for GDP growth per capita is that this variable is likely to be affected by financial development. At the same time growth will also have its consequences on inequality. Although there is a sound theoretical foundation to control for GDP per capita growth, I have a point of criticism in the way they control for it. They constructed this variable using the World Development Indicators (WDI) and Penn World Table 6.1. I have found out they control for GDP per capita growth using a completely different time span then their dependent and explanatory variables. The first year of observation for every country is 1958 and the latter is In my opinion this is a questionable choice. It would be better to control for GDP per capita growth only for the time span that is used in constructing the other variables. All countries have different lengths of time spans, some countries have variables with a time span of 10 years. Controlling for economic growth per capita over a period of 47 years in such a case is in my view questionable. I will come back to this point in my own empirical research. As stated before they also control for educational attainment. This is necessary for the reason that schooling impacts growth and inequality. These effects need to be filtered out. The source of this data is Barro and Lee (1996). In the growth in headcount regressions they additionally control for population growth and age dependency (the ratio of the population younger than 15 years or older than 65 years). The source of this data is WDI. 7

9 Kappel (2010) performs an empirical analysis regarding the relationship between financial development and income inequality with approximate the same results as in Beck et al. (2007). Her dataset consists out of a sample of 78 countries, using OLS and 2SLS as econometric methods. Instead of looking at the growth of -coefficient chooses to look at the level of. Her second dependent variable is headcount. Next to private credit as explanatory variable, she also looks to the development of a country its stock market. She also splits her sample in high- and low income countries. Financial development remains to have a negative and significant effect on inequality for medium- and high income countries, but there is no significant effect on low-income countries. In her analysis she controls for ethnic fractionalization, land inequality (a proxy for wealth), government spending and schooling. She finds that stock market development and private credit both decrease inequality. Her data is averaged over the period In my opinion this weakens her innovative idea to look at stock market development as a proxy for financial development, since stock market development is a rather recent development (Financial Structure Database). Observations only start in 1989 and for some counties much later. She reports that stock market development as the average of the timespan This implies that it must be calculated as the average between zero and the level of stock market development in It would be better to measure the effect of stock market development on income inequality over a shorter time span. Another criticism I have on this paper is that Kappel (2010) does not make any statements on how she collected data from WIID2b. It cannot be traced back how she picked her data points and thus we do not know anything about she dealt with potentially different income definitions. Nikoloski (2012) does find support for an inverted U-shaped relationship between finance and inequality. He looks at inequality and financial development in levels. The main reason difference between his conclusion and of others described above is that he has cleaned the data for income inequality using an algorithm that filters out consumption based s. He 8

10 also made a quality ranking of surveys and observations in WIID2c. This paper clearly demonstrates the importance for the usage of clean and comparable data. Batuo et al. (2010) find similar results as Beck et al. (2007) and Kappel (2010); however their study is limited to 22 African countries for the period In the conclusion they mention (without conducting any related) empirical research that the persistence of inequality could be reduced by aiming financial development on rural areas and on the lower incomes. Financial sector reforms should encourage this. Weak institutions, mainly in rural areas, are an important reason that commercial banks have high information asymmetries and thus high costs and rates of default. It would be better to use cooperative banks or microfinance institutions. Microfinance institutions are less dependent on governments and institutions and therefore more appropriate instruments to reduce poverty while improving growth. In line with these recommendations, Hisako and Hamori (2009) perform a cross country analysis regarding the effect of micro financing on inequality of 61 developing countries in the year Using very basic measures (number of micro finance institutions in a country and the number of people that use them) they find that micro finance institutions have a strong and negative effect on income inequality. In a sample of 30 countries (all member of Commonwealth of Nations) between 1995 and 2008, Batabyal and Chowdbury (2010) find that financial development has a negative effect on inequality; corruption has a positive effect. The most interesting finding in their paper is that income inequality in low- and middle-income countries is strongly affected by the interaction between corruption and financial development. This means that a policy addressed to both components will have a stronger and magnified effect on addressing inequality than implementing policies separately. Next to cross-country studies there are several country case studies. Liang (2006) finds a negative linear relationship between finance and inequality in a study focusing on urban China. Bittencourt (2006) finds a similar result for Brazil in the period Ang (2010) 9

11 finds the same results for India, although he states that financial liberalization (=deregulation) has led to an increase in inequality. Law and Tan (2009) find that there is no significant effect on income inequality in Indonesia. This thesis and other papers in the finance inequality nexus can be seen as a contribution to a more comprehensive strand of policy-oriented literature. The focus of this literature is on the relationship between inequality and growth. Several papers find that equality improves growth, among others: Persson and Tabellini (1994), Clarke (1995), Perotti (1996), Easterly (2001, 2002) and Barro (2008). Although there are papers that criticize above researches (For instance Forbes (2000)), there is general consensus that inequality hurts growth. Perotti (1996) draws an interesting conclusion that is worth highlighting. He finds that unequal societies tend to be more politically and socially unstable, resulting in lower investment and growth. Easterly (2001) finds that a country with a wealthier middle class tends to have higher growth. Most papers focus on redistributive policies to reduce inequality with the goal to improve growth. However, redistribution may create adverse incentives. If the wealthier are taxed more heavily their incentive to invest may decrease, resulting in a negative impact on growth. This is where the finance-inequality nexus steps in as an alternative policy approach. Financial development can be seen as a substitute for redistribution in pursuing growth, without potential incentive problems due to redistribution. The latter idea is retrieved from Beck et al. (2007). My empirical contribution (examining the effect of financial development on the income shares of all quintiles) will be valuable, since it can be used for a more integrated view on pursuing economic growth. In practice, financial policy and redistributive policies can be seen as complements. In the case where a developing country s government initially shapes it financial policy to promote growth, redistributive policies can be used to sustain or improve future growth. For instance, if growth results in a loss in income share of the third and fourth quintile, a certain policy of redistribution can be chosen in order create a wealthier middle class and hence amplify growth. 10

12 This thesis is also related to a new and growing strand of literature examining the channels by which financial development affects income inequality. This is very important for policy reasons, since it gives additional insights in the do s and don ts of reforming financial regulations. Beck, Levine and Levkov (2010) worked on identifying these channels. The paper itself is explorative. It looks on the effect of bank deregulation on income inequality in the United States of America between the 70 s and 90 s. During this period most states removed restrictions on intrastate banking. As a consequence this led to improved performance of banks, especially in thinly populated areas. The reason for this is that those banks were in monopoly positions, which particularly was at a disadvantage of the poor. The breakthrough of these monopolies due to deregulation improved bank performance and reduced income inequality. According to Jayaratne and Strahan (1996) it improved growth at the same time. According to Beck et al. (2010) there are three theoretical channels through which finance affects inequality. First of all we have the entrepreneurship-channel. Since the poor have little collateral and high borrowing cost, increased bank performance and competition will improve their possibilities to become an entrepreneur and thus increase in income. Deregulation makes capital markets more efficient, resulting in a disproportionate advantage for the poor (see section 2.1). Empirical tests do not support this theoretical Channel. The second channel is education and is also expected to have an inequality narrowing effect. Tertiary educational attainment increases somewhat due to improved financial access, but does not result in a significant decrease in income inequality. The third channel is the labour channel. The foundation behind this channel is retrieved from Jayaratne and Strahan (1998). Bank deregulation lowered the cost of capital of firms which gives firm the incentive to substitute capital for labor and expanding output. If the output effect is stronger, the demand for labor increases. This result is also verified empirically. Please note that above analysis only covers the USA. Luckily, De, Sarkar, Singh and Vij (2011) try to identify these channels in a crosscountry setting, covering 150 countries. They find weak evidence for the educational attainment channel and strong evidence for the labor channel. They also find that the 11

13 geographic penetration of bank branches (instead of just deepening the availability of credit) boosts the reduction of income inequality. 12

14 3. Empirical Analysis This part describes and summarizes data, econometric methods and results. The rationale behind my own empirical analysis is inspired and building on the work of Beck et al. (2007). Data description In constructing my own dataset and variables I use many different sources. None of these datasets give full coverage of observations over all the years for all countries for the period Therefore these variables need to be matched for each individual country, as in Beck et al. (2007). Since their dataset is publicly available I directly collect data for the same countries and over the same time spans as they did. Many variables are calculated as growth rates, so they are calculated using two observations: one at the first year of the time span and one at the last year. The minimum number of years between the observations must be 10 years. The countries studied and the accompanying time spans are presented in appendix B In my empirical analysis I will use two dependent variables. The first one is growth of. Growth of is calculated as the difference between the logarithms between the first and last observation. For calculating growth of I use the Standardized World Income Inequality Database (SWIID) by Solt (2011). This directly solves my criticism towards the usage of different income definitions as in Beck et al. (2007). The SWIID is constructed using an algorithm that makes s in UNU-WIDER World Income Inequality Database (WIID2c) comparable. They report s based on net income and s based on gross income. The concept of net income is after public redistribution. It includes all kinds of public distributions, like pensions (provided by the state) and taxes. Gross Income is the income before redistributions. Both of these variables have their shortcomings. In my opinion the concept of gross income is the cleaner when trying to establish the relationship between financial 13

15 development and income inequality, since it does not contain any distributional effects. I will analyze the effect of financial development on inequality using both s. The second dependent variable is the growth of income share of quintile i, where i is a number between 1 and 5. The first quintile corresponds to the poorest quintile and the fifth quintile corresponds to the richest quintile. This data is collected from the UNU-WIDER World Income Inequality Database (WIID2c) and Dollar and Kraay (2002). Specifically, I traced back all the observations in both these datasets used by Beck et al. (2007). When there is data missing I follow the same procedure in simulating income shares as they did: I assume a lognormal income distribution function, as suggested by Dollar and Kraay (2002). The standard deviation is calculated as. Where denotes the cumulative normal distribution function and G is the -coefficient between 0 and 100. The income share of the first quintile is calculated as, the income share of the second quintile is calculated as, the income share of the third quintile is calculated as and so forth. When I have collected this data I calculate the growth rate of a certain quintile s income share as the difference of the natural logarithms of the each country s first and last observation divided by the number of years in between. I will run regressions both with and without simulated data, since the usage of simulated data (25 out of 71 observations) might blur the picture. For the explanatory variable Financial Development I use the proxy of Private Credit to GDP. Private Credit equals all credit issued by financial intermediaries to the private sector and is calculated as the logarithm of average of the first and last observation. The source of this data is IMF s International Financial Statistics. I calculate financial development as the natural logarithm of the average between the first and last observation. In constructing this variable I am unfortunately unable to construct all the observations. I can only find usable data for 50 out of 72 countries. Since Beck et al. (2007) additionally used own calculations in constructing this variable I rely on their data and use this variable directly in my own dataset. 14

16 Control variables used are Growth of per capita GDP, Initial schooling, educational attainment, corruption, trade openness and ethnic fractionalization. Growth of GDP per capita is measured over the same period as Private Credit and the dependent variables. This solves one of my points of criticism towards Beck et al. (2007). The source of Growth of GDP is calculated using the most recent Penn World table. It Is calculated using the following formula: growth of GDP per capita = ( ). Controlling for growth of GDP per capita is necessary, since financial development affects growth (see literature review). Data regarding corruption is retrieved from the Corruption Perception Index (CPI) from Transparency International. Data is only available since For reasons of availability corruption is measured in the year This should not be a very large problem, since corruption tends to be a rather persisting phenomenon [see among others: Damania et al. (2004), Mauro (2004)]. The CPI measures corruption on a scale from zero to ten, where ten means that there is no perception of corruption in a country. I rescale this variable by deducting the CPI from 10. After that I take the natural logarithm. The variable initial schooling is retrieved from Barro and Lee (forthcoming). This dataset contains the average years of school attainment for the population of fifteen years or older in a country measured in fiveyear intervals. Therefore the first year of a country s sample period will be rounded downwards. For instance, the sample period for Argentina is For this country I collect data regarding initial schooling between for the year Trade openness is derived from the World Development Indicators (WDI). It is the sum of imports and exports divided by GDP. Trade openness is calculated as the natural logarithm of the average trade openness over the sample period. Ethnical fractionalization is retrieved from Alesina et al. (2003). They calculate fractionalization as. Where is the relative share of ethnic/religious group in country j. In my dataset fractionalization is calculated as the natural logarithm of fractionalization. Inflation is retrieved from WDI and calculated as the difference of the natural log of the first and last observation of the GDP deflator in a country divided by the number of years in between. 15

17 Table 1 (presented on page 29) presents summary statistics. Net income inequality increases, while gross income inequality decreases, on average. The average Private Credit to GDP is about 40%. The country with the highest Average Private Credit to GDP is Hong Kong, the lowest is Uganda. The least ethnically fractionalized country is the Republic of Korea; the most ethnically fractionalized country is Uganda. The countries with the lowest perception of corruption are New Zealand and Finland; the country with the highest perception of corruption is Bangladesh. The country with the lowest trade openness is the USA; the country with the highest trade openness is Singapore. Growth of GDP was highest in the Republic of Korea and lowest in Niger. Inflation was highest in Brazil, and lowest in Austria. Table 2 (presented on page 30) presents a correlation matrix. It is interesting to see that private credit is not significantly correlated to growth in gross and net s. Econometric Methods In this thesis I will employ ordinary least squares (OLS) as estimation method. The specification is as follows: [ Where [ ] is ] the growth rate of (gross or net) in country i or the growth rate of the income share of quintile j in country i. is the initial or the initial income share of quintile j, measured on the first year of the time span of country i. is the logarithm of the average level of financial development in country i over the whole time span. is a vector of control variables for country i at time span t. Results: Financial development and growth of net- and gross Table 3 (page 18) presents regression results with the growth rate of net as dependent variable. Specification 1 only controls for the initial net. A high initial net gini results in a reduction in the growth rate of net s. In other words, countries with higher income inequality have lower growth rates of net (i.e. tend to become more equal). 16

18 Private Credit turns out to be a highly significant variable. The economic significance is severe as well; a 1% increase in private credit results into a decrease of the annual growth rate of net with about 0,40%. Specification 2 additionally controls for inflation, trade openness and initial schooling. Inflation significantly increases income inequality. Private credit remains a highly significant variable. Specification 3 additionally controls for the Growth rate of GDP per capita, since financial development may affect economic growth. This control is not statistically significant. Regressions results remain largely unaltered. Specification 4 controls for the interaction between the growth rate of GDP and the initial net. It is necessary to control for this, since the relation between growth of and economic growth may be affected by the initial income distribution. The results are interesting; growth rate of GDP per capita and the interaction between initial net and growth rate of GDP per capita have both become significant variables. A higher growth rate of GDP per capita results in the reduction of inequality. The interaction term has a dampening effect. A higher initial net given the level of the growth rate of GDP per capita will on average result in a dampened reduction of income inequality. Private Credit remains a significant variable. Specification 5 additionally controls for corruption and ethnical fractionalization. Higher corruption significantly results in a higher net. Private credit has lost quite some of its significance, but remains significant at the 10% level. Ethnical fractionalization does not seem to be a significant determinant of changes in income inequality. When controlling for the interaction between private credit and corruption in specification 6, private credit and corruption both become insignificant. The interaction term turns out highly insignificant as well. The last model does not seem to be appropriate. In short, private credit turns out to be a robust determinant of the distribution of net income. Higher private credit is expected to result in more income equality. Note that the fit (R 2 ) of all models is very high. In unreported test I also investigated whether there was any evidence for an inverted U-shaped relationship between financial development and inequality, as suggested by Greenwood and Jovanovic (1990). The squared term of private credit turned out to be highly insignificant. 17

19 Table 3. Financial Development and Net Dependent variable is growth rate of net and calculated as the difference between the logarithms of net divided by the length of the time span in years. Private Credit is calculated as the log of the average private credit over the time pan. Initial net is the logarithm of the first observation. Inflation is the difference of the natural log of the first and last observation divided by the number of years in between. Average trade openness is calculated as the logarithm of the average trade openness over the sample period. Initial schooling is calculated as the logarithm of the average years of educational attainment of the population older than 15 years. Growth rate of GDP per capita is the logarithm of the first observation of GDP per capita divided by the last observation, divided by the number of years in between the observation. Ethnical fractionalization and corruption are calculated logistic. Corruption is calculated as the logarithm of the reversed corruption perception index. A high value indicates high corruption. Estimation method: Ordinary Least Squares. (1) (2) (3) (4) (5) (6) Growth Rate Net Growth Rate Net Growth Rate Net Growth Rate Net Growth Rate Net Growth Rate Net Private Credit *** ** * * (0.005) (0.022) (0.014) (0.079) (0.923) Initial Net *** *** *** *** *** *** Inflation ** ** ** * * (0.003) (0.003) (0.006) (0.013) (0.014) Trade openness (0.181) (0.190) (0.289) (0.260) (0.263) Initial Schooling (0.387) (0.375) (0.527) (0.707) (0.781) Growth Rate GDP per capita * (0.506) (0.040) (0.099) (0.097) Growth Rate GDP per * capita*initial Net (0.045) (0.100) (0.097) Ethnic Fractionalization (0.555) (0.592) Corruption (0.076) (0.289) Private credit*corruption (0.581) Constant *** *** *** *** *** *** Observations R p-values in parentheses + p < 0.10, * p < 0.05, ** p < 0.01, *** p <

20 Table 4. Financial Development and Gross Dependent variable is growth rate of gross and calculated as the difference between the logarithms of gross divided by the length of the time span in years. Private Credit is calculated as the log of the average private credit over the time pan. Initial gross is the logarithm of the first observation. Inflation is the difference of the natural log of the first and last observation divided by the number of years in between. Average trade openness is calculated as the logarithm of the average trade openness over the sample period. Initial schooling is calculated as the logarithm of the average years of educational attainment of the population older than 15 years. Growth rate of GDP per capita is the logarithm of the first observation of GDP per capita divided by the last observation, divided by the number of years in between the observation. Ethnical fractionalization and corruption are calculated logistic. Corruption is calculated as the logarithm of the reversed corruption perception index. A high value indicates high corruption. Estimation method: Ordinary Least Squares. (1) (2) (3) (4) (5) (6) Growth Rate Gross Growth Rate Gross Growth Rate Gross Growth Rate Gross Growth Rate Gross Growth Rate Gross Private Credit (0.451) (0.435) (0.611) (0.675) (0.328) (0.946) Initial Gross *** *** *** *** *** *** Inflation * (0.039) * (0.042) * (0.034) * (0.027) * (0.029) Trade openness (0.076) (0.081) (0.093) (0.093) (0.096) Initial Schooling (0.361) (0.366) (0.308) (0.833) (0.797) Growth Rate GDP per capita (0.643) * (0.014) (0.059) (0.059) Growth Rate GDP per capita*initial Gross * (0.015) (0.058) (0.058) Ethnic Fractionalization (0.592) (0.582) Corruption (0.453) (0.450) Private credit*corruption (0.777) Constant *** *** *** *** *** *** Observations R p-values in parentheses + p < 0.10, * p < 0.05, ** p < 0.01, *** p <

21 Table 4 (page 19) presents the estimation of private credit on the growth rate of gross. These specifications all have a very good fit, ranging from 62% till 73%. In none of the specifications private credit turns out to be a significant determinant of income inequality. Initial Gross is significant and negative in all specifications, this result is as expected. A higher intitial gross on average resulted into faster reductions of future gross coefficients. Inflation turns out to be a significant determinant of income inequality in all specifications. Trade openness also has an inequality increasing effect, which is significant in all specifications. Specification 3 does not show a significant effect of growth of GDP per capita on inequality, but when controlling for the interaction between initial net and growth of GDP per capita, the variable becomes significant and negative. The interaction tem is significant as well. The interpretation is equal as described above; a higher net given a level of average GDP per capita growth results in a dampened decrease of inequality. A higher growth rate of GDP decreases growth rates of gross s. According to specification 5 and 6, Corruption and ethnical fractionalization turn out to be insignificant. The relationship between table 3 and 4 is scientifically very interesting. The results raise many questions. Financial development seems to have influence on net income inequality, but not on gross income inequality. Both income concepts are inappropriate. Net income is the income after redistributions, enforced by the government, have taken place. Gross income is the income earned before any redistribution takes place. Pensions provided by the state are a part of net income, but not of gross income. Taxes (negative and positive) do impact net income, but are left out of account in the definition of gross income. Both definitions are imperfect, but in my opinion the effect of financial development should be most clear on the s using the concept of gross income. The channels identified by Beck et al. (2010) and De et al. (2011) by which financial development affects inequality are the motivation behind this. According to them, it is mainly the labor channel that decreases inequality. The disproportionate increase of the wages and working hours of the poor should definitely affect gross income. Since income from labor is the major component of income for the 20

22 individual/household, the results are truly puzzling. Using the dataset of Beck et al. (2007) I try to dive deeper into this. Table 5 (presented on page 33) is an exact replication of their regression table. Details about their data can be retrieved from my literature review, or from their original paper. In table 6 (presented on page 34) I replicate their regressions, but this time using dependent variables (both gross and net) retrieved from SWIID. I leave the rest of their database unaltered. The results are largely coinciding with my own results using my own dataset, but the fit of the model greatly improved. Private credit does have a significant effect on net s, but not on gross s. A potential explanation for this is that the variable private credit may not exactly measure what we want it to measure. Financial development is broader than the issuance of credit to the private sector. There are market-based financial systems and bank-based financial systems, as explained in Demirgüç-Kunt and Levine (1999). Markets are also able to mobilize savings, allocate capital, exert corporate control, facilitate risk management and ease transactions. The fact that a country it s banking sector issues relatively small amounts of credit does not imply that that a country is financially less developed than a country with a larger banking system. This might have disturbed the analysis. As discussed in the literature review, Kappel (2010) does see stock market development as an integral part of financial development. Her conclusion is that stock market development also reduces inequality. Therefore, the potential explanation of wrong measurement of financial development is unlikely to explain the counter-intuitive relationship between table 3 and 4. Results: Financial development and growth of the income share of quintile i The -coefficient summarizes inequality into one number. Several underlying observations can result into the same. This makes it interesting to look at the effect of financial development on the development of the income share of a certain quintile. In contrary to existing literature I examine the effect of financial development on the income share of all quintiles. Regressions are presented in table 7 till 11 (presented on page 35 till 39). The tables should be read simultaneously. The specifications are identical, but where 21

23 applicable the data behind it may be adapted to the adequate quintile like the variable initial income share. Specifications 1 till 4 include simulated quintile income shares, assuming lognormality (see data description). In specification 5 till 8 these simulations are dropped. Quintile 1 is the poorest quintile and quintile 5 is the richest. Specification 1 allows us to draw interesting conclusions. Quintiles 1 till 4 all see an increase in their income share. It is the fifth quintile that pays the relative price of financial development. The richer you are, the less you will relatively benefit from financial development. The poorer you are, the more you benefit. A 1% increase in private credit results in an annual growth rate of the poorest quintile its income share with about 0.86%. The second quintile will see an annual increase in growth rate of their income share with about 0.46%. The numbers for quintile four are 0.35% and 0.12%, relatively. A one percent increase in financial development is expected to result in the decrease of the growth rate of the income share of the fifth quintile with about 0.35%, annually. The model controls for the quintile its initial income share. The smaller a quintile s income share, the more likely it is to grow above average in the future. Specification 2 controls for initial schooling, trade openness and inflation. Private credit remains a highly significant determinant for the growth of the income share for all quintiles. Inflation most severely hurts the growth of the income share of the poorest quintile. The effect is more than twice as strong as the effect of inflation on the growth of the second quintile. The richest three quintiles do not face a statistically significant effect from inflation, assuming that only variables with p-values higher than 10% are considered to be significant. With a p-value of 0.196, inflation increases the income share of the richest quintile. This quintile is the only quintile with a positive coefficient. Specification 3 and 4 control for the growth rate of GDP per capita and the interaction between the growth rate of GDP per capita and the initial income share of the appropriate income share. All these controls turn out to be insignificant. In unreported tests ethnical fractionalization and corruption turned out to be highly insignificant determinants in the growth of income share of any quintile. Dropping observations that contain simulated income shares (specifications 5 till 8) results are largely similar. Private credit is a significant 22

24 determinant of the growth of the income shares for all quintiles, except for quintile 4. Results regarding the control variables are very comparable to the data including simulations. The simulations are unlikely to have blurred the picture. It is important to note that an important limitation of this analysis of the relationship between financial development and the growth of income shares of quintiles is that the quintiles belong to the coefficients in Beck et al. (2007). This means that mixed income definitions are used and therefore I doubt the reliability of above results. A limitation of all my empirical work is that I do not use an instrumental variable method to overcome potential problems of endogeneity. In all existing related literature OLS regressions and IV methods both find significant results, they never are contradicting. This is in favor of the validity of my results. 23