Policy Research Working Paper 11067 Rapid Economic Growth but Rising Poverty Segregation Will Viet Nam Meet the SDGs for Equitable Development? Hai-Anh H. Dang Shatakshee Dhongde Minh Do Cuong Viet Nguyen Obert Pimhidzai Development Data Group & Poverty and Equity Global Department February 2025 Policy Research Working Paper 11067 Abstract Viet Nam is widely regarded as a success story for its impres- particularly those with a larger ethnic minority population. sive economic growth and poverty reduction in the last few The analysis finds a beneficial impact of economic growth decades. Yet, recent evidence indicates that the country’s on poverty reduction, but this can depend on inequality economic growth has not been uniform. Compiling and levels. It also finds that greater inequality has had negative analyzing new, extensive province-level data from the Viet- effects on economic growth but varying negative effects on nam Household Living Standards Surveys spanning 2002 to different poverty indicators. The paper provides supportive 2020 and other data sources, this paper finds within-prov- evidence of the beneficial impact of economic transitions ince inequality to be much larger than between-province from agriculture to non-agriculture. The results suggest that inequality. Furthermore, this inequality gap has been rising policy makers in Viet Nam should focus on reducing spa- over time. Despite the country’s fast poverty reduction, tial disparities and income inequality to attain sustainable the poor were increasingly segregated in certain provinces, economic development. This paper is a product of the Development Data Group, Development Economics and the Poverty and Equity Global Department. 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:// www.worldbank.org/prwp. The authors may be contacted at hdang@worldbank.org and vietcuong@vnu.edu.vn. 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 Rapid Economic Growth but Rising Poverty Segregation: Will Viet Nam Meet the SDGs for Equitable Development? Hai-Anh H. Dang, Shatakshee Dhongde, Minh Do, Cuong Viet Nguyen, and Obert Pimhidzai * JEL: C15, D31, I31, O10, O57 Keywords: poverty, inequality, pro-poor growth, convergence, household surveys, Viet Nam * This is forthcoming with Review of Development Economics. Dang (hdang@worldbank.org; corresponding author) is a senior economist in the Living Standards Measurement Study Unit, Development Data Group, World Bank and is also affiliated with GLO, IZA, Indiana University, and London School of Economics and Political Science; Dhongde (shatakshee.dhongde@econ.gatech.edu) is a professor of economics, Georgia Institute of Technology; Do (minh.nn.do@gmail.com) is a visiting lecturer with University of Economics and Business, Vietnam National University, Hanoi; Nguyen (vietcuong@vnu.edu.vn; corresponding author) is a researcher in International School, Vietnam National University, and Thang Long Institute of Mathematics and Applied Sciences (TIMAS), Thang Long University, Hanoi, Vietnam; Pimhidzai (opimhidzai@worldbank.org) is a lead economist with the Global Poverty Practice, World Bank. We would like to thank the editors Sanghamitra Bandyopadhyay, Andy McKay, three anonymous reviewers, Judy Yang, World Bank colleagues, and participants at Review of Development Economics’ “Inequality in the 21st century: Symposium” workshop and University of Michigan’s Sustainability and Development Conference (Ann Arbor) for useful comments on an earlier draft. We are grateful to the UK Foreign Commonwealth and Development Office (FCDO) for funding assistance through the Data and Evidence for Tackling Extreme Poverty (DEEP) Research Program. 1. Introduction Poverty and inequality provide two closely related, but different, measures of household welfare. As rising global living standards have led to less poverty, increasingly more attention has been placed on whether the fruits of economic growth are equally distributed to different population groups in a society. The Sustainable Development Goals (SDGs) adopted by the United Nations (UN) offer a most notable example where the needs to reduce both poverty (SDG number 1) and inequality (SDG number 10) are emphasized by the international community. Indeed, a country with good economic growth but a (highly) unequal income distribution may neither be able to shrink its poor nor narrow the undesirable gaps between its population segments. Viet Nam is widely regarded as a success story for its impressive economic growth and poverty reduction in the last few decades. Its growth has been found to be pro-poor since the early 1990s (Glewwe and Dang, 2011; Nguyen and Pham, 2018; Pimhidzai and Niu, 2021). Yet, recent evidence indicates that the country’s economic growth was not uniform: while inequality gaps between urban and rural areas have been found to narrow over time, they have widened within urban and rural areas (Bui and Imai, 2019). Furthermore, poverty rates became concentrated spatially (Lanjouw, Marra, and Nguyen, 2017) and among ethnic minority groups (Benjamin, Brandt, and McCaig, 2017), suggesting that attention can be focused on these particular groups for more effective poverty amelioration. In this paper, we aim to provide a long-term (and mostly descriptive) review of trends in poverty, inequality, and economic growth in Viet Nam. We contribute to the existing literature in several ways. First, we offer a broad but early assessment of the intertwined relationship between poverty, inequality, and economic growth for the country over the past 20 years. Recent review studies suggest that there is an intricate and complex relationship between these three development 2 outcomes, which require further evidence for better understanding (Cerra et al., 2022; Ferreira et al., 2022). In fact, we offer the first study to investigate all three development outcomes of poverty, inequality, and economic growth for Viet Nam. Specifically, we investigate several research questions that are highly relevant to policy. Have average incomes across provinces converged (or diverged) over time? If yes, how were the poor segregated? Why were the poor spatially concentrated in some provinces and not others? Did factors such as inequality, urbanization, investment, and volume of government spending play a role in determining poverty reduction? Furthermore, what could we tell about the relationship between economic growth, inequality, and (different measures of) poverty? Second, our analysis spans the period 2002-2020, which represents the longest-running period for the country that has been examined in an academic study. We complete this task by analyzing 10 rounds of the nationally representative Vietnam Household Living Standards Surveys (VHLSS). Our findings on the trends of poverty, inequality and economic growth, which are further disaggregated into between-province and within-province and regional variations, offer a nuanced picture of the evolution of these outcomes over time. To our knowledge, these findings are new and not available in previous studies. 1 Finally, we construct a database of panel data at the province level, which allows us to offer more granular analysis than the urban-rural dichotomy analyzed in previous studies. We further 1 Among the three development outcomes of poverty, inequality, and economic growth that we analyze, existing studies focus on either a single outcome or two outcomes. Their analyses are mostly restricted to survey data before 2010. For example, analyzing earlier VHLSS data covering the early 1990s to the early 2000s, Le and Booth (2014) and Nguyen et al. (2007) found widening urban-rural gaps in inequality. Benjamin et al. (2017) examine household income inequality during 2002-2014 through several dimensions, including urban-rural, ethnicity, and work sectors. Bui and Imai (2019) examine urban-rural gaps in Viet Nam during 2008-2012 and find education and remittances to be important factors. For other studies that focus on the ethnic gaps in living standards, see Dang (2012) and Fujii (2018). 3 supplement this database with data on government spending that we collect from the central government and different provincial government websites. As a result, we can zoom in on provinces that are good (or bad) performers. Furthermore, this rich panel database also allows us to employ rigorous econometric modeling techniques to investigate the channels affecting the province-level income distributions. Our estimation results suggest that within-province inequality has steadily increased over time. Within-province inequality is also much larger than between-province inequality, with the former type of inequality being almost three times the latter type of inequality in 2020. Average incomes and expenditures appear to converge across provinces, while poverty significantly declines during this period. The remaining poor seem to be regionally segregated among provinces with a greater share of ethnic minority population. We find beneficial impact of economic growth on poverty reduction. While the relationship of inequality with headcount poverty appears weak and not statistically significant, inequality is positively and strongly statistically correlated with poverty gap and poverty severity and might impede economic growth. Provinces with greater population density and a larger share of urban population have faster growth, whereas the opposite result holds for those with a greater share of ethnic minority groups. The transitions from an agriculture-based economy toward wage-based and services-based economy could be beneficial for the country’s economic growth and poverty reduction. This paper consists of four sections. In the next section, we describe the data (Section 2.1), discuss the overall trends in economic growth, inequality, and poverty at the country level (Section 2.2) and at the regional level as well as analyze poverty segregation across provinces (Section 2.3). We subsequently investigate in Section 3 the relationship between economic growth, inequality, and poverty, including convergence in inequality. We finally conclude in Section 4. 4 2. Trends in economic growth, inequality, and poverty 2.1. Data We compile data from the Vietnam Household Living Standards Surveys (VHLSSs), which has been widely employed by the government, the international community, and academic researchers for poverty and inequality analysis for the country. The VHLSSs have been conducted by the General Statistics Office of Vietnam with technical support from the World Bank every two years since 2002. We compile data on all 58 provinces and five centrally controlled municipalities and supplement this data with other data that we collected. 2 In particular, provincial government spending data for the period 2018-2020 is currently unavailable for all the 63 provinces in any official document. To get the most updated data, we manually collected the state spending and investment spending data for 2018-2020 from several sources including the Ministry of Finance’s website, provincial finance departments’ websites, and relevant official documents. The VHLSSs contain detailed data on individuals and households. Household-level data are collected on durables, assets, production, income, and participation in government programs. Individual-level data are collected on demographics, education, employment, health, and migration. The 1999 Population and Housing Census was used as the sampling frame of the VHLSSs during 2002-2008, while the 2009 and 2019 Population and Housing Censuses were used as the sampling frame of the VHLSSs respectively for 2010-2016 and 2018-2020. Around 3,100 communes were chosen as the primary sampling units out of the list of 10,000 communes for the 2 These municipalities are Can Tho, Da Nang, Hai Phong, Hanoi, and Ho Chi Minh City. 5 whole country. A village was randomly selected from each commune, and about 15 households were selected randomly from the village. The large-sample VHLSSs have a sample size of about 46,000 households that are designed to be representative at the provincial level. While these large-sample surveys only collect income data, a sub-sample of the VHLSSs (the small-sample VHLSSs of around 9,000 households) collect data on both income and expenditure. Since policy discourse in the country is typically based on poverty and inequality analysis using expenditure data, we mostly analyze such data in this paper. We obtain the province-level expenditure using Elbers et al.’s (2003) small area (poverty-map) estimation method (see Appendix B for more detailed description for the data). For robustness checks, we also provide some alternate estimates using per capita income in Appendix A, which offer qualitatively similar results. 2.2. Overall trends in economic growth, inequality, and poverty We provide in Table 1 the country trends in (real) per capita income, expenditure, poverty, and inequality. Between 2002 and 2020, per capita incomes increased significantly from 4,565 thousand (Vietnamese) dong to 15,156 thousand dong. 3 Similarly, real per capita expenditure more than tripled from 3,476 thousand dong in 2002 to 14,251 thousand dong in 2020. 4 This rapid economic growth has been widely attributed to important policy changes that took place in Viet Nam in the last two decades (see, e.g., Justino and Litchfield, 2014, Benjamin et al., 2017 for a 3 On average, slightly more than 18,000 Vietnamese dong equals one US dollar in this period (World Bank, 2018a). 4 There is a large difference in income and expenditure values between the 2008 and the 2010 VHLSS (Table 1). This difference is mainly because of a change in the recall period in the questionnaires. From 2002 to 2008, the survey asked for household expenditure or income in the past 12 months. However, from 2010 onward, these values were asked for the past month and then multiplied by 12 to estimate annual values. As a result, there is a break in values in per capita expenditure between values before 2010 and those since 2010. We have not made any adjustment to our data following a change in the recall period; for more related discussion on this topic, see Deaton and Kozel (2005). 6 review). Specifically, the “Doi Moi” (renovation) policies introduced in 1986 helped transform Viet Nam from a centrally planned economy to an open, export-oriented economy with high volume of trade and foreign direct investment. In 2001, the United States and Viet Nam signed a Bilateral Trade Agreement, in which the United States granted Viet Nam the status of the Most Favored Nations. As a result, tariffs on Vietnamese exports to the U.S. reduced significantly, triggering export-led economic growth. Figure 1 plots the distribution of log of per capita expenditure over time. Between 2004 and 2020, the distribution shifted significantly to the right, indicating a rise in average per capita expenditures. 5 However there is not a marked decrease in the variance of the distribution, indicating that inequality did not change rapidly during this period. We estimate three different measures of inequality, namely the Gini index and the Theil L and T indices and show the results in Table 1. All the three measures show steady levels of inequality in per capita expenditure. The average value of the Gini index was 0.37 and close to the lower side of the typical range of 0.3- 0.5 for Gini values for per capita expenditures in developing countries (World Bank, 2005). The Gini index is derived from the Lorenz curve and cannot be written as the sum of a term summarizing within-group inequality and a term summarizing between-group inequality (Bourguignon, 1979). Unlike the Gini index, the Theil index is a generalized entropy (GE) measure, and it can be decomposed into within and between components. Table 2 shows the decomposition of the Theil L index (GE 0; also known as the mean log deviation) and the Theil T index (GE 1). We find that within-province inequality increased over time. Within-province inequality explained about 66% of total inequality in 2002, but more than 70% in 2020. This 5 The distribution for 2010 (not shown, to make Figure 1 less cluttered) also shifts to the right of that of 2008, which further supports the increase in living standards over time. This also helps lessen potential concerns about comparability issues with the changes in the questionnaires as discussed above. 7 translates into within-province inequality increasing from about twice higher than between- province inequality in the early 2000s, to about three times higher than between-province inequality in the late 2010s. Thus, within-province inequality has become much more significant over time than between-province inequality. 6 There were significant differences in inequality levels within the provinces (Appendix A, Table A.1). Compared with the national average of 0.37, the Gini index was greater than 0.40 in Lao Cai, Dien Bien, and Lai Chau in the northern mountain region. In 2020, these provinces still had some of the remarkably high poverty rates (Lao Cai: 21.5%, Dien Bien 46% and Lai Chau: 36%). Inequality and poverty levels were similarly high in the central highlands region (e.g., Kom Tum and Gia Lai had Gini values of 0.41 and 0.39 and poverty rates of 17% and 27% respectively). On the other hand, inequality was lower in Mekong River delta with a Gini index of about 0.31 in Long An and Tien Giang (where poverty rates hover around 1%) and 0.28 in Hung Yen and Thai Binh in the Red river delta (where poverty rates range around 1%). Finally, we also show in Table 1 the estimates of three Foster-Greer-Thorbecke (FGT) indices of poverty (Foster et al., 1984). The poverty rate (i.e., the headcount ratio) measures the incidence or the proportion of the poor in the population, the poverty gap measures the depth of poverty (i.e., the average income shortfall of the poor), whereas the poverty severity index (i.e., the squared poverty gap) takes into account inequality of the income distribution among the poor. All three measures are estimated using the national (expenditure-based) poverty line as well as the World Bank’s PPP $3.1 per day poverty line. In tandem with the rapid economic growth, we find that poverty rates declined significantly. Nationwide, the headcount poverty rate decreased from 29% 6 Interestingly, the world has witnessed rising within-country inequality in the past two decades as well, although within-country inequality still tends to be lower than between-country inequality (Gradin, 2021). 8 in 2002 to less than 10% and 5% in 2016 and 2020, respectively. Notably, the first goal of the United Nations’ Millennium Development Goals was to reduce extreme poverty rates by half between 1990 and 2015 but Viet Nam appears to have well exceeded the target. 7 More summary statistics are provided in Appendix A, Table A.2. 2.3. Regional distribution of poverty The country-level trends in economic growth, poverty and inequality do not reflect the regional variation in these indicators. In Table A.1 in the Appendix, we present the estimates for each of these indicators in 2020, for all 63 provinces. In the last two decades, although overall poverty declined rapidly in Viet Nam, poverty rates varied significantly across provinces. Poverty rates were lowest in Ho Chi Minh City (1.8% average over time) and neighboring Binh Duong province (3.2% average). Ho Chi Minh City is the largest city and the prime economic center in Viet Nam. The city has numerous export processing zones, industrial parks, colleges, and universities, as well as the largest international airport in the country. After Ho Chi Minh City, Binh Duong is the second highest recipient of foreign direct investment. Both Ho Chi Minh City and Binh Duong province are in the southeast region. Poverty was also lower in the Red River Delta, for instance in the capital city of Hanoi (5.4% average) and the port city of Hai Phong (5% average). 8 On the other hand, poverty rates were very high in northwest provinces of Son La (50% average), Dien Bien (62% average), Lai Chau (60% average) and Ha Giang (56% average) in the northeast. These 7 The poverty and inequality estimates in Table 1 can differ from those published by the World Bank because of different data sources used. The country’s performance is even more impressive if we consider that the poverty rate was 58% in 1992-1993 (World Bank, 1999). 8 These results concur with those in earlier studies. In particular, McCaig (2011) finds that provinces that were more exposed to the U.S. tariff cuts experienced faster decreases in poverty in the early 2000s. Pham and Mukhopadhaya (2018) also found that poverty rates were lower in the regions of Southeast and Red River Delta. 9 provinces lie in the inland, mountainous regions, bordering China and the Lao People’s Democratic Republic and have more than 80% of their population residing in rural areas. Furthermore, more than 70% of their population consists of ethnic minorities. Given the large variance in poverty levels across provinces, there is evidence suggesting that poverty has become spatially concentrated over time (Lanjouw et al., 2017). We measure disparity in the regional distribution of poverty by estimating a poverty segregation curve (Dhongde, 2017). 9 The poverty segregation curve is a useful graphical tool to analyze how the regional distribution of the poor changed over time. The curve compares a province’s share of the poor population with its share of the overall population. The poor are segregated when provinces’ share of the poor does not resemble their share in the overall population. Perfect integration (zero segregation) implies that each province has the same share in the poor and the overall population (poor and non-poor combined). Figure 2 plots the poverty segregation curve for 2002 and 2020. The diagonal line of equality shows that there is zero segregation of the poor. The 2002 curve lies above the 2020 curve and hence dominates the 2020 curve. In other words, there was unambiguous increase in the segregation of the poor in Viet Nam. Poverty segregation curves may often intersect or overlap, and thus fail to provide a complete rank ordering of inequality. In Table 3, we calculate two indices of segregation, namely the Dissimilarity index and the Gini index. The Dissimilarity index is equal to one-half the sum of the absolute difference between the proportion of the poor and the proportion of the population across provinces. The Gini index is equal to twice the area between the segregation curve and the diagonal 9 Also see Massey (2016) for a more general discussion on the segregation curve. 10 of equality. 10 Between 2002 and 2010, there was not a marked increase in segregation. However, since 2012, there was a steady rise in the spatial inequality in the distribution of the poor. Provinces with a cumulative share of about 50% of the total population had about 30% of the poor population in 2002, whereas only about 3% of the poor population in 2020. Over the years, poor provinces such as Dien Bien and Lai Chau saw their share of the poor population increase disproportionately. Despite a rapid decline in average poverty levels, we find that the remaining poor were increasingly segregated in certain provinces in the country. 3. Relationship between economic growth, inequality and poverty 3.1. Factors affecting provincial poverty We further analyze the VHLSS data to find which factors were highly correlated with provincial poverty levels. Following the existing literature that examines the effects of income and inequality on poverty (Bourguignon, 2003; Kraay, 2006; Chen and Ravallion, 2010; Ferreira, 2010; Marrero et al., 2024), we estimate the following province fixed effects (FE) model , = + 1 ( ) + 2 + ′ + + + (1) In equation (1), Pi,t is a poverty index of province i in year t, Yi,t is per capita expenditure, Git is the Gini index, Xit is a vector of explanatory variables including high school completion rates, population density (in logarithmic form), shares of the urban population and the ethnic minority population, share of the population with a high school diploma, and different types of investments such as state spending and investment spending. These are the control variables that are widely 10 The Gini index satisfies the properties of symmetry, scale invariance, and the regressive transfer principle, and provides a consistent ranking of the distributions whenever the segregation curves do not intersect. The dissimilarity index does not satisfy the principle of regressive transfer. See Dhongde (2017) for a detailed discussion on these properties. 11 used in previous studies. We also include in Xit the shares of the population with wage income and non-farm income to control for local economic development. The year dummy variables and the province dummy variables help capture unobserved trends occurring for the same years and provinces. is the error term. The two coefficients 1 and 2 that measure the correlation between per capita expenditure and inequality are of primary interest. 11 The summary statistics are provided in Appendix A, Table A.2. For better comparison, we use three model specifications for equation (1) where we sequentially add the explanatory variables to the two core variables (log of) per capita expenditure and Gini index. For the first model, we add the province FE. For the second model, we add to model 1 the year FE. Finally, for model 3, we add all the remaining Xit control variables, which is our preferred model for interpretation. Table 4 shows that the estimated results for per capita expenditure vary somewhat between model 1 and model 2, which is perhaps due to unobserved trends occurring in different years. But more importantly, once we control for both the province and year FE, overall, the results remain similar between model 2 and model 3. In particular, Table 4 shows that, holding all other factors constant, a 1 percent rise in a province’s per capita expenditure reduced its poverty rate by 0.26 percentage points (column 3), its poverty gap by 0.10 percentage points (column 6), and its poverty severity index by 0.05 percentage points (column 9). Interestingly, holding fixed the per capita expenditure levels, a 1 11 We also estimate a variant of equation (1) where we include the explanatory variables in lagged forms using both province FE and GMM models. The results are qualitatively similar (Dang et al., 2023). We further show the estimation results with additional control variables for public infrastructure such as the share of villages with year- round passable roads, the share of villages with a daily market, and the distance from the village to the nearest town in Appendix A, Tables A.3 and A.4. Since these variables are computed from the commune module of the VHLSSs, we do not have complete data for these control variables for the whole period of our study. Yet, these tables show that provinces with more villages that have year-round passable roads could reduce poverty and have faster economic growth. 12 percent rise in the Gini index led to a reduction in the poverty rate of 0.19 percentage points, but this estimate is not statistically significant. But a similar increase in the Gini index has stronger and statistically significant effects on both the depth and severity of poverty (columns 6 and 9). These results generally concur with findings for other countries, which suggest that while economic growth can be beneficial for poverty reduction, this relationship can change depending on inequality levels (Cerra et al., 2022; Ferreira et al., 2022). But our results shed new light on the intricate relationship between poverty versus inequality and growth using the Viet Nam context and further highlight that we should consider other measures of poverty beyond headcount poverty for a more nuanced understanding of these dynamics. For the other control variables, Table 4 shows that keeping other factors fixed, provinces with a larger share of urban population or ethnic minority population had greater poverty and more poverty depth and poverty severity. Despite being only marginally statistically significant at the 10 percent level, the positive sign of the estimated coefficient of state spending on all the three poverty indexes is a puzzle (columns 3, 6, and 9). A possible explanation is that state spending could have been mainly reserved for improving infrastructure, and poorer (and more remote) provinces were likely to have been allocated more investment. While population density could raise the poverty rate, higher shares of wage or non-farm income could significantly lower it. The magnitude of impact is comparable to that of the province’s per capita expenditure. A 1 percent increase in a province’s share of non-farm income and wage income reduced its poverty rate by 0.27 and 0.31 percentage points, respectively (column 3). We further examine factors that are related to a province’s poverty segregation, as measured by its dissimilarity index. Table A.5 in Appendix A indicates that richer provinces or provinces with a lower share of urban population tend to have less poverty segregation, while the opposite 13 holds for provinces with more population density, or provinces that have a higher share of wage or non-farm income. These results hold whether or not we control for the Gini index, the other measure of poverty segregation (columns 3 and 4). 3.2. Factors affecting provincial economic growth In the previous section, we found a strong negative relation between per capita expenditure levels and poverty and a positive relation between inequality and poverty. In this section, we analyze how poverty and inequality in turn affect economic growth. There is an extensive literature on the impact of inequality and poverty on economic growth, but there appears to be inconclusive evidence on the impact (see Baselgia and Foellmi (2022), Cerra et al. (2022), and Ferreira et al. (2022) for recent reviews). To analyze the inequality-growth and the poverty-growth links, Marrero and Serven (2022) employ the overlapping-generations model with learning-by-doing and knowledge spillovers (Aghion et al., 1999). In this model, poor people have initial endowment below a minimum consumption level. The poor do not save and do not contribute to the aggregate economic growth. In this setting, this study shows that aggregate income growth depends on the share of people below the poverty threshold (i.e., poverty) and on the distribution of endowments (i.e., inequality). Using VHLSSs data, we estimate their reduced form empirical model: ∆, = + 1 �,−1 � + 2 ,−1 + 3 ,−1 + ′,−1 + + + , (2) where ∆, = �, � − �,−1 �. Note that equation (2) is a combination of the standard empirical poverty and growth regression (Ravallion, 2012) and the standard inequality-growth regression (Forbes, 2000; Berg et al., 2018), 14 which allows for testing the impact of poverty and inequality on expenditure growth in the same equation. As Marrero and Serven (2022) point out, the identifying assumption for the main coefficients of interest ′s in this equation is that poverty is not an almost exact linear combination of expenditure and inequality. Equation (2) is also a standard model used to test conditional β-convergence in per capita expenditure (Barro and Sala-i-Martin, 1991). 12 Growth of per capita expenditure, which equals the difference between log of current per capita expenditure and log of lagged per capita expenditure, is regressed on lag of the control variables. This model is now modified by adding lagged values of inequality and poverty to the set of growth determinants. Clearly, we should be cautious and interpret the lagged Gini and poverty rate in equation (2) as having a correlational—rather than causal—relationship with income growth. Put differently, compared to equation (1), equation (2) presents a related, but different, hypothesis on the intertwined relationship between income growth, inequality, and poverty where the outcome is a change variable (i.e., growth of per capita expenditure) instead of a level variable (i.e., poverty indicators) as with equation (1). We estimate equation (2) using both province FE and GMM models, with the GMM model being our preferred model for interpretation. The province FE model can be biased if the log of lagged per capita expenditure is correlated with unobserved variables (but these can serve as useful robustness checks). We address this omitted variable bias as follows. Firstly, we also estimate the model of first-differenced variables, and the first difference transformation removes the time- invariant unobserved effect ( ). The Arellano–Bond tests for zero autocorrelation of the first order 12 Figure A.1 in Appendix A shows a scatter plot with initial expenditure levels and growth in expenditure in the following years. The scatter and the fitted line indicate that a negative relation between initial per capita expenditure levels and growth rates, suggesting β-convergence. Moreover, the correlation is higher for the long-term growth. It implies that in the long run provinces tend to be convergent in economic growth. 15 and second order in first-differenced errors are also reported, and these test statistics for autocorrelation present no evidence of model misspecification (i.e., the AR1 and AR2 test values of -6.41 and 4.25 correspond to p-values of less than 0.01; Table 5, column 6). Secondly, we apply the GMM estimator which was developed by Holtz-Eakin, Newey, and Rosen (1988) and Arellano and Bond (1991). The GMM-type instruments for the log of lagged per capita expenditure are higher order lags of the per capita expenditure variables. Although the exogeneity of these instruments may be questionable, we can perform the overidentification test to assess the validation of the instruments. The Sargan test statistics of overidentifying restrictions is performed and reported at the bottom of Table 5 (column 6). The null hypothesis that overidentifying restrictions are valid is not rejected in all the regressions. We use the same vector of explanatory variables Xit as with equation (1). We also estimate three model specifications, where we sequentially add the explanatory variables to the two core variables: lagged (log of) per capita expenditure and Gini index. That is, we add to the first model specification the province FE, and we add to the second model specification the year FE. Finally, for the third model specification, we add all the remaining Xit control variables, which is our preferred model for interpretation. We present the estimation results for equation (2) in Table 5. Similar to Table 4, the second model specification with both province and year FE provides intermediate estimation results between the other models: the absolute values of its estimated lagged expenditures (columns 2 and 5) are larger than those using the first model specification with province FE (columns 1 and 4) but smaller than those using our preferred third model specification with all the control variables (columns 3 and 6). We also re-estimate equation (2) using per capita income instead of per capita expenditure and show the estimates in Appendix A, Table A.6. 16 In almost all the different specifications in Table 5, the estimated coefficient on lagged log of per capita expenditure is negative and significant at the 1% level. Thus, there is evidence of conditional β-convergence of per capita expenditure across provinces. Provinces with higher expenditure experienced a lower growth rate. The absolute magnitude of the estimate of the lagged log of per capita expenditure is larger in models with control variables than in those without control variables. According to our preferred GMM model with control variables (column 6), if per capita expenditure in the current period increased by 1 percent, the growth rate of per capita expenditure in the next period is 0.60% lower. In other words, if per capita expenditure increases by 1 percent in the current period, per capita expenditure in the next period will increase by 0.40% (i.e., one minus 0.60). In addition to finding evidence on conditional β-convergence across provinces, Table 5 shows strong negative correlation between inequality and growth in per capita expenditures, suggesting that inequality is harmful for economic growth. Specifically, a 1 percent increase in the Gini index in the current period is associated with a 0.96% decrease in per capita expenditure growth rate in the next period (column 6). 13 However, poverty rates are not statistically significantly correlated with growth. However, when we replace the headcount poverty rate with poverty gap and poverty severity indexes, the estimated coefficient on the Gini loses statistical significance; the other two poverty indexes become strongly statistically significant instead (Appendix A, Tables A.7 and A.8, columns 6). Since these two other poverty indexes better account for inequality compared to the 13 While this impact may appear larger than those in recent cross-country studies, the difference can be due to different inequality levels and modelling options. For example, Berg et al. (2018) find that a 10-percentile increase in net Gini decreases growth on average by 0.48 percentage points per year, but the Gini index range in their study is around 0.30- 0.70, which is larger than ours at 0.26-0.48. On the other hand, Forbes (2000) finds the opposite result that a ten-point increase in a country’s Gini coefficient is correlated with a 1.3 percent increase in average annual growth over the next five years. 17 headcount poverty rate, this indicates that a combination of poverty and inequality could be harmful for economic growth. Regarding other control variables, provinces with more population density or larger shares of urban population or wage and non-farm income have faster growth. While the results with the population variables have opposite effects on poverty, the results with wage and non-farm income are consistent with those discussed earlier with Table 4. These suggest that the transitions from an agriculture-based economy toward wage-based and services-based economy could be beneficial for the country’s economic growth and poverty reduction. Our result is consistent with a key finding that movement of labor from agriculture to non-agriculture through structural transformation are a key driver behind economic development in recent studies for Viet Nam (Tarp, 2017; Liu et al., 2022), and other countries (Deininger et al., 2022). As further robustness checks, we modify equation (2) by adding the interaction terms between lagged per capita expenditure (and income) and lagged values of control variables (state spending, investment spending, population density, share of urban population, share of population with high- school diploma and share of ethnic minority) and show the estimation results in Appendix A, Tables A.9 and A.10. We still find strong evidence on conditional β-convergence in both tables. Furthermore, the interaction terms of expenditure and income with poverty is negative and strongly statistically significant in both tables, and the interaction term of income and inequality is negative in Table A.10. These results suggest that, for the same level of economic development, higher levels of poverty and inequality are generally not conducive to economic growth. 4. Discussion and conclusion 18 In this paper, we provide a comprehensive analysis of poverty and inequality trends in Viet Nam in the past two decades. We compile new, extensive panel data at the province level over the past two decades using the Vietnam Household Living Standards Surveys and other data sources. We find that although average incomes between provinces tended to converge, there was a significant rise in within-province inequality over time. Within-province inequality was three times larger than between-province inequality in 2020. While economic growth helped reduce poverty, greater inequality could negatively affect poverty gap and poverty severity as well as economic growth. Although poverty levels in the country declined significantly, the poor were increasingly segregated in certain provinces. In particular, provinces with a larger share of ethnic minority groups had greater poverty levels. Economic transitions toward wage and service economies could have beneficial impacts on growth and poverty reduction but could also increase poverty segregation. SDG 1 calls for an end to poverty in all its forms. Certainly, the goal refers to a broader notion of poverty. Admittedly a limitation of our analysis is that we can focus only on monetary (i.e., consumption and income) poverty and do not measure changes in multi-dimensional poverty in Viet Nam. However, goal 1 also emphasizes the need to address poverty in specific underserved geographic areas within each country. To that effect, our analysis examines a new aspect of poverty in the country and reveals a rise in the segregation of its poor. This new finding lends further support to those in the existing studies that poorer population groups are both spatially and ethnically concentrated (Benjamin, Brandt, and McCaig, 2017; Lanjouw, Marra, and Nguyen, 2017; Fujii, 2018). As such, although Viet Nam has made substantial progress in reducing overall poverty, geographically targeted policies can be designed to reach out to the remaining poor. 19 Another of our contributions is to highlight the importance of reducing inequality. SDG 10 urges policy makers to accelerate progress towards lowering inequality within countries. Although inequality in Viet Nam is lower compared to many other developing countries, we found a rising share of inequality within provinces. Greater inequality levels were negatively correlated with economic growth and positively correlated with poverty levels. The recent Covid-19 pandemic can offer an illustrative example for both the importance of reducing inequality and poverty segregation. Analyzing Labor Force Surveys data spanning the pandemic, Dang, Nguyen, and Carletto (2023) find that the pandemic had far stronger effects on low-wage workers; specifically, it increased the proportion of below-minimum wage workers by 32% and worsened various wage equality indexes. This study also finds that the pandemic effects were smaller in provinces with greater openness to the global economy (as measured by the share of exports and imports in provinces’ GDP). Reviewing the literature that studied the pandemic’s effects on vulnerable population groups in Viet Nam, Dang and Do (2023) also find that poor households, informal workers, people in poor health, ethnic minority groups, and other disadvantaged groups have been most affected by Covid-19. These results further highlight that inequality in the country can be deepened during times of crisis, affecting the more vulnerable groups that are most in need of assistance. Indeed, geographical disadvantages could explain most of the non-farm participation gap in disadvantaged communities (World Bank, 2019). Furthermore, while national target programs that specially support poorer communes have sustained high commune level investments, a smaller share went to the poorest communes (Pimhidzai and Niu, 2021). As such, place-based poverty interventions can help effectively target and mitigate the disadvantages of fewer economic activities in lagging areas, which typically have less population density. Investment in both digital 20 and physical infrastructure, such as building better Internet connections and roads, is beneficial for integrating these communities into the national (and global) economy. Promising directions for future research include investigation of the impacts of interventions such as place-based poverty programs in Viet Nam as well as experimenting with different analytical frameworks that can provide more insights into the intricate relationship between poverty, inequality, and economic growth. 21 References Aghion, P., Caroli, E., & Garcia-Penalosa, C. (1999). Inequality and economic growth: the perspective of the new growth theories. Journal of Economic Literature, 37(4), 1615-1660. Arellano, M., & Bond, S. (1991). Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Review of Economic Studies, 58(2), 277-297. Barro, R. J. (2012). Convergence and modernization revisited (No. w18295). National Bureau of Economic Research. Barro, R. J., Sala-i-Martin, X., Blanchard, O. J., & Hall, R. E. (1991). Convergence across states and regions. Brookings Papers on Economic Activity, 107-182. Baselgia, E., & Foellmi, R. (2022). Inequality and growth: a review on a great open debate in economics. WIDER Working Paper Series, (wp-2022-5). Benjamin, D., Brandt, L., & McCaig, B. (2017). Growth with equity: income inequality in Vietnam, 2002–14. Journal of Economic Inequality, 15(1), 25-46. Berg, A., Ostry, J. D., Tsangarides, C. G., & Yakhshilikov, Y. (2018). Redistribution, inequality, and growth: new evidence. Journal of Economic Growth, 23, 259-305. Bourguignon, F. (1979). Decomposable income inequality measures. Econometrica, 47(4): 901- 920. Bourguignon, F. (2003). The growth elasticity of poverty reduction: explaining heterogeneity across countries and time periods. In T. S. Eicher & S. J. Turnovsky (Eds.), Inequality and Growth: Theory and Policy Implications. MIT Press. Bui, T. P., & Imai, K. S. (2019). Determinants of rural-urban inequality in Vietnam: Detailed decomposition analyses based on unconditional quantile regressions. Journal of Development Studies, 55(12), 2610-2625. Cerra, M. V., Lama, M. R., & Loayza, N. (2022). “Links between growth, inequality, and poverty: a survey”. In Cerra et al. (2022). (Eds.). How to Achieve Inclusive Growth. United Kingdom: Oxford University Press. Chen, S., & Ravallion, M. (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. Dang, H. A. (2012). “Vietnam: A widening poverty gap for ethnic minorities. Indigenous Peoples, Poverty and Development”. In Gillette Hall and Harry Patrinos. (Eds.) Indigenous Peoples, Poverty and Development. Cambridge University Press. 22 Dang, H. A., & Do, M. N. (2023). COVID-19 Pandemic and the Health and Well-being of Vulnerable People in Vietnam. In Handbook of Social Sciences and Global Public Health. Cham: Springer International Publishing. Dang, H. A., Nguyen, C., & Carletto, C. (2023). Did a Successful Fight against COVID-19 Come at a Cost? Impacts of the pandemic on employment outcomes in Vietnam. World Development, 161, 106129. Dang, H. A., Dhongde, S., Do, M., Nguyen, C. V., & Pimhidzai, O. (2023). Rapid Economic Growth but Rising Poverty Segregation: Will Vietnam Meet the SDGs for Equitable Development? IZA Discussion paper no. 15196. Deaton, A., & Kozel, V. (2005). Data and dogma: the great Indian poverty debate. World Bank Research Observer, 20(2), 177-199. Dhongde, S. (2017). Measuring Segregation of the Poor: Evidence from India. World Development, 89, 111-123. Elbers, C., Lanjouw, J. O., & Lanjouw, P. (2003). Micro-level estimation of poverty and inequality. Econometrica, 71(1), 355-364. Ferreira, F.H. (2010). “Distributions in Motion”. In Oxford Handbook of the Economics of Poverty. Ferreira, I. A., Salvucci, V., & Tarp, F. (2022). Poverty, inequality, and growth: trends, policies, and controversies (No. wp-2022-43). World Institute for Development Economic Research (UNU-WIDER). Forbes, K. J. (2000). A reassessment of the relationship between inequality and growth. American Economic Review, 90(4), 869-887. Foster, J., Greer, J., & Thorbecke, E. (1984). A class of decomposable poverty measures. Econometrica, 52(3): 761-766. Fujii, T. (2018). Has the development gap between the ethnic minority and majority groups narrowed in Vietnam?: Evidence from household surveys. World Economy, 41(8), 2067-2101. Glewwe, P., & Dang, H. A. H. (2011). Was Vietnam’s economic growth in the 1990s pro-poor? An analysis of panel data from Vietnam. Economic Development and Cultural Change, 59(3), 583-608. Gradín, C. (2021). Trends in global inequality using a new integrated dataset (No. 2021/61). WIDER Working Paper. Holtz-Eakin, D., Newey, W., & Rosen, H. S. (1988). Estimating vector autoregressions with panel data. Econometrica, 1371-1395. 23 Justino, P., & Litchfield, J. (2003). Poverty dynamics in rural Vietnam: winners and losers during reform. Poverty Research Unit at Sussex Working Paper, 10. Kraay, A. (2006). When is growth pro-poor? Evidence from a panel of countries. Journal of development economics, 80(1), 198-227. Lanjouw, P., Marra, M., & Nguyen, C. (2017). Vietnam’s evolving poverty index map: Patterns and implications for policy. Social Indicators Research, 133(1), 93-118. Le, H. T., & Booth, A. L. (2014). Inequality in Vietnamese Urban–Rural Living Standards, 1993– 2006. Review of Income and Wealth, 60(4), 862-886. Liu, Y., Barrett, C. B., Pham, T., & Violette, W. (2020). The intertemporal evolution of agriculture and labor over a rapid structural transformation: Lessons from Vietnam. Food Policy, 94, 101913. Massey, D. S. (2016). “Segregation and the perpetuation of disadvantage”. In Oxford Handbook of the Social Science of Poverty, 16, 369-405. Marrero, G. A., & Servén, L. (2022). Growth, inequality and poverty: a robust relationship? Empirical Economics, 63(2), 725-791. Marrero, Á. S., Marrero, G. A., & Servén, L. (2024). Poverty convergence clubs. Review of Income and Wealth. Doi: https://doi.org/10.1111/roiw.12688. McCaig, B. (2011). Exporting out of poverty: Provincial poverty in Vietnam and US market access. Journal of International Economics, 85(1), 102-113. Nguyen, C. V., & Pham, N. M. (2018). Economic growth, inequality, and poverty in Vietnam. Asian‐Pacific Economic Literature, 32(1), 45-58. Nguyen, B. T., Albrecht, J. W., Vroman, S. B., & Westbrook, M. D. (2007). A quantile regression decomposition of urban–rural inequality in Vietnam. Journal of Development Economics, 83(2), 466-490. Pham, A. T. Q., & Mukhopadhaya, P. (2018). Measurement of poverty in multiple dimensions: The case of Vietnam. Social Indicators Research, 138(3), 953-990. Pimhidzai, O., & Niu, C. (2021). Shared gains: How high growth and anti-poverty programs reduced poverty in Vietnam- Vietnam poverty and shared prosperity update report. Washington, D.C.: World Bank Group. Ravallion, M. (2012). Why don't we see poverty convergence? American Economic Review, 102(1), 504-523. 24 Tarp, F. (Eds). (2017). Growth, Structural Transformation, and Rural Change in Viet Nam: A Rising Dragon on the Move. Oxford University Press. World Bank. (1999). Vietnam development report 2000: attacking poverty. World Bank in Vietnam. ---. (2005). Poverty Manual. Chapter 6. http://siteresources.worldbank.org/PGLP/Resources/PovertyManual.pdf ---. (2018a). World Development Indicators Online. ---. (2018b). Piecing together the Poverty Puzzle. Washington DC: World Bank. ---. (2019). Better opportunities for all: Poverty reduction and shared prosperity in Vietnam. Washington, D.C.: World Bank Group. 25 Table 1: National trends in per capita expenditure, inequality, and poverty Years Indicators 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020 4565.9 5335.9 6176.6 6414.9 9305.1 9680.7 10600.6 12044.0 14861.7 15156.3 Per capita income (‘000 D) (88.1) (142.6) (140.5) (171.4) (225.8) (184.7) (144.5) (172.9) (223.3) (197.1) 3476.1 4009.7 4519.2 4560.0 8520.3 8902.6 9586.5 11577.7 12376.8 14250.9 Per capita expenditure (‘000 D) (63.0) (113.1) (113.9) (93.6) (147.5) (121.9) (128.9) (190.7) (170.4) (218.8) Poverty estimate using the national expenditure poverty line 28.8 19.5 16.0 14.5 20.7 17.2 13.5 9.8 7 4.7 Poverty rate (%) (0.6) (1.0) (0.8) (0.8) (0.6) (0.6) (0.5) (0.5) (0.4) (0.3) 0.069 0.047 0.038 0.035 0.059 0.045 0.037 0.027 0.02 0.011 Poverty gap index (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.001) 0.024 0.017 0.014 0.012 0.024 0.017 0.015 0.010 0.008 0.004 Poverty severity index (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.000) Poverty estimate using the poverty line of 3.1$ PPP/day 56.2 45.3 34.7 35.7 9.6 6.5 5.7 4.3 3.2 1.4 Poverty rate (%) (0.8) (1.7) (1.5) (1.4) (0.4) (0.4) (0.3) (0.4) (0.3) (0.2) 0.187 0.141 0.103 0.100 0.023 0.014 0.013 0.010 0.007 0.003 Poverty gap index (0.003) (0.006) (0.005) (0.005) (0.001) (0.001) (0.001) (0.001) (0.001) (0.000) 0.081 0.061 0.042 0.040 0.008 0.005 0.005 0.003 0.001 0.023 Poverty severity index (0.002) (0.003) (0.002) (0.002) (0.001) (0.000) (0.000) (0.000) (0.000) (0.000) Gini 0.370 0.370 0.358 0.356 0.393 0.357 0.348 0.381 0.357 0.368 (0.003) (0.004) (0.004) (0.005) (0.006) (0.004) (0.004) (0.005) (0.004) (0.005) Theil L index 0.221 0.224 0.212 0.208 0.259 0.213 0.205 0.246 0.219 0.231 (0.008) (0.007) (0.006) (0.006) (0.009) (0.006) (0.006) (0.008) (0.006) (0.007) Theil T index 0.249 0.241 0.227 0.227 0.294 0.230 0.216 0.265 0.226 0.247 (0.010) (0.007) (0.008) (0.008) (0.014) (0.009) (0.009) (0.013) (0.007) (0.011) Note: i) Authors’ estimation from VHLSSs. ii) Standard errors in parentheses. iii) Per capita income and expenditure in thousand VND, adjusted to constant prices in 2002 26 Table 2: Percent share of inequality in per capita expenditures within and between provinces Years Inequality index 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020 Theil L Within 65.8 68.8 73.8 76.4 72.1 77.9 78.4 72.3 77.4 73.2 Between 34.2 31.2 26.2 23.6 27.9 22.1 21.6 27.7 22.6 26.8 Theil T Within 64.6 67.4 73.7 76.2 73.5 78.6 79.1 72.9 78.1 74.1 Between 35.4 32.6 26.3 23.8 26.5 21.4 20.9 27.1 21.9 25.9 Note: i) Authors’ estimation from VHLSSs 27 Table 3: Indices measuring provincial segregation Year Dissimilarity Index Gini Index 2002 0.21 0.30 2004 0.26 0.36 2006 0.28 0.38 2008 0.23 0.32 2010 0.24 0.34 2012 0.28 0.38 2014 0.32 0.45 2016 0.40 0.53 2018 0.50 0.66 2020 0.50 0.61 Note: i) Authors’ estimation based on VHLSSs data 28 Table 4: Factors related to provincial poverty Dependent variable is the poverty indexes of provinces Poverty headcount Poverty gap index Poverty severity index Explanatory variables ` Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Model 9 (1) (2) (3) (4) (5) (6) (7) (8) (9) -0.420*** -0.328*** -0.261*** -0.036*** -0.104*** -0.099*** -0.013*** -0.047*** -0.047*** Log of per capita expenditure (0.013) (0.061) (0.054) (0.003) (0.015) (0.015) (0.001) (0.009) (0.009) Gini index -0.507** -0.350* -0.125 0.180** 0.085 0.136** 0.086** 0.040 0.066* (0.203) (0.201) (0.166) (0.073) (0.055) (0.055) (0.041) (0.036) (0.037) Log of state spending 0.002* 0.001* 0.000* (0.001) (0.000) (0.000) 0.001 0.002 0.002 Log of investment spending (0.005) (0.002) (0.001) 0.293*** -0.007 -0.013 Log of population density (0.099) (0.021) (0.011) 0.003** 0.001** 0.000** Share of urban population (0.001) (0.000) (0.000) Share of population with high- 0.060 -0.091* -0.047 school diploma (0.297) (0.053) (0.031) Share of ethnic minority 0.001*** 0.000*** 0.000*** population (0.000) (0.000) (0.000) Share of wage income -0.310*** -0.026 -0.005 (0.106) (0.031) (0.016) Share of non-farm income -0.273*** -0.044 -0.019 (0.102) (0.027) (0.015) Constant 4.078*** 3.355*** 3.160*** 0.311*** 0.898*** 0.795*** 0.112*** 0.400*** 0.343*** (0.100) (0.479) (0.419) (0.025) (0.119) (0.106) (0.013) (0.073) (0.063) Year FE No Yes Yes No Yes Yes No Yes Yes Province FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Observations 630 630 630 630 630 630 630 630 630 Number of provinces 63 63 63 63 63 63 63 63 63 R2 0.873 0.892 0.158 0.550 0.716 0.843 0.507 0.647 0.746 0.0584 0.0609 0.316 0.0430 0.0335 0.0218 0.0217 0.0173 0.0129 0.0724 0.0609 0.0545 0.0247 0.0175 0.0167 0.0129 0.0101 0.00968 0.394 0.501 0.971 0.752 0.786 0.632 0.739 0.746 0.640 Notes: i) Authors’ estimation from VHLSSs using province FE models, ii) Robust standard errors in parentheses clustered at the province level, iii)*** p<0.01, ** p<0.05, * p<0.1. 29 Table 5: Relation between expenditure growth, poverty, and inequality Dependent variable is the growth of per capita expenditure (Log Yt – Log Yt-1) FE GMM Explanatory variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 (1) (2) (3) (4) (5) (6) -0.195*** -0.819*** -0.852*** -0.149*** -0.187*** -0.595*** Lagged log of per capita expenditure (0.033) (0.058) (0.062) (0.044) (0.049) (0.091) Lagged Gini index -0.631* 0.100 0.078 -0.581*** -0.853*** -0.960*** (0.362) (0.146) (0.156) (0.203) (0.161) (0.166) Lagged poverty rate -0.170** -0.070 -0.069 -0.055 0.055 -0.111 (0.073) (0.068) (0.071) (0.102) (0.100) (0.078) 0.002** 0.001 Lagged log of other State spending (0.001) (0.002) 0.001 0.000 Lagged log of investment spending (0.005) (0.008) Lagged log of population density -0.067 0.046*** (0.071) (0.014) -0.001 0.003*** Lagged share of urban population (0.001) (0.001) 0.051 0.060 Lagged share of population with high-school diploma (0.283) (0.171) 0.001*** -0.000 Lagged share of ethnic minority population (0.000) (0.000) Lagged share of wage income 0.252** 0.373*** (0.095) (0.111) Lagged share of non-farm income 0.149 0.238** (0.103) (0.096) Constant 2.101*** 6.704*** 6.732*** 1.655*** 1.861*** 5.082*** (0.369) (0.469) (0.496) (0.390) (0.422) (0.703) Year FE No Yes Yes Province FE Yes Yes Yes Observations 567 567 567 567 567 567 R2/ Chi2 0.093 0.894 0.903 2298 3926 7831 Number of provinces 63 63 63 63 63 63 30 Arellano-Bond test for AR(1) -7.365 -7.437 -6.411 Arellano-Bond test for AR(2) 2.939 2.837 4.253 Sargan test 311.1 310.8 135 Note: i) Authors’ estimation from VHLSSs using province FE model (columns 1 to 3) and GMM model (columns 4 to 6), ii) Robust standard errors in parentheses clustered at the province level, iii)*** p<0.01, ** p<0.05, * p<0.1. 31 Figure 1: Density of log of per capita expenditure over time .8 .6 Density .4 .2 0 4 6 8 10 12 14 Log of per capita expenditure VHLSS 2004 VHLSS 2012 VHLSS 2020 Note: i) Authors’ estimation based on VHLSSs data 32 Figure 2: Segregation of the poor across provinces in Viet Nam: 2002-2020 1 Equality 2002 0.9 2020 0.8 0.7 0.6 cumulative proportion of poor 0.5 0.4 0.3 0.2 0.1 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 cumulative proportion of population Note: i) Authors’ estimation based on VHLSSs data 33 Supporting Information for review and online publication only Appendix A: Additional tables and figures Table A.1: Average income, inequality, and poverty rates among provinces in Viet Nam in 2020 VHLSS 2004 VHLSS 2012 VHLSS 2020 Province Per capita Per capita Poverty Gini Per capita Per capita Poverty Gini Per capita Per capita Poverty Gini Province name income expend. rate (%) income expend. rate (%) income expend. rate (%) code (thousand (thousand (thousand (thousand (thousand (thousand VND) VND) VND) VND) VND) VND) Red River Delta Hà Nội 1 7,569 6,151 9.0 0.318 34,292 34,666 6.7 0.350 70,183 68,586 0.4 0.321 Quảng Ninh 22 8,061 5,411 10.3 0.342 30,562 27,687 13.7 0.368 53,041 47,940 3.7 0.319 Vĩnh Phúc 26 4,847 3,534 17.8 0.270 22,186 22,806 13.7 0.310 58,206 49,875 0.5 0.300 Bắc Ninh 27 5,891 4,482 8.4 0.279 29,435 26,730 6.0 0.306 68,768 54,120 0.0 0.281 Hải Dương 30 5,414 4,175 11.1 0.266 24,565 23,870 8.3 0.294 57,274 49,202 0.3 0.297 Hải Phòng 31 6,484 5,565 8.1 0.363 30,426 27,745 6.5 0.311 67,333 58,427 0.3 0.302 Hưng Yên 33 5,156 3,551 15.9 0.268 21,704 23,985 8.8 0.290 51,696 44,467 0.8 0.286 Thái Bình 34 4,586 3,676 14.8 0.259 20,488 21,297 12.3 0.295 56,373 44,839 1.1 0.275 Hà Nam 35 4,284 3,431 23.3 0.283 21,282 22,668 13.1 0.303 51,912 49,058 1.1 0.311 Nam Định 36 4,860 3,680 18.1 0.303 21,864 22,558 9.5 0.294 57,699 48,829 0.9 0.313 Ninh Bình 37 4,436 3,813 15.5 0.291 20,385 22,692 11.7 0.314 50,993 47,557 1.0 0.306 Northern Mountain Hà Giang 2 2,965 2,395 57.2 0.371 10,131 10,403 70.5 0.381 22,811 22,916 41.1 0.466 Cao Bằng 4 3,344 3,009 43.0 0.350 13,386 14,279 49.2 0.388 27,336 29,094 24.6 0.424 Bắc Kạn 6 3,265 2,730 42.7 0.312 14,306 15,161 42.2 0.356 27,778 28,930 18.7 0.391 Tuyên Quang 8 4,085 3,311 34.4 0.339 14,666 16,586 38.2 0.366 37,344 32,360 8.1 0.317 Lào Cai 10 3,362 2,937 53.5 0.400 13,831 15,817 51.6 0.441 30,663 29,258 21.5 0.426 Điện Biên 11 2,690 2,058 68.8 0.355 10,445 9,985 72.7 0.383 21,853 21,397 45.6 0.458 Lai Châu 12 2,579 1,949 70.9 0.339 9,424 9,619 74.4 0.400 24,449 21,207 35.9 0.385 Sơn La 14 3,326 2,416 59.0 0.372 12,526 12,048 60.4 0.361 27,329 24,156 30.4 0.393 Yên Bái 15 3,935 3,190 36.1 0.343 13,772 16,261 43.9 0.389 32,735 32,504 18.2 0.397 Hoà Bình 17 3,491 2,816 48.9 0.365 14,762 16,561 37.2 0.375 34,479 33,871 9.1 0.385 Thái Nguyên 19 4,761 4,005 23.6 0.340 21,266 22,855 16.5 0.341 43,535 42,082 4.0 0.309 Lạng Sơn 20 4,184 3,217 38.0 0.359 14,451 15,290 46.5 0.392 30,544 30,678 10.4 0.340 Bắc Giang 24 4,709 3,553 20.9 0.301 19,870 19,337 23.4 0.319 50,049 42,191 1.7 0.283 Phú Thọ 25 4,441 3,454 25.9 0.313 18,987 20,477 18.8 0.331 42,208 42,264 2.6 0.315 Central Coast Thanh Hoá 38 3,727 3,040 34.9 0.313 15,455 18,394 27.3 0.332 47,451 37,421 4.5 0.331 Nghệ An 40 3,750 3,087 33.2 0.304 16,561 19,018 26.2 0.343 49,245 36,426 8.9 0.348 34 VHLSS 2004 VHLSS 2012 VHLSS 2020 Province Per capita Per capita Poverty Gini Per capita Per capita Poverty Gini Per capita Per capita Poverty Gini Province name income expend. rate (%) income expend. rate (%) income expend. rate (%) code (thousand (thousand (thousand (thousand (thousand (thousand VND) VND) VND) VND) VND) VND) Hà Tĩnh 42 3,690 3,218 31.6 0.305 16,259 20,775 17.9 0.318 42,545 38,686 4.7 0.342 Quảng Bình 44 3,617 3,158 33.9 0.308 17,270 20,239 23.3 0.349 43,326 37,591 8.7 0.353 Quảng Trị 45 3,657 2,956 37.5 0.313 16,140 18,386 30.5 0.362 36,003 36,802 12.8 0.391 Thừa Thiên Huế 46 4,578 3,945 22.1 0.340 19,724 22,144 14.7 0.340 41,907 40,520 4.6 0.328 Đà Nẵng 48 8,043 6,859 2.9 0.305 34,565 35,850 3.2 0.336 61,791 70,096 0.3 0.321 Quảng Nam 49 3,924 3,295 28.4 0.300 17,434 18,872 22.5 0.312 44,941 42,968 4.5 0.320 Quảng Ngãi 51 4,048 3,499 22.6 0.292 15,684 18,303 25.7 0.324 40,594 38,232 8.0 0.361 Bình Định 52 5,021 4,107 21.1 0.333 20,673 21,349 14.6 0.313 43,066 44,340 2.1 0.336 Phú Yên 54 4,516 3,627 22.5 0.303 17,413 19,321 16.9 0.297 39,255 38,622 6.4 0.319 Khánh Hoà 56 5,676 4,555 16.5 0.340 21,138 24,903 14.3 0.348 40,098 44,819 3.8 0.318 Ninh Thuận 58 4,667 4,494 20.8 0.374 17,061 20,129 21.9 0.325 36,271 39,373 7.9 0.369 Bình Thuận 60 5,338 4,924 10.9 0.339 21,012 22,304 14.1 0.298 49,947 43,601 2.1 0.310 Central Highlands Kon Tum 62 4,084 3,015 47.8 0.402 17,072 16,731 47.6 0.422 36,165 33,221 16.8 0.411 Gia Lai 64 4,431 3,047 46.1 0.381 22,246 17,663 36.3 0.375 29,994 27,393 26.7 0.391 Đắk Lắk 66 4,622 3,211 37.0 0.372 20,877 18,844 30.4 0.365 35,057 33,144 14.1 0.349 Đắk Nông 67 4,257 2,975 32.0 0.295 20,059 18,240 23.9 0.314 35,779 36,117 7.5 0.333 Lâm Đồng 68 5,324 3,813 26.6 0.339 22,353 24,150 18.7 0.345 47,473 43,901 5.2 0.358 Southeast Bình Phước 70 5,848 4,502 14.4 0.319 26,671 22,980 11.8 0.291 50,678 48,115 3.4 0.362 Tây Ninh 72 5,721 4,627 11.3 0.312 23,193 21,192 14.0 0.284 53,535 47,888 0.8 0.312 Bình Dương 74 9,335 6,379 3.4 0.307 42,197 27,204 6.2 0.294 84,979 53,861 0.0 0.287 Đồng Nai 75 8,140 5,571 6.8 0.313 29,374 25,884 8.9 0.310 70,667 59,240 0.9 0.319 Bà Rịa - Vũng Tàu 77 7,868 6,372 6.2 0.345 33,363 29,922 5.2 0.318 54,245 68,820 0.2 0.344 Hồ Chí Minh 79 13,943 9,490 1.9 0.307 40,295 34,009 3.7 0.321 77,646 85,497 0.0 0.329 Mekong River Delta Long An 80 6,000 4,460 9.7 0.287 23,282 21,779 11.3 0.290 54,063 42,739 1.7 0.308 Tiền Giang 82 5,750 4,399 10.7 0.292 22,983 23,416 10.9 0.303 54,663 41,873 0.6 0.305 Bến Tre 83 4,999 4,121 12.8 0.295 20,071 20,014 15.5 0.305 47,149 37,194 1.7 0.288 Trà Vinh 84 4,755 3,614 20.3 0.301 18,860 17,416 27.8 0.331 50,631 33,305 8.3 0.335 Vĩnh Long 86 5,077 4,032 15.0 0.288 20,972 22,020 14.5 0.321 43,349 39,023 2.0 0.293 Đồng Tháp 87 5,687 4,492 16.7 0.458 19,400 19,866 19.3 0.321 52,289 34,445 4.5 0.289 An Giang 89 6,218 3,883 20.3 0.297 22,574 17,921 20.3 0.292 44,669 33,414 6.5 0.323 Kiên Giang 91 6,130 3,819 22.4 0.331 23,627 18,770 28.1 0.362 62,005 32,432 10.6 0.335 Cần Thơ 92 6,292 4,886 13.4 0.352 28,090 23,662 13.4 0.333 64,735 44,047 1.8 0.326 Hậu Giang 93 5,387 3,486 22.0 0.286 19,450 18,489 27.9 0.333 53,440 35,402 5.3 0.302 35 VHLSS 2004 VHLSS 2012 VHLSS 2020 Province Per capita Per capita Poverty Gini Per capita Per capita Poverty Gini Per capita Per capita Poverty Gini Province name income expend. rate (%) income expend. rate (%) income expend. rate (%) code (thousand (thousand (thousand (thousand (thousand (thousand VND) VND) VND) VND) VND) VND) Sóc Trăng 94 4,742 3,437 24.7 0.304 17,860 17,636 29.4 0.349 47,319 32,997 7.7 0.360 Bạc Liêu 95 5,610 3,583 22.8 0.305 24,444 19,485 22.8 0.326 51,605 34,829 7.2 0.357 Cà Mau 96 6,176 3,950 19.4 0.318 19,811 18,571 22.3 0.317 39,848 34,263 6.8 0.323 Note: i) Authors’ estimation from the 2020 VHLSS. 36 Table A.2: Summary statistics Variables Obs Mean Std. Dev. Min Max Headcount poverty 630 0.249 0.252 0.000 0.895 Poverty gap index 630 0.059 0.063 0.000 0.345 Poverty severity index 630 0.024 0.030 0.000 0.188 Log of per capita expenditure 630 8.724 0.563 7.411 10.148 Gini index 630 0.331 0.040 0.257 0.475 Log of state spending 630 15.085 1.761 0.000 18.180 Log of investment spending 630 13.927 1.130 8.397 17.275 Log of population density 630 -1.261 0.999 -3.413 1.482 Share of urban population 630 25.337 16.499 5.023 87.511 Share of population with high-school diploma 630 0.072 0.036 0.006 0.237 Share of ethnic minority population 630 20.851 27.157 0.000 97.066 Share of wage income 630 0.369 0.107 0.145 0.694 Share of non-farm income 630 0.207 0.061 0.045 0.559 37 38 Table A.3: Factors related to provincial poverty, with additional control variables Dependent variable is the poverty indexes of provinces Poverty headcount Poverty gap index Poverty severity index Explanatory variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Model 9 (1) (2) (3) (4) (5) (6) (7) (8) (9) -0.420*** -0.328*** -0.274*** -0.036*** -0.104*** -0.103*** -0.013*** -0.047*** -0.049*** Log of per capita expenditure (0.013) (0.061) (0.057) (0.003) (0.015) (0.015) (0.001) (0.009) (0.009) Gini index -0.507** -0.350* -0.090 0.180** 0.085 0.144*** 0.086** 0.040 0.069* (0.203) (0.201) (0.158) (0.073) (0.055) (0.052) (0.041) (0.036) (0.035) Log of state spending 0.001* 0.000 0.000 (0.001) (0.000) (0.000) 0.002 0.003* 0.002** Log of investment spending (0.005) (0.002) (0.001) 0.290*** -0.010 -0.015 Log of population density (0.102) (0.021) (0.011) 0.003*** 0.001*** 0.000*** Share of urban population (0.001) (0.000) (0.000) Share of population with high-school 0.143 -0.041 -0.017 diploma (0.299) (0.047) (0.027) 0.001*** 0.000*** 0.000*** Share of ethnic minority population (0.000) (0.000) (0.000) Share of wage income -0.316*** -0.032 -0.009 (0.104) (0.031) (0.017) Share of non-farm income -0.258** -0.042 -0.020 (0.113) (0.030) (0.016) Share of village with year-round passable -0.064 -0.040*** -0.024*** road (0.046) (0.011) (0.007) Share of village with daily market 0.039 0.018 0.010* (0.041) (0.011) (0.006) Distance from village to the nearest town -0.007 -0.001 0.000 (0.011) (0.003) (0.002) Constant 4.078*** 3.355*** 3.238*** 0.311*** 0.898*** 0.814*** 0.112*** 0.400*** 0.348*** (0.100) (0.479) (0.440) (0.025) (0.119) (0.099) (0.013) (0.073) (0.056) Year FE No Yes Yes No Yes Yes No Yes Yes 39 Province FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Observations 630 630 628 630 630 628 630 630 628 Number of provinces 63 63 63 63 63 63 63 63 63 R2 0.873 0.892 0.160 0.550 0.716 0.839 0.507 0.647 0.741 0.0584 0.0609 0.317 0.0430 0.0335 0.0213 0.0217 0.0173 0.0140 0.0724 0.0609 0.0542 0.0247 0.0175 0.0163 0.0129 0.0101 0.00946 0.394 0.501 0.971 0.752 0.786 0.631 0.739 0.746 0.686 Notes: i) Authors’ estimation from VHLSSs using province FE models, ii) Robust standard errors in parentheses clustered at the province level, iii)*** p<0.01, ** p<0.05, * p<0.1. 40 Table A.4: Relation between expenditure growth, poverty, and inequality, , with additional control variables Dependent variable is the growth of per capita expenditure (Log Yt – Log Yt-1) FE GMM Explanatory variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 (1) (2) (3) (4) (5) (6) -0.195*** -0.819*** -0.863*** -0.149*** -0.187*** -0.816*** Lagged log of per capita expenditure (0.033) (0.058) (0.065) (0.044) (0.049) (0.110) Lagged Gini index -0.631* 0.100 0.115 -0.581*** -0.853*** -0.233 (0.362) (0.146) (0.160) (0.203) (0.161) (0.305) Lagged poverty rate -0.170** -0.070 -0.073 -0.055 0.055 -0.203 (0.073) (0.068) (0.077) (0.102) (0.100) (0.127) 0.002* -0.001 Lagged log of other State spending (0.001) (0.002) 0.001 -0.012 Lagged log of investment spending (0.005) (0.012) Lagged log of population density -0.064 0.058*** (0.072) (0.021) -0.001 0.005*** Lagged share of urban population (0.001) (0.001) 0.096 -0.388 Lagged share of population with high-school diploma (0.291) (0.386) 0.001*** 0.001 Lagged share of ethnic minority population (0.000) (0.000) Lagged share of wage income 0.244** -0.050 (0.096) (0.192) Lagged share of non-farm income 0.156 -0.237 (0.106) (0.193) Share of village with year-round passable road -0.052 0.680*** (0.038) (0.215) Share of village with daily market 0.036 0.180 (0.032) (0.194) Distance from village to the nearest town -0.015 -0.094 41 (0.012) (0.069) Constant 2.101*** 6.704*** 6.835*** 1.655*** 1.861*** 7.207*** (0.369) (0.469) (0.524) (0.390) (0.422) (0.924) Year FE No Yes Yes Province FE Yes Yes Yes Observations 567 567 565 567 567 565 R2/ Chi2 0.0767 0.420 0.289 2298 3926 4210 Number of provinces 63 63 63 63 63 63 Arellano-Bond test for AR(1) -7.365 -7.437 -5.325 Arellano-Bond test for AR(2) 2.939 2.837 1.197 Sargan test 311.1 310.8 62.81 Note: i) Authors’ estimation from VHLSSs using province FE model (columns 1 to 3) and GMM model (columns 4 to 6), ii) Robust standard errors in parentheses clustered at the province level, iii)*** p<0.01 ** p<0.05, * p<0.1. 42 Table A.5: Factors related to provincial dissimilarity index Dissimilarity Index Explanatory variables Model 1 Model 2 Model 3 Model 4 (1) (2) (3) (4) 0.005*** -0.018*** -0.014** -0.012** Log of per capita expenditure (0.001) (0.006) (0.006) (0.006) Gini index 0.029 0.066*** 0.052*** (0.022) (0.022) (0.018) Log of state spending -0.000 0.000 (0.000) (0.000) -0.000 -0.000 Log of investment spending (0.001) (0.001) 0.030*** 0.030*** Log of population density (0.007) (0.007) -0.000*** -0.000*** Share of urban population (0.000) (0.000) -0.017 -0.019 Share of population with high-school diploma (0.020) (0.021) 0.000 0.000 Share of ethnic minority population (0.000) (0.000) Share of wage income 0.016** 0.021*** (0.007) (0.007) Share of non-farm income 0.015** 0.021** (0.007) (0.008) Constant -0.040*** 0.129*** 0.146*** 0.147*** (0.096) (0.419) (0.350) (0.024) Year FE No Yes Yes No Province FE Yes Yes Yes Yes Observations 630 630 630 630 Number of provinces 63 63 63 63 R2 0.112 0.0189 0.0339 0.0226 sigma_u 0.0118 0.0137 0.0264 0.0286 sigma_e 0.00573 0.00508 0.00461 0.00471 rho 0.810 0.880 0.970 0.974 43 Notes: i) Authors’ estimation from VHLSSs using province FE models, ii) Robust standard errors in parentheses clustered at the province level, iii)*** p<0.01, ** p<0.05, * p<0.1. Table A.6: Relation between income growth, poverty, and income inequality Dependent variable is the growth of per capita income FE GMM Explanatory variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 (1) (2) (3) (4) (5) (6) -0.169*** -0.589*** -0.579*** -0.156*** -0.119 -0.456*** Lagged log of per capita income (0.025) (0.050) (0.055) (0.037) (0.082) (0.091) -0.298 -0.258 -0.224 -0.347** -0.607** -0.607*** Lagged Gini index (0.222) (0.180) (0.167) (0.170) (0.270) (0.218) -0.236*** -0.032 0.027 -0.240** -0.091 -0.217 Lagged poverty gap (0.070) (0.096) (0.099) (0.094) (0.152) (0.210) -0.002 -0.003 Lagged log of other State spending (0.002) (0.002) 0.005 0.004 Lagged log of investment spending (0.007) (0.009) -0.279*** 0.054*** Lagged log of population density (0.072) (0.013) 0.001 0.003*** Lagged share of urban population (0.001) (0.001) -0.124 -0.024 Lagged share of population with high-school diploma (0.317) (0.305) 0.001 -0.000 Lagged share of ethnic minority population (0.000) (0.000) 44 0.188 0.083 Lagged share of wage income (0.117) (0.123) -0.060 -0.212 Lagged share of non-farm income (0.134) (0.165) 1.809*** 5.143*** 4.536*** 1.721*** 1.471** 4.306*** Constant (0.272) (0.432) (0.489) (0.348) (0.746) (0.789) Year FE No Yes Yes Province FE Yes Yes Yes Observations 567 567 567 567 567 567 0.137 0.485 0.515 2390 2632 2023 R2/ Chi2 Number of provinces 63 63 63 63 63 63 -5.118 -4.641 -5.017 Arellano-Bond test for AR(1) 2.257 1.338 0.710 Arellano-Bond test for AR(2) 144.9 82.10 74.92 Sargan test Note: i) Authors’ estimation from VHLSSs using province FE model (columns 1 to 3) and GMM model (columns 4 to 6), ii) Robust standard errors in parentheses clustered at the province level, iii)*** p<0.01, ** p<0.05, * p<0.1. 45 Table A.7: Relation between expenditure growth, poverty gap, and inequality Dependent variable is the growth of per capita expenditure (Log Yt – Log Yt-1) FE GMM Explanatory variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 (1) (2) (3) (4) (5) (6) -0.123*** -0.813*** -0.909*** -0.119*** -0.252** -0.856*** Lagged log of per capita expenditure (0.009) (0.062) (0.065) (0.013) (0.102) (0.134) Lagged Gini index -0.531 0.141 0.210 -0.711** -0.573 -0.103 (0.351) (0.149) (0.160) (0.338) (0.395) (0.276) Lagged poverty gap -0.080 -0.165 -0.638*** 0.105 -0.351 -2.192*** (0.260) (0.183) (0.209) (0.254) (0.611) (0.554) 0.002** 0.002 Lagged log of other State spending (0.001) (0.002) 0.003 0.011 Lagged log of investment spending (0.005) (0.008) Lagged log of population density -0.092 0.039** (0.068) (0.015) -0.001 0.004*** Lagged share of urban population (0.001) (0.001) Lagged share of population with high- 0.030 0.018 school diploma (0.266) (0.178) Lagged share of ethnic minority 0.002*** 0.001* population (0.000) (0.000) Lagged share of wage income 0.258*** 0.379*** (0.094) (0.107) Lagged share of non-farm income 0.144 0.184* (0.102) (0.109) Constant 1.396*** 6.613*** 7.100*** 1.413*** 2.367*** 6.864*** (0.124) (0.494) (0.507) (0.076) (0.773) (1.008) Year FE No Yes Yes Province FE Yes Yes Yes Observations 567 567 567 567 567 567 R2/ Chi2 0.091 0.894 0.904 2234 4069 7018 Number of provinces 63 63 63 63 63 63 Arellano-Bond test for AR(1) -7.365 -6.458 -6.183 Arellano-Bond test for AR(2) 2.939 4.202 3.671 Sargan test 311.1 136.2 115.4 Note: i) Authors’ estimation from VHLSSs using province FE model (columns 1 to 3) and GMM model (columns 4 to 6), ii) Robust standard errors in parentheses clustered at the province level, iii)*** p<0.01, ** p<0.05, * p<0.1. 46 Table A.8: Relation between expenditure growth, poverty severity, and inequality Dependent variable is the growth of per capita expenditure (Log Yt – Log Yt-1) FE GMM Explanatory variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 (1) (2) (3) (4) (5) (6) -0.116*** -0.804*** -0.891*** -0.113*** -0.241*** -0.782*** Lagged log of per capita expenditure (0.008) (0.060) (0.062) (0.012) (0.092) (0.128) Lagged Gini index -0.592* 0.131 0.180 -0.855** -0.605 -0.263 (0.352) (0.148) (0.160) (0.353) (0.390) (0.269) Lagged poverty severity 0.387 -0.194 -1.002*** 0.522 -0.600 -3.668*** (0.516) (0.314) (0.329) (0.575) (1.196) (1.093) 0.002** 0.002 Lagged log of other State spending (0.001) (0.002) 0.003 0.010 Lagged log of investment spending (0.005) (0.009) Lagged log of population density -0.099 0.039*** (0.068) (0.015) -0.001 0.004*** Lagged share of urban population (0.001) (0.001) 0.034 -0.019 Lagged share of population with high-school diploma (0.265) (0.175) 0.002*** 0.001 Lagged share of ethnic minority population (0.000) (0.000) Lagged share of wage income 0.270*** 0.399*** (0.093) (0.106) Lagged share of non-farm income 0.150 0.207* (0.102) (0.109) Constant 1.343*** 6.535*** 6.925*** 1.399*** 2.276*** 6.260*** (0.112) (0.477) (0.468) (0.072) (0.682) (0.948) Year FE No Yes Yes Province FE Yes Yes Yes Observations 567 567 567 567 567 567 R2/ Chi2 0.0744 0.420 0.246 1883 4033 6803 47 Number of provinces 63 63 63 63 63 63 Arellano-Bond test for AR(1) -7.372 -6.495 -6.190 Arellano-Bond test for AR(2) 2.990 4.226 3.693 Sargan test 310.5 135.8 114.9 Note: i) Authors’ estimation from VHLSSs using province FE model (columns 1 to 3) and GMM model (columns 4 to 6), ii) Robust standard errors in parentheses clustered at the province level, iii)*** p<0.01, ** p<0.05, * p<0.1. 48 Table A.9: Relation between expenditure growth, poverty, and inequality with interactions Dependent variable is the growth of per capita expenditure (Log Yt – Log Yt-1) Explanatory variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 -0.659*** -0.767*** -0.491*** -0.525*** -0.505*** -0.486*** -0.519*** -0.505*** Lagged log of per capita expenditure (0.180) (0.179) (0.074) (0.074) (0.072) (0.069) (0.088) (0.071) 0.011 Lagged log of per capita expenditure * Lagged log of other State spending (0.009) 0.016* Lagged log of per capita expenditure * Lagged log of investment spending (0.009) 0.001 Lagged log of per capita expenditure * Lagged log of population density (0.008) 0.000 Lagged log of per capita expenditure * Lagged share of urban population (0.000) 0.461* Lagged log of per capita expenditure * Lagged share of population with high-school diploma (0.252) 0.001* Lagged log of per capita expenditure * Lagged share of ethnic minority population (0.000) Lagged log of per capita expenditure * Lagged Gini 0.114 index (0.261) Lagged log of per capita expenditure * Lagged -0.188** poverty rate (0.077) -0.097 0.002 0.000 0.001 0.001 0.001 0.001 0.001 Lagged log of other State spending (0.083) (0.002) (0.002) (0.002) (0.001) (0.002) (0.002) (0.002) 0.003 -0.129 0.003 0.000 0.001 0.000 0.001 -0.002 Lagged log of investment spending (0.007) (0.082) (0.007) (0.007) (0.007) (0.007) (0.007) (0.007) 0.035*** 0.035*** 0.030 0.039*** 0.034*** 0.030*** 0.033*** 0.031*** Lagged log of population density 49 Dependent variable is the growth of per capita expenditure (Log Yt – Log Yt-1) Explanatory variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 (0.010) (0.011) (0.073) (0.011) (0.011) (0.010) (0.011) (0.010) 0.002*** 0.002*** 0.002*** -0.002 0.002*** 0.003*** 0.002*** 0.002*** Lagged share of urban population (0.001) (0.001) (0.001) (0.005) (0.001) (0.001) (0.001) (0.001) Lagged share of population with high-school 0.175 0.163 0.063 0.203 -3.784* 0.102 0.145 0.177 diploma (0.156) (0.155) (0.138) (0.146) (2.155) (0.135) (0.146) (0.159) 0.000 -0.000 -0.000 0.000 0.000 -0.006** -0.000 -0.000 Lagged share of ethnic minority population (0.000) (0.000) (0.000) (0.000) (0.000) (0.003) (0.000) (0.000) -0.887*** -0.902*** -0.873*** -0.882*** -0.934*** -1.008*** -1.924 -0.789*** Lagged Gini index (0.133) (0.148) (0.155) (0.149) (0.143) (0.156) (2.299) (0.175) -0.135** -0.183** -0.117 -0.145** -0.127* -0.073 -0.116* 1.312** Lagged poverty rate (0.062) (0.075) (0.077) (0.071) (0.067) (0.071) (0.069) (0.600) 0.280*** 0.302*** 0.333*** 0.291*** 0.267*** 0.326*** 0.300*** 0.353*** Lagged share of wage income (0.098) (0.100) (0.097) (0.099) (0.095) (0.098) (0.096) (0.101) 0.217** 0.203** 0.141 0.151 0.161* 0.147 0.141 0.213** Lagged share of non-farm income (0.097) (0.091) (0.097) (0.094) (0.086) (0.096) (0.090) (0.092) 5.718*** 6.568*** 4.252*** 4.582*** 4.439*** 4.257*** 4.547*** 4.416*** Constant (1.534) (1.522) (0.585) (0.610) (0.567) (0.536) (0.772) (0.550) Year FE Yes Yes Yes Yes Yes Yes Yes Yes Province FE Yes Yes Yes Yes Yes Yes Yes Yes 567 567 567 567 567 567 567 567 Observations 6983 7747 8511 9686 7710 8503 7664 6568 R2/ Chi2 63 63 63 63 63 63 63 63 Number of provinces -6.628 -6.502 -6.668 -6.661 -6.734 -6.800 -6.654 -6.715 Arellano-Bond test for AR(1) 4.146 4.120 4.194 4.309 4.235 4.299 4.293 4.109 Arellano-Bond test for AR(2) 198.4 186.1 225.8 214 210.9 209.5 222.8 219.9 Sargan test Note: i) Authors’ estimation from VHLSSs using province GMM models, ii) Robust standard errors in parentheses clustered at the province level, iii)*** p<0.01, ** p<0.05, * p<0.1. 50 Table A.10: Relation between income growth, poverty, and inequality with interactions Dependent variable is the growth of per capita income (Log Yt – Log Yt-1) Explanatory variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 -0.399*** -0.179 -0.312*** -0.254*** -0.339*** -0.378*** -0.355*** -0.326*** Lagged log of per capita income (0.099) (0.273) (0.063) (0.052) (0.091) (0.068) (0.062) (0.057) 0.004 Lagged log of per capita income * Lagged log of other State spending (0.007) -0.009 Lagged log of per capita income * Lagged log of investment spending (0.016) -0.008 Lagged log of per capita income * Lagged log of population density (0.012) -0.003*** Lagged log of per capita income * Lagged share of urban population (0.001) -0.339 Lagged log of per capita income * Lagged share of population with high-school diploma (0.381) -0.000 Lagged log of per capita income * Lagged share of ethnic minority population (0.000) Lagged log of per capita income * Lagged Gini -0.102*** index (0.036) Lagged log of per capita income * Lagged poverty -0.038** rate (0.017) -0.043 -0.004* -0.005** -0.004* -0.005** -0.004 -0.003 -0.005* Lagged log of other State spending (0.067) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) 0.006 0.089 0.005 0.003 0.004 0.006 0.000 -0.000 Lagged log of investment spending (0.008) (0.145) (0.007) (0.007) (0.008) (0.008) (0.007) (0.007) 0.042*** 0.043*** 0.110 0.062*** 0.049*** 0.044*** 0.034*** 0.042*** Lagged log of population density 51 Dependent variable is the growth of per capita income (Log Yt – Log Yt-1) Explanatory variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 (0.010) (0.009) (0.112) (0.012) (0.011) (0.010) (0.012) (0.009) 0.002*** 0.002*** 0.002*** 0.034*** 0.002*** 0.002*** 0.003*** 0.002*** Lagged share of urban population (0.001) (0.001) (0.001) (0.009) (0.001) (0.001) (0.001) (0.001) Lagged share of population with high-school 0.111 0.100 0.105 0.067 3.171 0.108 0.071 0.152 diploma (0.192) (0.176) (0.188) (0.177) (3.373) (0.205) (0.166) (0.213) -0.001** -0.000 -0.001* -0.000 -0.001** 0.000 -0.000 -0.000 Lagged share of ethnic minority population (0.000) (0.000) (0.000) (0.000) (0.000) (0.004) (0.000) (0.000) -0.409** -0.390** -0.419** -0.574*** -0.460** -0.432** -0.192 -0.421** Lagged Gini index (0.188) (0.171) (0.201) (0.122) (0.183) (0.202) (0.208) (0.208) -0.074 -0.056 -0.047 0.091 -0.032 -0.117 -0.098 0.148 Lagged poverty rate (0.096) (0.121) (0.106) (0.083) (0.115) (0.111) (0.091) (0.149) 0.091 0.116 0.095 0.034 0.079 0.088 0.209** 0.037 Lagged share of wage income (0.108) (0.100) (0.105) (0.110) (0.106) (0.108) (0.099) (0.099) -0.223* -0.221 -0.258** -0.260** -0.202 -0.227* -0.224 -0.254** Lagged share of non-farm income (0.122) (0.139) (0.131) (0.131) (0.147) (0.137) (0.142) (0.129) 3.750*** 1.794 3.011*** 2.487*** 3.224*** 3.544*** 3.529*** 3.273*** Constant (0.904) (2.396) (0.536) (0.434) (0.784) (0.578) (0.551) (0.456) Year FE Yes Yes Yes Yes Yes Yes Yes Yes Province FE Yes Yes Yes Yes Yes Yes Yes Yes 567 567 567 567 567 567 567 567 Observations 3446 3460 3666 3055 3255 3335 3043 2985 R2/ Chi2 63 63 63 63 63 63 63 63 Number of provinces -4.913 -4.915 -4.998 -5.048 -5.176 -5.363 -4.898 -5.006 Arellano-Bond test for AR(1) 1.172 1.243 1.234 1.131 1.193 1.031 1.402 1.503 Arellano-Bond test for AR(2) 173.5 175.8 195.9 151.5 161.6 147.5 174.1 194.9 Sargan test Note: i) Authors’ estimation from VHLSSs using province GMM models, ii) Robust standard errors in parentheses clustered at the province level, iii)*** p<0.01, ** p<0.05, * p<0.1. 52 Figure A.1: Growth of per capita expenditure and initial expenditure level Panel A. Two-year growth of per capita expenditure Panel B. Four-year growth of per capita expenditure 1 1 Growth over four years Growth over two years .5 .5 0 0 -.5 -.5 7.5 8 8.5 9 9.5 10 7.5 8 8.5 9 9.5 10 Log of per capita expenditure Log of per capita expenditure Provinces Fitted values Provinces Fitted values Panel C. Ten-year growth of per capita expenditure Panel D. 18-year growth of per capita expenditure 1.8 1.2 1.6 1 Growth over ten years Growth over 18 years 1.4 .6 .8 1.2 .4 1 7.5 8 8.5 9 9.5 7.5 8 8.5 9 Log of per capita expenditure Log of per capita expenditure Provinces Fitted values diff_pcexp9 Fitted values Note: i) Authors’ estimation using VHLSSs data 53 Appendix B: Further data description The Vietnam Household Living Standard Surveys (VHLSS) are conducted biennially, covering approximately 45,000 households in each survey round. The VHLSSs are designed to be representative at the provincial level (63 provinces). Within the VHLSSs, a sub-sample of around 9,000 households is selected to collect expenditure data. However, this smaller sample is only representative at the regional level (6 regions). To obtain estimates that are representative at the provincial level, the full sample of 45,000 households should be utilized. Thus, we estimate per capita expenditure of households in the sample of 36,000 households using the “poverty mapping” imputation method (Elbers et al., 2003). Estimating per capita expenditure consists of two steps. First, we estimate an expenditure model using the small-sample VHLSs (9,000 households). The dependent variable is the per capita expenditure, and the explanatory variables consist of household characteristics including demographics, ethnicity, education of household heads and household members, durables, housing conditions, and region dummies. We estimate separate models for urban and rural areas. The variables are selected using stepwise regressions. Only variables that are statistically significant at the 1% level and demonstrate reasonable signs are used in the final models. Second, we apply this expenditure model to the sample of 36,000 households (using the same variables that were employed in the expenditure model based on the small-sample VHLSSs) and predict per capita expenditure for these households. As a result, we have per capita expenditure data for the full sample of 45,000 households, and we use this data to estimate the per capita expenditure of provinces. 54