The effect of financial development on the distribution of income

The effect of financial development on the distribution of income

Does financial development disproportionately benefit the poor?

Prepared by

Leontien Rendering

July 2012

Abstract

This paper looks into the effects of financial development on the income distribution within a county. Unlike many other papers researching this relation we will use several different measures to reflect financial development. In addition we will allow country characteristics to affect the relationship between financial development and both poverty and inequality, by incorporating an interaction term between financial development and educational attainment. Evidence is found that financial development reduces both poverty and inequality, irrespective of the measure used to reflect financial development. In addition we found the relationship between financial development and poverty to be dependent on the average level of educational attainment within a country.

Keywords: financial development, inequality, poverty, education.

Master thesis for MSc International Economics and Business Student number: 1561774

Email: l.rendering@student.rug.nl

Supervisor: dr. R. Inklaar - Department of Global Economics and Management Second Supervisor: dr. B. Los – Department of Global Economics and Management Rijksuniversiteit Groningen

2 1. Introduction

In the past two decades we have observed rapidly increasing levels of financial development worldwide. Both advanced and less-advanced countries see financial development as a growth strategy, as a well-developed financial system is believed to accelerate economic growth. However, what are the implications for the income distribution within societies of this on-going financial development? Does it lead to a reduction of inequality as it provides the poor with new opportunities? As worldwide an estimated 2.5 billion people live of less than $2.00 a day it is important to analyse what the effect of financial development will be on the income distribution within countries.

However, for policy makers it may not be so attractive to invest in the reduction of poverty, and especially inequality, as it is believed that inequality gives peoples an incentive to work hard.“…conventional textbook approach is that inequality is good for incentives and therefore good for growth” from Aghion et al. (1999). However, there is a fair amount of research that refutes this claim; for example see: Alesina and Rodrik (1994). Alesina and Rodrik (1994) state that financial market imperfections obstruct the efficient allocation of capital (as in: Aghion and Bolton, 1997 and Galor and Zeira, 1993) and as a result, growth levels are lower than they could have been if capital was allocated efficiently.

Can financial development help to bring about a more equal distribution of income? There is a continuing debate regarding the question who benefits from increased financial development. Empirical research regarding this relationship is inconclusive. Several scholars have found that financial development leads to both poverty and inequality reduction (Beck et al. 2004); however the results are far from conclusive, as others believe that the effect of financial development on inequality will be neutral (Dollar and Kraay, 2002) or even inequality enhancing (Greenwood and Jovanovic, 1990).

However, there are two major issues concerning previous research regarding financial development and inequality. For one, previous research has many times failed to include more than one measure in order to reflect financial development. As concluded by Honohan (2004): “It becomes evident that summarizing the development of a financial system by a single measure of the scale of its banking is not likely to fully capture variations in the degree and effectiveness with which it performs these functions. Regulatory and information infrastructures in the economy may also evidently be important” Several scholars have made suggestions regarding the incorporation of more qualitative measures of financial development in research, like accounting standards or the ability to enforce contracts. In order to develop a more complete picture, this research has incorporated five different measures to reflect financial development.

The second issue with previous research is the inability to make a distinction between countries. General effects for a large group of countries may not give us a clear indication of the effect of financial development on poverty and inequality in different types of countries. In order to be able to distinguish between countries we have incorporated an interaction term between financial development and education in which education is the moderating variable.

3 variable that is noted many times to be an important channel through which poverty reduction can come about. For example; indivisible investment in human capital (Galor and Zeira, 1993), or the ability of individuals to properly use and / or adopt a new technology (Helpman & Rangel, 1999), are noted as important channels through which poverty reduction, as a result of financial development, will take place.

In a study rather similar to this paper Beck et al. (2004) experimented with incorporating an interaction term between private credit and education before, however they did not find any significant effects. As Beck et al. (2004) used different dependent variables, reflecting the growth rates of both poverty and inequality; we feel it is legitimate to incorporate the interaction term again and to expect different results.

We have concluded on two key findings in this paper. One, financial development reduces the percentage of the population living of less than $2.00 a day irrespective of the measure used to reflect financial development, however this effect varies with the level of educational attainment present in a country. Two, financial development does reduce inequality; however we do not find conclusive evidence that this effect varies with the level of educational attainment within a country.

This paper will proceed as follows; in order to carefully analyse the effects of financial development on the income distribution within countries, we will use several different measures to reflect financial development. These measures will reflect both quantitative and qualitative measures of financial development. In line with Beck et al. (2004) we will also use both relative and absolute poverty measures. We will regress the poverty and inequality measures on financial development, educational attainment, their interaction term and on a set of conditioning information. By using an instrumental variable approach we will try to establish the causal direction of the relationship between financial development and both poverty and inequality.

2. Financial development, Inequality and Poverty

2.1 The effect of Financial Development on the Income Distribution

Even though many scholars are convinced that financial development leads to faster economic growth, this does not necessarily mean the financial development will also reduce poverty or diminish inequality. If financial development increases average growth only through increasing the incomes of the rich, poverty will not be reduced and inequality will increase.

4 Some theoretical models imply that financial development stimulates growth and at the same time reduces inequality. In these models financial market imperfections are believed to be disproportionally binding on the poor as they lack both collateral and credit history (Galor & Zeira, 1993). These models state that financial development will reduce poverty as it relaxes credit constraints on the poor. By relaxing credit constraint on the poor, income will be distributed more equally as a result of financial development (Galor and Zeira 1993). Suggestions similar to Galor and Zeira (1993) are made by Banerjee and Newman (1993). Following this line of reasoning, one expects a linear and negative relationship between measures of financial development and inequality.

On the other end of the spectrum, theorists suggest that financial development may increase inequality. These models emphasize the importance of informal family ties when the poor try to gather capital, improvements in the ‘formal’ financial system will therefore (especially in the beginning) only benefit the rich (Greenwood & Jovanovic, 1990). Following this line of reasoning scholars expect a linear and positive relationship between measures of financial development and inequality. Greenwood and Jovanovic (1990) do state that when the financial system becomes fully developed an economy will see a stable distribution of income across society, so the effect of financial development will eventually become neutral.

2.2 Empirical findings on financial development and income distribution

As we already stated in the introduction there is a continuing debate regarding the question who benefits from increased financial development and that the empirical research has not provided us with any conclusive results. Here we will give a short overview of some of the previous empirical research regarding the relationship between financial development and the income distribution.

Li et al. (1998) empirically researched the previous discussion and their findings were not in line with the negative expectations of Greenwood and Jovanovic (1990). Li et al. (1998) find that inequality is largely determined by factors within countries that change only slowly, like the financial system. Li et al. (1998) find that financial deepening contributes to the reduction of inequality, as the average income of the lower 80% of the population increases as a result of financial deepening. However, as they address only the average income of the lower 80% of the population this does not give us an idea as to what is happening with poverty in absolute terms, or who exactly benefits from financial deepening.

Honohan (2004) finds that financial depth, measured as banking depth, is negatively related to the poverty head count ratio, an absolute measure of poverty. In his conclusion Honohan (2004) emphasizes the importance of taking into account the legal,- and regulatory systems within a given country when one wants to analyse the effectiveness of the financial system present in a specific country.

5 As shown above much research has found both poverty and inequality reducing effects from financial development; however none of these studies, besides Beck et al. (2004), allow for a differentiation between countries and if allowed for no results were found.

However, Evans et al. (2002) also incorporated an interaction term between financial development and education while researching the growth effects of financial development; they did find that the interaction term was positively related to economic growth. This means economic growth resulting from financial development is faster for higher levels of educational attainment. As economic growth is one of the two channels through which poverty reduction is believed to take place, one would expect the interaction term between financial development and education also to be related to both poverty and inequality.

Though not all studies find inequality and poverty reducing effects; Dollar and Kraay (2002) find that growth, resulting from financial development, is on average neutral which implies that the rich benefit more from every increment of growth than do the poor id est inequality will not be reduced as a result of this growth; however it does indicate that absolute poverty is declining (Dollar and Kraay 2002). Greenwood and Jovanovic (1990) found that at early stages of economic development financial development widens income inequality, they do find that when the financial system matures it no longer affects the distribution of income. Clarke et al. (2006) also find empirical results that inequality indeed widens when financial sector development increases at low levels of overall financial development.

Based upon these findings we still expect financial development to reduce both inequality and poverty at least through its growth effects. However, there is a possibility that negative redistribution effects may offset the positive effects of economic growth. We therefore do not discard the possibility that the coefficients found for the respective measures of financial development could be positive when regressed on our inequality indicators.

In the light of the suggestions of Honohan (2004) that it may not be sufficient to reflect financial development by a single measure, we progress by looking into how to develop a clearer picture of financial development. A second issue we will address in the following section is the incorporation of the interaction term. Fact is that in many theoretical models human capital is suggested as channel through which poverty reduction is likely to occur (Galor and Zeira 1993). In the light of the inconclusive findings between Beck et al. (2004) and Evans et al. (2002), we believe incorporating educational levels as a moderating variable may shine a new light on the effect of financial development on both poverty and inequality.

3. Measuring Financial Development and Education

6 Suggestions regarding the inclusion of measures reflecting the quality of the financial system have been made by several scholars like Rajan and Zingales (1998) and Levine et al. (2000). For example, Rajan and Zingales (1998) suggest that quality of financial development can be measured by using accounting standards as an indicator for financial development. Higher accounting standards are believed to enhance a firm’s ability to raise funds, as they allow for more transparency and monitoring possibilities for investors (Rajan and Zingales, 1998). Levine et al. (2000) found that countries have better functioning financial intermediaries if they possess a legal and regulatory system that gives high priority to creditor rights. In their article they come up with several measures that can be used to reflect both the legal and regulatory system in a given country. Even though the quantity and quality of the financial system reflect how well the system performs its tasks, this may not be the only thing that matters when researching the relation between financial development and poverty and inequality. Honohan (2008) points out that having access to financial services may be of a greater importance to the poor, than the amount they can loan or the quality of the services. Honohan (2008) constructed a measure reflecting access to financial services and found the measure to be negatively related to poverty. Honohan (2008) points out that actual access to savings, the ability to pool risk or access to payment services, which are all tasks performed by the formal financial sector, may largely influence poverty and inequality.

For this research we will, for one, incorporate a measure suggested by both Rajan and Zingales (1998) and Levine et al. (2000) being accounting standards, second we will incorporate a measure constructed by Levine et al. (2000) being, enforcement and thirdly, we will incorporate the access to financial services measure constructed by Honohan (2008). We believe these three measures will give us a good indication of the quality of the financial system as we can observe transparency, the possibility of risk reduction through legal strength, and the accessibility of the system. We will complement these qualitative measures by two more quantitative measures that are frequently used to reflect the level of financial development, being liquid liabilities and private credit. All five measures will be thoroughly discussed in the data section.

Up till now the majority of empirical research has been concerned with characterizing stylized facts about growth, or with testing a single theory of growth. Relatively little research has tried to integrate rival theories, in particular those emphasizing human capital and finance (Evans et al. 2002). This gives us the opportunity to further explore this relation, as we find it hard to believe that the effect of financial development on inequality is not affected by country characteristics like the average educational level.

One reason to believe the educational level may affect the effect of financial development on poverty and inequality is that “An educated labor force is better at creating, implementing, and adopting new technologies, thereby generating growth.” as suggested by Benhabib and Spiegel (1994). This suggestion is made at a relatively aggregated level, however at a less aggregated level we believe individuals to be better able to grasp the opportunities presented to them if they are better educated. In line with Benhabib and Spiegel (1994) we will postulate as well that human capital will directly influence productivity.

7 suggestion is made along the lines of the findings of Barro (2000). Barro (2000) found growth to be positively related to school attainment at secondary and higher levels, he also found workers with this educational background are complementary to new technologies. As a result one may also expect that education also to be complementary to financial development, as found by Evans et al. (2002), speeding up economic growth might lead to a faster reduction of poverty and possibly inequality.

However education may not only affect productivity, it may also affect competition and cultural characteristics within a country. According to Honohan (2008) extending loans typically suffers from selection bias, also at an individual client level. Individuals that are more energetic are likely to experience growing loan and deposit balances (Honohan 2008), so the more confident individuals are (because of education) the more likely it is for them to receive a loan. In the light of increasing levels of education (likely not to be distributed perfectly even across society), it is likely that the competition for a loan amongst the poor will intensify as a result of increasing levels of education, as most loans will flow to those individuals that have received education. Honohan (2008) also depicted that in advanced economies the poor are likely to become excluded from the formal financial system for several reasons can which can be product features (like a high minimum covers for a product or service), more material obstacles (for example: when having a fixed address is a pre-requisite in order to be able to open an account) or this group may be screened-out because of the risk characteristics they possess. This suggests that the poor in countries with on average higher educational levels are either outcompeted, or excluded, from the formal financial sector. Second, Guiso et al. (2004) higher levels of education to be related to higher levels of trust prevalent within a country. In countries with on average low trust levels it is, according to Guiso et al. (2004) normal to access finance through informal channels, however as trust levels start increasing the accessibility of finance through informal channels is strongly reduced. Putting the findings of Honohan (2008) and Guiso et al. (2004) together we expect that the poor in more developed countries are unlikely to benefit from financial development as both the formal,- as well as the informal financial system will become inaccessible to them.

Without an interaction term incorporated in the regressions, only the average effect of financial development on both poverty and inequality can be measured. However, in the light of the previous discussion we believe there is good reason to suspect the relation between financial development and both poverty and inequality might be affected by the average level of educational attainment prevalent within a country.

4. Data and Model

4.1 Basic description of the data

8 Because of data limitations we first considered all countries in the world and then limited down the selection based on the availability and the quality of the data. After doing so a total of 108 countries remained1. As the data on our financial development indicators became available from 1960 onwards, 1960 became the starting point of our analyses. Given limited availability of several variables in the early years of our dataset we decided to shorten the dataset. The final range of the dataset became 1980 to 2009 in 5-year intervals. We constructed the intervals to range from, for example: 1980 to 1984, for this period we averaged the available data per variable and labelled it 1980. The dataset can be defined as being ‘short and wide’ as N>T, which poses some limitations on the possibility of using certain estimation techniques for panel data.

Table 1 - Description of Variables

Indicator Reflects Variable Data Source

GINI (IP)

Inequality GINI-coefficient. UNU-WIDER data set, Deininger

and Squire (1996). PovHCR2

(IP)

Poverty Poverty Head Count Ratio at $2.00

a day (PPP) (% of population).

World Bank WDI & GFD database.

PovGAP2 (IP)

Poverty Poverty Gap at $2.00 a day (PPP)

(%).

World Bank WDI & GFD database.

yrseduc (EA)

Educational Attainment

Average years of schooling for persons aged 15 or above.

Barro-Lee educational attainment dataset.

M3 (FD)

Financial Development

Liquid Liabilities (M3/GDP) World Bank WDI & GFD database.

Privatecredit (FD)

Financial Development

Domestic credit to private sector (% of GDP)

World Bank WDI & GFD database.

Accounting (FD)

Financial Development

Standards of financial disclosure. The Center for International Financial Analysis and Research. Enforcement (FD) Financial Development Standards of enforceability of contracts. Levine et al. (2000) Access (FD) Financial Development

The percentage of the population with access to financial services.

Honohan (2008) GDPpcGrowth

(control variable)

Control GDP per capita growth .Control

variable as suggested by Beck et al. (2004).

World Bank WDI & GFD database.

Inflation

(control variable)

Control Yearly rate of inflation. Control

variable as suggested by Beck et al. (2004) and Clarke et al. (2006)

World Bank WDI & GFD database.

Government (control variable)

Control Government final consumption

expenditure (% of GDP). Control variable as suggested by Clarke et al. (2006)

World Bank WDI & GFD database.

4.2 Data on Inequality and Poverty 4.2.1 Inequality

A widely used measure to represent income inequality is the GINI-coefficient; the GINI-coefficient is based on the Lorenz-curve which plots the share of population against the share of income received. The GINI-coefficient measures the inequality among values of a frequency distribution, in this case income levels. The GINI-coefficient is a value ranging from zero to one (or hundred in a percentile

9 scale), with zero expressing perfect equality which implies that all people have equal income. As GINI-coefficient increases the level of equality decreases, a GINI-coefficient of one expresses maximal inequality. As the GINI-index is widely reported in many official sources and many studies included the measure (Deininger & Squire, 1996), it will be used here as well to represent income inequality. The data will be drawn from the UNU-WIDER data set, developed by Deininger and Squire (1996). Within the UNU-WIDER data set Deininger and Squire (1996) have labeled their data with quality labels. The quality labels range from 1 to 4, with 1 representing the highest quality of data. In line with Barro (2000) we will construct two separate sets of the data based upon the quality labels.

i) Quality2: Represents data points drawn from the UNU-WIDER data set which are classified by either quality-label 1, or quality-label 2.

ii) Quality3: Represent data points drawn from the UNU-WIDER data set which are classified by quality-label 1, quality-label 2, or quality label 3.

Using the GINI-coefficient does have some limitations, for example if income is redistributed from the top to the middle class, this will have a similar effect on inequality as an income transfer from the middle,- to the lower class. In the first case, the top class will move closer to the lower class as income is transferred away from the top,- to the middle class. In the second case, the lower class moves upwards, closer to the top class as income is transferred from the middle,- to the lower class. Clearly the effect of both transfers on inequality is the same, the only thing different is who benefits from the transfer. It now becomes clear the inequality reduction does not necessarily mean that the poor are getting richer (Deininger & Squire, 1996).

4.2.2 Poverty

To be able to analyse the effect of Financial Development on the poor we also have included two measures of absolute poverty. In line with Beck et al. (2004) we will use two different measures of absolute poverty besides the inequality measures, being the poverty head count ratio and the poverty gap.

iii) PovHCR2 this measure reflects the poverty head count ratio in a given country at a $2.00 per day level (PPP), as a percentage of the population. Put simply, this indicator reflects the percentage of the population living of a budget of less than $2.00 per day. Data will be drawn from the World bank WDI & GFD data base.

iv) PovGAP2 this measure reflects the poverty gap in a given country at $2.00 per day. The poverty gap reflects the difference between an individuals’ actual income and the poverty line ($2.00) and is expressed as a percentage of the poverty line. Non-poor score zero as they have zero shortfall from $2.00. As opposed to the head count ratio, this measure reflects both the depth of poverty as well as its incidence. Data will be drawn from the World bank WDI & GFD data base.

10 equal amounts from financial development one would find a reduction in the poverty indicators, but the inequality indicator will remain unchanged.

Table 4 – Correlation Matrix for the Poverty and Inequality indicators

Quality2 Quality3 povHCR2 povGAP2

Quality2 1.0000

Quality3 0.9785** 1.0000

povHCR2 0.0616 0.1021 1.0000

povGAP2 0.0599 0.1407* 0.9599*** 1.0000

** Correlation is significant at the 0.01 level * Correlation is significant at the 0.05 level

Observations labelled quality 3 by Deininger and Squire (1996) have been included in our dependent variable Quality3 (in addition to the observations labelled quality1 and quality2), this appears to have slightly deteriorated the quality of that data as the correlation between Quality2 and Quality3 is not exactly one. However, even though the data is of a lower quality, the incorporation of observations labelled quality 3 does enhance the number of observation available for our research. The two absolute poverty measures also show close correlation, this finding is similar to the findings of Beck et al. (2004). The inequality measures do not show close correlation with the poverty measures, this tells us it is useful to assess the effect of financial development on both absolute and relative measures of poverty. Measurements of inequality generally do not include individuals administered in pensions, hospitals or nursing homes, homeless individuals and undocumented immigrants, at least not for high income countries (Gottschalk and Smeeding, 2000).

As in this paper we will incorporate an interaction term between financial development and education, similar to what is done by Beck et al. (2004) we want to emphasize here that our dependent variables are rather different from the dependent variables used by Beck et al. (2004). Beck et al. (2004) analysed the effect of financial development on the growth rates of their poverty and inequality measures whereas we assess the effect of financial development on their absolute numbers2. The correlation between the measures used in this research and the measures used by Beck et al. 2004 is rather low; therefore we expect our results may differ from the findings of Beck et al. (2004)

Table 5 – Correlation Matrix for the Poverty and Inequality indicators

povGAP2 PGgrowth povHCR2 PHgrowth

povGAP2 1.0000

PGgrowth 0.1603** 1.0000

povHCR2 0.9599*** -0.0295 1.0000

PHgrowth 0.0229 0.9774**** 0.0324 1.0000

4.3 Data on Financial Development

Financial intermediary indicators are usually imperfect measures of how well financial intermediaries do in; researching firms, monitoring managers, mobilize savings, pooling risks and easing transactions. There is little agreement on how financial development can be appropriately measured, however we selected five measures which we believe to reflect both quantity and quality of financial development.

11 4.3.1 Quantitative measures of Financial Development

Quantitative:

i) Liquid Liabilities: This measure resembled by M3/GDP which equals the liquid liabilities of the financial system. Liquid liabilities are a typical measure of financial debt as it is believed to resemble the size of the financial sector ((Levine et al. 2000) and (King & Levine, 1993)). Levine et al. (2000) do suggest this commonly used measure has some short comings as it may not accurately measure the effectiveness of the financial sector with regards to the overcoming of informational asymmetries and easing transaction costs. Data will be drawn from the World bank WDI & GFD data base.

ii) Private Credit: This measure resembles the domestic credit extended to the private sector. It refers to financial resources provided to the private sector, such as loans, non-equity securities, trade credits and other accounts receivable. Similar measures are used by King and Levine (1993), Levine and Zervos (1998) Levine et al. (2000) and Rajan and Zingales (1998) and Galor and Zeira (1993). The data will be drawn from the World bank WDI & GFD data base.

4.3.2 Qualitative measures of Financial Development Qualitative:

iii) Accounting: This measure reflects the standards of financial disclosure in a country. The higher the standards of financial disclosure in a country, the easier it should be for firms to raise funds from investors id est the better the functioning of the financial system. Information about corporations is critical for: exerting corporate governance, identifying the best investment and it will simplify contracting. The Center for International Financial Analysis and Research created an index for different countries by rating the annual reports of at least three firms in every country. These reports where rated on the inclusion, or omission, of 90 items. Accounting standards as a measure of the level of financial disclosure in a country is used before by several scholars; Francis et al. (2005) Levine et al. (2000) and Rajan and Zingales (1998) to enrich research on financial development. As the measure is based on the assessments of company reports in 1990, in line with Levine et al. (2000) we will use this measure for a restricted regressions only based on the years 1980 to 1995. However, Levine et al. (2000) have found similar results for regressions base upon a longer time frame being 1960 to 1995.

12 found similar results for regressions based upon a longer time frame, we will not extend the timeframe.

v) Access: This measure reflects the percentage of adults that has access to financial services in a given country. The estimated access percentage is generated from the estimated number of bank accounts and the average deposit size in a country3.The timeframe underlying the several components of the access to financial services measure range around 2000, we will use this measure on our entire timeframe 1980-2009. We believe access to financial services to be an increasing phenomenon. As the data underlying the composite measure of Honohan (2008) is gathered around 2000, this means that the actual access to financial services level before 2000 was probably lower than the value that we will be using. This means we risk the relationship between access to financial services and both poverty and inequality to be understated.

Levine et al. (2000) show that differences in creditor rights, the ability to enforce contracts and accounting standards explain a significant amount of the cross-country differences in the levels of financial development. They find liquid liabilities to be more closely correlated with enforcement and creditor rights (not included in our analyses), whereas private credit appears to be closely linked to both enforcement and accounting standards.

Table 6 – Correlation Matrix for the Financial Development Indicators

Liquid Liabilities Private Credit Accounting Enforcement Access Liquid Liabilities 1.0000

Private Credit 0.8568** 1.0000

Accounting 0.2836** 0.4800** 1.0000

Enforcement 0.5014** 0.5958** 0.5940** 1.0000

Access 0.5827** 0.6936** 0.6052** 0.8863** 1.0000

** Correlation is significant at the 0.01 level * Correlation is significant at the 0.05 level

In line with Levine et al. (2000) we find that liquid liabilities are more closely correlated to enforcement as to accounting. We also find private credit to show a relatively higher correlation with both accounting and enforcement. With regards to the access to capital measure the only clearly different correlation is found between enforcement and access.

4.4 Data on Educational Attainment

Educational Attainment: Both literacy rates and enrolment rates have proven to be popular measures for scholars to use in empirical research; however, they might not properly reflect reality. Benhabib and Spiegel (1994) and Barro (2000) suggest it may be useful to take into account both the distribution and the level of education, especially when related to income distribution effects.

3

13 As we want to measure both the distribution and the level of education we opted to use the ‘average years of total schooling’ measure constructed by Barro and Lee (2010). As Barro and Lee (2010) explain; “The data on the distribution of educational attainment among the population, combined with the information for each country on the duration of school at each level, generate the number of years of schooling achieved by the average person at various levels and at all levels of schooling combined.”. So, the measure ‘average years of total schooling’ gives us some insight, though limited, into both the distribution and quantity of education in a given country. Taken together, an equal distribution and a longer duration of the several levels of education could be considered to give some indication about the quality of education as well. So, we therefore assume that an increase in the average years of education reflects both an increase in the quality as well as the quantity of the education.

i) Yrseduc will depict average per capita years of education for a person aged fifteen and over. Data will be drawn from the Barro and Lee (2010) database on educational attainment.

As the average years of education per capita depicts completed education it is not necessary to include a lagged variable for the average years of education, as the education is ‘ready use’ as opposed to measures like enrolment rates, which depict educational levels that are not readily available to the country.

5. Research Methodology

Before we move on the methodology, we will depict our econometric model, which will be as follows:

IP

= β

+ β

FD

+ β

EA

+ β

FD

* EA

+ γX

ε

In this regression IP can be reflected by any of the inequality or poverty measures, being Quality2, Quality3, povHCR2 and povGAP2, as depicted before. FD is our measure for financial development which can either be reflected by, Liquid Liabilities, Private Credit, Accounting and Enforcement. EA is the measure for educational attainment which is reflected by the average years of education in a country. X is a set of conditioning information, here we will control for per capita GDP growth, inflation and government size.

The interaction term between FD and EA is the product of two continuous variables and alters the relationship between each of these variables and the dependent variable.

5.1 Estimation Methods

For this research a panel dataset was constructed, therefore we’ll shortly discuss several ways of conducting panel data research in the following section

14 estimation methods (Hill et al. 2008). Using the method of seemingly unrelated regressions would be an interesting option as it would allow us to assume contemporaneous correlation, unfortunately, it is not recommended to use this technique on a short and wide dataset (Hill et al. 2008). This leads us to use either Pooled regression estimator, or a Fixed Effects,- or Random Effects estimator. Pooled regression is a method for pooling time series and cross-sectional data, this approach is used when the groups to be pooled are relatively similar, or homogenous. If running the model yields large standard errors this could be signal that the groups are not homogenous and a more advanced approach, like the Fixed or the Random Effects model, are more appropriate estimation techniques to use (Hill et al. 2008).

In order to determine which test to use we tested for the presence of random effects by performing a Lagrange multiplier-test. When random effects where found we could be sure the pooled OLS model was not the appropriate model to use. Therefore we continued to test the suitability of either the Fixed,- or Random Effects model. The suitability of either model is based upon the assumptions underlying the model and is dependent on the dataset used; in order to make a well informed decision on the best estimation method to use we have performed the Hausman-test, which was ran for all regressions. The Hausman-test allows us to distinguish between using either the Fixed Effects or the Random Effects estimation method. Based on the results of the Hausman-test, we used either the Fixed Effects or the Random Effects estimators for the regression incorporating the quantitative indicators of financial development (being: Liquid Liabilities and Private Credit), depending on which estimation method was found to be the most appropriate.

For the regressions incorporating the qualitative measures of financial development (being: Accounting and Enforcement) which are invariable over time, the Fixed Effects estimation method may not be convenient as it excludes all variables that are time invariable. In this case we opted for either a (pooled) OLS regression or a Random Effects regression. As we already tested for the presence of random effects by performing a Lagrange multiplier-test as discussed in the previous indentation. We concluded a Random Effects regression would be the most appropriate for all regressions which included the qualitative measures of financial development.

5.2 Data analyses 5.2.1 Collinearity

Before running the regression we checked the data for collinearity by looking at the correlation matrices. If, in the correlation matrix explanatory variables show high correlation the variables are said to be collinear, this causes a problem called collinearity. In this case the data is not information ‘rich’. However, as long as the correlation is not found to be exactly one the Gauss-Markov theorem is not violated and we can continue our analysis (Hill et al. 2008). As we do not observe high values for the correlation between the explanatory variables, collinearity is not likely to be a problem in the model4.

4

15 Looking at the correlation matrices can also provide us with some other basic information. Overall, we found that all four measures used to reflect financial development are negatively and significantly correlated with both inequality and poverty. Only accounting appears to be positively correlated to poverty. This indicates that on average countries with more and/or better developed financial intermediaries will experience a faster reduction in inequality and the number of individuals living in poverty (living of less than $2.00 a day). Education is also negatively and significantly related to both inequality and poverty which indicates that countries with higher educational levels will experience a faster decline in both inequality and poverty.

5.2.2 Outliers & Robustness

After checking the correlation matrices we also checked for outliers, which can easily be done by using statistical software. By running the ‘extremes’ command (Stata) on the data for similar variables, for example the financial development measures. Using this approach we can monitor whether an extreme value is found for just one of the variables per countries or that for some reason the values of all variables are high for the entry year (for example as a result of an economic shock). Some values where found that could be labelled as outliers. A common way to deal with outliers is to not exclude outliers right away, probably because of limited data availability, generally regressions are ran first with all data points included. Then, in order to make sure outliers do not ‘drive’ the results these observations are excluded later on and the regressions are run again. This method is, for example, used by Beck et al. (2004). We have followed the same procedure and concluded that our results are not altered much after the exclusion of the outliers5. The regressions based upon the poverty measures are fairly robust against the effects of possible outliers. Overall, after exclusion of the outliers, the signs and size of the coefficients remained largely the same as well as the significance. Some changes appeared for the regression based on the two inequality measures (Quality2 and Quality3), where some observations lost significance because of the exclusion of the outliers, other observations became stronger. Overall, we believe the outliers do not largely affect our findings.

5.3 Diagnostic Tests

As noted before, for the regressions which incorporated the quantitative Financial Development measures we have used either the Fixed Effects or the Random Effects estimation method. After running the regression the normality of the residuals was checked, this was done by looking at the kernel density plots of the residuals. After looking at the kernel density plot a simple test is conducted to check for kurtosis and skewness. When we found the residuals not to be normally distributed we tested for both autocorrelation and heteroskedasticity. If we found autocorrelation and heteroskedasticity to be present they were controlled for these problems by constructing robust and clustered (by country) standard errors.

For the regressions which contain the qualitative Financial Development variables we have used the Random Effects estimation. Again we checked the normality of the residuals after running the regressions by checking the kernel density plots of the residuals and checking for kurtosis and skewness. Here as well we found the residuals not to be normally distributed and therefore we

5

16 checked for both autocorrelation and heteroskedasticity; if present we re-ran the regressions while controlling for autocorrelation and/or heteroskedasticity by constructing robust and clustered (by country) standard errors.

6. Causality

There is the potential for reversed causality in our analyses, for example a simple demand pull relation, between poverty reduction and financial development. A demand pull relation would imply that poverty reduction may simply enhance the demand for financial services, in this way poverty reduction would lead to financial development. Or inequality reduction may alter the political arena demanding more efficient financial markets that channel the funds to their most productive uses, instead of investments decisions based upon political considerations (Beck et al. 2004). In order to establish the direction of the relationship between financial development and both poverty and inequality, we used the instrumental variable approach in which legal origin will function as our possible instrumental variable.

La Porta et al. (1997) explain that most countries can be divided by their legal origins. Legal origins are typically obtained through colonization or occupation and are therefore believed to be exogenous. Most countries possess English, French, German or Scandinavian legal origins. La Porta et al. (1997) show that the legal,- and regulatory environment, which governs financial transactions, is strongly influenced by legal origin. La Porta et al. (1997) argue that the legal and regulatory system will influence the capability of the financial system to offer high quality financial services. As legal origin is exogenous and it can be used to explain variations in the financial system, we suspect legal origin may be a good instrument to use in order to determine the direction of the relationship between financial development and both poverty and inequality. Many other studies have used legal origin as instrumental variables for financial development; among which: Levine et al. (2000) and Lundberg and Squire (2003).

Before using legal origin as an instrumental variable several test were conducted in order to establish the strength and most importantly the validity of the instruments. One can test the validity of the instruments by testing for over-identifying restrictions this can be done by performing the Sargan-test. This test assumes the variables used to instrument the endogenous variable are valid, id est uncorrelated to the error term, if the instruments are not valid, H0 will be rejected. If H0 is rejected it means that the instrumental variables are correlated with the error term, and cannot be used to perform iv-regressions (Hill et al. 2008). After running careful analyses we found that legal origin suffered from over-identifying restrictions when used to instrument for financial development in our inequality regressions (quality2 and quality3). Though, we did find legal origin to be a valid instrument to use in order to determine the causal relationship between financial development and poverty6. Therefore we will use an iv-estimation approach for our regressions on the de poverty measures and a either fixed,- or random effects estimation approach for our inequality regressions.

Below the first stage regression results are shown. The second regression for each financial development measure includes all exogenous variables we will be using in our regression.

17 Table 20 – Instrumental variable regressions for the quantitative measure of financial development.

The first regression solely incorporates the legal origin dummies whereas the second regression includes our moderating variable as well as our control variables.

Liquid Liabilities Liquid Liabilities Private Credit Private Credit legor_uk -29.245*** (10.803) -11.730 (10.249) -41.943*** (7.929) -16.177 (7.619) legor_fr -41.936*** (10.603) -18.333* (10.440) -57.136*** (7.732) -29.630*** (7.602) legor_ge 18.599 (13.126) 65.812*** (14.065) -20.402** (8.917) -14.315 (8.383) yrseduc 5.232*** (1.003) 7.233*** ( .651) GDPpcgrowth -.575 ( .988) -.282 ( .547) inflation .001 ( .001) -.002*** ( .002) government .689 ( .493) .225 ( .314) _cons 86.889*** (10.002) 25.023 (15.552) 93.136*** (7.376) 17.611 (10.394) No. Obs. 214 190 591 535 Prob > F 0.000 0.000 0.000 0.000

* = p-value < 0.10, ** = p-value < 0.05, *** = p-value <0 .01

Legor_uk = English legal origin Legor_fr = French / Napoleonic legal origin. Legor _ge = German legal origin. Scandinavian legal origin is the omitted category.

Table 21 – Instrumental variable regressions for the qualitative measure of financial development.

The first regression solely incorporates the legal origin dummies whereas the second regression includes our moderating variable as well as our control variables.

Accounting Accounting Enforcement Enforcement Access Access Legor_uk -4 (2.683) -.568 (2.750) -2.955*** ( .516) -1.441*** ( .426) -51.233*** (4.980) -26.786*** (4.208) Legor_fr -20*** (2.513) -13.024*** (2.8620 -3.073*** ( .509) -.813* ( .446) -55.127*** (4.815) -29.071*** (4.132) Legor_ge -13.25*** (3.158) -10.450*** (3.604) -.820 ( .644) 1.01* ( .570) -20.55*** (5.487) -16.834*** (4.547) Yrseduc .823* ( .436) .326*** ( .058) 5.319*** ( .360) GDPpcgrowth .370 ( .446) .149** ( .064) .2132 ( .295) Inflation -.004 ( .003) -.000 ( .000) -.001** ( .001) Government .444*** ( .175) .140*** ( .025) .808*** ( .175) _cons 74*** (2.233) 54.744** (5.007) 9.723*** ( .455) 2.960*** (.736) 93.8 (4.610) 23.783*** (5.803) No. Obs. 128 115 152 135 621 521 Prob > F 0.000 0.000 0.000 0.000 0.000 0.0000

* = p-value < 0.10, ** = p-value < 0.05, *** = p-value <0 .01

Legor_uk = English legal origin Legor_fr = French / Napoleonic legal origin. Legor _ge = German legal origin. Scandinavian legal origin is the omitted category.

18 least well-developed financial intermediaries. When we performed the instrumental variable regressions on our poverty measures the results confirmed that the relationship runs through financial development id est financial development leads to poverty reduction, as the results of the instrumental variable regressions were rather similar (in sign and significance) to our fixed,- and random effects regressions7. Though we do find more significant results for our IV-regressions than we find for our fixed,- and random effects estimations. This results from the way in which he standard errors are estimated8.

7_{These results can be found in appendix F.} 8

19 7. The effects of Financial Development on Poverty

7.1Basic results

As we found legal origin to be a valid instrument to use in order to instrument for financial development when regressed on poverty below the instrumental variable regression results for the poverty regressions will be depicted9. The first regression set (table 10) is regressed on the poverty head count ratio and the second regression set (table 11) is regressed on the poverty gap. In these regressions the interaction term between financial development and education has not yet been included.

Table 10 - Empirical results for the instrumental variable regressions on the Poverty Head Count Ratio (the interaction term excluded) Liquid Liabilities PovHCR2 Private credit PovHCR2 Accounting PovHCR2 Enforcement PovHCR2 Access PovHCR2 Fin. Dev. -.676*** ( .120) -.256*** ( .047) .356 ( .331) -14.046*** (1.632) -.526*** ( .086) Yrseduc -7.111*** ( .960) -8.176*** ( .542) -7.190*** (2.420) -7.973*** (1.107) -6.977*** ( .595) GDPpcgrowth .379 ( .887) 1.305*** ( .447) .411 ( .920) 1.510** ( .633) .923** ( .449) Inflation -.008 ( .005) -.006* ( .004) -.003 ( .006) -.009*** ( .002) -.009** ( .004) Government .966 ( .620) -.387 ( .266) .257 ( .834) 1.418* ( .713) -.537** ( .266) _cons 61.895*** (10.285) 88.997*** (3.941) 68.0145*** (20.517) 44.548*** (12.280) 82.527*** (4.629) No. Obs. 94 285 42 53 290 Prob > F 0.0000 0.0000 0.0057 0.0000 0.0000

* = p-value < 0.10, ** = p-value < 0.05, *** = p-value <0 .01

1) Financial development (reflected by the measure indicated above the regression) is instrumented by legor_uk, legor_fe and legor_ge also in the interaction term. 2) The results are obtained by substituting the financial development measure in an OLS regression (excluding the interaction term) and obtaining the fitted variables.3) All standard errors are made robust to control for heteroskedasticity.

20 Table 11 – Empirical results for the instrumental variable regressions on the Poverty Gap (the interaction term excluded)

Liquid Liabilities PovGAP2 Private credit PovGAP2 Accounting PovGAP2 Enforcement PovGAP2 Access PovGAP2 Fin. Dev. -.436*** ( .075) -.168*** ( .027) .159 ( .132) -6.439*** ( .699) -.296*** ( .047) Yrseduc -3.808*** ( .553) -4.394*** ( .322) -3.236*** (1.150) -4.114*** ( .520) -3.788*** ( .345) GDPpcgrowth -.257 ( .481) .451** ( .229) -.227* ( .455) .246 ( .266) .262 ( .237) Inflation -.005** ( .002) -.003** ( .001) -.002** ( .003) -.004*** ( .001) -.005*** ( .002) Government .684** ( .339) .005 ( .141) .101 샿샿.355) .734** ( .294) -.085 ( .146) _cons 27.031*** (5.506) 41.613** (2.259) 30.230*** (9.950) 21.217*** (5.302) 38.895*** (2.615) No. obs. 94 285 42 53 289 Prob > F 0.0000 0.0000 0.0057 0.0000 0.0000

* = p-value < 0.10, ** = p-value < 0.05, *** = p-value <0 .01

1) Financial development (reflected by the measure indicated above the regression) is instrumented by legor_uk, legor_fe and legor_ge also in the interaction term. 2) The results are obtained by substituting the financial development measure in an OLS regression (excluding the interaction term) and obtaining the fitted variables.3) All standard errors are made robust to control for heteroskedasticity.

The coefficients for the financial development variables in both table 10 and 11 are all but the coefficient found for accounting, negatively and significantly related to poverty. This tells us that financial development reduces poverty. These findings are in line with the findings of Beck et al. (2004) and Honohan (2004). The overall effect of accounting standards on poverty appears not to be significantly different from zero. However this may be the result of the very limited number of observations included in this regression.

21 Table 12- Empirical results for the instrumental variable regressions on the Poverty Head Count Ratio (the interaction term included)

Liquid Liabilities Private credit Accounting Enforcement Access

PovHCR2 PovHCR2 PovHCR2 PovHCR2 PovHCR2

Fin. Dev. -2.232*** ( .299) -1.238*** ( .137) -5.894*** (1.306) -14.188*** (5.093) -1.631*** ( .235) Yrseduc -4.141*** ( 1.04) -5.696*** ( .605) -.862 (1.780) -7.934*** (1.976) -5.533*** ( .658) Yrs*FD .274*** ( .049) .136*** ( .018) .931*** ( .194) .023 ( .853) .152*** ( .029) GDPpcgrowth .359 ( .798) 1.491*** ( .449) .693 ( .852) 1.512** ( .638) 1.005** ( .435) Inflation -.006 ( .004) .000 ( .004) .003 ( .003) -.009*** ( .003) -.006* ( .004) Government .674 ( .533) -.170 ( .248) .608 ( .648) 1.417* ( .728) -.635** ( .255) _cons 47.767*** ( 9.339) 67.690*** (5.120) 20.277 (14.070) 44.325*** (14.670) 71.367*** (5.186) No. obs. 94 258 42 53 290 Prob > F 0.0000 0.0000 0.0000 0.0000 0.0000

* = p-value < 0.10, ** = p-value < 0.05, *** = p-value <0 .01

1) Financial development (reflected by the measure indicated above the regression) is instrumented by legor_uk, legor_fe and legor_ge also in the interaction term. 2) The results are obtained by substituting the financial development measure in an OLS regression (excluding the interaction term) and obtaining the fitted variables. 3) These fitted variables were multiplied by the average years of education in order to be able to incorporate the interaction term. 4) All standard errors are made robust to control for heteroskedasticity.

Table 13 – Empirical results for the instrumental variable regressions on the Poverty Gap (the interaction term included)

Liquid Liabilities Private credit Accounting Enforcement Access

povGAP2 povGAP2 povGAP2 povGAP2 povGAP2

Fin. Dev. -1.422*** ( .219) -.817*** ( .086) -2.445*** ( .614) -10.723*** (2.378) -1.067*** ( .141) Yrseduc -1.794*** ( .520) -2.714*** ( .324) -.560 ( .759) -2.942*** ( .747) -2.798*** ( .338) Yrs*FD .1769*** ( .033) .090*** ( .010) .389*** ( .090) .683* ( .367) .107*** ( .017) GDPpcgrowth -.211 ( .406) .582*** ( .224) -.113 ( .467) .294 ( .263) .325 ( .221) Inflation -.003** ( .001) .001 ( .003) .000 ( .001) -.004*** ( .001) -.003** ( .002) Government .522** ( .247) .163 ( .127) .239 ( .289) .683** ( .271) -.152 ( .134) _cons 16.929*** (4.469) 27.103*** (2.694) 10.200 (6.692) 14.604** (5.706) 31.232*** (2.610) No. obs. 94 285 42 53 289 Prob > F 0.0000 0.0000 0.0000 0.0000 0.0000

* = p-value < 0.10, ** = p-value < 0.05, *** = p-value <0 .01

22 As one can see in the tables 12 and 13, the interaction term between financial development and education has been included. We find the interaction term to be positive and significant in almost all regressions. The main effects of financial development and education on poverty should now be interpreted in combination with the interaction term. Apparently an increase in educational attainment will offset the poverty reducing effects of financial development. These findings are not in line with the findings of Beck et al. (2004) as they did not find any significant results when incorporating the interaction term between financial development and education. However, Beck et al. (2004) did use the growth rates of both head count ratio, - and poverty gap as their dependent variables, these

measures are not closely related to our dependent variables we have shown in the data section (section 4.2).

So, the marginal effect of financial development on poverty does depend on the level of educational attainment within a country. The level of education should be multiplied by the coefficient of the interaction term and added to the coefficient of financial development in order to determine the marginal effect of financial development on poverty for a certain level of education. In order to form a clearer picture of this effect we will extensively discuss the effect and its possible underlying reasons in the following section.

7.2 The marginal effects of Financial Development on Poverty

In this section we will analyse how the relation between financial development and poverty is altered for different levels of educational attainment. As we saw in the previous section financial development and education appear to be substitutes, this means that an increase in education will offset the poverty reducing effects of financial development. Why does educational attainment offset the poverty reducing effect of financial development and for what levels of educational attainment does the poverty reducing effect of financial development become zero? These questions will be answered in the following section. The graph below shows the marginal effect of financial development on poverty while average years of education increase10.

23 Graph 1 – The marginal effect of Financial Development of Poverty

-. 4 -. 2 0 .2 d Po v G A P2 /d Pr iv a te C re d it 2 4 6 8 10 12

Average years of Education

[95% Conf. Interval} dIP/dFD increasing average years of education

The marginal effect of Financial Development (PC) on Poverty (povGAP2)

As can clearly be observed in graph 1, financial development reduces poverty; at least for countries with low average levels of education. For countries with on average higher levels of education (education>=8) the effect of financial development on poverty is no longer significantly different from zero. As the upper bound of our confidence interval becomes > 0 the effect of financial development on inequality is no longer significantly different from zero. This mea ns that when the level of educational attainment becomes 8 years or higher, financial development will no longer have a poverty reducing effect. So an average educational level of less than 8 years appears to be the threshold value for financial development to reduce poverty.

How does financial development reduce poverty and why is this effect different for countries with high levels of educational attainment?

In order to answer this question one should think about the following:

24 inequality.) The poor may use the loans extended to them to: either invest in human capital (get educated, as suggested by many theoretical models) or to start their own (small business) and becomes self-employed. Going back to the fundamentals of this research the second option is the most likely to be underlying our results, as we have not incorporated financial development as a lagged variable11. Therefore poverty relieving effect of financial development must lie in the present. As getting educated is a process of which the results cannot be enjoyed until later this effect cannot be underlying our results.

As we saw financial development reduces poverty in countries with on average low levels of educational attainment. In many less developed countries (countries with on average lower levels of educational attainment), many people are self-employed and work in the tertiary sector. They have small businesses in which they sell good (crafts, ice, fruits or do other handy work like shoe shining) to their richer counter parts or tourists. McKernan (2002) finds that supplying financing to the poor can exert a large and positive impact on self-employment profits. Because of the characteristics of these jobs the results of such loans will be rather immediate, which therefore can explain our findings. For a review of literature regarding microcredit one should see Brau and Woller (2004).

How come these poverty reducing effects fade away when average years of education increase? Basic textbooks tell us that credit will flow to its most productive use. As we have postulated that education increases productivity (Benhabib and Spiegel 1994) it is likely that the loans are extended to individuals who have benefited from this increase in education, leaving the poor behind. So, an increase in education may alter the (limited) flow of capital which no longer flows to the poorest individuals (individuals living of less than $2.00 a day), but only to the lower-middle, middle or higher class who are likely to be relatively more educated than ‘the poor’. However, this may not be the only reason why the poor no longer appear to benefit from financial development as average level educational attainment increases.

In low and middle-income countries exclusion from financial services is next to normal for the bulk of the population therefore, these groups mostly use informal channels when it comes to financing (Honohan, 2008). According to Honohan (2008) in advanced economies a small group of (poor) individuals does not have access to any financial services (like a transaction,- or a savings account) and this group most certainly does not have the ability to loan from a formal intermediary. Honohan (2008) explains this group can become excluded for several reasons; these can be product features (like a high minimum covers for a product or service), more material obstacles (for example: when having a fixed address is a pre-requisite in order to be able to open an account) or this group may be screened-out because of the risk characteristics they possess. This sounds logical and would explain why the poor cannot benefit from financial development through formal channels, but how about

11

25 informal channels? We suspect there might be a relationship between the level of education in a country and being excluded from informal financial system.

This suspicion is confirmed by the findings of Guiso et al. (2004). Guiso et al. (2004) analysed the access to finance through both formal and informal channels, for many countries with different overall levels of development. Guiso et al. (2004) for one found that low trust areas are often characterized by a more intense reliance on transactions within small subgroups, subgroups like family and friends. Second, they also found that the likelihood of receiving a loan from a relative reduces with the level of trust prevalent in the area. Thirdly, Guiso et al. (2004) found that the number of individuals using the formal financial system increases with average years of education. In their research Guiso et al. (2004) associate low trust areas with low educational levels and high trust areas with high educational levels. This tells us that one’s ability to benefit from financial development through informal ties reduces as average levels of education increase. This can clearly explain why we found financial development not to benefit the poor (not even through informal channels) in countries that possess high levels of educational attainment. These findings support the assumptions of Greenwood and Jovanovic (1990) as they believe the poor benefit from financial development mainly through informal family ties.

Summing up, we found financial development to reduce poverty; at least for countries with low levels of educational attainment. We believe the reason the results differ for countries with on average higher levels of educational attainment has to do with the channels through which individuals can access finance, either formal or informal channels and whether individuals can access finance through these channels. In addition there still is the fact that finance will always be looking for the most productive opportunity, therefore an increase in average educational levels can easily alter the (limited) flow of capital. This alteration of the flows of capital and the formalization of the financial sector will render the poor excluded from the financial sector; therefore these individuals will no longer benefit from financial development.

8. The effects of Financial Development and Education on Inequality

8.1Basic results

The two regression sets below show the result of regressing the different measures of financial development on inequality12. The first regression set is regressed on the Quality2 and the second regression set is regressed on the Quality3. In these regressions the interaction term between financial development and education has not yet been included. We would expect to see rather similar results for both regression sets as the measures Quality2 and Quality3 resemble the same inequality measure, however as explained in section 4.2 they do reflect different levels of reliability of the underlying data.