WPS7968 Policy Research Working Paper 7968 Distribution of Consumption Expenditure in East Asia La-Bhus Fah Jirasavetakul Christoph Lakner Development Research Group Poverty and Inequality Team February 2017 Policy Research Working Paper 7968 Abstract Using a new database of household surveys, this paper exam- now almost entirely explained by within-country differences, ines inequality among all individuals living in developing while gaps in average income across countries have become East Asia regardless of their country of residence. The East unimportant. This reversal has been driven by rising national Asian Gini index increased from 39.0 in 1988 to 43.3 in inequality especially in populous countries, counteracted by 2012. Inequality increased during the initial decade, regard- catch-up growth in average incomes, particularly in China. less of the choice of inequality measure. The trend appears Interpersonal differences in income at the regional level to have reversed in the mid-2000s. Regional inequality is have thus become internalized within national boundaries. This paper is a product of the Poverty and Inequality Team, Development Research Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at ljirasavetakul@imf.org and clakner@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Distribution of Consumption Expenditure in East Asia LA-BHUS FAH JIRASAVETAKUL AND CHRISTOPH LAKNER1 JEL Codes: D31, D63 Keywords: inequality, global inequality, East Asia, expenditure distribution 1 Jirasavetakul: The International Monetary Fund (e-mail: ljirasavetakul@imf.org), Lakner: The World Bank (e- mail: clakner@worldbank.org). Both authors are Research Associates at the Centre for the Study of African Economies, University of Oxford. This paper is a background paper for Ruggeri Laderchi et al. (2017). We thank Shaohua Chen, Reno Dewina, Carolina Diaz-Bonilla, Yumeka Hirano, Bryan Bonsuk Koo, Caterina Ruggeri Laderchi, Prem Sangraula and Nikola Spatafora for help with the data and comments. The opinions expressed in the paper are the authors’, and should not be attributed to the authors’ organizations. I. INTRODUCTION East Asia is home to a number of populous countries that have seen steep increases in national inequality over the last quarter century. Figure 1a shows long-run changes in the national Gini index for a number of countries in the region for which long-run data are available. China’s Gini index increased sharply by some 10 points between 1990 and the early 2000s. Indonesia experienced an increase of a similar magnitude, beginning a decade later. However, inequality did not increase in all countries; Thailand has managed a continued reduction in the Gini index, which has also been observed in Malaysia (albeit at a slower pace). For the average country in the region, the national Gini fluctuated considerably, as shown by the solid blue line in Figure 1b. In the latest period, however, there appears to be evidence of falling national inequality in the average country. Furthermore, the Gini index fell by more than 1 point in five out of seven countries with comparable surveys between 2008 and 2013.2 Figure 1: National inequality in East Asia a. Long-run changes in national Gini index b. Average national Gini index Note: The welfare aggregate in Malaysia is income; in Note: Averages for the unbalanced sample of all other countries it is consumption expenditure. countries. Global averages shown for comparison. Source: Calculations based on World Bank (2016) and PovcalNet. In this paper, we adopt a different perspective for studying inequality in East Asia. We ignore national boundaries and examine the East Asia-wide distribution of consumption expenditure among individuals, and how it changed between 1988 and 2012. By combining national household surveys from as many East Asian countries as possible, we can measure the inequality among all East Asian citizens regardless of their country of residence. Our analysis puts the disparities in living standards that exist among persons in East Asia into context with the disparities that exist within and between East Asian countries. 2 In China, the Gini changed by less than 1 point, which is unlikely to be significant, while inequality increased in the Lao People’s Democratic Republic. For comparison, between 1993 and 2008, the Gini index increased (fell) by more than 1 point in five (three) out of nine countries. See World Bank (2016) for more detailed results. 2 Implicit in our analysis is an East Asian social welfare function which treats persons irrespective of national borders. It is closely related to analyses of the global income distribution, which presumes a cosmopolitan social welfare function.3 In opinion surveys, East Asia shows some of the highest levels of concern over inequality being too high (Ruggeri Laderchi et al., 2017). Similarly, an informal survey of policy makers in Asia shows a rising concern with inequality (Kanbur and Zhuang, 2012). This appears to be inconsistent with the within-country results presented in Figure 1, which showed a stable Gini index for the average country. However, as Alvaredo and Piketty (2014) argue in their analysis of inequality in the Middle East, inequality at the regional level could be important for explaining these growing concerns. This paper aims to provide the latest evidence on this regional inequality for East Asia. Over the 25-year period, the East Asia and Pacific Gini index increased by approximately 4.2 points (or 10.8%), reaching 43.3 in 2012. On the one hand, the strong growth in average consumption in China had an equalizing effect on (between-country) East Asian inequality. On the other hand, the increasing region-wide inequality trend is explained mainly by rising within-country inequality in the two most populous countries, China and Indonesia. This is consistent with the trend in the population-weighted average Gini index in Figure 1b (dotted blue line), which captures national inequality for the average person in East Asia. In other words, while inequality has been approximately flat for the average country, it has increased strongly for the average person and, as this paper shows, for the region as a whole. Considering how growth was shared along the distribution, we find that the growth incidence was broadly pro-rich – in line with the observed increase in inequality. The pattern of growth was such that the richest 5% of East Asians received more than 20% of the gain in regional income, compared with less than 1% for the poorest 5%. In the most recent period, however, the upward trend in regional inequality appears to have reversed, as within-country inequality has stopped increasing. The paper is structured as follows. First, we explain the construction of our database from household surveys. Second, we study the regional distribution, its inequality and the contributions of within- and between-country differences. Third, we examine both relative and absolute gains along the distribution using growth incidence curves. Fourth, we analyze relative movements of specific country groups in the regional distribution, and the resulting changes in the regional composition. The Appendix includes further methodological details, and robustness checks. 3 The literature on global inequality is reviewed in Milanovic (2005), Atkinson and Brandolini (2010), Anand and Segal (2015) and Milanovic (2016). Similar to our paper, Ravallion (2014a) estimates inequality in the developing world. 3 II. DATA CONSTRUCTION The analysis focuses throughout on the developing countries in East Asia and the Pacific (EAP, or “East Asia” for brevity), as defined by the World Bank.4 The starting point of our data set is a set of harmonized household survey data by the World Bank, which we supplement with additional data from PovcalNet.5 In total, we use 60 surveys, with 31 surveys directly from the microdata and the remainder from PovcalNet. Because most countries do not conduct annual surveys, we match these surveys to a number of benchmark years – 1988, 1992, 1997, 2002, 2007, and 2012.6 Due to limited availability of microdata, all information for the first two benchmark years is entirely based on PovcalNet. For China, no microdata are available, so all data come from PovcalNet. Our database includes around 10 surveys per benchmark year out of a total of 19 EAP countries, and 92% of the surveys are consumption surveys (Table 1).7 Given our interest in estimating the EAP distribution of consumption expenditure, it is of course important that the surveys included in our database cover as much of the region as possible. On average, we cover about 98% of regional GDP, and at 97% a slightly lower share of the regional population (Table 1).8 While we have a good coverage of EAP as a whole, our coverage of Pacific countries is relatively low and unstable.9 The Appendix includes a robustness check using only those countries that are available in all benchmark years. Each national distribution is approximated by the average consumption of the 100 percentile groups. For the microdata, percentile groups are computed directly from the household survey. PovcalNet also reports percentile groups whenever it has access to micro data. However, in some cases it only has access to grouped data, such as 10 or 20 coordinates of the Lorenz curve. The reliance on grouped data has diminished over time, but for some countries, such as China, 4 That is, our analysis includes Cambodia, China, Fiji, Indonesia, Kiribati, Lao PDR, Malaysia, Marshall Islands, Myanmar, Micronesia, Mongolia, Palau, Papua New Guinea, the Philippines, American Samoa, Solomon Island, Thailand, Timor-Leste, Tonga, Tuvalu, Vanuatu, and Vietnam. High-income countries in the region, such as Japan; the Republic of Korea; or Taiwan, China, are not included in the analysis (see p. 49 of World Bank (2016) for the full list of high-income countries). 5 Harmonized household survey data: EAPPOV created by the World Bank’s EAP Team for Statistical Development, accessed on 18 January 2016. PovcalNet: online tool developed by the World Bank’s research department, http://iresearch.worldbank.org/PovcalNet/, accessed on 25 November 2016. 6 Similar to Lakner and Milanovic (2016), surveys need to be within two years of a benchmark year, and surveys in consecutive benchmark years must be at least three and no more than seven years apart from each other. Because of poor availability of surveys in earlier years, we cannot start before 1988. 7 Throughout the analysis, income surveys are used for Malaysia, while all other countries use consumption. Similar to Lakner and Milanovic (2016), we mix income and consumption surveys across countries and make no adjustment. While we are fully aware of the important differences between the two concepts, income and consumption surveys cannot reliably be inferred from one another (Anand and Segal, 2015). Also, in an abuse of terminology, we refer to income and consumption interchangeably in this paper. 8 We use the World Bank’s income groups from 1 July 2015. Countries are classified according to their GNI per capita in 2014 using exchange rates based on the Atlas method. 9 There are no surveys available for Pacific countries prior to 1997, and the coverage for 2012 is particularly low. Between 1997 and 2007, on average, our surveys cover about 60% of both the population and the total GDP of Pacific countries. 4 PovcalNet continues to rely on such data.10 To obtain a consistent number of observations per country-year, we fit a parametric Lorenz curve to the quantile groups and build a data set of percentile groups throughout.11 The Appendix includes a robustness check which uses the raw quantile data without fitting a parametric Lorenz curve. Each percentile group is population-weighted in the analysis, so every person is assigned the average consumption expenditure of her percentile in the country-year distribution. Welfare is measured using per capita consumption expenditure (or income), expressed in 2011 PPP- adjusted USD (see methodological details in the Appendix). Because we compare consumption levels across countries it is important to account for differences in purchasing power across countries using purchasing power parity (PPP) exchange rates. For the same reason, welfare aggregates could also be adjusted for spatial price differences within countries, as recently discussed by Ferreira et al. (2016) in the context of estimating global poverty. In the baseline results, we follow the approach adopted in PovcalNet (Chen and Ravallion, 2010) and adjust for spatial price differences within China and Indonesia (see Appendix), but not in other countries in the region.12 In the Appendix, we present robustness checks with further within- country spatial price adjustments between 2002 and 2012, as well as using the older set of 2005 PPP exchange rates. 10 The number of quantiles per country-year ranges from 9 to 200, with an average of 71 across all years (and 36 on average in 1988). 11 We fit a log-normal Lorenz curve to the quantiles and use it to generate a distribution of 10,000 points. We use the ‘ungroup’ Stata routine (Abdelkrim and Duclos, 2007) which implements the Shorrocks and Wan (2008) method. As described in more detail in Lakner et al. (2014), this is very similar to the approach used by PovcalNet in the measurement of global poverty. Fitting a parametric Lorenz curve is more robust than fitting a kernel density (Minoiu and Reddy, 2014). 12 We thus have a consistent spatial price adjustment over time. One reason why PovcalNet adjusts for spatial price differences in China is that rural and urban surveys were conducted separately until recently. 5 Table 1: East Asia and Pacific sample summary statistics Benchmark year Total 1988 1992 1997 2002 2007 2012 Number of surveys 5 8 9 11 16 11 60 Years between survey year and benchmark year (%, by benchmark year) ‐2 0.00 0.00 0.00 0.00 6.25 0.00 ‐1 20.00 12.50 22.22 9.09 12.50 22.22 0 40.00 37.50 33.33 63.64 31.25 33.33 1 20.00 37.50 33.33 9.09 37.50 33.33 2 20.00 12.50 11.11 18.18 12.50 11.11 Within +/‐ 1 of benchmark 80.00 87.50 88.88 81.82 81.25 88.88 Income vs. Consumption surveys (%, by benchmark year) Consumption 80 87.5 88.89 90.91 93.75 100.00 91.67 Income 20 12.5 11.11 9.09 6.25 0.00 8.33 GDP (in 2011 PPP‐adjusted USD) (% of regional GDP represented in the database) EAP 95.00 99.02 99.55 99.38 99.48 96.30 98.12 East Asia 95.78 99.80 99.79 99.61 99.50 96.56 98.51 CHN, IDN, PHL only 100.00 100.00 100.00 100.00 100.00 100.00 100.00 Other East Asia 76.92 99.00 98.79 97.19 95.83 60.71 88.07 Pacific islands 0.00 0.00 55.88 32.71 92.36 3.98 30.82 LIC 0.00 100.00 0.00 100.00 100.00 100.00 66.67 LMIC 84.08 97.82 99.66 96.49 96.73 94.43 94.87 UMIC 98.81 99.45 99.79 99.97 99.99 96.59 99.10 Population (% of regional population represented in the database) EAP 94.39 99.47 99.21 96.92 97.30 95.31 97.10 East Asia 94.75 99.86 99.32 97.28 97.31 95.81 97.39 CHN, IDN, PHL only 100.00 100.00 100.00 100.00 100.00 100.00 100.00 Other East Asia 48.10 98.61 93.49 79.28 79.69 68.68 77.97 Pacific islands 0.00 0.00 72.53 20.06 95.73 3.41 31.95 LIC 0.00 100.00 0.00 100.00 100.00 100.00 66.67 LMIC 76.25 98.42 99.74 87.23 89.10 87.64 89.73 UMIC 99.75 99.75 99.93 99.99 99.99 97.91 99.55 Notes: The last column is the (unweighted) average over the benchmark years 1988 to 2012. 6 III. EAST ASIAN DISTRIBUTION AND ITS INEQUALITY Figure 2 shows the EAP distribution of consumption expenditure and how it evolved over time. It moves towards a unimodal distribution which looks approximately log-normal although it has a long right-hand tail. The (almost) twin peaks observed in the earlier benchmark years smoothed out over time. As shown by the rightward movement of the distribution, there was particularly strong growth since 2002, which is also illustrated by the positive Growth Incidence Curves (GIC) in Section IV (Figure 7). At the same time as the distribution moved rightwards, it also stretched out suggesting an increase in inequality. As a first step towards explaining these changes in the overall distribution, we split up the overall density by country and sub-region (Figure 3).13 It is immediately apparent, that China dominates the regional distribution, accounting for around 70% of the regional population. It is also clear that China has seen very fast growth, followed by the rest of East Asia – excluding Indonesia and the Philippines. While in 1988 the modal Chinese income was below the EAP mode, in 2012 it was clearly above. These compositional changes are analyzed in more detail in Section V (see Figure 10). Compared with within-country inequality in the region, the Gini index of the EAP-wide distribution of consumption expenditure is high, at between 39 and 45 across the benchmark years (Table 2, Panel A). For example, in benchmark year 2012, the (unweighted) average within-country Gini is around 38 (Table 2, Panel I), and none of the included EAP countries have a Gini greater than the EAP-wide Gini.14 It is important to note that the trend of the average within-country Gini changes from declining to increasing, once population weights are used, implying an increasing Gini in populous countries (also see Figure 1 in the introduction). Over the entire period of analysis, the EAP-wide Gini index has increased by almost 11%. By contrast, the (unweighted) average within-country Gini has fallen somewhat by around 2%. In other words, within-country inequality for the average EAP country declined slightly, while the region-wide inequality increased. The Lorenz curve for 1997 is uniformly to the right of the 1988 curve without crossing (Figure 4), so inequality increased between 1988 and 1997 irrespective of the choice of inequality measure (Atkinson, 1970). This appears to have reversed in the most recent period (2007-2012) when the Lorenz curves moved inwards. Between 1997 and 2012, however, there is no unambiguous ranking of Lorenz curves, so the direction of change in inequality depends on the choice of inequality measure. Similarly, when looking at the entire period between 1988 and 2012, the ranking in terms of inequality depends on the choice of inequality measure. That is, the Lorenz curve for 2012 crosses to the left of the 1988 curve at around the 95th percentile – suggesting that inequality increased at the bottom. This is also confirmed by the alternative GE inequality measures in Table 2, Panel A. The bottom-sensitive GE(0) measure increases by 26% between 1988 and 2012, while the top-sensitive GE(2) declines by around 36%. At 13 Lakner and Milanovic (2016, fn. 31) explain in detail how these stacked densities are created. 14 For the earlier benchmark years, there are only a few countries with a Gini higher than the EAP-wide Gini: Malaysia (although it uses income surveys), the Philippines and Thailand before the 1997 financial crisis, Papua New Guinea in 1997 and Solomon Islands in 2007. 7 the same time, the Gini index, which is sensitive around the middle of the distribution, increased by almost 11%, as already discussed. Figure 2: The EAP distribution of consumption expenditure Figure 3: The EAP distribution of consumption expenditure, by sub-region 8 However, it is important to point out that the 1988 and 2012 Lorenz curves cross at a very high income level. It is well established that the GE(2) measure is sensitive to extreme values (Cowell and Flachaire, 2007). Hence it is important to understand whether the disagreement over the direction of change in inequality happens over a meaningful range of preferences. This can be done by examining the direction of change in the Atkinson index , another inequality measure, for different values of , the inequality aversion parameter. The gap between the Atkinson indices in 1988 and 2012 falls with (not shown), i.e. with very low aversion, inequality levels are very similar in the two years. The minimum value for which is typically reported is around 0.25 (e.g. by the US Census Bureau, see Proctor et al., 2016). Even with 0.1, which is a very low level of inequality aversion (and thus a low weight is attached to lower incomes), the Atkinson index increased by 5% between 1988 and 2012.15 Therefore, while there is no strict Lorenz dominance between 1988 and 2012, the standard measures all agree that inequality increased. Overall EAP inequality can be decomposed into two components, which arise from income differences within and between countries, respectively. Between 1988 and 2012, the between component of the GE(0) measure declined, while the within component increased both in levels and as a share (Table 2, Panel B).16 Therefore, the increase in overall EAP inequality appears to be driven by increasing inequality within countries, in particular among those with a high population share, which is consistent with changes in the population-weighted Gini index. As a result, the contribution of consumption differences within countries to total inequality rose from 68% to 92% (Table 2, Panel C). This evolution of within-country inequality is stronger than the results at the global level, where within-country inequality also increased but between-country differences remain the dominant source of inequality (Lakner and Milanovic, 2016). It is perhaps not surprising that between-country differences matter more at the global level compared with a regional analysis, as one might expect countries’ average income or consumption levels to be more similar within a region. A similar analysis of regional inequality in Sub-Saharan Africa (Jirasavetakul and Lakner, 2016) shows a higher level of inequality than in East Asia, with the African Gini index increasing from 52 in 1993 to 56 in 2008. In contrast to EAP, Africa’s within-country component gradually declined, or put differently the increase in overall Africa inequality has been driven mainly by increasing inequality between countries (i.e. differences in growth rates 15 Atkinson (1975) provides the following intuition in terms of a transfer between two people, one of whom has twice the other’s income. For someone whose aversion parameter 0.1, a transfer of $1 from the rich to the poor person would only be acceptable if $0.93 reaches the poor. With 0.25, the requirement would be $0.84 for every $1 transfer. 16 The measures of inequality in the GE-class are additively decomposable by subgroup. The decomposition of GE(0) uses population shares, while the decomposition of the GE(1) index is in terms of consumption shares. Therefore, the within-country component of GE(0) can be interpreted as the residual inequality after equalizing average consumption across EAP (Anand and Segal, 2008). 9 across countries).17 But similar to EAP, within-country inequality still explains most of overall Africa-wide inequality (around 60%). The GE(0) country-decomposition is the sum of the within and between contributions of every country, so we can analyze these contributions separately, as done in Figure 5. The total level of inequality, which has increased over time (also see Table 2, Panel A), is given by the total width of the bars. For some of the smaller countries, we have combined the contributions together. The within component is always positive, but the between component is negative when the country’s mean exceeds the overall EAP-wide mean. It is clear that within-country inequality has increased in China and Indonesia, while it is harder to tell for smaller countries. 77% (13%) of the increase in within-country inequality from 0.17 in 1988 to 0.29 in 2012 (Table 2, panel B) was driven by China (Indonesia). At the same time, due to its growth in average consumption, China’s contribution to differences between countries has declined strongly (Figure 5, orange bars) – so much that it became negative in the most recent year. Both Other East Asia (Figure 5, green bars) and China have moved closer to regional average consumption, but from opposite directions – the excess of Other East Asia in terms of average consumption has declined.18 By contrast, Indonesia’s gap to the regional mean has increased (Figure 5, blue bars), because it could not keep up with the growth in the regional average. Taken together, it is clear that the growth in average consumption in China has had an equalizing effect on EAP-wide inequality, while the accompanying increase in inequality in China had the opposite effect. Specifically, China’s within-component increased from 0.13 to 0.22, while its between component declined from 0.14 to -0.04. Therefore, China’s contribution to total GE(0) inequality actually declined (from 0.26 to 0.18, or from 104% to 58% of total regional inequality), which is entirely driven by its growth in the mean. This equalizing effect from China was more than counteracted by the other countries, resulting in an increasing GE(0) overall. These results, specifically on the direction of regional inequality as well as its country decomposition, are robust to a number of methodological checks. In the Appendix, we explain and estimate four robustness checks – (1) within-country price differences for a larger set of countries, (2) 2005 PPPs, (3) a balanced sample of countries, and (4) the raw data without a parametric Lorenz curve. 17 As explained in Jirasavetakul and Lakner (2016), this comparison of the between-country contribution is robust to using the method introduced by Elbers et al. (2008), which is particularly useful in settings where the number of groups and their relative sizes vary across comparators. 18 Of course, this does not imply that average consumption of Other East Asia has actually fallen. However, as Table 2, Panel H, shows, average consumption in Other East Asia has grown slower than the EAP-wide average – growth of 43%, compared with 265% for EAP. 10 Figure 4: EAP Lorenz curve Figure 5: EAP inequality decomposition, by sub-region 11 Table 2: East Asia and Pacific inequality change  change  change  change  change  1988‐2012  1988 1992 1997 2002 2007 2012 (%, pp) (%, pp) (%, pp) (%, pp) (%, pp) change (%, pp) A. Regional inequality Gini index (%) 39.04 6.19 41.46 2.99 42.70 1.17 43.19 3.34 44.64 ‐3.06 43.27 10.84 GE(0) (Theil‐L) 0.25 13.04 0.29 6.23 0.30 2.83 0.31 8.23 0.34 ‐5.74 0.32 25.99 GE(1) (Theil‐T) 0.32 12.81 0.37 4.90 0.38 ‐10.44 0.34 5.97 0.36 ‐8.89 0.33 2.33 GE(2) 0.75 16.01 0.87 5.12 0.91 ‐38.48 0.56 4.23 0.58 ‐17.87 0.48 ‐35.78 B. Country decomposition of GE(0): level of within‐ and between‐country inequality GE(0) (Theil‐L) (within) 0.17 18.74 0.20 3.62 0.21 31.28 0.28 6.55 0.30 ‐1.09 0.29 70.22 GE(0) (Theil‐L) (between) 0.08 0.91 0.08 12.80 0.09 ‐62.67 0.03 21.84 0.04 ‐38.70 0.03 ‐68.26 C. Country decomposition of GE(0): within‐country contribution in % (change is in percentage points) GE(0) within contribution 68.06 3.43 71.48 ‐1.76 69.72 19.29 89.01 ‐1.38 87.63 4.33 91.95 23.90 D. Absolute measures of inequality Absolute Gini index 312 26.12 393 25.78 495 30.25 644 55.45 1001.86 26.02 1262.58 304.79 E. Average annual consumption expenditure per capita (in 2011 PPP‐adjusted USD), by percentiles Bottom 10% 246 8.59 267 15.90 309 11.03 343 35.31 465 31.30 610 148.27 P40‐P50 538 12.37 605 21.71 736 28.10 943 48.22 1,398 35.09 1,888 250.80 P50‐P60 624 16.48 727 19.84 872 31.73 1,148 50.46 1,727 33.97 2,314 270.66 P60‐P70 753 17.53 885 18.78 1,051 35.73 1,427 50.71 2,150 32.47 2,849 278.30 P80‐P90 1,192 18.74 1,416 23.81 1,753 40.74 2,467 52.73 3,768 29.51 4,879 309.24 P90‐P95 1,581 21.83 1,926 31.93 2,541 35.59 3,445 55.54 5,358 28.54 6,887 335.72 P95‐P99 2,500 28.07 3,201 33.44 4,272 37.88 5,890 44.40 8,505 23.70 10,521 320.87 Top 1% 7,712 24.87 9,630 20.52 11,606 25.15 14,525 20.51 17,505 16.32 20,361 164.01 H. Average annual consumption expenditure per capita (in 2011 PPP‐adjusted USD), by region EAP 799 18.77 949 22.13 1,159 28.75 1,492 50.43 2,245 30.00 2,918 265.19 East Asia 799 18.77 949 22.13 1,159 28.73 1,492 50.74 2,249 29.75 2,918 265.21 CHN, IDN, PHL only 702 16.95 821 22.41 1,005 38.91 1,396 50.64 2,103 34.62 2,831 303.28 Other East Asia 2,621 ‐20.64 2,080 22.02 2,538 ‐9.81 2,289 50.33 3,441 8.75 3,742 42.77 Pacific islands n/a n/a n/a n/a 1,114 57.09 1,750 ‐21.37 1,376 31.10 1,804 n/a Low income n/a n/a 1,238 n/a n/a n/a 1,428 17.30 1,675 18.03 1,977 n/a Lower middle income 844 16.23 981 9.07 1,070 27.57 1,365 17.51 1,604 31.55 2,110 150.00 Upper middle & high income 790 18.73 938 26.23 1,184 29.05 1,528 59.55 2,438 30.35 3,178 302.28 I. Comparison with within‐country inequality: Average within‐country Gini index Avg. Gini (%) 38.55 1.76 39.23 1.67 39.89 ‐6.00 37.49 4.50 39.18 ‐3.48 37.81 ‐1.91 Avg. Gini (%) [pop‐weighted] 32.46 8.57 35.24 1.35 35.72 13.30 40.47 2.62 41.53 ‐0.59 41.28 27.17 Notes: For the decompositions by country, changes between benchmark years are in percentage points. For all other rows, changes are measured in percent (not annualised). Observations are weighted using  population. Using dataset of 100 percentile groups per country‐year, based in part on a parametric Lorenz curve. 12 IV. EAST ASIAN GROWTH INCIDENCE CURVE Over the past two decades and a half, average consumption in EAP as a whole increased by more than three times (or about 5.5% per annum). The EAP GIC over the period 1988 to 2012 (Figure 6) shows how this growth was distributed along the distribution.19 For example, it shows how the average consumption of the bottom 5% in 1988 compares with the average consumption of the bottom 5% in 2012. With the exception of the very top ventile group, the growth incidence was pro-rich. More specifically, the annual growth of the bottom half was lower than the growth in the mean, while the annual growth rates were relatively high among the 12th to 19th ventile (60th to 95th percentile) groups. Growth dropped significantly for the top ventile group. This is largely explained by its relatively low growth during the Asian financial crisis period, as shown by the GICs for the separate 5-year periods between benchmark years (Figure 7). The GICs are upward sloping for almost all periods. However, it is interesting that in the most recent period, which spans the global financial crisis, growth appears to have been slightly pro-poor. It is important to stress that the discussion so far has focused on growth rates, or relative gains, which can imply very different absolute gains in expenditures. Figure 8 shows the EAP GIC but with absolute gains (per year) in expenditure on the vertical axis. It is thus equivalent to the difference between the Pen’s Parades (or quantile functions) in 2012 and 1988.20 The mean expenditure of the richest 5% of East Asians rose by nearly PPP-2011 $400 per annum between 1988 and 2012 (or more than PPP-2011 $7,000 over the entire period), compared with less than PPP-2011 $30 for the bottom 20% and around PPP-2011 $60 for the median group. Figure 9 shows the same information somewhat differently – it shows how the total increase in consumption between 1988 and 2012 was distributed among the different ventile groups. For example, the top 5% received more than 20% of the gain, compared with less than 1% for the bottom 5%. It is striking, that the drop observed at the very top in the standard GIC (Figure 6) disappears completely when we consider absolute gains. In other words, the distribution of absolute gains has been strongly pro-rich. The important distinction between absolute gains and growth rates can also be illustrated with a numerical example. The eleventh ventile (between 50th and 55th percentile) grew by 5.55% per year between 1988 and 2012, slightly faster than the top 5% (5.45%) (Figure 6). But the initial income of the eleventh ventile in 1988 was only PPP$601 compared with PPP$3,507 for the top 5%. By 2012, these two groups received PPP$2,196 and PPP$12,537 respectively. In other words, while both groups tripled their expenditures, one group gained PPP$1,595 while the other gained PPP$9,030, leading to very different improvements in livelihoods. 19 Our definition of the GIC follows Lakner and Milanovic (2016) but is slightly different from the original definition by Ravallion and Chen (2003) who plot the growth rate of consumption of a particular fractile (e.g. the 5th percentile) not the fractile group (e.g. the poorest 5%). 20 The Pen’s Parade is the inverse cdf, i.e. it plots income on the vertical axis for various quantiles along the horizontal axis (Ravallion, 2014b). Instead of fractile incomes, we show the average incomes of a fractile group. 13 Figure 6: EAP Growth Incidence Curve (GIC), 1988-2012 Figure 7: EAP GICs, between six benchmark years 14 Figure 8: EAP absolute gains, 1988-2012 Figure 9: Beneficiaries of the gains in total consumption 15 The discussion of absolute gains also has direct implications for the choice of inequality measure. The standard measures of inequality (e.g. Table 2, panel A) are relative measures, with the corresponding GIC in terms of relative gains (Figure 6). Relative measures are invariant to a proportional transformation, such as a doubling of all incomes, while an absolute measure is invariant to an additive change, such as adding $100 to all incomes (Kolm, 1976). Panel D of Table 2 reports the absolute Gini index, defined as , where is the relative Gini and is the mean. This absolute inequality measure rose fourfold between 1988 and 2012, consistent with Figure 8. In sum, depending on one’s view of inequality in relative or absolute terms, one would conclude that inequality rose moderately (e.g. Gini rose by 11%) or very strongly (e.g. absolute Gini increased by 305%).21 21 With a relative view, you might even conclude that inequality has fallen, since there is no (relative) Lorenz dominance. The empirical evidence as to whether people perceive inequality as absolute or relative is mixed (Amiel and Cowell, 1999). In experimental studies, university students in Germany, Israel, the United Kingdom, and the United States are approximately evenly split between relative and absolute views of inequality (Ravallion, 2016). 16 V. CHANGING REGIONAL COMPOSITION The GICs analyzed thus far are fully anonymous, i.e. they compare the average consumption of EAP-ventiles which could be composed of quite different country-percentiles. Figure 10 breaks the EA-wide distributions in 1988 and 2012 into 20 ventile groups, each accounting for 5% of the regional population.22 The height of the bars indicates the population composition for each of these ventiles, e.g. in 2012 77% of the richest 5% of East Asians are Chinese, while 3% are from Indonesia and 2% from the Philippines. The composition of the top ventiles changed substantially over the 24-year period, with China moving up. It is these compositional changes which underlie the EAP-wide GIC (Figure 6). From the chart, it is also clear that China has grown faster than Indonesia and the Philippines, whose relative share at the bottom increased. Figure 10: Regional composition of the EAP distribution of consumption expenditure We can also directly evaluate the relative success of different country-percentiles by comparing their position in the EAP-wide distribution and how it changed over time. In Figure 11 the within-country position is shown on the horizontal axis, while the vertical axis plots the position in the regional distribution. For example, in 2012 the 20th percentile group in Thailand is around the median in the regional distribution. We have shown this for a number of countries and years, while this can obviously be repeated for every country-year. The China line shows 22 For ease of interpretation, the figure does not show Pacific countries as a separate group due to the low population coverage although they are still included in the calculations. Furthermore, the chart does not account for changes in country-coverage of the sample. For example, another reason for the increasing share of Other East Asia at the bottom is the higher survey coverage of low- and lower-middle income countries in 2012. 17 up as an almost straight line, i.e. the national percentiles are almost identical to the regional percentiles. This is because China dominates the regional percentiles given its population size. The percentile groups from Thailand have maintained their positions at the top of the regional distribution. By contrast, the Philippines, Indonesia and Cambodia have all lost out relative to the rest of the region (and relative to China, as we already saw in Figure 10). For example, in 1992 all percentile groups in Cambodia were richer than the corresponding Chinese groups. By 2012, this has changed completely, e.g. the Cambodian median is around the 30th regional percentile. On the other hand, Figure 11 shows a remarkable success story for Vietnam during the period 1992-2007.23 In 1992 it was in a similar, or even slightly worse position than Cambodia. But with every year it appears to have moved towards the upper left corner (while Cambodia moved in the opposite direction during 2002 and 2007), so much so that in 2007 the bottom quintile in Vietnam seem better off than their Chinese comparators. Figure 11: Regional composition of the East Asia distribution of consumption expenditure 23 Due to a different welfare aggregate (applied after 2010), we do not show the 2012 results for Vietnam although they are still included in the aggregate results. 18 VI. CONCLUSION By combining as many national household surveys as possible, this paper has studied the distribution of consumption expenditure among all individuals in (developing) East Asia. It thus complements analyses of inequality that are typically focused on inequality among individuals within the same country. The East Asian Gini index increased by more than 4 points (or almost 11%) between 1988 and 2012. The increase in inequality has been most pronounced during the initial decade (between 1988 and 1997). Regional inequality can be decomposed into differences among individuals residing in the same country, and differences between countries. The within-country component has increased over this period, driven by increasing national inequality in some populous countries – China initially and Indonesia a decade later. On the other hand, differences between countries have declined, i.e. average incomes have converged across countries. This can be explained by the rapid growth in China closing the gap with the regional average income, which counteracted the inequality-increasing effects of some lagging countries such as the Philippines. China thus had both equalizing and disequalizing effects on East Asian inequality, through its growth in average consumption and national inequality, respectively. But the rapid growth in average income dominated, thus leading to a net equalizing effect. As a result of these changes, the share of regional inequality that is explained by within-country inequality increased from two-thirds to close to 90%. At the global level, there has also been an increase in the within-country share, but at a much lower level – it increased from 20% to 35% over the same period (World Bank, 2016). Therefore, the situation within East Asia is already very close to an internalization of regional inequality within countries, in contrast to the global level.24 However, in the most recent period (2007-2012), which spans the Great Recession, regional inequality appears to have fallen slightly, consistently across different inequality measures. For the first time, the level of within-country inequality has stopped increasing, or might have even fallen slightly. But it is important to point out that there are at least two reasons why our data might be underestimating levels and increases in within-country inequality.25 First, household surveys tend to underestimate top incomes. The limited evidence that is available for the region suggests that top incomes might be rising quickly in recent years (Lakner and Ruggeri Laderchi, 2017). Second, our use of consumption expenditure might underestimate living standards at the top. The rising national savings rate in China (Ma and Yi, 2010) suggests that inequality in incomes would have increased faster than in terms of consumption. Therefore, while our results are robust to a number of methodological checks, we leave (at least) two important issues for future research and caution that the recent trend reversal should be interpreted carefully. 24 Bourguignon (2015) and Milanovic (2016) write that if the trends of falling between-country and increasing within-country inequality continue, one could get a situation where within-country inequality accounts for global inequality entirely, i.e. global inequality becomes “internalized” within countries. This would resemble the pattern of global inequality in the early 19th century (Bourguignon and Morrisson, 2002). 25 The effect on between-country inequality is a priori ambiguous. 19 REFERENCES Abdelkrim, A. and J. Y. Duclos: 2007, “DASP: Distributive Analysis Stata Package”, PEP, World Bank, UNDP and Université Laval. Alvaredo, F. and T. Piketty: 2014, “Measuring top incomes and inequality in the Middle East: Data limitations and illustration with the case of Egypt”, Working Paper 832, ERF. Amiel, Y. and F. Cowell: 1999, “Thinking about Inequality”, Cambridge University Press. Anand, S., and P. Segal: 2008, “What Do We Know about Global Income Inequality?” Journal of Economic Literature 46 (1): 57–94. Anand, S. and P. Segal: 2015, “Chapter 11 – The Global Distribution of Income”, In Atkinson A. B. and F. Bourguignon, Eds. “Handbook of Income Distribution”, Elsevier. Atkinson, A.B.: 1970, “On the Measurement of Inequality”, Journal of Economic Theory, 2(3), 244-263. Atkinson, A. B.: 1975, The Economics of Inequality. Clarendon Press Atkinson, A.B. and A. Brandolini: 2010, “On Analyzing the World Distribution of Income”, The World Bank Economic Review, 24(1), 1-37. Bourguignon, F.: 2015, The Globalization of Inequality. Princeton University Press. Bourguignon, F. and C. Morrisson: 2002, “Inequality among World Citizens: 1820-1992”, The American Economic Review 92(4), 727-744. Chen, S. and M. Ravallion: 2010, “The Developing World is Poorer than We Thought, But No Less Successful in the Fight against Poverty”, Quarterly Journal of Economics, 125(4), 1577-1625. Cowell, F. A. and E. Flachaire: 2007, “Income distribution and inequality measurement: The problem of extreme values”, Journal of Econometrics 141(2), 1044-1072. Elbers, C., P. Lanjouw, J.A. Mistiaen, and B. Özler: 2008, “Reinterpreting between-group inequality,” Journal of Economic Inequality, 6(3), 231-245. Ferreira, F.H.G., S. Chen, A. Dabalen, Y. Dikhanov, N. Hamadeh, D. Jolliffe, A. Narayan, E. B. Prydz, A. Revenga, P. Sangraula, U. Serajuddin, and N. Yoshida: 2016, “A Global Count of the Extreme Poor in 2012: Data Issues, Methodology and Initial Results”, Journal of Economic Inequality, 14(2), 141-172. Jirasavetakul, L.F., and C. Lakner: 2016. “The Distribution of Consumption Expenditure in Sub-Saharan Africa: The Inequality Among All Africans,” World Bank Policy Research Working Paper 7557. Jolliffe, D. and E. Prydz: 2015, “Global poverty goals and prices: how purchasing power parity matters”, World Bank Policy Research Working Paper Series 7256, The World Bank: Washington DC. Kanbur, R. and J. Zhuang, 2012. “Confronting Rising Inequality in Asia.” In: Asian Development Outlook 2012. Washington: Asian Development Bank. Kolm, S. C.: 1976, “Unequal Inequalities I”, Journal of Economic Theory, 12(3), 416-42. Lakner, C. and B. Milanovic: 2016, “Global Income Distribution: From the Fall of the Berlin Wall to the Great Recession”, World Bank Economic Review, 30(2), 203-232. Lakner, C., M. Negre and E.B. Prydz: 2014, “Twinning the Goals: How Can Shared Prosperity Help to Reduce Global Poverty?”, World Bank Policy Research Working Paper Series 7106. Lakner, C., M. Negre and E.B. Prydz: 2015, “The role of inclusive growth in ending extreme poverty,” Paper presented at Sixth ECINEQ Meeting, July 2015. 20 Lakner, C. and C. Ruggeri Laderchi, 2017, “Pulling apart? The growth of the super-rich in East Asia and Pacific and its implications for inclusive growth”, World Bank Policy Research Working Paper. Ma, G. and W. Yi: 2010, “China’s high saving rate: myth and reality”, BIS Working Papers 312, Bank for International Settlements. Milanovic, B.: 2005, “Worlds Apart: Measuring International and Global Inequality”, Princeton University Press. Milanovic, B.: 2016, Global Inequality: A New Approach for the Age of Globalization. The Belknap Press of Harvard University Press. Minoiu, C. and S. Reddy: 2014, “Kernel Density Estimation on Grouped Data: The Case of Poverty Assessment”, The Journal of Economic Inequality, 12(2), 163-189. Proctor, B.D., J.L. Semega and M.A. Kollar, 2016: “Income and Poverty in the United States: 2015,” U.S. Census Bureau, Current Population Reports, P60-256(RV). Ravallion, M.: 2014a, “Income inequality in the developing world”, Science, 344, 851-855. Ravallion, M.: 2014b, “Are the World’s Poorest Being Left Behind?”, NBER Working Papers 20791, National Bureau of Economic Research. Ravallion, M.: 2016, The Economics of Poverty: History, Measurement, and Policy. New York: Oxford University Press. Ravallion, M. and S. Chen: 2003, “Measuring pro-poor growth”, Economics Letters, 78(1), 93-99. Ruggeri Laderchi, C., N.L. Spatafora, S. Shetty and S. Zaidi: 2017, Richer and Fairer: An East Asian Miracle for the 21st Century, Washington, DC: World Bank. Shorrocks, A. and G. Wan: 2008, “Ungrouping Income Distributions”, Working paper 2008/16, UNU-WIDER. World Bank. 2016. Poverty and Shared Prosperity 2016: Taking on Inequality. Washington, DC: World Bank. 21 APPENDIX A. Methodological Details PovcalNet reports incomes in 2011 PPP-adjusted USD. For the microdata, which reports consumption expenditure in local currency units, we apply the same method as PovcalNet: We use the local consumer price index (CPI) to deflate consumption to 2011 domestic prices, and the 2011 PPP conversion factors for private consumption to convert into 2011 PPP-adjusted USD. We obtain the CPI from EAPPOV created by the World Bank’s EAP Team for Statistical Development, and the PPP exchange rate from the World Development Indicators. Adjustment for spatial price differences For China and Indonesia, PovcalNet reports rural and urban distributions separately, and we fit separate log-normal Lorenz curves to each. We create percentiles for the national distribution by combining these Lorenz curves, weighting by the appropriate rural/urban populations, and using PPP exchange rates which adjust for spatial price differences. For China we follow this procedure for all years, while for Indonesia we directly use the microdata (with spatial adjustment) in the last three benchmark years. Lakner et al. (2015) describe how to derive urban and rural PPP exchange rates in general. We use the following 2011 PPP exchange rates, which are implicit in the World Bank’s poverty monitoring (Ferreira et al., 2016) (national PPP exchange rates are shown for comparison): China: rural: 3.04, urban: 3.90, national: 3.70; Indonesia: rural: 3,666.16, urban: 4,360.48, national: 4,091.94. The corresponding 2005 PPP exchange rates, which are used in the robustness check below, are as follows: China: rural: 2.98, urban and national: 4.09; Indonesia: rural: 3399.68, urban: 4795.88, national: 4192.83. In some years, PovcalNet also reports national distributions for China and Indonesia (using 2011 PPP), which are constructed using a similar approach. We do not use these distributions because (1) they are not available in all years, and (2) we present results for both 2005 PPP (robustness check) and 2011 PPP (baseline) so we need to adopt a flexible approach. Comparing our results with PovcalNet’s national distributions, we obtain somewhat lower within-country inequality in China, and larger inequality in Indonesia: In China, across the first five benchmark years for which PovcalNet reports national distributions, our Gini is on average 0.5 points smaller. In Indonesia, between 1988 and 1992 our Gini is on average 1.9 points greater. These differences could be explained by the different functional forms for the Lorenz curve. PovcalNet chooses between a General Quadratic and Beta Lorenz depending on the goodness of fit. Instead, we use a lognormal Lorenz curve, which Shorrocks and Wan (2008) have shown to fit better. Estimates using 2005 PPP exchange rates The conversion from 2005 to 2011 PPP-adjusted USD is given by Lakner and Milanovic (2016) and Jolliffe and Prydz (2015) as 11 05 (1) 22 , where 11 ( 05 ) is consumption expenditure in 2011 (2005) PPP-adjusted USD. is the CPI in 2011, 11 is the PPP conversion factor in 2011, and the 2005 terms are defined accordingly. The observations from PovcalNet, which are reported in 2011 PPPs, can be converted into 2005 PPPs by rearranging Equation (1). For Micronesia, Mongolia and Tuvalu, 2005 PPP exchange rates are unavailable. Therefore, we have extrapolated the 2011 PPPs backwards to 2005 as follows 2005 2011 2005 2011 (3) 2005 2011 , where is the US CPI. B. Results of Robustness Checks We include four robustness checks on our main estimates of East Asian inequality. First, we allow for within-country price differences for a larger set of countries. Second, we check for the robustness to using the older set of PPP exchange rates for 2005. Third, we use a balanced sample of countries, i.e. those countries that are present throughout the period. Fourth, we use the raw data without assuming a parametric Lorenz curve. In the main text, we follow PovcalNet and adjust for spatial price differences only within China and Indonesia. While this approach produces a consistent time series, it is arguably inconsistent to spatially adjust in some but not other countries. In Table A1 panel A, we control for within- country price differences in additional country-years, using the consumption aggregate that is used by the World Bank in its regional poverty monitoring. We can only do this for the micro- data which become available in 2002 and hence there is a break in the time-series. In other words, for (say) the Philippines the first three benchmark years are drawn from PovcalNet and unadjusted, while the last three benchmark years stem from micro-data and can thus be adjusted for spatial price differences. Some countries are never adjusted, either because no spatial adjustment is available or because they are always drawn from PovcalNet. For Thailand, Solomon Islands and Papua New Guinea (in 1997), no spatially deflated consumption aggregate is available. The most important countries used from PovcalNet (without an adjustment) are Malaysia (all years) and Mongolia (all years except the last year which is taken from micro-data). The effects of the additional spatial adjustment are very small. As explained above, the first three benchmark years are identical since no additional adjustments are available.26 In 2012, the EAP Gini index falls from 43.3 to 43.2 when using the additional spatial price adjustment. As expected, the level of within-country GE(0) inequality falls. But all the results regarding the direction of regional inequality, as well as its country decomposition, remain robust. New PPP exchange rates become available periodically, reflecting new collections of international price data. Recently, the World Bank adopted the 2011 PPP exchange rates for its measurement of global poverty. Ferreira et al. (2016) provide a detailed explanation of how 26 For these years, almost all surveys come from PovcalNet. The only country with micro-data is Papua New Guinea in 1997, but no spatial adjustment is available. 23 the World Bank adopted these new PPP exchange rates, and in particular how the international poverty line was updated. Panel C provides a robustness check of our regional inequality results when we use the older set of 2005 PPP exchange rates. Using the 2005 PPPs has quite a large (predominantly downward) effect on the level of regional inequality, especially in the initial years.27 The EAP Gini index drops by around 2 points during the first three benchmark years. In contrast, the EAP Gini actually increases in the last year. Taken together, its increase between 1988 and 2012 becomes even stronger at around 20%. The within-country component (of GE(0) inequality) falls very slightly due to the change in the relative urban-rural price levels for China and Indonesia.28 But as expected the change in PPP exchange rates mostly affects the between-component. It falls strongly in the first three benchmark years but remains unchanged in 2007 and even increases in 2012. As a result the within-contribution is higher in the first three years but increases at a slower pace. We obtain similar results when we use the balanced samples – regional inequality is lower especially in the earlier years, so the increase over time becomes even stronger. We present results for two balanced samples. Panel D only uses countries that are present in every year during the entire period 1988-2012 which includes China, Indonesia, the Philippines and Thailand. Panel E lowers this requirement to the period 1992-2012, which adds Lao PDR and Vietnam to the sample. In sum, these results suggest that these sets of countries explain a substantial part of overall inequality, or that the omitted countries are not systematically different. In the final robustness check (panel F), we use the raw PovcalNet data of quantile groups without fitting a parametric Lorenz curve. Some national distributions are now represented by (say) ventile groups instead of percentiles in the baseline. This would tend to reduce within- country, and thus regional, inequality. Using the raw data, the regional Gini is lower by around 1 point in 1988 and 1992, and around 0.1 points after that. The effect is largest in the earlier years, because PovcalNet contributes a larger number of countries.29 Given that the two series are converging, the increase in the Gini index is larger in the raw data. In the baseline, the Gini increases by 10.8% between 2012 and 1988, while in the raw data it increases by 13.1%. The shape of the growth incidence curve (not shown) is also very similar in the two versions of the database. 27 It is interesting to note that the 2005 PPP have the opposite effect on the global Gini index (Lakner and Milanovic, 2016), i.e. the global Gini using 2005 PPPs exceeds the one using 2011 PPPs. This difference can be explained by how the change in the PPPs affects the between-country component regionally and globally. In other words, the regional effect depends on how (say) China’s price level moves relative to other East Asian countries, while the global effect depends on how it moves compared with the global mean price level. 28 Using the PPP exchange rates given above, it is clear that the relative urban-rural price level has fallen between 2005 and 2011 (from 1.37 to 1.29 for China and 1.41 to 1.19 for Indonesia), i.e. prices increased faster in rural than urban areas. Without a change in the relative urban-rural price levels, within-country inequality would have remained unchanged. This is because in China and Indonesia urban and rural incomes would have been multiplied by the same factor, leaving a scale-invariant measure such as GE(0) unchanged. 29 In addition, the number of fractile groups reported in PovcalNet increased over time. The average number of groups was 36 in 1988 compared with 71 across all years. 24 Table A1: East Asia and Pacific inequality ‐ robustness checks change (%,  change (%,  change (%,  change (%,  change (%,  1988‐2012  1988 1992 1997 2002 2007 2012 pp) pp) pp) pp) pp) change (%, pp) A. Regional inequality (from Table 3) Gini index (%) 39.0 6.2 41.5 3.0 42.7 1.2 43.2 3.3 44.6 ‐3.1 43.3 10.8 GE(0) (Theil‐ L) 0.25 13.0 0.29 6.2 0.30 2.8 0.31 8.2 0.34 ‐5.7 0.32 26.0 GE(0) within level 0.17 18.7 0.20 3.6 0.21 31.3 0.28 6.6 0.30 ‐1.1 0.29 70.2 GE(0) within contribution (%) 68.1 3.4 71.5 ‐ 1.8 69.7 19.3 89.0 ‐1.4 87.6 4.3 92.0 23.9 B. Regional inequality ‐ Additional adjustment for within‐country spatial price differences Gini index (%) 39.0 6.2 41.5 3.0 42.7 0.7 43.0 3.8 44.6 ‐3.5 43.1 10.4 GE(0) (Theil‐ L) 0.25 13.0 0.29 6.2 0.30 1.9 0.31 9.1 0.34 ‐6.5 0.32 24.9 GE(0) within level 0.17 18.7 0.20 3.6 0.21 30.1 0.28 6.7 0.29 ‐1.3 0.29 68.6 GE(0) within contribution (%) 68.1 3.4 71.5 ‐ 1.8 69.7 19.3 89.0 ‐2.0 87.0 4.8 91.8 23.8 C. Regional inequality ‐ 2005 PPP‐adjusted USD Gini index (%) 36.4 6.8 38.9 4.6 40.7 3.5 42.1 5.2 44.3 ‐1.7 43.6 19.6 GE(0) (Theil‐ L) 0.22 14.0 0.25 10.3 0.28 6.9 0.29 13.2 0.33 ‐2.5 0.32 48.5 GE(0) within level 0.17 18.1 0.20 4.5 0.20 30.6 0.27 6.9 0.29 ‐0.9 0.28 70.8 GE(0) within contribution (%) 75.7 2.7 78.4 ‐ 4.2 74.3 16.5 90.8 ‐5.1 85.7 1.4 87.1 11.4 D. Regional inequality ‐ Balanced sample (1988 to 2012) Gini index (%) 36.3 8.4 39.3 2.3 40.2 6.1 42.7 2.7 43.9 ‐0.8 43.5 19.9 GE(0) (Theil‐ L) 0.22 18.2 0.26 4.4 0.27 14.2 0.30 6.9 0.33 ‐1.1 0.32 49.0 GE(0) within level 0.17 18.8 0.20 3.1 0.21 34.8 0.28 6.8 0.30 ‐0.9 0.30 74.6 GE(0) within contribution (%) 78.4 0.4 78.8 ‐ 1.0 77.8 14.0 91.8 ‐0.1 91.7 0.2 91.9 13.5 E. Regional inequality ‐ Balanced sample (1992 to 2012) Gini index (%) 39.2 2.2 40.0 6.4 42.6 2.4 43.6 ‐0.7 43.3 10.6 GE(0) (Theil‐ L) 0.25 4.1 0.26 14.7 0.30 6.3 0.32 ‐0.7 0.32 25.9 GE(0) within level 0.20 2.9 0.21 33.6 0.28 6.5 0.29 ‐0.4 0.29 45.8 GE(0) within contribution (%) 79.6 ‐ 0.9 78.6 13.0 91.6 0.2 91.8 0.3 92.1 12.5 F. Regional inequality ‐ Raw data without parametric Lorenz curve Gini index (%) 38.1 5.9 40.4 5.5 42.6 1.1 43.1 3.3 44.5 ‐3.0 43.1 13.1 GE(0) (Theil‐ L) 0.24 12.9 0.28 9.2 0.30 2.6 0.31 8.0 0.33 ‐5.6 0.32 29.0 GE(0) within level 0.16 18.9 0.19 7.8 0.21 31.1 0.28 6.2 0.29 ‐0.6 0.29 77.1 GE(0) within contribution (%) 67.0 3.5 70.5 ‐ 1.0 69.6 19.3 88.9 ‐1.5 87.4 4.6 92.0 25.0 Notes: For the decompositions by country, changes between benchmark years are in percentage points. For all other rows, changes are measured in percent (not annualised). Observations are weighted  using population. Using dataset of 100 percentile groups per country‐year, based in part on a parametric Lorenz curve, except in panel F. In panel E, the last column refers to changes between 1992 and  25