1

Analyzing the Relationship between Income Inequality and Economic Growth: Does the Kuznets Curve Exist in Thailand?

Paravee Meneejuk1 and Woraphon Yamaka1

1. Introduction

The relationship between income disparity and economic growth has been a critical issue in macroeconomics over several decades. Thailand is one of the countries that need economic growth to improve standard of living, reduce poverty, and also to survive in the modern economy. But in the meantime a rise in growth can also cause the inequality which may lead to a long-standing problem which in turn can become destructive to the growth. Alan Greenspan, the chairman of the Federal Reserve of the United States from 1987 to 2006, gave notion on the negative consequences of too much income equality on the economic stability and the democratic capitalist system that “Income inequality is where the capitalist system is most vulnerable. You can’t have the capitalist system if an increasing number of people think it is unjust.” (Greenspan, 2007).

While we seem to consider the inequality as a consequence of economic growth, some economists tend to argue that inequality may affect the growth at the same time and perhaps it is necessary to propel the growth. ‘Two sides of the same coin’ is what Ostry and Berg, IMF staff, used to define the effect of inequality on economic growth through their paper in 2011. They reckoned that the impact of unequal distribution of income is split into two sides. Inequality can influence the growth negatively and become a problem for society. For example, it can obstruct the human capital accumulation especially in education. That is because individuals having higher income can invest more in education than poor people can and have lots of choices in occupation. Recently, IMF has released one more discussion paper about inequality and growth entitled ‘Causes and Consequences of Income Inequality: A Global Perspective’ (Dabla-Norris et al., 2015). One part of this paper refers to a negative effect of inequality on growth and reckons that inequality can also lead to political and economic instability, resulting in corruption, resource misallocation, and nepotism. In

1 Faculty of Economics, Chiang Mai University, Chiang Mai, Thailand Address: mparavee@gmail.com, woraphon.econ@gmail.com

Abstract. This study attempts to explore the relationship between income inequality and economic growth, and particularly the existence of the Kuznets curve in Thailand’s economy. The estimate results show that the Kuznets hypothesis does not hold in the case of Thailand but we can discover an alternative way to reduce income inequality that is the inflation and the share of private sector. Additionally, this study applies the same algorithm for investigating the Kuznets curve to the regions to answer the question ‘Are there regional Kuznets Curves in Thailand?’ Our model shows some interesting results about income distribution in the different regions that may challenge the conventional theory of Kuznets.

2

contrast, inequality can also be good for economy since it provides incentives for people to invest and save for their future livelihood and also for entrepreneurs to run their businesses successfully (Lazear and Rosen, 1981). Inequality does not instantly propel economy but without inequality economy is like socialism where the growth of economy hardly exists.

Trade-off between increasing growth and reducing inequality

While growth is needed, inequality still matters. Some economists suggest that an improvement in income distribution and economic growth can be achieved at the same time (Boonyamanond, 2007). Simon Kuznets, an American economist who received the Nobel Memorial Prize in Economic Sciences in 1971, hypothesizes that as a country develops the economic inequality would first increase and then later decrease (Kuznets, 1955). This hypothesis is graphed as an inverted-U shape called ‘Kuznets Curve’ where the horizontal x-axis is income per capita and the vertical y-axis is economic inequality. This inverted-U shape curve implies that income inequality following the curve is gradually increased resulting from economic evolution. That is, at the initial level of low economic growth, income inequality is also low. Then, as countries seek to increase the rate of economic growth, industrial sector often plays a key role to achieve the goal. This event leads to a big gap between rich and poor, and hence income inequality increases. According to the Kuznets Hypothesis, the level of inequality keeps rising as long as the economic growth increases. However, as the same hypothesis, inequality will decrease after the growth reaches some certain level, a threshold, then the economy is developed efficiently resulting in an increase in average income, an improvement in social welfare of society, higher educated populations, and so on.

Motivated by this reasoning, this paper aims to investigate whether the Kuznets hypothesis holds in Thailand’s economic system where Thai government has been trying to promote the economic growth and especially with a concern about inequality at the same time. This is crucially important because, if we could prove that the hypothesis holds and the result could illustrate the threshold accurately, it means we can find the proper rate of economic growth that will not hurt income distribution in the long term. Our findings should be useful for the Thai government and policy makers and can contribute to longer-run benefits for society.

2. Literature Reviews

As in the literature, many studies aimed to measure the effect of economic growth on the income inequality and paid attention on an investigating of the Kuznets hypothesis. We found the differential results depending on the country’s or region’s characteristic, the data used (type and period of time), and the method for estimation. However, the results can be separated into two groups. The first group is the empirical evidences that quite clearly accept a Kuznets hypothesis in favor of the inverted U-shaped relationship between income inequality and economic growth. We found the existence of the Kuznets hypothesis in many countries such as the study of Heshmati (2004), Khalifa and El Hag (2010) and Ogus (2012). However, there are still some studies that reject this hypothesis or could not find the inverted U-shaped relationship between these two factors. For example, Hsing and Smyth (1994) studied the inverted U-shaped curve of income inequality and economic growth and found an insignificant effect of these two variables through the U.S. data. Moreover, the studies of Deininger and Squire (1996), Easterly (1999) and Dollar and Kraay (2002) also showed some evidences about the nonexistence of the Kuznets curve.

As, we mention above, the heterogeneous results were obtained in the previous literatures. We cannot instantly conclude that the Kuznets hypothesis is held for Thailand. In

3

the case of Thailand, the effect of economic growth on income inequality has been investigated by various studies, for example Ikemoto and Uehara (2000), Noppakhun (2007), and more recent Preechametta, (2015). Some of them could find a little support for the Kuznets hypothesis in Thailand; however, most of the studies were limited by the lack of the efficient statistical tools. These studies typically rely on linear regression in which some of them decide to split the observation into different group and estimate different coefficients. These previous algorithms may face to a bias result easily and, sometimes, cannot capture the complexity of the process. We suspect that the lack of the efficient tools and the estimation method are the important reason that leads to the various results. First and foremost it has to be said that there is no such thing as the right measurement of the effect of economic growth on inequality. The question which method should be chosen is dependable on how suitable the method is explaining the given problem. We found that many empirical studies are now based on the linear rigid quadratic regression model such as the autoregressive distributed lag model (ARDL) in the study of Cheng and Wu (2014) and error correction model (ECM) in the study of Shahbaz (2010). However, the regression results cautioned about estimation of only parametric forms which produce strictly concave functions to test Kuznets hypothesis. Therefore, non-Parametric and semi-Parametric approaches are proposed to examine how income inequality both within and across countries has evolved at different levels of development (see, Frazer, 2006). For a brief review of the methodology used to detect the Kuznets hypothesis, refer to Ganaie and Kamaiah (2015). However, the strong assumption of simple linear relationship between inequality and subsequent growth may lead the low accuracy of the result and it might not be appropriate to investigate the relationship between inequality and growth under the Kuznets hypothesis. Banerjee and Duflo (2003) examine the data without imposing a linear structure and it becomes clear that the data does not support the linear structure and exist the non-linearities in this relationship. Recently, few studies have investigated relationship between inequality and subsequent growth. One of the exceptions is the study carried out by Savvides and Stegnos (2000). They employed a threshold regression model and perform a test of inverted-U hypothesis. The advantage of this model is that we can achieve our test without splitting the sample to estimate different coefficients for poor and rich countries or regions and for different time periods.

In this paper, the non-linearity relationship is taken into account, we employ a Kink regression approach with unknown threshold of Hansen (2015) instead of using threshold regression model since the functional form of the model is discontinuity. Moreover, the Kink regression is extended to smooth Kink regression in order to provide a better fit to inverted U-shaped relationship between inequality and growth. This new model will become one of the contributions of this study.

3. Methodology: Econometric Modeling Framework

Investigating the inverted U-shaped curve or the so-called Kuznets curve is a profoundly complicated problem and difficult to do in the real world because economic growth and income inequality themselves actually depend on many factors. (We will discuss about the factors affecting growth and income inequality later in the fifth part: model specification and variables.) Therefore, an appropriate tool for investigating this phenomenon with special concern about the real mechanism of growth and income inequality becomes difficult, but important. As in the literature, numerous studies failed to find the Kuznets curve especially the empirical studies on time series, perhaps, due to a lack of an effective tool (See e.g. Gallup (2012)).Therefore, this paper is going to present a novel tool for investigating the Kuznets hypothesis that is ‘the simultaneous kink equation (SKE) model’.

4

As we are dealing with a simultaneous relationship between income inequality and economic growth where each equation contains several specific factors, and we speculate that some covariate variables can produce effects both negative and positive since many studies promote the idea that the economy behaves differently if values of certain variables lie in one state rather than in another. Therefore, we modify a sudden switches kink regression model with unknown threshold of Hansen (2015) to smooth switches kink regression model and apply to the simultaneous equation model in order to capture our considerations –identifying the unequal effect of growth on the distribution of income or the so-called Kuznets hypothesis. What is the kink regression? Briefly, it is simply a regression discontinuity model. The regression function itself is continuous but the slope is discontinuous at a threshold point, thereby a kink and a smooth transition between regimes is often more convenient and realistic than just sudden switches transition. With that in mind, we have the simultaneous smooth transition kink equation (SKE) model as an econometric model for this work.

3.1 Introduction to the Simultaneous Smooth Transition Kink Equation (SKE)

The SKE model with m -simultaneous equation can take the form as the following equation where 1( ,...., )t mty y and 1( ,...., )t mt are 1T vector of dependent variable and error, respectively. The term jX is m k matrix of exogenous regressors, and iY is a T G matrix of G endogenous regressors on the right-hand side of the thi equation. 1( ,..., )m is the sigma parameter of margin 1,…, m and 1( ,...., )t mt are the margin error that follows some distributions.

1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

1 1 1 1

1 1 1

(1 F(r ,s )) F(r ,s ) (1 F(r ,s )) F(r ,s )

(1 F(r ,s )) F(r ,s ) (1 F(r ,s )) F(r ,

G G k k

t i i i i i i i i j j j j j j j j ti i j j

G G k

mt m mi mi mi i mi mi mi i mj mj mj j mj mji i j

y Y Y X X

y Y Y X

1s )

k

mj j m mtj

X

(1)

Following Teräsvirta (1998), we separate each regressor variable into two regimes using the functional form of the transition function which takes the form as

( )1F(r, s)

1 ir s Xe

, (2)

where r is a kink or threshold parameter, s is smooth transition parameter and thus F(r, s) has a monotonously increasing function. For thi equation, the coefficients with respect to the right hand side variables equal to 1( ,.., )i mi and 1( ,..., )j mj for value of iY and iX times (1 F(r, s)) . Conversely, they are equal to 1( ,.., )i mi and 1( ,..., )j mj for value of iY and iX times (1 F(r, s)) . Yet the regression function is continuous in all variables. Importantly, this model is assumed to have some correlation between the errors of the m -equations, so that the Copulas are employed to model the nonlinear dependence structure for our model. Here we have the Copula based SKE model with different marginal distributions as suggested by Wichitaksorn, Choy, and Gerlach (2006) and Pastpipatkul, et al. (2016). Therefore, in the next section, the general idea of copula approach will be presented in brief.

5

3.2 Copulas

Copulas are parametrically specified joint distributions which are generated from any given marginal distributions (Trivedi and Zimmer, 2007). The main advantage of copulas is that it can separate the marginal behavior and the dependence structure of the variables from their joint distribution function (Mahfoud and Michael, 2012). This paper conducted both bivariate Elliptical Copulas and Archimedean Copulas to model the dependence structure of the SKE model ( 2)m . Thus, let 1u and 2u be the marginal distribution of 1 and 2 , with

1 1( )u F and 2 2( )u G , then according to Sklar’s Theorem there is a copula function C such that:

1 2 1 2( , ) ( ( ), ( ))C u u C F G (3)

where 1u and 2u are bound on [0,1] and are uniform distribution

Here, the Elliptical class consists of the symmetric Gaussian and t-Copula. The Archimedean Copula consists of the asymmetric Gumbel, Fank, Clayton, and Joe. (See Cherubini, Luciano and Vecchiato, 2004 and Wichitaksorn, Choy, and Gerlach, 2006)

3.3 Estimation of the Copula based SKE model

In this paper, two equations SKE model is considered. Thus, the bivariate Copula is conducted for joining these two equations together. Prior to the model estimation, we need to check the stationarity of our data using the Augmented Dickey-Fuller unit roots test.

The estimation technique that we employ here is a maximum likelihood method where the log likelihood function of our model can be defined as:

1 1 2 2 2 2 1 21 1 2 2 1 2 1 2 1 1 1 1 1log ( ) (log ( ) log( , , ) log ( , ), , , , , , , , , , ,TiL l y Y X c u uy Y y Y X X u u y Y X (4)

where 1 1 1 1 1 1 1 1 1 1, , , , ,s ,s , ,i i i j i j j jr r , 2 2 2 2 2 2 2 2 2 2, , , , ,s ,s , ,i i i j i j j jr r and are the estimated parameters of equation 1, equation 2, and copula function, respectively. In this study, we proposed four different marginal distributions consisting Normal, Student-t, skewed Normal, and skewed Student-t distributions. Copula families, i.e. Gaussian, Student-T, Clayton, Gumbel, Frank, and Joe are proposed to join those marginals in order to construct the joint distribution function of a bivariate random variable with the univariate marginal distribution. 4. Simulation Study

We conduct a simulation study to explore the performance and accuracy of our proposed model ‘Copula based SKE’. We apply the Student-t copula to model the dependence structure of the bivariate SKE model and set the true value for the correlation coefficient at 0.5 while the additional degrees of freedom parameter c is equal to 4. Then, the obtained uniform data are transformed to be 1t and 2t using the quantile function of marginal distribution choices where 1 2and are equal to 1, the additional degrees of freedom parameter of quantile function of the Student-t distribution, , is equal to 6. Thus, we can write our simulated Student-t copula based SKE with normal and Student-t margins as in the following equations:

6

1 1 11 11 11 1 11 11 11 1 1 1

2 2 21 21 21 1 21 22 22 1 2 2

(1 F(r ,s )) F(r ,s )(1 F(r ,s )) y F(r ,s ) y

t t t t

t t t t

y X Xy

.

To generate the dependent 1ty of the first equation, the independent variable 11X are randomly simulated from (15,10)N where the kink point or threshold value for 11X is 11r =15, the smooth parameter 11s =0.5, and the true coefficients for 1 1 , 11 0.5 and 11 2 . To simulate 2ty of the second equation, independent 1ty are simulated from the first equation, the kink point or threshold value for 1ty is 21r =20, the smooth parameter 22s =0.5, and the true coefficients for 2 1 , 21 0.1 and 21 2 .

Table 1. Simulation Results of Copula-based SKE with normal and student-t margins Copula

True Estimate S.E. Copula

True Estimate S.E.

Gaussian

1 1 1.184 0.150 Student-t

1 1 0.770 0.189 11 2 2.115 0.056 11 2 2.014 0.087 11 -0.5 -0.554 0.292 11 -0.5 -0.479 0.026 1 1 0.975 0.071 1 1 1.074 0.077 2 1 1.205 0.115 2 1 0.991 0.140

21 0.5 0.429 0.041 21 0.5 0.495 0.034 21 3 2.978 0.032 21 3 3.040 0.029 2 1 1.588 1.068 2 1 0.868 0.076 2v 6 5.977 0.032 2v 6 6.089 0.810

11r 3 2.491 0.292 11r 3 3.031 0.364 22r 2 1.478 0.244 22r 2 2.373 0.215 11s 0.5 0.429 0.038 11s 0.5 0.490 0.045 22s 0.5 0.381 0.030 22s 0.5 0.516 0.041 0.5 0.621 0.065 0.5 0.535 0.084

Joe

1 1 0.817 0.142

Clayton

1 1 0.927 0.103 11 2 2.039 0.082 11 2 1.948 0.033 11 -0.5 -0.482 0.037 11 -0.5 -0.487 0.015 1 1 1.011 0.065 1 1 0.869 0.057 2 1 1.052 0.136 2 1 1.087 0.104 21 0.5 0.506 0.046 21 0.5 0.525 0.019 21 3 2.987 0.034 21 3 2.987 0.015 2 1 1.008 0.116 2 1 0.905 0.072

2v 6 4.317 1.579 2v 6 8.961 5.386 11r 3 2.592 0.302 11r 3 3.051 0.126 22r 2 2.067 0.246 22r 2 2.089 0.131 11s 0.5 0.429 0.044 11s 0.5 0.536 0.026 22s 0.5 0.495 0.047 22s 0.5 0.472 0.023 2 2.085 0.263 3 2.735 0.477

Gumbel

1 1 1.048 0.106

Frank

1 1 0.932 0.150 11 2 2.028 0.026 11 2 1.931 0.064 11 -0.5 -0.524 0.015 11 -0.5 -0.511 0.034 1 1 0.829 0.057 1 1 0.965 0.069 2 1 0.975 0.100 2 1 0.938 0.174 21 0.5 0.499 0.022 21 0.5 0.530 0.048 21 3 3.022 0.016 21 3 3.008 0.037 2 1 0.780 0.054 2 1 1.031 0.081

2v 6 6.913 0.280 2v 6 8.104 2.570 11r 3 2.927 0.127 11r 3 3.131 0.260 22r 2 2.170 0.121 22r 2 2.215 0.283 11s 0.5 0.467 0.021 11s 0.5 0.499 0.051 22s 0.5 0.503 0.003 22s 0.5 0.586 0.068 3 2.723 0.286 0.5 2.780 0.702

Source: Calculation.

7

Table 1 shows the results of the Monte Carlo simulation investigating the maximum likelihood estimation of the Copula based SKE model. We found that our model can perform very well through simulation study. It is found that the mean parameters are very close to their true values with low and acceptable standard errors. The results demonstrate the accuracy of estimation. For example, Table 1 representing Normal and Student’s t margins shows that the mean value of the coefficient 1 is 1.184 with standard error equals 0.15 while the true value is 1. Overall, the Monte Carlo simulation suggests that our proposed model ‘Copula based SKE’ is reasonably accurate.

5. Model Specification and Variables

Apart from the difficulty of econometric framework, this part is about an economic model and some discussion about the variables of the model. To test a relationship between income inequality and economic growth, we use the following two equations which are modeled simultaneously. The first equation represents the economic growth in which we consider not only the rate of increase in GDP but also many other factors influencing the growth, such as monetary base, human capital accumulation, institutional change, and government debt. These variables will be discussed later. The second equation represents inequality in income distribution. We measure the inequality using the best known index ‘Gini coefficient2’ and consider all possible factors that tend to have a major impact on distribution of income, such as subsidies, private sector share of GDP, inflation, government expenditure on education, and rate of unemployed labor. A key point of this model is that the growth equation containing Gini coefficient and the rate of growth is also entered into the Gini equation in order to find a direct link between these two factors.

1 11 12 13 14

15 16 1

ln( ) ln( ) ln( )1GDP GDP M HC Instt t t t t

Gini Debt ut t

(5)

2 21 22 23 24

25 26 2

ln(Sub ) ln( ) ln( )Gini GDP PS Inft t t t t

Edu Unemp ut t

(6)

At this stage, we construct this model and decide which economic variables should be included by combining a lot of information from expert advices and literatures. The first part of the growth equation includes change in nominal GDP lagged by one period ( 1tGDP ) and monetary base ( tM ). The monetary base refers to a part of the money supply which is highly liquid including notes, coins, and commercial bank deposits with the central bank. Economist John Taylor of Stanford University has quantitatively shown the positive impact of the monetary base growth on GDP growth through the U.S. data. The work of Kimura, Kobayashi, Muranaga, and Ugai (2003), the staff of the Bank of Japan, also supports this idea. They found that an increase in the monetary base can stimulate Japan’s economy at zero interest rate. We also consider this variable and expect to see the significant role of the monetary base in stimulating the output in Thailand’s economy. The next variable is human capital accumulation ( tHC ) which comes from an argument from Gregory, Romer, and Weil (1992). They argued that the traditional framework of Solow growth model should include human capital accumulation, in terms of education and training, health status, and experience, since they experimentally found that human capital has a greatly positive impact on economic growth and may induce sustainable growth. Also, institution ( tInst ) should be considered as a 2The Gini coefficient is a number on a scale measured from 0 to 1, where 0 represents perfect equality and 1 represents total inequality (Ray, 1993).

8

cause of the long-run growth (Acemoglu, Johnson, and Robinson, 2005). The institution in a society is commonly related to such social mechanisms as democracy, religion, property rights, and rule of law. The authors greatly explained through their work that institution determines the incentives and creates constraints in the economy. For example, as the impact of very different economic and political institutions, South Korea has much higher economic growth rate and better standard of living than North Korea. The last included variable in the growth equation is change in government debt ( tDebt ). Ghazanchyan, Stotsky, and Zhang pointed through their recent work in 2015 that high government spending may induce growth, but the excessive amount of spending may cause debt, and then lead to lower growth. Similarly, Bivens and Irons (2010) also comprehensively explained in the same way that debt matters for economic growth, as found from their experiment that 90% of debt to GDP will lead to a negative growth effect and slower economic growth.

Next, we will discuss about potential sources influencing income inequality. It is not easy to choose which factors should be included in the inequality equation because inequality is somewhat complicated and often unclear to define. Therefore, following the study of Anneli Kaasa (2003), we are able to separate the factors into four groups. The first group is economic growth and general development of a country in which we select the growth in GDP ( tGDP ) to represent this group and to describe a particular country’s development level. In addition, but importantly, they are chosen just to act in keeping with the Kuznets Hypothesis. The next variable is subsidies ( tSub ) or the transfer. OECD (2012) reported that the transfer systems can help reduce overall income inequality in all countries. For example, on average across the OECD, three quarters of the reduction in inequality is due to the transfers. And hence, we take into account this variable and attempt to see the effect of this variable in the case of Thailand. The second group comprises the political factors which are defined as privatization and the public sector [Anneli Kaasa (2003)]. Privatization is believed to cause wealth concentration which finally leads to income inequality. Similarly in the public sector factor, Gustafsson and Johansson (1999) pointed out that the wage gap in the public sector is typically lower than in the private sector. Therefore, the bigger share of private sector ( tPS ) may imply the higher income inequality; hence we should have a special concern and include this factor into our inequality equation. The third group is macroeconomic factors. We especially consider inflation rate ( tInf ) and the number of unemployed people (

tUnemp ) to represent macroeconomic influence on inequality. The effect of inflation is found to increase inequality since high inflation leads to a redistribution of resources with fixed nominal income. To support this idea, we refer to an empirical study on income distribution by Bulir (1998) which indicated that inflation affects levels of income inequality as measured by Gini coefficient. The influence of unemployment is commonly understood in its own right as it obviously has inequality increasing effect. Finally, the fourth group is demographic factors; here we focus on some part of demographic development in which education is our special concern. We decide to use a variable expenditure on education ( tEdu ) to illustrate this group because education is greatly important for reducing uneven income. It is found that a country with high investment in education can significantly reduce income inequality through a development in level of human capital (Debla-Norris et al., 2015).

6. Data

We use a quarterly data set of all considered variables, spanning from 1993:Q1 to 2015:Q4. The economic growth data include a variety of different factors as described previously: GDP, human capital index, institution index, monetary base, and the government

9

debt. In addition, the income inequality data include the Gini coefficient, the government subsidy and transfer payment, private sector share of GDP, inflation rate, the government expenditure on education, and the rate of unemployed labor. We collected all the data from Thomson Reuters DataStream, from Financial Investment Center (FIC), Faculty of Economics, Chiang Mai University, except the Gini coefficients which were collected from the National Statistical Office of Thailand and World Bank data. The regional Gini coefficients were collected from Office of the National Economic and Social Development Board (NESDB), covering the years 1997 to 2013. Note that, human capital index, Gini coefficient and institution index were disaggregated from the annual to the quarterly series by the temporal disaggregation method of Chow and Lin (1971).

Table 2. Data description

Variable GDP EDU Inf Unemp Sub PS Mean 0.009 0.009 1.041 0.031 0.036 14.204 Median 0.010 0.009 1.131 -0.018 0.030 14.199 Maximum 0.096 0.187 2.425 1.275 0.163 14.601 Minimum -0.063 -0.122 -2.303 -0.375 -0.096 13.755 Std. Dev. 0.021 0.036 0.758 0.296 0.050 0.245 Skewness -0.008 0.517 -1.856 1.708 0.109 0.083 Kurtosis 6.813 10.569 9.860 7.271 3.237 1.646 Variable Inst Gini HC M Debt Mean 4.174 0.409 0.855 0.017 -0.013 Median 4.174 0.418 0.800 0.018 0.000 Maximum 4.272 0.451 1.751 0.068 0.695 Minimum 3.478 0.345 0.711 -0.162 -0.680 Std. Dev. 0.098 0.025 0.215 0.027 0.180 Skewness -4.691 -0.873 3.054 -3.138 -0.077 Kurtosis 32.089 3.295 11.741 23.413 7.664

Source: Calculation.

The descriptive statistics of all data series are presented in Table 2. In addition, as we use time series data, testing for stationarity of data is then required. Here we apply the Augmented Dickey-Fuller (ADF) unit roots test to the subsets of data to examine whether or not the data series contains a unit root. We do not show the result here in this paper but we found that all variables passed the test at level with probability equal to zero, meaning all of them are stationary

7. Empirical Results

In this section, we will present the results of a set of experiments sequentially. The experiment begins with plotting the simple link between income inequality and economic growth and finding the current position of Thailand among all countries around the world. Then, the estimated results regarding the question of whether or not there is the Kuznets curve in Thailand’s economy will be indicated in the following part. We are going to start with the test for a kink effect on each variable, and then, as a sequential estimation method, we will select the best-fit Copula for our model before obtaining the estimated results in the last part. Additionally, in the last two parts of this empirical result, we will extend this algorithm for investigating the Kuznets curve to the regions to answer the question ‘Are there regional Kuznets Curves in Thailand?’ In addition, we will evaluate the efficiency of some Thailand’s economic and social development plans by looking at their accomplishments of boosting the economy and reducing income inequality.

10

7.1 The simple link between inequality and growth: What is the position of Thailand?

In this section, the Gini coefficient and the GDP growth in the year 2015 of 144 counties around the world including Thailand are used to indicate the simple link between inequality and economic growth and also investigate the position of our country. To achieve these goals, the fuzzy c-means algorithm introduced by Gustafson and Kessel (1978) is conducted to partition the Gini coefficient and GDP growth in a flexible manner. We need to determine the partition of the variables regarding the number of clusters. These clusters have “fuzzy” boundaries, that is, each data value belongs to the cluster to some degree. Their modelling technique involved clustering the data into fuzzy sets, estimating the relationship of Gini and GDP growth over each set, and then combining each sub-model into a single overall model by using the membership functions associated with the clustered data.

According to the result shown in Figure 1 and Table 3, the analysis reveals that the countries can be separated into two clusters. The different cluster can reflect the economic stability and the membership status within the world. We define Cluster 1 (Red dot) as the countries with low income inequality and low economic growth and Cluster 2 (Blue dot) as the countries with high income inequality and high economic growth.

Fig. 1. Plot of Gini coefficients against GDP growth, 2015

Table 3. Cluster description Fuzzy Observation Mean of Growth Mean of Gini

Cluster 1 (Red dot) 71 1.98 0.33 Cluster 2 (Blue dot) 63 4.59 0.46

Source: Calculation.

The result shows that Thailand locates in the cluster 1 (based on this data set), meaning that Thailand is in the group of countries with lower income inequality and low economic growth. To compare the ability of Thailand with other countries, we generate the horizon dashed line meaning the same level of GDP growth and the vertical dashed line meaning the same level of Gini coefficients as Thailand. We found that there are some countries, e.g. Cambodia, Lao, Indonesia, and Turkey, have higher level of economic growth when they have the same level of income inequality as Thailand. On the other hand, some countries e.g. Japan, Liberia, Netherland, and Belgium, have lower income inequality when they have the same level of economic growth as Thailand. This results indicate that Thailand seems not to be at a good position of both inequality and economic growth when compare with the other countries.

11

7.2 Exploring the Relationship between Growth and Inequality in Thailand

In this subsection, we will present the results sequentially. The first part is the test for a kink effect on each variable. The second part is about selecting the best-fit Copula or the joint distribution to be a linkage between economic growth and income inequality models. Then, the estimates of the growth and income inequality models will be shown in the last part of this subsection.

7.2.1 Testing for a Kink (Threshold) Effect

The important question is whether or not the kink (or threshold) exists with respect to our model. Even though we are given a clue by the expert and literatures that some covariates have both negative and positive effects on economic growth and even on income inequality (See Section 5), we still need to do an experimental test to verify this hypothesis. Recall, the kink effect means (or refers to) the unequal effect of the covariate on the dependent variable. It can be viewed as a regression discontinuity model where the regression function itself is continuous but the slope is discontinuous at a threshold point, thereby a kink. Thus, we expect to obtain the kink effect of GDP growth on the distribution of income or the so-called Kuznets hypothesis.

For this purpose, we conduct the Likelihood ratio test (LR-test) to test the kink effect following a recommended algorithm of Hansen (2000), the originator of the kink regression. That is, the kink effect on a relationship between dependent variable, y, and its covariate, xi, is examined as a single equation. This algorithm is kept using for each pair of the covariate and the dependent variable. We assume the null hypothesis is linear regression and the alternative is kink regression. The results are shown in Table 4.

Table 4. Results of Likelihood Ratio Test Covariate

Growth Equation tGDP

1tGDP In( )tM In( )tHC In( )tInst tGini tDebt 1.872 -0.171 0.031 0.388 0.989 0.112

Inequality Equation tGini

tGDP In( )tSub In( )tPS In( )tInf tEdu tUnemp 1.036 0.415 3.415* 15.28*** 0.863 0.028

Source: Calculation. Note: “*”and “***” denote rejections of the null hypothesis at the 10% and 1% significance levels, respectively.

Assuming the null hypothesis is true, Table 4 indicates that only the covariates In( )tPS and In( )tInf reject the null hypothesis of linearity at the 10% and 1% levels, respectively, and hence their outcomes are said to be in favor of the kink regression for the inequality equation. On the other hand, we found that other covariates cannot reject the null hypothesis of linearity, therefore, they are said to be in favor of the linear regression for the growth equation.

As the result, we failed to find the significant kink effect of GDP growth on the income inequality, meaning that the Kuznets hypothesis does not hold for Thailand’s economy through this data set. However, there still is a kink effect on income inequality due to the changes in the share of private sector ( In( )tPS ) and inflation ( In( )tInf ). Therefore, our simultaneous kink equation (SKE) model with specified kink points can take the form as:

12

1 1

1 11 12 13 14 15

16

ln( ) ln( ) ln( )1

t

GDP GDP M HC Inst Ginit t t t t t

Debtt

(7)

2 2

2 12 22 23 23 23 23 23 23

24 24 24 24 24 24 25 26

ln( ) ln( ) ln( )

ln( ) ln( )

(1 F(r ,s )) F(r ,s )

(1 F(r ,s )) F(r ,s ) t

Gini GDP sub PS PSt t j t t

Inf Inf Edu Unempt t t t

. (8)

The model is set similarly to the equations (7) and (8) in which growth in GDP and the Gini coefficient are the dependent variables influenced by a variety of independent variables. The parameters 1i and 2 j can be explained by the same manner as the conventional regression, but the terms 23 23 24, , and 24 represent the coefficients of negative and positive parts in the kink regression. Here, we treat the parameters 23r and 24r as the kink points which have to be estimated. This model is said to be the SKE model because these two equations are simultaneously estimated where the dependent variable of one equation also plays a role as the independent variable in the other equation, but the slope with respect to the share of private sector ( In( )tPS ) and inflation ( In( )tInf ) are discontinuous since they have the kink at

23In( )tPS r and 24In( )tInf r , respectively.

7.2.2 Selecting Copulas for the SKE Model

As a sequential estimation method, the Copula has to be selected before estimation. Therefore, this part is about selecting the best-fit Copula or the joint distribution to be a linkage between economic growth and income inequality equations. Given a set of Copula families i.e. Gaussian, Student’s t, Clayton, Gumbel, Frank, and Joe, as described in Section 3.2, we then select the Copula using a classical way of selecting model, the Akaike Information Criterion (AIC). In addition, we also employ another measure, the Bayesian Information Criterion (BIC) to strengthen the result. We assume four different distributions for the marginals, namely Normal, Student’s t, skewed Normal, and skewed Student’s t distributions. Therefore, we have got sixteen cases for a pair of marginal distributions of the growth and inequality equations, respectively. The results are illustrated in Table 5.

According to the results shown in Table 5, the minimum values of AIC and BIC are -924.537 and -852.366, respectively, from which Frank Copula is chosen to be a linkage between skewed Student-t of the growth equation and skewed normal margin of the inequality equation.

13

Table 5. AIC and BIC Criteria for Model Choice

Marginal Distribution

Family of Copula

Gaussian Student-t Clayton Gumbel Frank Joe [1] Normal/Normal -820.9 -858.6 -779.5 -291.8 -860.0 -816.4

-756.2 -791.4 -707.3 -219.6 -787.8 -744.2 [2] Normal/Student-t -877.5 -740.1 -408.2 -824.7 -306.3 -834.3

-802.8 -662.9 -333.6 -750.0 -381.0 -759.6 [3] Normal/Skewed Normal

-660.0 -791.4 -748.4 -764.7 -814.0 -785.3 -585.3 -714.3 -673.8 -690.1 -739.4 -710.6

[4] Normal/Skewed Student-t

-487.5 -242.2 -819.5 -776.4 -362.8 -251.4 -410.3 -162.5 -742.4 -699.2 -285.6 -174.2

[5] Skewed Normal/Normal

-857.4 -768.6 -748.7 -768.6 -858.0 -789.6 -782.8 -691.5 -674.0 -691.5 -783.3 -714.9

[6] Skewed Normal/ Student-t

-281.8 -298.1 -299.6 -272.5 -321.3 -300.3 -204.7 -218.5 -222.4 -195.3 -244.2 -223.1

[7] Skewed Normal/ Skewed Normal

-751.0 -761.5 -741.3 -722.5 -757.0 -758.1 -673.8 -681.9 -664.1 -645.3 -679.9 -690.0

[8] Skewed Normal/ Skewed Student-t

-215.7 -110.9 -323.6 -472.2 -404.8 -134.8 -136.1 -28.8 -244.0 -392.5 -325.1 -55.1

[9] Student-t/Normal -807.2 871.2 -787.7 -845.1 -882.5 -845.1 -737.5 -794.0 -713.0 -770.4 -807.8 -770.4

[10] Student-t/Student-t -915.8 -890.8 -801.6 -876.5 -910.6 -834.0 -838.6 -811.2 -724.4 -799.4 -833.4 -756.8

[11] Student-t/Skewed Normal

-815.0 -812.0 -766.5 -781.2 -831.3 -809.3 -737.8 -732.4 -689.3 -704.1 -754.2 -732.1

[12] Student-t/Skewed Student-t

-817.6 -800.7 -797.4 -777.1 -799.4 -800.6 -738.0 -718.6 -717.8 -697.5 -719.8 -720.9

[13] Skewed Student-t/ Normal

-879.1 -890.8 -817.1 -839.9 -868.8 -191.0 -801.9 -811.1 -791.7 -762.7 -791.7 -113.9

[14] Skewed Student-t/ Student-t

-903.5 -888.7 -828.6 -363.2 -924.5 -776.1 -823.9 -806.6 -748.9 -316.7 -852.4 -696.4

[15] Skewed Student-t/ Skewed Normal

-840.6 -767.6 -340.2 -817.8 -850.4 -818.2 -760.9 -685.5 -260.5 -738.1 -770.8 -738.6

[16] Skewed Student t/ Skewed Student-t

-816.4 -785.5 -802.9 -764.4 -779.0 -783.7 -734.2 -700.9 -720.8 -682.3 -696.9 -701.5

Source: Calculation.

7.2.3 Estimates of the Growth and Income Inequality models

To explore the relationship between growth and income inequality in Thailand, equations (11) and (12) are estimated by MLE and the results are presented in Table 6. In general, the model can perform well through this data set and it can capture a nonlinear effect of some variables. Most of the parameters are rightly signed and statistically significant at the conventional levels.

The estimates of the growth equation The estimates of the growth equation, shown at the beginning of Table 6, reveal that only some variables suggested by the empirical growth literatures are significant in the case of Thailand. It is found that Thailand’s economic growth depends significantly on the institution, Gini coefficient, and the government debt, but the impact of other variables on the economic growth are not significantly found.

14

Table 6 illustrates a negative relationship between GDP growth and the institution index in which the coefficient is significantly equal to -0.0166. Typically, the institution is considered as a cause of the long-run growth (Acemoglu, Johnson, and Robinson, 2005). It is commonly related to such social mechanisms as democracy, religion, property rights, and rule of law and it also determines the incentives and creates constraints in the economy. The different type of economic (and political) institution can bring about the different level of outcome of the economy. For example, as the impact of very different economic institutions, South Korea has much higher economic growth rate and better standard of living than North Korea. The negative correlation between GDP growth and the institution index happened in Thailand may imply that the current economic (and political) institution can be destructive to the growth.

Additionally, Table 6 also shows the positive effect of the government debt on the GDP growth in which the coefficient is equal to 0.0638, meaning that an additional 1% of the government debt leads to 0.064% increase in economic growth. This result provides numerical evidence consistent with what some economists have found in the literatures, for example the works of Ghazanchyan, Stotsky, and Zhang (2015) and Bivens and Irons (2010). We know that the government spending can help induce the economy as well as the government debt, but the debt can become destructive to the growth if it is excessive. Bivens and Irons (2010) found from their experiment that 90% of debt to GDP will lead to a negative growth effect and slower economic growth, but before that certain level the debt can be good for the economy. However, Thailand’s data provides just one way of this consequence; it is found that Thailand government debt creates only the positive effect on the economic growth.

Importantly, we also found a positive relationship between growth and the Gini coefficient. The coefficient is equal to 0.0225, meaning that an increase in the income inequality by 1% leads to an increasing in GDP growth in Thailand over 0.02%. This result implies that during the studied period (the years 1993-2015), the income inequality –measured by Gini coefficient- can create a positive impact on Thailand’s economy. The possible reason for this situation is that inequality may provide the incentives for people to invest, to make money, and to save for their future livelihood and also for progress that drive the capitalist system. Inequality can create the incentives for innovation and productivity that provide the broader economic and social benefits of capitalism. In fact, inequality may not instantly propel the economy but without inequality economy is like socialism where the growth of economy hardly exists.

The estimates of the income inequality As also reported in Table 6, income inequality in Thailand depends significantly on GDP growth, share of private sector, inflation, and the expenditure on education. We found that both share of private sector and inflation have nonlinear impacts on income inequality. These happen due to the kink effect and we will discuss completely about these effects later.

As we failed to find the kink effect of the GDP growth on income inequality across this data set (See Section 7.2.1 Testing for a Kink Effect), we may conclude that the Kuznets hypothesis does not hold for Thailand’s economy. However, we still found a linear relationship among these two variables, the GDP growth is found to be positively correlated with income inequality. The coefficient of the GDP growth is 0.0139, meaning that an additional 1% of the growth rate of GDP leads to 0.014% increase in Gini coefficient. This estimate result confirms the previous works and also the empirical literatures (as reported in Section 5), that is a rise in GDP growth can bring about the income inequality.

15

Table 6. Estimation results of Copula based SKE model

Variables Estimated Value S.E. ( 210 )

Growth Intercept term 0.0359* 2.16 GDP growth lagged by one period 0.0596 6.72 Monetary base 0.0384 2.62 Human Capital Accumulation 0.0119 0.93 Institution -0.0166*** 0.11 Gini Coefficient 0.0225** 0.95 Government debt 0.0638* 3.97 Sigma ( 1 ) 0.0355** 1.97 Degree of freedom Student-t 2.4739*** 50.83 Degree of freedom Skewed Student-t 0.5666*** 19.71 Income Inequality Intercept term 0.1888 17.97 GDP growth 0.0139* 0.63 Government subsidy and transfer payment -0.0194 2.04 Share of Private Sector (positive part) 0.0025*** 0.10 Share of Private Sector (negative part) -0.0031*** 0.12 Kink parameter of Share of Private Sector 14.2516*** 165.70 Smooth parameter of Share of Private Sector 13.4745*** 112.46

Inflation (first positive part) 0.1105* 45.90 Inflation (second positive part) 0.0145* 0.71 Kink parameter of Inflation 1.2809*** 41.39 Smooth parameter of Inflation 0.7581** 25.14 Expenditure on Education -0.0393* 2.41 Unemployment 0.0001 0.42 Sigma ( 2 ) 0.0181* 1.06 Degree of freedom Skewed Student-t 2.4833** 118.64 Joint Estimate Frank Copula Dependency ( ) 1.9833*** 7.36

Source: Calculation. Note: “*,” “**,” and “***” denote rejections of the null hypothesis at the 10%, 5%, and 1% significance levels, respectively.

Additionally, Table 6 shows that the expenditure on education is also important for reducing uneven income. The coefficient is equal to -0.0393, meaning that an additional 1% of the expenditure on education leads to 0.04% decrease in Gini coefficient of Thailand. Our result provides numerical evidence consistent with what IMF has pointed out through the discussion paper about inequality and growth entitled ‘Causes and Consequences of Income Inequality: A Global Perspective’ (Dabla-Norris et al., 2015). The expenditure on education can play a pivotal role in reducing income inequality because education determines occupational choice, access to jobs, and the level of pay. However, we failed to find significance for other important variables across this data set, i.e. the government subsidy and transfer payment and the growth rate of unemployed worker.

The Kink effects of the share of private sector and inflation on income inequality A specific capability of our method allows us to preserve the nonlinear impacts of the share of private sector and inflation on income inequality. The impacts are split into two

16

groups based on kink points. In Figure 2 (a) we display the relationship between Gini coefficient and the share of private sector (% of GDP) through a regression line (red line), on which the horizontal axis is the share of private sector and the vertical axis is income inequality measured by the Gini coefficient. We can see that the regression shows a small positive slope for low share of private sector with a significant kink around 14.2516 (14.25%), displayed as the blue dot, switching to a negative slope for the share of private sector beyond that kink value. This means share of private sector displays a positive impact on the Gini coefficient when the share is below 14.25%. The coefficient of this positive part is 0.0025, meaning that an additional 1% of share of private sector leads to 0.0025% increase in Gini coefficient.

To explain this positive correlation, we refer to the literature on the private sector and the distribution of income. The literature shows that privatization can cause wealth concentration which finally leads to income inequality (Anneli Kaasa, 2003). To make it clearer, if we compare the wage gap of the private sector with public sector, we will find that earnings inequality in the public sector is typically smaller than in the private sector. The public sector administratively sets wages, thereby the low income inequality within the sector, but in the private sector whose income usually depends on, for example, investment, production, and output. Therefore, the distribution of earnings within the private sector is usually unequal. This process is likely to increase income inequality due to the wage differential between public and private sectors and the unequal distribution of income within the private sector. So that is why the share of private sector should deserve a special attention from the policy maker.

Apart from the previous knowledge of the private sector, our study argues that the growing size of the private sector beyond the kink value can negatively affect income inequality. As shown in the same Figure 2 (a), the regression line is switching from the positive slope to the negative slope for the share of private sector beyond the kink value around 14.25%, displayed as the blue dot. The coefficient of the share of private sector in the negative trend is -0.0031, meaning that an increase in the share of private sector by 1% can cause a reduction in Gini coefficient by 0.0031%. According to this evidence, the share of private sector may no longer be destructive the distribution of income. Instead, the proper level of the share of private sector can help reduce uneven income in Thailand.

(a) Share of private sector (b) Inflation Fig. 2. Plot of Gini coefficient and its covariates and the trend lines estimated by

Copula based Kink regression

13.8 14.0 14.2 14.4

0.36

0.38

0.40

0.42

0.44

Share of private sector and Gini smooth kink plot

fitted line

kink point

0.5 1.0 1.5 2.0 2.5

0.36

0.38

0.40

0.42

0.44

Inflation and Gini smooth kink plot

fitted linefitted line

Gin

i

17

In Figure 2 (b) we illustrate the relationship between Gini coefficient and inflation through another regression line. We can see that the regression shows a steeply positive slope for low inflation with a significant kink around 1.2809 (1.28%), displayed as the blue dot, switching to a smaller positive slope for the inflation beyond that kink value. The result observes the positive relationship between inflation and income inequality, in which the higher inflation the more income inequality. However, our work can do more than that; we find a salient result that the impacts of inflation on inequality are split into two levels based on the kink point. In the first stage, at any level of inflation below 1.28%, an additional 1% of inflation leads to 0.11% (0.1105) increase in Gini coefficient. But in the second stage, the effect is much less than the first stage. We found that an additional 1% of inflation just leads to 0.015% (0.0145) increase in Gini coefficient. To explain this difference, we refer to the discovery of Chris Crowe, CEP researcher and IMF’s officer who first mentions about positive relationship between inflation and inequality. Inflation here is a consequence of raising government revenue by printing money. That is a central bank issues new currency in order to let the government collect more tax rather than collecting tax paid out of the existing money stock, but this event can cause higher inflation rate. Not all people suffer from the same inflation tax rate because the tax hits only a group of people holding more nominal assets such as cash as a fraction of total income, who are typically poor and middle class. As a result, inflation below that kink value creates a big gap between the wealthy and poor people, and then leads to the greater income inequality.

We can see that income inequality keeps rising as long as inflation increases, but beyond that kink value the effect of inflation seems to be less than before, but still exists. Chris Crowe suggests that income inequality in this second stage is influenced by democratic institutions, and happens among the wealthy people. The greater inequality in this stage means greater inequality in political participation in which the political voices depend on income. The wealthiest people have high power in adoption of policy, such as inflation, more favorable to their own business, and this leads to higher income inequality among those wealthy people.

But, from our opinions, we think in the other way round. It is true that the greater inflation may cause the higher income inequality in both stages; but what we suggest here is to use the proper level of inflation to reduce income inequality in the case of Thailand. Our experiment found that in the first stage, under the kink value 1.28%, the elasticity of the Gini coefficient with respect to the inflation is low; therefore a small decrease in the inflation rate results in a large decrease in the Gini coefficient. Unlike the second stage, the elasticity is much higher, meaning that even the central bank provides a big decrease in the inflation rate; we can observe just a minor decrease in the Gini coefficient. Hence, it is reasonable to preserve the inflation rate below that kink value.

7.3 Are there the Kuznets Curves in the regions of Thailand?

In this section, we extend the same algorithm of investigating the Kuznets curve to the regions to test whether the Kuznets Hypothesis holds in some regions of Thailand. We consider five regions (i=5) consisting of Middle, North, Northeast, and South regions, and Bangkok since each region’s characteristic is dissimilar, and hence the relationship between GDP growth and income inequality may be different across regions. We conduct the Likelihood ratio test (LR-test) to test the kink effect on a relationship between the regional Gini coefficient (dependent variable) and the regional GDP growth (covariate). This algorithm is kept using for each pair separately. Again, we assume the null hypothesis is linear regression and the alternative is kink regression. The results are shown in Table 7.

18

Table 7. Results of Likelihood Ratio (LR) test of each region Dependent variable

Covariate GDPi

Bangkok BGini

Middle MGini

North NGini

Northeast NEGini

South SGini

21.46*** 10.79*** 3.23*** 0.0023 0.0015 Source: Calculation. Note: “***” denotes rejection of the null hypothesis at the 1% significance levels.

Table 7 indicates that the Middle region, the North regions, and Bangkok can reject the null hypothesis of linearity at the 1% levels, meaning that the relationship between GDP growth and Gini coefficient of these three regions have the kink effect.

The estimated results are shown in Table 8. It is found that the regional Gini coefficients depend positively on the regional GDP growth in which the Middle region, the North regions, and Bangkok have the nonlinear impacts of GDP growth on income inequality due to the kink effect. As we found the existence of the kink effect, we expect to obtain the Kuznets curve in these three regions. However, the truth that we cannot find the Kuznets curve in Thailand’s economy is also true for the regions. As we can see, the impacts are split into two levels based on the kink point, whereas all of them are positive. The Northeast region has a positive linear relationship between GDP growth and Gini coefficient. Its coefficient is equal to 1.322, meaning that an additional 1% of GDP growth leads to 1.322% increase in Gini coefficient of the Northeast region. Note that the parameter of the South regions is not statistically significant.

Table 8. Estimate results for each region

Dependent variable Covariate

GDPi Bangkok

BGini Middle

MGini North

NGini Northeast

NEGini South

SGini

Intercept term 0.337*** 0.334*** 0.356*** 0.351*** 0.364*** GDPi 1.322*** 0.247

GDPi (first stage) 0.002* 0.491*** 0.029*

GDPi (second stage) 0.031*** 2.515** 0.022**

Kink parameter of GDP (%) 0.723*** 0.805** 0.945**

Smooth parameter of GDP 1.025* 0.755** 2.554**

Source: Calculation. Note: “*,” “**,” and “***” denote rejections of the null hypothesis at the 10%, 5%, and 1% significance levels, respectively.

As shown in Table 8 and Figure 3, the impacts of the growth on the inequality are split into two levels based on the kink point. For instance, the result of Bangkok shows that at any level of GDP growth below 0.723% per quarter, an additional 1% of the growth leads to 0.002% increase in Bangkok’s Gini coefficient. But in the second stage, the effect is much larger than the first stage. We found that an additional 1% of the growth can lead to 0.031% increase in the Gini coefficient. The results are the same for the Middle and North regions. These results argue the conventional theory of Kuznets –the inverted U-shaped curve of income concentration- that the decreases do not necessarily happen, and in fact, inequalities

19

have grown sharply in these three regions after reaching those kink points. Our finding is rather linked with the work of Thomas Piketty, a well-known French economist. Piketty's work refers to the capital-output ratio and explains that the rate of capital return (r) in developed countries is persistently greater than the rate of economic growth (g), and this will cause income (wealth) inequality to increase in the future.

Motivated by this reasoning, we think that the positive nonlinear relationship between the growth and inequality of the three regions can be explained through the same idea of Piketty. As these three regions play a key role in Thailand’s economy especially Bangkok. Bangkok and the surrounding areas is the economic centre of our country, it is the heart of the country's investment and development as well as Chiang Mai which is in the North. It is viewed as the second-largest province of Thailand that contributes a lot to Thailand’s economy. This causes the rate of capital returns of these three regions to be high and greater than the growth rate of GDP, and hence it leads to an increasing in the income inequality.

Fig. 3. Plot of Gini coefficient against GDP growth and the trend lines

estimated by Copula based Kink regression

-2 -1 0 1 2

0.32

0.34

0.36

0.38

0.40

Bangkok

GDP growth (%)

Ban

gkok

Gin

i coe

ffici

ent

Kink Point

fitted line

-2 -1 0 1 2 3

0.33

0.34

0.35

0.36

0.37

Center

GDP growth (%)

Cen

ter G

ini c

oeffi

cien

t

Kink Point

fitted line

-1 0 1 2

0.35

0.36

0.37

0.38

0.39

0.40

0.41

North

GDP growth (%)

Nor

th G

ini c

oeffi

cien

t

Kink Point

fitted line

20

7.4 Policy effect on the economic growth-Gini coefficient

In this section, we attempt to examine whether Thailand’s economic and social development plans, covering the plans 7-11, can accomplish boosting the economy and reducing income inequality at the same time. We focus on the overall country and the five regions of Thailand, consisting of Middle, North, South, Northeast, and Bangkok. First of all, let consider the overview impact of economic growth on the Gini coefficients of Thailand as shown in Table 9. The estimated regression coefficients indicate that GDP growth creates the positive effect on the Gini coefficient, whereas many parameters are not statistically significant. The finding also illustrates that the income inequality tends to increase along the Seventh Plan to the Eleventh Plan. This results show the evidence of a relevant impact of GDP growth rate to income inequality during the Eighth Plan and the Eleventh Plan.

Table 9. Economic plan summary

Economic Plan Intercept Slope

Plan 7 (1992-1996) 0.4335*** (0.0023)

0.0739 (0.0922)

Plan 8 (1997-2001) 0.4218*** (0.0016)

0.0016* (0.0701)

Plan 9 (2002-2006) 0.4171*** (0.0031)

0.0312 (0.1829)

Plan 10 (2007-2011) 0.3916*** (0.0022)

0.0166 (0.0869)

Plan 11 (2012-2015) 0.3606 (0.0054)

0.4148* (0.2154)

Source: Calculation. Note: “*,” and “***,” denote rejections of the null hypothesis at the 10 and 1% significance levels, respectively. The numbers in blankets are the standard errors.

Furthermore, it would also be interesting to investigate in detail the relationship between inequality and GDP growth in five different regions or areas of Thailand. A time-varying regression model is adopted to estimate time-varying effects of economic growth on income inequality. By using this model, it allows the effects to evolve as the level of income inequality changes along our sample study. Note that the study uses Gini coefficient of each region as a proxy for income inequality, and the GDP growth of each region as a proxy for economic growth. The result of time varying coefficient of the GDP growth effect on Gini for each region is illustrated in Figure 4. In most of the regions, the effects of GDP growth are either positive for the whole sample period or turn from negative to positive, except for South region. In this figure, we can observe that North, Northeast, and Middle regions present a similar movement of the time varying coefficients and mostly take place above the zero line. This indicates that there is a positive effect of economic growth on Gini coefficients during the implementation of the Eighth Plan to the Eleventh Plan. In the case of BKK and South regions, we observe that there is a negative effect of economic growth on inequality during the Eighth Plan before moving to positive effect when approaching the Ninth Plan for BKK and the Tenth Plan for South. We find an interesting main issue that is an economic growth can reduce the inequality during the Eighth Plan in some regions. We found that the Eighth Plan has adjusted the development concept, shifting from growth orientation to people-centered development. In the past plans, the national and social development concept has focused on the acceleration of economic growth utilizing comparative advantage in terms of natural resources and low-cost labor to produce goods and services for export. However, the success in economic growth might not mean that the Thai people are enjoying a greater

21

wealth and quality of life. Thus, the Eighth plan aimed to improve the well-being rather than the economic growth. And this might be the reason that economic growth could bring a lower income inequality during this plan.

In contrary to the Eighth Plan, the effect of economic growth on inequality is positive along the Ninth to the Eleventh Plan. How could it have happened? To answer this question, we look back to at the main points of those plans and found that these plans adopted the Sufficiency Economy philosophy to guide the development and administration of the country, at the same time as continuing the holistic approach of people-centered development from the Eighth Plan. Moreover, these plans also built an economy with strong internal foundations and flexible to external changes and aiming for balanced development with respect to people, society, economy, and environment in order to achieve sustainable development and the well-being of the Thai people. The main idea of these three plans seem to be similar to the Eight plan but the time varying coefficients results provide a positive effect of growth on inequality. We expect that Thailand had encountered more complicated domestic and global changes, and fluctuations that would present a threat to national development. Thailand was in transition from a rural to an urban society. As a result of the compartmentalized development of urban and rural areas, there is imbalance in the development of rural communities. These, in turn, had weakened the economy, the society and democracy. The NESDB (2012) reported that the distributions of income by region and the returns to the factors of production have been skewed. Returns to labor have declined continuously, from 30.2 percent during the Eighth Plan to 28.9 percent during the first three years of the Tenth Plan, while returns to capital, entrepreneurship and land have increased. These evidences indicate the lack of economic and social opportunity, access to resources, bargaining power, and fairness with respect to rights among groups of people that affects both income and accumulation of wealth. In addition, major obstacles have remained, including political unrest, environmental and ecological degradation, low educational quality, and severe drug problems.

Fig. 4. Time-varying effects (coefficients) of GDP growth on

income inequality of each region3

3 Due to the estimation method of time varying regression model need the initial data to compute a time varying coefficient therefore, the time varying coefficients in Seventh plan cannot be obtained.

Q3/1998 2002 2007 2012 Q4/2013

22

Considering the magnitude of time-varying coefficients, we observe that the effect of economic growth on income inequality in Bangkok had the lowest effect relative to other regions while Northeast region had the highest effect of economic growth on income inequality. These findings indicate the lack of economic, difference in human capital and social opportunity, access to resources, bargaining power, and fairness with respect to rights among groups of Thai people. These factors will result in different returns in the labor market and contribute to income inequality in Thailand.

8. Conclusion and Policy Implications

This study attempts to explore the relationship between income inequality and economic growth and investigate the Kuznets hypothesis with special focus on Thailand. To accomplish this desired goal, we propose Copula based the simultaneous smooth transition kink equation (SKE) model to investigate this relationship. We consider a variety of different variables as suggested by the empirical growth and inequality literatures, and add all the possible variables into our model. The estimate results show that Thailand’s economic growth depends significantly on the institution, Gini coefficient, and the government debt. On the other hands, income inequality in Thailand depends significantly on the GDP growth, share of private sector, inflation, and the expenditure on education in which both share of private sector and inflation have the kink effect on income inequality.

Let us explain about our considerable issue, income inequality and growth. We found that the Gini coefficient creates a positive impact on the economic growth, and conversely. There is no evidence supporting the Kuznets hypothesis through this data set. However, this paper can discover an alternative way to reduce income inequality in Thailand that is the share of private sector (% of GDP) and the inflation. We experimentally found that the share of private sector beyond the kink value 14.25% can reduce income inequality in Thailand. But more than that, this study also found that at any levels of inflation under the kink value around 1.28%, a small decrease in the inflation rate results in a large decrease in the Gini coefficient. Therefore, it is reasonable to preserve the inflation rate below that kink value.

Additionally, this study also provides the experiment on the regions of Thailand. We found that income inequalities of the Middle, North, and Bangkok tend to increase in the future. Finally, the result of time varying coefficients indicates that the Eighth plan which conducted a people-centered development approach has a potential to improve the well-being of the society since a negative effect of the growth on the Gini coefficient is obtained in almost regions. In addition, we consider the magnitude of the time varying coefficients and found that there is heterogeneous magnitude of the effect of economic growth, especially in Bangkok and the Northeast region. We suspect that the difference in human capital and social opportunity, access to resources, the lack of bargaining power, and fairness with respect to rights among the groups of Thai people in each region can lead to the income disparity.

Policy Recommendation

The overall results of this study indicate that the problem about income inequality still matters for Thailand’s economy as well as the regions. The economic growth is necessary for our country but it brings about the uneven income at the same time. The existence of the Kuznets hypothesis seems to be the only one solution for this problem, but unfortunately it is not held for our country. Our finding provides an alternative way for reducing the income inequality that is the proper levels of the share of private sector and inflation as described previously. However, apart from these two alternative ways, this study would recommend the

23

policy maker to pay special attention to the root of this problem that is ‘Education’. We think that the education policy is a key for this problem. It can determine the occupational choice, the access to jobs, and the level of earnings. We believe that improving education quality, for instance providing the training or apprenticeship programs and providing the financial support to higher education can help reduce the income disparity and also help boost skill levels and let the future generations be able to cope with the future unpredictable changes. However, importantly, these education policies should at least promote more equal access to the basic education.

.References

1. ดร. ตรีเทพ นพคุณ (2550). การศึกษาผลกระทบของโลกาภิวัตน์และอัตราการเจริญเติบโตทางเศรษฐกิจต่อความเหลือมลาํระหว่าง ภูมิภาค

ในประเทศไทย พ.ศ. 2524-2546 :กรณีศึกษาภูมิภาคภายในและภูมิภาคชายฝัง

2. Acemoglu, D. (1998). Why do new technologies complement skills? Directed technical change and wage inequality. Quarterly journal of economics, 1055-1089.

3. Acemoglu, D., Johnson, S., & Robinson, J. A. (2005). Institutions as a fundamental cause of long-run growth. Handbook of economic growth, 1, 385-472.

4. Barro, R. J. (1999). Inequality, growth, and investment (No. w7038). National Bureau of Economic Research.

5. Bhandari, R., Pradhan, G., & Upadhyay, M. P. (2010). Another empirical look at the Kuznets curve. International Journal of Economic Sciences and Applied Research, 3(2), 7-19.

6. Bivens, J., & Irons, J. (2010). Government debt and economic growth. Economic Policy Institute Briefing Paper, 271.

7. Boonyamanond, S. (2007). Can Equality and Growth Be Simultaneously Achieved?. Chulalongkorn Journal of Economics, 19(2), 135-160.

8. Bulir, M. A. (1998). Income inequality: does inflation matter?. International Monetary Fund. 9. Charoenphandhu, N., & Ikemoto, Y. (2014). INCOME DISTRIBUTION AND THE MEDIUM-TERM

TREND OF INEQUALITY IN THAILAND. Ekonomski i socijalni razvoj, 1(1), 43-51. 10. Cheng, W. (2014). Understanding the Kuznets Process: An Empirical Investigation of Income Inequality in

China 1978-2011 (No. 12-14). Monash University, Department of Economics. 11. Cherubini, U., Luciano, E., & Vecchiato, W. (2004). Copula methods in finance. John Wiley & Sons. 12. Chow, G. C., & Lin, A. L. (1971). Best linear unbiased interpolation, distribution, and extrapolation of time

series by related series. The review of Economics and Statistics, 372-375. 13. Crowe, C. W. (2006). Inflation, Inequality, and Social Conflict (No. 6-158). International Monetary Fund. 14. Dabla-Norris, M. E., Kochhar, M. K., Suphaphiphat, M. N., Ricka, M. F., & Tsounta, E. (2015). Causes and

consequences of income inequality: a global perspective. International Monetary Fund. 15. Deininger, K., & Squire, L. (1996). A new data set measuring income inequality. The World Bank Economic

Review, 10(3), 565-591. 16. Dollar, D., & Kraay, A. (2002). Growth is Good for the Poor. Journal of economic growth, 7(3), 195-225. 17. Easterly, W. (1999). Life during growth. Journal of economic growth, 4(3), 239-276. 18. Gallup, J. L. (2012). Is there a Kuznets curve. Portland State University. 19. Ganaie, A. A., & Kamaiah, B. (2015). Kuznets Inverted U Hypothesis of Income Inequality: Looking Inside

the Available Economic Literature. Available at SSRN 2591284. 20. Gustafson E.E., Kessel W.C., 1978. Fuzzy clustering with a fuzzy covariance matrix. Proceedings of the

IEEE Conference on Decision and Control, pp. 761-766. 21. Ghazanchyan, M., Stotsky, J. G., & Zhang, Q. (2015). A New Look at the Determinants of Growth in Asian

Countries. IMF Working Paper. 22. Gregory, M. N., Romer, D., & Weil, D. N. (1992). A contribution to the empirics of economic

growth. Quarterly Journal of Economics, 107(2), 407-437. 23. Gustafsson, B., & Johansson, M. (1999). In search of smoking guns: What makes income inequality vary

over time in different countries?. American Sociological Review, 585-605. 24. Hansen, B. E. (2000). Sample splitting and threshold estimation.Econometrica, 68(3), 575-603. 25. Hansen, B. E. (2015). Regression kink with an unknown threshold. Journal of Business & Economic

Statistics, (just-accepted), 00-00. 26. Heshmati, A. (2004). Growth, inequality and poverty relationships. 27. Hsing, Y. and D. J. Smyth (1994), “Kuznets's Inverted-U Hypothesis Revisited”, Applied Economics Letters,

1, 111-113.

24

28. Ikemoto, Y., & Uehara, M. (2000). Income inequality and Kuznets’ hypothesis in Thailand. Asian Economic Journal, 14(4), 421-443.

29. Kaasa, A. (2003).Factors influencing income inequality in transition economies. University of Tartu Economics and Business Administration Working Paper Series, (18).

30. Khalifa, S., & El Hag, S. (2010). Income disparities, economic growth, and development as a threshold. Journal of Economic Development, 35(2), 23.

31. Kimura, T., Kobayashi, H., Muranaga, J., & Ugai, H. (2003, September). The effect of the increase in the monetary base on Japan’s economy at zero interest rates: an empirical analysis. In Monetary Policy in a Changing Environment, Bank for International Settlements Conference Series (Vol. 19, pp. 276-312).

32. Kuznets, S. (1955). Economic growth and income inequality. The American economic review, 45(1), 1-28. 33. Lazear, E., & Rosen, S. (1981). nRank order tournaments as optimal salary schemeso. Journal of Political

Economy, 89(5), 841. 34. Mahfoud, M., & Michael, M. (2012). Bivariate Archimedean copulas: an application to two stock market

indices. BMI Paper. 35. OECD 2012, “Income inequality and growth: The role of taxes and transfers”, OECD Economics

Department Policy Notes, No. 9. January 2012. 36. Ogus Binatli, A. (2012). Growth and income inequality: a comparative analysis.Economics Research

International, 2012. 37. Ostry, J. D., & Berg, A. (2011). Inequality and unsustainable growth: two sides of the same coin? (No.

11/08). International Monetary Fund. 38. Pastpipatkul, P., Maneejuk, P., Wiboonpongse, A., & Sriboonchitta, S. (2016). Seemingly Unrelated

Regression Based Copula: An Application on Thai Rice Market. In Causal Inference in Econometrics (pp. 437-450). Springer International Publishing.

39. Preechametta, A. (2015). Dynamics of Income Inequality under Social Identity Hypothesis. Applied Economics Journal, 22(2), 1-24.

40. Piketty, T. (2014). Capital in the twenty-first century. Harvard University Press. 41. Ray, D. (1998). Development economics. Princeton University Press. 42. Shahbaz, M., & Islam, F. (2011). Financial development and income inequality in Pakistan: an application

of{ARDL} approach. 43. Sørensen, P. B., & Whitta-Jacobsen, H. J. (2010). Introducing advanced macroeconomics: growth and

business cycles (pp. 1-2). McGraw-Hill higher education. 44. Teräsvirta, T. (1998), “Modelling Economic Relationships with Smooth Transition Regression,”

inHandbook of Applied Economic Statistics. New York, Marcel Dekker, pp. 507–572. 45. Trivedi, P. K., & Zimmer, D. M. (2007). Copula modeling: an introduction for practitioners. Now

Publishers Inc. 46. Wichitaksorn, N., Choy, S. T. B., & Gerlach, R. (2006). Estimation of bivariate copula-based seemingly

unrelated Tobit models. Discipline of Business Analytics, University of Sydney Business School. NSW.