The Great Chinese Inequality Turnaround

China’s high income and wealth inequality has long attracted the interest of policymakers and re-searchers, yet surprisingly little has been done since 2010 on inequality trends. Given China’s evolving economic structure and the government’s adoption of new policy tools in recent years, we revisit the latest data on Chinese inequality and assess the impacts of economic and policy changes on income distribution. After a quarter century of rapid, sustained increase, we see Chinese inequality plateauing and even diminishing. To verify this finding, we draw upon a range of data sources and measures of inequality. We examine inequality trends through decomposition by income source and population subgroups, and consider possible explanations such as policy shifts and structural trans-formation of the Chinese economy. The findings suggest that the narrative on Chinese inequality today should focus on clarifying the factors driving this apparent inequality turnaround.


Introduction
Alongside the spectacular growth and extraordinary reductions in poverty, perhaps the most dramatic in human history, the evolution of Chinese income inequality since the start of the reform process in 1978 has been a focus of interest among analysts and policy makers. Table A.1 in the appendix gives a flavor of this interest by summarizing the most significant studies concentrating on the evolution of income inequality. In their study of the evolution of inequality in China focusing on spatial inequality over the long run, from 1952 to 2000, Kanbur and Zhang (2005) identified two phases of inequality change after the start of reforms in 1978. After an initial short phase of falling inequality as rural incomes rose in the wake of the liberalization of the personal responsibility system, inequality rose inexorably as China opened up to the world and explosive growth took place in the coastal regions.
This increase in inequality became an integral part of the narrative on Chinese development, 1 with some commentators arguing that this was the inevitable price to be paid for the high rates of growth, with others warning of the social consequences of rising gaps. In any event, "harmonious society" was given center stage at the 2005 National People's Congress and among rising policy concerns on inequality. As more data has accumulated, greater attention has turned to an examination of the evolution of inequality in China in the 2000s, including in the present decadethe years after 2010. A number of studies which used data from the mid-2000s onward began to argue that the rise in inequality was being mitigated, and inequality was possibly plateauing and perhaps even turning down. 2 This paper attempts to provide a comprehensive assessment of what the data show, a deeper look into the patterns of inequality change, and preliminary explanations for the trends observed.
Our basic conclusion is that there does indeed appear to be a turnaround taking place in Chinese inequality, and that the explanations lie in policy changes and in the nature of structural transformation in China.
The plan of the paper is as follows. Section 2 sets out the data sources on Chinese inequality on which any assessment will have to be based. Section 3 then presents the basic trends over the 20- year period from 1995 to 2014. Section 4 examines the patterns of inequality change by looking, respectively, at decomposition by income source and by population subgroup. Section 5 presents some preliminary explanations for the observed trends. Section 6 concludes.

Data
In this study, we use two kinds of data: household-level data from household surveys and provinciallevel data from the National Bureau of Statistics (NBS  : 1988, 1995, 2002, 2007, 2008, and 2013 We use household survey data to analyze the evolution of household income inequality and the attributes of different income sources, as these data provide rich information about the various income components in each household. For the analysis of regional inequality evolution and its decomposition, we make use of the provincial-level data. Each dataset is described below in greater detail.
The household-level data we use are taken from CHIP 1995, 2002, and 2007 and CFPS  There are some differences between CHIP and CFPS in terms of the items included in each income source. 4 For example, the rental value of housing equity is included in CHIP 1995 but not in other surveys, and medical expenses paid by a collective or the government are included in transfer income in CHIP but not in CFPS, and so on. To ensure as much consistency as possible, we broke down the different sources of income in CHIP and reconstructed them with the items that are included in CFPS only. In addition, there is no "other income" category in CHIP 2007, but we constructed it following the CFPS definition. In our decomposition by income source, we present two results, one with the original household income from CHIP and CFPS, and the other with adjusted income from CHIP that is consistent with the CFPS definition.
Another data-related issue we need to address is the missing data in income sources. We assume that there exists a fixed hidden distribution for household income, for both rural and urban categories. We approximate the hidden distribution for rural and urban categories from the existing non-missing data. Then we sample new pseudo-value from this approximated distribution to fill in the missing entries. The pseudo-value is a random number drawn from the sample distribution. This approximation for distribution requires a sufficiently large sample size, which is a condition not satisfied using a county-level sample. Provincial distribution is not suitable either because the CFPS is not representative on the province level. Hence we use the national distribution.
In addition to the two issues addressed above, there are some observations for which the sum of all income components does not equal the household net income in CFPS. This is due to the fact that for households that did not report their annual net income, the household's net income is estimated according to its consumption. To deal with this issue, we rescale each income source using the proportion household net income sum of all the income sources .
Although the two household surveys both include rich information about household income, their geographical coverage differs. Moreover, CFPS's sampling is not representative on the provincial level. Because of these limitations, we could not apply regional decomposition to the household survey data. Therefore, in our analysis of regional inequality, we use provincial-level income and population data from the NBS.
As Gibson (2013, 2016) have noted, Chinese yearbooks previously reported provincial population and per capita economic outputs based on households registered, that is, the hukou population rather than the residential population. This resulted in a distortion of the estimate of This is the data base for our assessment of Chinese inequality trends over the last 20 years.
We proceed now to a description of the overall trends and the decomposition patterns in the data.

Trends
We estimate various inequality measures using household survey data from CHIP and CFPS for six points of time covering the 20-year period between 1995 and 2014. income construction methods, we see that the Gini coefficient has an inverted U shape pattern with 5 The generalized entropy indices are a popular class of measure for inequality. They are derived from information theory as a measure of redundancy in data.
, where y i is the income of observation i and µ is the mean of income with the distribution F(y).  (0), the peak appears in 2012, while for GE(1) and GE (2) it is in 2010. The differences in the turning patterns of each index could be because that each inequality index captures different characteristics of inequality. For the generalized entropy indices GE(c), the greater c is, the more sensitive it is to the top income groups. That is to say, GE(0) is more sensitive to the bottom income groups, while GE(2) is more sensitive to the top income groups.  To provide a more detailed picture of income distribution, quartile and decile income shares are    The combination of CHIP and CFPS data gives us six observations spanning the period from 1995 to 2014 based on household surveys. An alternative data perspective, useful for capturing long-term annual trends, was introduced in Zhang (1999, 2005). This method uses NBS data on provincial consumption per capita, broken down by rural and urban areas for each province. Combining this with rural-urban population data for each province (see the discussion on population data in Section 2), we can construct a synthetic national consumption distribution which suppresses inequality within the rural areas and urban areas of each province. Clearly, this is an understatement of the level of inequality, but the trend over time may nevertheless convey information on the evolution of inequality.  Overall, then, a careful assessment of the best data sources seems to suggest a plateauing of inequality, with a possible turning point around or just before 2010. To begin building an explanation of the trend, we decompose inequality by income sources and population groups.

Decompositions
To unpack the patterns of inequality change, we proceed to decompose inequality, first by income source and then by population subgroup.

Decomposition by income source
To understand the role of different income sources in the evolution of overall inequality, we decompose the Gini coefficient by income source following Lerman and Yitzhaki's (1985) rule.
where S k = µ k /µ is the share of kth income component in total income, ��� is the "pseudo-Gini," 8 is the Gini correlation of component k with total income, and G k is the Gini of income component k. The absolute contribution of income source k to total income inequality is ( ) = . (2) Its proportion of the total inequality is where Y i is the income of household i and Y ki is the income from source k of household i. 9 The marginal effect of income source k is The share of property income was small, at less than 10 percent, throughout the period under study, while its Gini coefficient was very high and remained above 0.96. The proportionate contribution 8 The pseudo-Gini is different from the conventional Gini because the weight attached to corresponds to the rank of individual i in the total income distribution, which is, in general, not the same as his or her rank in the distribution of income source k. 9 We weighted household income by family size in all calculations. to the total Gini coefficient of each income source, � ( ), and its marginal effects, η k ( ), are reported in Tables 4.3 and 4.4, respectively. The largest contribution is from wage income, which ranged between 0.7 and 0.8 over the years, followed by transfer income, which ranged between 0.13 and 0.19. The contributions of other income source are less than 0.1. In addition to its high contribution to the overall Gini coefficient, wage income also has the largest marginal effect. Note: To be as consistent as possible across the two datasets, we excluded some components from the Chinese Household Income Project (CHIP) that are not in the China Family Panel Studies (CFPS) survey. In addition, the income sources are recalculated in CHIP according to CFPS definitions. Wage income is labor income including bonuses, allowances and subsidies, and remittances from migrant worker family members. Operational income includes net income from the sale of farm products, net income from private enterprises, and gross value of self-consumption of farm products. Property income is income from rental or sales of properties. Transfer income includes social security, pension, subsidies, etc. Other income is mainly money and gifts from relatives or friends.   Given the importance of wage income, the trends shown in Table 4.2 are central in understanding the forces underlying the overall inequality trend. Inequality of wage income has fallen sharply, as has inequality of transfers. These are the dominant factors in total income, and thus their declining inequality is the dominant factor in inequality change and accounts for the decrease in inequality.
To see the sensitivity of the results, we follow Paul's (2004) extension of the Gini decomposition to decompose the Theil's T index, 10 that is, GE(1), by income sources.
where µ is the mean of population income.
The absolute contribution to income inequality of income source k is When expressed as a proportion of total inequality, it can be written as The marginal effect of income source k on the Theil's T index is where S ki is the share of income source k in the total income of i-th household. The decomposition results for the Theil's T index are presented in Tables 4.5 and 4.6. The results are quite consistent with what we find in the Gini decomposition. In addition to the level of inequality, the change in inequality over time can also be expressed as a weighted average of over time changes in each income source, as stated in Paul, Chen, and Lu (2017). Denote , +1 = ( +1 − )/ , which is the proportionate change in household income inequality between year t and year t + 1. It could be written as where � ( ) serves as a weight, and ̇� , +1 � =  the Gini coefficient from wage income, operational income, and property income were quite comparable. However, for the Theil's T index, wage income served as the top inequality-reducing component. Overall, then, these accounting exercises are consistent with the hypothesis that it is the narrowing of the wage distribution and the role of transfers that are important in beginning an understanding of the Chinese inequality turnaround.

Decomposition by subgroups
An alternative perspective on patterns of inequality change is provided by decomposition by population subgroup. Unequal income distribution between urban and rural sectors is a common feature in developing countries, and China is no exception. In addition to the unequal development between rural and urban regions, the disparity between the coastal areas in the east and inland areas in the middle and west is also enormous (Fan, Kanbur, and Zhang 2011). To understand these components of inequality, we use the data underlying Table 4.9, the synthetic distribution constructed from rural and urban per capita consumption and population. We further decompose the Theil's T index by rural-urban subgroups and coastal-inland subgroups, respectively, as in equation (10).
where N is the total number of individuals and k is an indicator for groups, for example, rural or urban. The first term is the within-group component of the Theil's T index and the second term is the between-group component.
The rural-urban between component and the coastal-inland between component are reported in Table 4.9 and graphed in Figure 4.1. There are three peaks for the rural-urban between component, in 1995, 2000, and 2004. After the third peak, the rural-urban between component maintained a declining trend. Notice that 2005 is the year when regional inequality and rural-urban between components turned downward. That is the year when, it has been argued, China passed the "Lewis turning point" (Zhang, Yang, and Wang 2011). That is also the year when the agriculture

Some explanations
Our main task in this paper has been to establish the key trends in Chinese inequality over the past 20 years. Based on a number of perspectives, it does seem as though there was a turnaround in Chinese inequality about 10 years ago, with inequality plateauing and even declining after a long period of sharp increase. Explanations for this evolution will have to await detailed investigation from researchers focusing on a range of factors in depth. However, in this section we present a broad framework for such explanations.
A simple way to think of the evolution of national income distribution is to divide the economy up into key sectors and to look at inequality within and between sectors. Given the importance of the structural transformation which is under way in China just now, we can begin our discussion in terms of two sectors-rural and urban. The national income distribution is a weighted sum of the rural income distribution and the urban income distribution, with the weights being the population shares of the two sectors. Overall inequality will then depend on (1) the inequality within each of the two sectors, (2) the gap between the means of the two sectoral distributions, and (3) the population share of each sector.
As an illustration, for the GE(0) index, also known as the mean log deviation, denoted L, national inequality can be decomposed as follows: where subscripts 1 and 2 denote rural and urban, respectively; x is the population share of the urban sector; and k is the ratio of the urban mean to the rural mean. The evolution of national inequality is then composed of (i) the evolution of L1 and L2, (ii) the evolution of k, and (iii) evolution of x.
With this framework, we can relate the inequality turnaround to basic economic forces and to policy. First, as Zhang, Yang, and Wang (2011) have argued, China has now reached the Lewis turning point, where rural-to-urban migration begins to tighten rural labor markets and hereby mitigate the rural-urban wage differential. In addition, heavy government investment in infrastructure in the rural sector and in lagging regions, a feature of Chinese policy from the 2000s onward (Fan, Kanbur, and Zhang 2011), will also raise economic activity and incomes in these areas. This will surely lower k in equation (11) and hence, ceteris paribus, overall inequality. This is consistent with the evolution of the rural-urban component of inequality shown in Table 4.9, and it is further consistent with the observed reduction in inequality in the national wage distribution as shown in The narrowing of the wage distribution and the increasing equality of the transfer distribution shown in Table 4.2 can also be associated with policy changes. For example, in 2004 the Ministry of Labor and Social Security issued a "Minimum Wage Regulations" law and the next decade saw rising minimum wage standards coupled with substantial improvements in compliance (Kanbur, Li, and Lin 2016). Further, a number of social programs were introduced and strengthened from the 2000s onward. Since 2004, for example, China has introduced new rural cooperative medical insurance, currently covering more than 95 percent of the rural population. Rural social security has also been rolled out since 2009. Although the benefits for rural medical insurance and social security are still much lower than their urban counterparts, the programs have provided some cushion to rural residents against health risk and elderly care. Tightening labor markets in rural areas, combined with inequality-mitigating transfer and regulation regimes in urban and rural areas, acted through channels (i) and (ii) to reduce inequality.
The impact of x on L, as seen through equation (11), is quite complex. With all other factors constant, it can be shown (Kanbur and Zhuang 2013) that under certain conditions the behavior of L as a function of x has an inverse-U shape, as hypothesized by Kuznets (1955). Up to a certain point, urbanization increases inequality, and beyond this point further urbanization will decrease inequality.
This "Kuznets turning point" sets out the effect of urbanization pure and simple on inequality. The turning point itself depends on the other inequality parameters, but it is shown by Kanbur and Zhuang (2013) that Chinese urbanization has now crossed the Kuznets turning point-and further urbanization will reduce inequality through channel (iii) above.
Of course, each of these potential explanations needs to be investigated more fully and in greater depth. But they appear to us to be consistent with underlying economic and policy forces which can explain the inequality turnaround we see in the data.

Conclusion
We have argued in this paper that the long period of inequality increase in China is coming to an end.
The data, seen from different perspectives, seem to indicate a turnaround towards the latter part of the 2000s. The explanations for this turnaround need to be explored further, but there is prima facie evidence of economic forces and government policy tightening labor markets in rural areas, together with government transfers and social policy mitigating inequality in urban and rural areas, which may explain the observed trends. This of course, raises the further question of why government policy changed over a 20-year period from allowing inequality to increase to mitigating it. The political economy of the Chinese state (Wong, 2011) may provide an explanation, but that takes us beyond our present remit. Although China's inequality has come to a turnaround, the level is still rather high compared with many countries. More efforts are still needed to keep the momentum.