An investigation into the determinants of income inequality and testing the validity of the Kuznets Hypothesis

Transcription

1 Mälardalen University Västerås, School of Sustainable Development of Society and Technology (HST) Bachelor Thesis in Economics Tutor: Dr. Johan Lindén An investigation into the determinants of income inequality and testing the validity of the Kuznets Hypothesis Evaluating its relevance within Japan and China over time Yasir Khan Meenal Javed

2 Date: Level: C-thesis in Economics, 15 hp Title: An investigation into the determinants of inequality and the validity of the Kuznets hypothesis Authors: Yasir Khan and Meenal Javed Supervisor: Johan Lindén Abstract This study deals with testing some of the most widely discussed determinants of inequality in literature within China and Japan, a developed and a developing country. The second part of the paper deals with testing the Kuznets hypothesis in the two countries. First, we consider the relevant literature on the topic and then employing at least three different functional forms of our regression models, we conduct tests of the inverted-u relation for both countries. The results of our study show that secondary education has a statistically significant and a negative relationship with inequality while none of the determinants considered proved to be significant for Japan. A Kuznets type relationship is also confirmed for China by using GDP and GDP 2 as independent variables while we find that the Japanese trend in inequality is better explained through a cubic hypothesis suggested by Tachibanaki (2005). In our conclusion we speculate about a possible recurring trend of income inequality following the pattern of: inequality-equality-inequality. 2

3 Contents Abstract... 2 Section 1: Introduction Introduction Aim Method... 6 Section 2: Why Japan and why China? Reasons for focusing our study on these two countries Background of China and Japan... 9 Section 3: Data Data collection: Delimitations of data and method: Section 4: Theoretical Framework Kuznets Hypothesis: The Gini Coefficient as a measure of inequality: Gross domestic product (GDP) as a measure of development: Section 5: Determinants of inequality: Why does income inequality exist? The Determinants to be investigated Average income per capita (GDP per capita) Foreign Direct Investment (FDI) inflows International openness indicator Urbanization ratio Urban-rural income gap Average years of secondary and tertiary schooling Percentage of population aged 65 and above Empirical Analysis of Determinants China Japan Determinants of inequality: results summary:

4 Section 6: Kuznets Hypothesis The debate on Kuznets Hypothesis: a review of relevant literature Empirical Analysis for Kuznets Curve China Japan Analyzing and graphing the derived curves China Japan Summary of results for Kuznets tests: Section 7: Conclusion Appendix: References:

5 Section 1: Introduction 1.1 Introduction According to the IMF (International Monetary Fund, 1999), global output has grown by more than 4% per year over the course of the past decade. World Bank estimates show that over the period of , the number of people living on less than $1.25 per day has fallen from 1.4 to 1.9 billion. However, over roughly the same period from the global inequality has increased from 64.8 Gini points to 70.8 Gini points; an increase of approximately 9.3 % (World Bank, 2009). Therefore, despite progress, the benefits of development have not been experienced evenly. Regional disparities in standards of living and income inequalities are mounting issues in both the developing and the developed world. This has raised serious questions for policy makers globally on how to tackle disparities. There has been a heated debate on the trends of inequality and its relationship with growth in different countries. The pivotal point in this debate rests on one question that holds significant importance for those countries that have been battling poverty and inequality: Is there a trade-off between equity and growth or are they complementary? (IMF, 1999) This question is incorporated into a proposition formally known as the Kuznets hypothesis. Before delving into the subject of causes and analysis of inequality trends it is useful to consider why inequality should be regarded worthy of investigation and what can be gained by studying it: Apart from being an interest area for policy makers who value the moral aspects of equality such as social justice and fairness, inequality is widely researched due to its implications for poverty reduction. It is known that for a given level of average income, the more unequal the income distribution is the more people will live below the poverty line (Weil, 2009). Due to the close relationship between income distribution and poverty; policies that promote equity can help tackle income inequality. In a recent study, Fosu (2010) has suggested that inequality affects the transformation of growth to poverty reduction, implying that a high level inequality would thwart the increased benefits of growth. Regarding the responsiveness of poverty to economic growth, Ravallion (1997) showed that growth elasticity of poverty decreases with inequality. However, the mainstream literature on the topic maintains that the average level of GDP is the most important determinant of well being of the poor (Dollar and Kraay, 2002). One argument that rejects growth 5

6 as being always beneficial for the poor is represented by the Kuznets Hypothesis. It suggests that in the early period of development income distribution worsens with growth and then improves as the wider population catches up with the rise in income. If this were true it would suggest that growth can be detrimental for the poor and the strategies for battling poverty would see a significant change. Thus, explaining the causes of inequality and its trends are truly worthy of investigation. It is exactly these subjects which will be made the focus of this essay. The famous Kuznets hypothesis will also be subjected to debate and tested in order to prove/disprove its validity within the case of Japan and China; in addition the causes of income inequality in Japan and China will also be touched upon in this paper. 1.2 Aim The aim of our paper is to first consider the determinants of inequality in both China and Japan and to test the relevance of some of the most widely investigated causes of inequality in literature. The second part of our paper tests the validity of the Kuznets Hypothesis in the two countries. This is especially interesting since China is a developing country and Japan is an industrialized, developed country. Finally, we hope to draw reasonable conclusions regarding the following issues: Which is the most important determinant of inequality in case of each country? Is the trend of inequality in both China and Japan characterized by the Kuznets Hypothesis or does an alternate explanation exist? 1.3 Method We have first provided a foundation for our analysis by laying down the theoretical framework for our investigation, through discussing the measurement of inequality and development which are the main components of Kuznets hypothesis. Then we have reviewed the relevant literature regarding the most extensively investigated causes of inequality which are then tested in the two countries for their relative relevance; these include: Gross Domestic Product (GDP), Foreign Direct Investment (FDI), International openness indicator, urbanization ratio, urban-rural income gap, secondary and tertiary schooling and population aged 65 and above. In order to test the effect of these control variables on inequality, a multiple regression model was used with the figures for these variables obtained from World Bank, Ministry of Health, Welfare and Labor of Japan and NBS (National Bureau of statistics of China); spanning the period for China and for Japan. 6

7 Then our paper tests the much debated Kuznets Hypothesis which has gained significant popularity in the field of development economics as a universal socio-economic phenomenon. Most of the studies in the past decades have concentrated on testing its validity in a cross-section; however, we argue in light of the most recent literature that a cross-sectional test at a single point in time is inappropriate for testing this hypothesis. In the past 10 years, it has become increasingly possible to test Kuznets relationship over time, as many recent studies have, owing to the availability of better quality and longer time series data thanks to the efforts of organizations like World Bank. Thus, we have undertaken the task of testing the Kuznets relationship over time in both Japan and China. For this purpose we have used three possible functional forms of a multiple regression model to test the Kuznets relationship in the two countries. 7

8 Section 2: Why Japan and why China? 2.1 Reasons for focusing our study on these two countries Simon Kuznets (1955) suggested that the level of economic development was the underlying factor responsible for determining income inequality trends in a specific country. The position of a country on Kuznets inverted-u curve trajectory depends on the stage of development that a country has acquired. Thus, we have chosen to investigate China as a developing country and Japan as an advanced, industrialized nation. According to the Kuznets process, a developing country in the early stages of economic development must be in the increasing phase of the inverted-u curve while the developed country should be in its decreasing phase. Thus, by selecting a typical developed and a developing country we can investigate whether this proposition is true. China is especially an interesting choice as it is not only one of the fastest growing economies in the world but also one with a rapidly increasing level of inequality; in 1990 the Gini index was 35.5 and it reached 44.7 in 2004 which is a drastic rise of 26%. According to the Gini index in World Bank Report 2004, China ranked 85 th out of 120 economies. Of the 35 countries ranking below China, 13 had negative GDP per capita growth in , (Xiaolu, 2006). Thus, it would be interesting for our investigation to consider the Chinese case of rising inequality and infer whether it will reach its maximum and start to decline as the Kuznets process suggests. A lack of such a maximum would show that we cannot expect economic growth to correct inequality by itself. Interestingly, Japan has also displayed an increasing trend in income inequality in the past decade despite being an advanced, industrialized nation. This can be observed from the difference between the highest and lowest level of inequality experienced by Japan over the past three decades; the Gini index of primary income was in 1981 at its lowest and in 2002 at its highest, thus, yielding a difference of around 0.15 index points which amounts to a substantial change in inequality (Tachibanaki, 2005). Japan is currently experiencing the same situation as many other European developed countries such as Netherlands, Norway and UK where the level of inequality has been rising since 1980s (Atkinson, Rainwater and Smeeding, 1994). Therefore, in the case of Japan we can show how this recent increase has changed the direction of the overall inequality trend and our study can infer whether the Kuznets relation is dead or alive in case of a typical developed country. If rejected, we will suggest what alternate explanations exist. 8

9 2.2 Background of China and Japan We pause here to consider the major changes in Chinese and Japanese economies in relation to growth and economic development as both have under gone some drastic reforms. China is currently the second largest economy after the United States with a Nominal GDP of trillion US$ in 2009 (World Bank). The GDP growth of China averaged around 9.5% for the last decade, making China one of the fastest growing economies. The GDP per capita of China was 3,744 Yuan in The potential depicted in China s economy was not always like this, more than 30% of Chinese lived under poverty in 1978 (Naughton, Ravallion and Chen, 2007). The GDP per capita in 1978 was 314 Yuan, (Wei & Chao, 1982). The Chinese economy went under a complete overhaul in 1978 and new economic reforms were implemented to fight poverty. The Economic Reforms of 1978 saw the implementation of a dual-track system to grow out of the planned economy and move towards market economy. The dual-track system, the most distinctive element of the reforms, was to have a coexistence of traditional planned and market channel system (Naughton, 1996). Apart from that an Open-Door Policy was implemented and changes in foreign trade policy system were made to enhance the foreign direct investment in China (Wei and Chao, 1982). These economic reforms paid dividends. In 1993 only 56% of the total labour force worked in the agriculture sector, as compared to 71% in The overall average growth for the same period was 9.5%. The integration of China into the world economy was equally dramatic: Trade(exports plus imports) rose from 10% of GNP in 1978 to 36% in 1993 and the foreign direct investment was $28 billion in 1993 compared to $2 billion in 1978 (Woo, Parker and Sachs, 1997). Now, turning to Japan that has the third largest economy in the world, with a nominal GDP of $5.068 trillion and a per capita GDP of $5.068 trillion in Japan has also experienced drastic growth in the past century due to technological, industrial and structural changes but this has recently slowed (CIA, 2011). The start of Japan s historical growth and development can be tracked backed to the famous Meiji Restoration of This initiated many important reforms where the feudal system was stamped out and western legal and educational systems were adopted. However, the most significant change came to Japan in the after math of the Second World War; structural reforms 9

10 were implemented during the US occupation which changed the course of the Japanese society and economy (Minami, 1998). These reforms included: the dissolution of ziabatsu monopolies, decline in landlordism due to land reform of , unionization of labor (which reached 50% in 1940s), increase in farm income due to government policies of farm price support and progressive tax reforms of (which reduced inequality), (Minami, 1998). This resulted in a fall in urban-rural disparity in post-war Japan and led to an era of rapid economic growth which spanned the 1950s until the 1980s; during this period Japan experienced one of highest economic growth rates in the world that averaged 10% in the rapid growth era of the 1950s-1960s (Tachibanaki, 2005). However, the economic growth slowed down dramatically when the bubble economy collapsed in the 1990s. In the 1990s GDP grew at a rate of 1% yearly compared to the 4% yearly growth in 1980s. Japanese economy later fell into one of the worst recessions in 2008, reporting the GDP growth for that year of -5.2% in 2009 (CIA, 2011). The slow growth and recession experienced after the end of the bubble economy has completely changed the lives of the Japanese people. In Japan the number of the unemployed has risen to 4 million and the number of homeless people has gone over 30,000 (Tachibanaki, 2005). Section 3: Data 3.1 Data collection: The data on income inequality for China was obtained from WIID2c database provided by UNU-WIDER (United Nations University World Institute for Development Economics Research) and the values for some years were also used from a study undertaken by Wu and Perloff (2004) calculating the Gini measure of inequality from the data provided by surveys carried out by NBS China (National Bureau of Statistics of China). The WIID2c database contains a high quality data-set which has been subjected to various adjustments to facilitate comparability. The Gini index values for income inequality range from The WIID2c database estimates for China are also based on surveys carried out by NBS China. The data on income inequality for Japan has been obtained from the Income Redistribution Survey (IRS) for the years , organized by the Ministry of Health, Labor and Welfare for Japan. However, in the empirical analysis, when a longer time trend using available observations from over the course of last century was to be considered; the Deininger and Squire (1996) database provided by World Bank was used. The data from this database was 10

11 used to provide values from the pre-war era and the early post-war era for which the IRS values do not exist, which is the period spanning ; the database values were also used to extend the time series data beyond this period. The World Bank online database provided data for GDP per capita, FDI (foreign direct investment), Indicator of Openness, Population and Education. Data used for China was from and for Japan it was The time series data for Gross domestic Product (GDP) in local currency units (LCU) was obtained from World Bank online database for Japan and China. These values were deflated using the GDP deflator provided for each country by World Bank; in order to control for inflationary effects. The resulting real GDP was then divided by the total population of each country to obtain the average income per capita (GDP per capita). The international openness indicator has been calculated as the ratio of export plus imports to GDP; the values of exports and imports were obtained from World Bank in local currency units (LCU). These were deflated using the GDP deflator for the respective countries before calculating the final value of the indicator. The figures for the percentage of population aged 65 and above was calculated by dividing the figure for population aged 65 and above by the total population. The urbanization ratio has been calculated as a percentage of urban population to the total population. The urban-rural income gap has been calculated by taking the ratio of per capita annual disposable income of urban households and per capita annual net income of rural households. This ratio was only calculated for China, as the data was readily available; while in case of Japan because data regarding the average income in urban and rural households was not available separately, therefore this variable was not included. The values used to calculate the ratio were taken from NBS China for the period spanning 1985 to Delimitations of data: The limitations in our data lie in the fact that in order to prolong the time series observations for Gini index for both Japan and China we have used figures from two different sources which might have an impact in terms of level of comparability of these values; since we have not attempted any adjustments ourselves. Barro (2000) has also voiced these concerns: Differences in method of measurement arise due to aspects such as: whether the data is for individuals or 11

12 households, whether inequality is calculated for income gross or net of taxes or for expenditure rather than income, (Barro, 2000, p.17). Again, it s the trend we are interested in, not the exactness of measurement. Section 4: Theoretical Framework 4.1 Kuznets Hypothesis: Simon Kuznets (1955), a Nobel laureate, presented his now famous inverted-u hypothesis which formally came to be known as the Kuznets Hypothesis. This proposition is one of the most enduring arguments in the history of social sciences. It was first made public in Kuznets Presidential Address to American Economic Association in 1954 and later discussed in his famous paper in 1955 called: Economic growth and income inequality (Moran, 2005). In his discourse on the size distribution of income and how it varies with the level of development, Kuznets attempted to explain the decrease in income inequality in the 1920s in developed countries such as USA, England and Germany after a period of stability in inequality trends. This was accompanied by increases in average per capita income. He called this decrease a puzzle and suggested that there were at least two forces supporting an upward trend in inequality; the concentration of savings at the top of the income distribution and urbanization, as a consequence of industrialization (Kuznets, 1955). Kuznets attempted to explain the latter part of the puzzle by taking a simplistic approach, he divided the economy into two sectors: agricultural/rural and industrial/urban. He further suggested that the sectoral change, where there is a shift away from agriculture and towards other urban sectors, should cause an increase in inequality. He explained the reason for this to be that the exacerbation of the income gap between urban and rural citizens due to the shift from rural to urban sectors. This expected worsening of income gap was justified in the following way: the distribution of income in the urban sector is wider than in the rural sector; thus a shift of population from a more equal (rural sector) to a more unequal (urban sector) would increase the weight of the more unequal sector, thereby increasing inequality. Also, the average per capita income is higher in the urban sector than the rural sector, thus increasing the urban-rural income gap as more rural population shifts to the urban sectors. This should result in a rise in overall inequality according to Kuznets. However, through a numerical illustration based on several assumptions and inference based on time series data on income distribution, Kuznets suggested that inequality rises at the early 12

13 Income Inequality stages of development, then after reaching its peak, it levels off and then it starts to decline in the mature stages of economic growth. He reasoned that the decrease in inequality in the advanced stages of development is caused due to the rise in income of the poor in the non-agricultural sector, as they benefits from urban facilities of education and health etc. Kuznets called this phenomenon a long swing in inequality (Kuznets, 1955). This theory implies that graphing the level of inequality as a function of the level of GDP per capita would give an inverted-u relationship, formally known as the Kuznets curve. This hypothesis sparked research in the area of growth and inequality as many economists have tried to prove, disprove or explain the hypothesis (Weil, 2009). However, Kuznets admitted to the lack of empirical data for testing this proposition. In his paper he states the paper is perhaps 5% empirical information and 95% speculation, some of it possibly tainted by wishful thinking, thus acknowledging the speculative nature of his proposition (Kuznets, 1955). Now, nearly 55 years later the theoretical and empirical standing of Kuznets proposition is still debatable and subject to uncertainty. Kuznets Curve GDP per capita 4.2 The Gini Coefficient as a measure of inequality: The most extensively used measure of inequality is the Gini coefficient. In order to construct the Gini coefficient for income inequality, the data on the incomes of all the households (or a representative sample of households) is utilized. By first arranging the households form lowest to highest income; we can then calculate the fraction of total income earned by the poorest 1%, 2% and so on (Weil, 2009). This information is then used to graph the cumulative percentage of 13

14 households (from lowest to highest income) on the horizontal axis and the cumulative percentage of income (or expenditure) on the vertical axis. Such a graph is known as the Lorenz curve (Haughton and Khandker, 2009). The Lorenz curve has a bowed shape owing to the level of income inequality. If the income were distributed perfectly equally, the Lorenz curve would be a straight line with a gradient of 1. This line is also known as the line of perfect equality. The degree to which the Lorenz curve is bowed represents the level of inequality and this aspect is made the basis for calculating the Gini Coefficient. The Gini coefficient is calculated by measuring the area between the Lorenz curve and the line of perfect inequality and then dividing this area by the total area under line of perfect equality. For an income distributed perfectly equally the Gini coefficient will have a value of 0 and if the income is distributed perfectly unequally then the Gini coefficient will have a value of 1 (Weil, 2009). Since, either income or expenditure can be used to calculate the Gini index it is important to keep in mind income is more unequally distributed than expenditure. Thus when comparing inequality between different countries one must either use the Gini index based on household expenditure or household income but not to mix the two. Cumulative Percentage of household income Line of perfect equality Lorenz curve Cumulative percentage of households Based on the criteria that form a good measure of inequality, the following are some of attractive features of the Gini index: Gini coefficient is mean independent i.e. if all incomes are doubled, the Gini value would not change. The Gini index is also independent of the population 14

15 size. In addition to that, Gini index satisfies the Pigou Dalton Transfer sensitivity which suggests that the transfer of income from rich to poor should reduce inequality. However, the drawback of using the Gini coefficient lies in its lack of decomposability. Inequality may need to be broken down by population groups, income sources or other dimensions for this purpose the Gini index cannot be used as it does not allow decomposability into additive groups in this case measures like the Theil s index are more suitable. Nevertheless, for this paper due to the sake of ease and availability of data; the Gini index will be used as a measure of inequality (Haughton and Khandker, 2009). 4.3 Gross domestic product (GDP) as a measure of development: The level of inequality is deeply related to level of economic development in a country as initially emphasized by Simon Kuznets (1955) in his famous Kuznets Hypothesis. When examining an inequality trend in a single country over time, its variation with economic growth could provide some useful insight. The level of development in a country can be measured through different indicators; however, the most widely used measure is the Gross domestic Product (GDP). GDP represents the value of all goods and services produced in a country in a year. It can be calculated either as the value of output produced in a country or equivalently as the total income in the form of wages, rents, interest and profits earned in a country. Thus, GDP is known as output or national income, synonymously (Weil, 2009). In our paper we have employed GDP as a measure of development owing to its wide use. However, GDP is not a perfect measure of economic development. Many aspects of economic welfare are not measured by GDP. This has also been pointed out by Simon Kuznets (1973) in his paper Modern economic Growth. He stressed that the conventional measures of GNP (Gross national product) or GDP (Gross domestic product) are not representative of the costs of social and economic structural changes that a country faces in the process of economic growth. In addition to that it also fails to account for other positive effects such as education and health improvements etc. (Kuznets, 1973). Section 5: Determinants of inequality: However, before turning to explaining the trends in income inequality and testing the Kuznets hypothesis in China and Japan, it is useful to consider the respective causes of income inequality and why it exists. 15

16 5.1 Why does income inequality exist? The reason why income inequality exists is because people differ from each other in aspects that have an effect on their incomes. Such differences can occur in possession of human capital, physical capital, location of residence, specific skills etc. While considering the sources of income inequality across countries, it is important to focus on the distribution of different economic characteristics among a population and (about) how different characteristics translate into different levels of income, (Weil, 2009, p.380). Therefore, inequality in a given country could change over time (both) because of a change in the way characteristics are distributed or (and) rewarded, (Weil, 2009, p.381). 5.2 The Determinants to be investigated In our study, we have selected a few explanatory variables that have been widely discussed in literature in relation to inequality. These variables have been tested to evaluate their importance for income inequality in Japan and China. These factors can be roughly divided into three categories: the first category consists of factors related to economic growth, the second category relates to provision of public goods and the third category relates to demographics. For the first group of growth related determinants, the following factors have been chosen to assess their effect on inequality in China and Japan: average per capita income (GDP per capita) and its square, foreign direct investment (FDI), urbanization ratio, international openness indicator and urban-rural income gap Average income per capita (GDP per capita) GDP per capita and its square are used as development measures in order to examine their relationship with inequality and (in a later section) to test the validity of the Kuznets hypothesis. According to the Kuznets Hypothesis, the expected signs of the relationship between inequality and GDP per capita and its square should be positive and negative respectively Foreign Direct Investment (FDI) inflows The relationship between foreign direct investment and inequality has been extensively investigated under the studies dealing with effects of globalization. Typically, income inequality has been found to be positively related to FDI. Evans and Timberlake (1980) state that dependence on foreign capital exacerbates income inequality by distorting the occupational 16

17 structure of developing economies, bloating the tertiary sector and producing highly paid elite and large groups of marginalized workers, (Lee, Nielson and Alderson, 2007). Alderson and Nielson (1999) have found an inverted-u shaped relationship of inequality with FDI stock per capita. They have explained this relationship by systematically linking the FDI inflows and outflows with a country s level of development. As suggested by Dunning (1981), less developed countries have little inward and outward FDI, countries at intermediate development stage have excess inflows over outflows and developed countries have excess of outflows over inflows. The curve therefore, portrays declining dependence on foreign investment. Thus, we would expect a positive relationship between inequality and FDI inflows for China and a negative relationship for Japan International openness indicator The standard trade theory suggests that the effect of opening up an economy to international trade on the income distribution depends on the factor endowments. For countries that are highly endowed in human and physical capital, trade expansion would tend to lower the relative wages of unskilled labor and thereby increase inequality. This would involve an increase in imports of products intensive in unskilled labor and increase in exports of products intensive in human and physical capital. For countries that are highly endowed with unskilled labor, international openness would raise the relative wages of unskilled labor and lead to a lower degree of inequality. This view suggests that openness would raise inequality in rich countries and lower it in poorer countries (Barro, 2000). However, this is in conflict with the general views of the popular debate on globalization which suggests that international openness would mostly benefit the well off groups in society; this effect would be more pronounced when the average income is low. Therefore, openness would increase inequality in poor countries. Barro (2000) has shown that openness has a statistically significant, positive relationship with inequality and that the openness ratio has a more pronounced positive relation in poorer countries and gets weaker as the country gets richer (Barro 2000). Thus, we expect a significant, positive relationship of openness with inequality for China and a very weak positive relationship for Japan. 17

18 5.2.4 Urbanization ratio Kuznets (1955) has pointed out in his hypothesis the inequality inducing effects of urbanization in the earlier stages of development. During industrialization the migration from agricultural sector to non-agricultural and urban sectors can cause low income groups to rise, causing rising urban inequality without simultaneously reducing rural inequality. This factor is expected to be more important for China which facing a rise in urbanization compared to Japan which is largely an urban nation with only 4% employed in agriculture. Thus, we would expect a positive relationship between inequality and urbanization, at least in China Urban-rural income gap This factor has only been considered for China due to the availability of data and its relative importance for China. Urban-rural income gap contributes to the overall inequality; this has been shown to be true for China according to recent studies. Sicular, Yue, Gustafsson and Li (2006) concluded that in 2002 the urban/rural income gap was the source of ¼ of the overall inequality. Thus, we expect a positive relationship of income inequality with urban-rural income gap. In the second category that relates to the provision of public goods, we have chosen to examine the effect of secondary schooling on income inequality Average years of secondary and tertiary schooling The provision of public education has been found to have a positive effect on efforts for inequality reduction; this is because education helps increase the stock of human capital within middle- and low- income groups and improves their chances of gaining employment (Xiaolu, 2009). In a recent study, Barro (2000) found a negative relationship (although not significant) between inequality and average years of secondary schooling for aged above 15. While higher education was found to have positive relationship with inequality (Barro, 2000). Thus, we expect a negative relationship between inequality and secondary education and a positive relationship between inequality and tertiary/higher education. The third category, in which our last factor falls into, deals with demography. This factor is based on studies, such as the one by Gustafsson and Johansson (1999), which argue that the change in population structure can affect income inequality on the household level. 18

19 5.2.7 Percentage of population aged 65 and above If there are a large number of children or elderly people in a household the mean disposable income would be low. Thus, a rise in number of elderly or young people would mean a greater level of inequality. However, the government policies on welfare such as: pension policy and family policy can affect the relationship between age and equivalent income. We have chosen percentage of population aged 65 and above to investigate the effects of demographic structure on inequality only in Japan; since this factor is more Japan specific. Japan s share of aging population has been on the rise; in 1999 the percentage of population aged 65 and over was 16.7% and in 2009 it was 21.9%. Gustafsson and Johansson (1999) found a negative relationship for developed countries between the percentage of population aged 65 and above and income inequality. However, the negative coefficient could be owing to the policy measures related to income redistribution; therefore it is not clear whether the expected relationship is negative or positive in advance. 5.3 Empirical Analysis of Determinants In Section 5.2 we have discussed the variables which could affect income inequality; in addition to that specific variables for China and Japan were also mentioned. This Section would show the relationship of these variables with income inequality (measured by the Gini coefficient) by using regression analysis. Recall that the factors affecting income inequality were: 1. Gross Domestic Product per Capita (GDPPC). 2. Foreign Direct Investment (FDI) 3. Indicator of Openness (IO) 4. Average years of Tertiary Schooling (TEDU) 5. Average years of Secondary Schooling (SEDU) 6. Urban Population (UP) 7. Urban/Rural Income Gap(U/R Y)[for China Only] 8. Population aged 65 and above (OAGE)[for Japan only] The Regression analysis would be done by dividing the section in two parts. First the results will be discussed for China and then for Japan. 19

20 5.3.1 China To explore the impact of the previously stated variables on China s income inequality, the following regression model is estimated: As explained in Section 5.2; Gross Domestic Product per Capita (GDPPC), Foreign Direct Investment (FDI), Indicator of Openness (IO) and Urban Population (UP), Urban/Rural Income Gap (U/R Y) and Tertiary Education (TEDU) are expected to have a positive effect on income inequality in China, indicated by the positive values of their Betas. The Secondary Education (SEDU) is expected to have a negative impact on income inequality. The data used to estimate the regression model spans the period which is provided in table (A) added in the appendix. Table 1 (below) shows the results of the test we had done with the Gini coefficient as the dependant variable and other several factors; where we have added and dropped some of the variables in each test. Both the tables in this section have some values marked with a single star (*) and double stars (**). A single star (*) indicates that the p- value attained for the relevant variable is relevant at 10% significance level, double star (**) shows a significance level of 5%. The p-values for the variables are stated in brackets under the coefficients of the variables. Variables Test 1 Test 2 Test 3 Test 4 Urban/Rural income gap * ( ) ** ( ) ** (0.007) ** ( ) real GDPPC (LCU) 1.27E-05 ( ) 1.35E-05 ( ) 1.2E-05 ( ) 1.9E-05** ( ) Urban population ( ) ** ( ) ** ( ) ** (1.3E-06) Average years of secondary schooling ( ) ( ) ** ( ) ** ( ) Foreign direct investment ( ) ( ) (0.1399) Indicator of openness ( ) ( ) Average years of tertiary schooling ( ) 20

21 R Square Standard Error F in ANOVA Observations Table 1: The values stated under Test columns are the coefficients for the stated variables in the respective rows. The values stated in brackets under the coefficients in italic are the corresponding P- values for the variables. Before interpreting the results of the analyses we must run a global test. The purpose of running this test is to infer whether the independent variables, all of them, used in the regression model effect the dependent variable at all. Stating it simply, we would test whether all the betas are zero or not (Lind, Marchel and Wathen, 2010). The Null Hypothesis Test is: H₀: β₁ = β₂ = β₃ = β₄ = β₅ = β₆ = β₇ = 0 The Alternative Hypothesis is: H₁: Not all the Betas are zero. Using the F distribution table at 5% significance level, we find that the corresponding value for df (7,13) is 2.83, for Test 1,meaning that if the value of F in ANOVA table is smaller than that, then the null hypothesis test is not rejected. But since from the table (1) we can see that the value of F in ANOVA is , i.e. is greater than 2.66, the null hypothesis is rejected. The co-efficient of determination, stated as R-Squared in Regression Statistics, is defined as The percentage of variation in the dependent variable explained, or accounted for, by the set of independent variables, (Lind, Marchel and Wathen, 2010). In the case of Test 1, R-Square value is , which states that 99.96% of the income inequality is due to the independent variables we used for this model. The R square s value could only lie between 0 and 1. The higher the value is the better the model is (Lind, Marchel and Wathen, 2010). Up until now, the model looks really good but the p-values given in brackets in Table 1 under every coefficient value suggests otherwise. For example p-value of real GDPPC is in 21

22 Test 1. This value is interpreted as there is 65.7% chance that the co-efficient of real GDPPC might deviate from the stated value. Looking at the results only the Urban/Rural Income GAP would be acceptable at 10% significance level. So this result is rejected and no further interpretation is required for these results. If you see at Test 1, the variable with the highest p-value is Average years of Tertiary Education. For Test 2 we ran the regression analysis without this determinant, so that its removal might decrease the p-values of the other determinants and consequently make the results more reasonable. The new estimated model for Test 2 is: Before interpreting the data, we would follow the same prerequisites as we did for Test 1. The global test indicates that we should reject the Null Hypothesis Test, as F distribution Value for df (6,14) is 2.84 and less than value of F in ANOVA i.e The R-Square is the same as it was for Test 1, so the removal of Average years of Tertiary Education did not affect this model at all. By now it would be clearly evident from the table that the p-values for many variables have decreased in high proportion. The Urban/Rural Income Gap and Urban Population are now at 5% level of significance. But still other four variables have p-values that are statistically insignificant. We would now remove the Indicator of Openness, the variable with the highest p- value. We would create a new model for Test 3; This model easily passes the global test as the value of F in ANOVA is and the F distribution Value for df(5,15) is 2.9. Again we see that the removal of the variable does not change the relevance of our model at all, the R-Square is the same as it was in Test 1 and 2. But two of the variables used in this model still have p-values which do not fall in the 10% or 5% level of significance. However, it is obvious that the p-values for all the variables have started to show significance. A new model, which excludes the variable with the highest p-value, is formed for Test 4. 22

23 This model is acceptable, as the table shows that all the variables are marked with double stars (**). The F value in ANOVA is also very high. The R Square also suggests that 99.9% of the change in China s Gini-coefficient is due to the independent variables used in this model. The equation attained is: Looking at the equation above, all the signs of the coefficients are exactly what we expected them to be. GDP per capita (GDPPC), Urban/Rural Income Gap (U/R Y), Urban Population (UP) have a positive impact on China s Gini coefficient, where as Secondary Education (SEDU) has a negative impact on the Gini of China. It is worth mentioning here that as discussed previously, the equation reflects the aspects shared by a growing economy. Since our data consists of observations over time, we will conduct the Durbin-Watson test for auto-correlation. Auto-correlation can cause the results of regression analysis to be inaccurate and this problem mostly occurs with time series data (Lind, Marchel and Wathen, 2010). To conduct the test for auto-correlation the null and alternate hypothesis are: H 0 : No residual correlation H 1 : Positive residual correlation The decision rules for the Durbin-Watson test are: values less than d L lead to rejection of the null hypothesis, values greater than d U will result in the hypothesis not being rejected and values between d L and d U suggest that the result is inconclusive. At a 5% significance level, k (number of independent variables) =4 and (sample size) = 20 observations (for test 5 in table 1); the critical values for d are: d L =0.90 and d U= The value obtained for the Durbin-Watson test statistic is > 1.83= d U. Thus, the hypothesis is not rejected and there is no confirmation of auto- correlation Japan The analyses for Japan would follow the same pattern as for China. The first test would include all the determinants, if the results are insignificant we would remove the variable with the 23

24 highest p-value, in order to obtain the model which is statistically significant. The original model for Japan with all the estimated variables is: The Gross Domestic Product per Capita (GDPPC), Foreign Direct Investment (FDI) and Indicator of Openness (IO), all three of these variables could either have a negative or a positive beta. The co-efficient of the respective betas depend on which stage of development the Japanese economy is in. This has been explained in detail in section 5.2. Both the indicators of education would behave in the same way as they did for China, i.e. tertiary positive and secondary negative. Population aged 65 years and above and Urban Population are expected to increase Japan s Gini coefficient when they increase. The data used to estimate the regression equation spans which is provided in table B added in the appendix. The summary of regression analyses is given in table 2 (pg 25). Japan Test 1, includes all the determinants of inequality, has F in ANOVA at 995 whereas F distribution value for df(7,13) is So the global test is passed. The R Square is at but all of these indicators are statistically insignificant because not even one variable falls in 5% significance level, all the determinants have very high p-values. For Test 2 we exclude Foreign Direct Investment, as it had the highest p-value in Test 1. The new model is: Variables Test 1 Test 2 Test 3 real GDPPC (LCU) -4.5E-08 ( ) -3.9E-08 ( ) -1.5E-08 ( ) Average years of tertiary schooling Average years of secondary schooling ( ) ( ) Urban population ( ) ( ) ( ) (0.2577) ( ) ( ) ( ) 24

25 Indicator of openness (0.506) ( ) ( ) Population ages 65 and above ( ) ( ) Foreign direct investment ( ) R Square Standard Error F in ANOVA Observations Table 2, The values stated under Test columns are the coefficients for the stated variables in the respective rows. The values stated under the coefficients in italic are the corresponding P- values for the variables. Just as we had in Test 1; Test 2 passes the global test with F in ANOVA, 1247, is greater than F distribution value for df(6,14), which is 2.84, the R square is at , still none of the determinants are statistically significant at 5% level. The results for test 3 are shown in the table. This model passes the global test. It excludes the variable with the highest p-value in the prior test, i.e. Population age 65 and above. Still none of the variables fall under 5% significance level. After Test 3 was concluded GDPPC was rendered the most insignificant variable with the p-value of 0.727, which would suggest the exclusion of one of the most important determinants of inequality, therefore the tests were concluded. So it was clear that, according to the data none of the variables proved to be statistically significant as determinants of the Gini coefficient in Japan. A possible reason for this might be that these factors are either not important determinants of inequality in Japan or it could be due to inaccuracy from using a simple estimated regression model for time series data. Perhaps, the cause was auto correlation? We can compute the Durbin-Watson statistic for test 1 in table 2; at 5% significance level, 20 observations and 7 independent variables, the critical values are: d L =0.595 and d U = The value computed for the Durbin-Watson test statistic in this case was >2.339 =d U. Thus, the null hypothesis is not rejected and the residuals are not autocorrelated. Then perhaps these determinants are not important for Japan. 25

26 5.4 Determinants of inequality: results summary: The tests for the determinants of inequality show that in China real GDPPC, urban-rural income gap and percentage of urban population are significant determinants of inequality and they all affect it positively. While secondary education has a negative effect on inequality in case of China. The determinant which has the most influence on the Gini Index for China is secondary education; an increase in the average years of secondary schooling by one year would decrease the Gini index by This might seem a bit drastic but this negative relationship has been confirmed for China even by Xiaolu (2006) who found it to be an important factor. In case of Japan no statistically significant results were obtained indicating that perhaps these determinants are not relevant in case of Japan and that other factors which we have not been included might be at play. All of these results have been tested for auto-correlation and no evidence of it has been found. Section 6: Kuznets Hypothesis 6.1 The debate on Kuznets Hypothesis: a critical review of relevant literature The intellectual history of the Kuznets hypothesis can be broken down into three periods: the first period spans from where the hypothesis came to be treated as a black box, an undisputed fact, forming the foundation of the expanding field of development economics. In the second period from , the inverted U-curve hypothesis was challenged leading to contradictory findings and an inconclusive debate; also described as the opening of the black box. This era led to a change in the way inequality was interpreted and understood in context of development economics. The third period from the 1990s till today is shaped by continuing debate on the validity of the Kuznets hypothesis, where some are still persistent on its soundness while others have provided alternate explanations (Moran, 2005). As stated previously, Kuznets inverted-u hypothesis can either be tested in a cross-section (a group of several countries) at a single point in time or over time within a country. The discussion below considers the debate which revolves around these methods used for the test. During the first period spanning from , (due to the limited availability of overtime observations) there was a surge in the number of cross-sectional studies which confirmed the U-curve hypothesis; the Gini Index as a dependent variable was regressed with a quadratic term in 26