Household Inequality and Remittances in Rural Thailand: A Lifecycle Perspective Richard Disney*, Andy McKay + & C. Rashaad Shabab + *Institute of Fiscal Studies, University of Sussex and University College, London. + University of Sussex 22/12/2017
What does this paper do? Documents declining income inequality over the lifecycles of a panel of Thai Households. Demonstrates that decline is not explained by the standard lifecycle factors (individual earnings, household composition). Presents evidence that differences in the receipt of remittances from adult children of the household head living outside the village account for the entirety of observed convergence. Studies the features of the distribution of remittances which enable remittances to reduce inequality among the households of origin.
Why is a Decline in Income Inequality Interesting? Income inequality in any fixed membership group, such as a balanced panel or a cohort does not usually decrease. If innovations to income have a permanent component that is imperfectly correlated across households, income inequality will diverge. U.S.A., U.K. and Taiwan (Deaton and Paxson, 1994), Australia (Chatterjee, Singh, and Stone, 2015), Germany (Bonke, Corneo and Luthen, 2015), Italy (Rosati, 2003 and Jappelli and Pistaferri, 2010) and Japan (Yamada, 2009). Potentially different inequality dynamics in developing countries: Agricultural shocks are mainly transitory in nature (Deaton 1989, 1991). Agricultural shocks exhibit a strong covariate component (Rosenzweig and Wolpin, 1989; Udry 1994; Morduch, 1994; Townsend, 1994 and Dercon 2006, among others). Multiple generations cohabit, insuring household income against lifecycle-related productivity dips (Deaton, 1989; Ehrlich and Lui, 1991; Banerjee, Meng and Qian, 2010; Oliveira, 2016)
Literature: Inequality and Remittances Cohabitation is not the only option available to insure against lifecycle productivity dips; children can leave the village of origin and remit back. Inequality and Remittances: Lipton (1980): Remittance flows disproportionately benefitted households that were better off to begin with and so exacerbated rural inequality (multiple countries). Stark, Taylor and Yitzhaki (1986) found Gini coefficients were lower with remittances, suggesting diffusion of information (Mexican villages). Adams (1989): Predict household income if migrants had stayed back. Remittances increase inequality compared to this counterfactual (Egyptian villages). McKenzie and Rapoport (2007) allow for multiplier effects and general equilibrium effects, and find that migration reduces inequality, if there is enough past migration (Mexican villages).
The Data: The Townsend Thai Project I use data from the Townsend Thai Project (Townsend, 2011) 15 years (1997-2011) of panel data on income, and remittances from children living outside the village, 64 villages. Balanced panel of 609 households. Unbalanced panel of 14,163 observations in 15 years. Especially suitable to study intra-family, intergenerational transfers because of dedicated section on children living outside the village. Reasonably high quality income data for a developing country context (validation, inclusion of gifts and home production).
The Townsend Thai Project on Inequality or Migration Pawasutipaisit and Townsend (2011) document declining wealth inequality in the monthly series of the Project. This decline is driven by differential savings rates, and differential returns on assets rather than remittances. However, this finding is not robust to the annual data (p. 57) Yang (2004) studies differences between inequalities in productivity and income at the provincial level. The focus is not on the dynamics of income inequality, or on cohorts of households as it is here. Paulson (2000) shows that migration plays an important insurance function in rural Thailand, but is not primarily interested in the effect on inequality.
Declining Inequality in the Balanced Panel Suggests that a robust prediction of the lifecycle model does not hold, but: Younger household heads may be replacing older ones. Younger cohorts may be less unequal than older ones (Hall, 1978; Dickens, 2000, etc.) Important to break this down into year of birth cohorts.
Inequality is declining within cohorts: Income inequality declines over the lifecycle for every cohort for which we have reasonable cohort-year cell sizes. There does not appear to be much evidence that initial inequality varies systematically between cohorts as it does in other studies (e.g. Blundell, Pistaferri and Preston, 2008). There may be heterogeneity in the rate at which inequality is declining. We test these observations using: σ "# = α + β " ) t + + γ " ) t + u "#
Declining inequality, limited heterogeneity We cannot statistically distinguish between the initial levels of inequality between the three younger cohorts. The oldest cohort is significantly more unequal than the others at the beginning of the panel. For every cohort we reject the hypothesis that g = 0 in favour of the alternative that g < 0. No statistical evidence that the rates of decline differ systematically between cohorts (I also test this formally and fail to reject the null that they are the same).
Robustness: Different Measures of Inequality For every inequality measure I reject the hypothesis that g = 0 in favour of the alternative that g < 0, for every cohort. The result is thus robust to a wide range of commonly used measures of inequality. Therefore, it is not driven by an implicit choice over different social welfare functions.
Is The Result Driven by Differences Between Villages? We regress household income on a fully interacted set of village and time fixed effects. This absorbs all between village dynamics and allows us to focus on variation within villages, between households. We then repeat the analysis on the residuals from this regression.
Possible drivers of convergence Convergence in the distribution of individual earnings over the lifecycle. Changes in the composition of households over the lifecycle. Differences in the receipt of transfers from outside the household over the lifecycle.
Are Individual Earnings Converging? Falling household income inequality may be due to convergence in the earnings of individuals that comprise the household. Ideally, I would disaggregate household income into the contributions of individual members, but this cannot be done unambiguously (potential unobserved heterogeneity). So restrict attention to the 26% of working household members who are in wage labor. Daily wages are subject to labor supply decisions and availability of hours, whereas monthly wages are not. So I only present the results for monthly wages here (daily wage analysis available in the paper)
Monthly Wage Inequality Over the Lifecycle Some evidence that older cohorts are more unequal. (standard finding, e.g. Dickens, 2000 and Blundell et al. 2008) No evidence of convergence.
Might Cohabitation Reduce Inequality? Cohabitation of adult children in their parent s household has important implications for the applicability of the lifecycle hypothesis to developing countries (Deaton, 1989; Ehrlich and Liu, 1991; Banerjee et al., 2010; among others). Poorer households may exhibit higher fertility, so that more children contribute to household income later in the lifecycle. A given child of a poorer household may be more likely to stay on in their parents household into adulthood. Either of these factors would imply that later on in the lifecycles of the heads, poorer households would have more income earners, explaining convergence in the distribution of household income.
Higher rates of cohabitation of adult children in richer households On average, the number of children in poorer households starts to decline slightly later in the lifecycle of the heads. Early on, a household in the bottom income quartile has one-third more resident children than a household in the top quartile. But the children of poorer households continue to leave until this trend is reversed when the heads reach their mid-50s. For the remainder of the lifecycle, richer households have on average one child of the head resident with them. At this stage, poorer households have roughly one third fewer resident children.
Increasing Importance of Remittances Over the Lifecycle Remittances increase in importance from when the heads of household reach their mid-forties until they reach their late fifties. After the heads of household are in their fifties, remittances account for between one quarter and one third of household income. This effect may vary across the income distribution.
Importance of Remittances by Decile of Permanent Income Remittances are a greater proportion of the incomes of poorer households than richer ones. Cohort-year cell sizes are too small to plot this over the lifecycle for every decile. So, I split the sample into relatively rich and relatively poor households.
Comparing the Importance of Remittances Between the Rich and the Poor All indications are that these differences would be more pronounced at the extremes of the income distribution, if we had the data to observe them.
Inequality Dynamics of Income Not Remitted by Children Inequality in the component of household income that is not remitted by non-resident children increases in the standard way. So inequality in this component of household income is indeed increasing over the lifecycle.
Can we Argue Causation? Adams (1989) if household members had not migrated and remitted, they would have been in some other form of employment. Here we use matching techniques to identify the counterfactual dynamics of inequality that would have prevailed if children stayed in the household. Matching is usually used to identify counterfactual levels, not inequality. Errors in matching will overstate counterfactual inequality; matches to extreme values will exhibit mean reversion understating counterfactual inequality. We assume these errors are uncorrelated with the age of the head of household.
Matching households Ideally match remittance receiving to otherwise similar non-receiving households. Only 83/609 households never receive remittances. So we match low (<10% of household income) remittance receiving households to high (>10%) remittance receiving households. We use a Probit to model the probability of households receiving high remittances over the duration of the panel conditioning on household characteristics in 1997. Characteristics are sex, age, education, total number of adult children of the head of household; and the average number of years of education of the adult children in the household Use observed incomes for low remittance households and matched incomes for high remittance households.
Counterfactual Inequality Dynamics The declines in counterfactual inequality are significantly less sharp than for observed inequality for the three earlier cohorts. The resulting t-statistics for the cohorts born in the 1930s, 1940s, 1950s and 1960s are 15.15, 19.33, 15.11 and 1.68 (1% critical value 2.39) Thus we can conclude that remittances have caused an acceleration in the decline in inequality. The cohort born in the 1960s is relatively young and therefore unlikely to have working children, explaining the exception.
Conclusions: Inequality within villages, between household is declining over time among these Thai households. Remittances from children constitute a larger share of the incomes of poorer households. They become important only later in the lifecycle of the recipients. Together, these forces account for the ability of remittances to reduce income inequality over time, and over the lifecycle. Counterfactual income distribution suggests that remittances have caused an acceleration in the decline in inequality for three out of four cohorts. Can we dig deeper? Is it more children remitting, or each child remitting more? Does gender play a role? Read our paper!
Thanks!
The Quantity of Remitters by Income Level Households in the top two deciles receive remittances from significantly fewer children. Households in the bottom two deciles of permanent income receive remittances from 3.5 children, on average. Those in the top two deciles receive remittances from less than 2.5 children on average.
The Quality of Remitters by Income Level The remittances of each child constitute a larger proportion of household income for poorer households than for richer ones. This representation reduces statistical power because the data have been collapsed down into deciles to ease visual interpretation. In a regression with remittance per child as a proportion of household income as the dependent variable, and a household s percentile in the distribution of permanent income is significant. A 10 percentage point movement up the distribution of permanent income is associated with a 0.537% (t = 4.86) reduction in the proportion of household income that is accounted for by remittance per child.
Non-Resident Female Children and Income Poorer households have more daughters who live outside the village. However, women may be less likely to migrate for economic reasons than men.
Number of Female Remitters and Income There is no significant difference in the number of female remitters across the distribution of income.
Daily Wage Inequality Over the Lifecycle Evidence for daily wages is mixed. Inequality does appear to be decreasing for the middle three cohorts, though this is accompanied by a great deal of noise. Certainly not the clear declines documented in Figure 2.
Table 1: Summary Statistics Variable Observations Mean Standard deviation Minimum value Maximum value Net household 14,163 157,570.5 274,221 419.66 12,050,222 income Individual 2,929 10,561.66 7,805.43 136.43 85,744.91 monthly wages Individual 6,695 180.63 59.82 11.99 1231.19 daily wages Year of birth of 14,244 1949.17 13.304 1903 1989 household head Number of 14,263 1.39 1.14 0 10 resident Children Remittances 9,570 21,374.46 43,037.53 0 1,096,907 from children Number of 13,907 2.35 2.29 0 13 children living outside village Number of children who remit 14,574 1.131124 1.561793 0 12
Table 4: Cohort Year Cell Sizes for Household Income Decade of birth 1900s 1910s 1920s 1930s 1940s 1950s 1960s 1970s 1980s 1997 2 33 87 210 194 258 150 12 0 1998 2 27 88 208 195 253 148 17 0 1999 0 23 83 200 195 265 145 18 0 2000 0 20 84 196 203 273 148 19 0 2001 0 19 84 192 200 270 154 14 1 2002 0 17 86 181 202 269 158 21 2 2003 0 17 82 181 206 274 160 21 2 2004 0 16 74 183 210 270 161 26 2 2005 1 10 57 188 205 270 172 35 4 2006 2 9 51 181 208 276 174 37 3 2007 0 6 41 160 208 288 182 55 7 2008 0 5 29 138 209 288 208 63 8 2009 0 5 20 135 210 281 216 74 10 2010 0 4 18 120 207 286 223 83 11 2011 0 2 17 106 203 285 232 95 12
Table 6: F-tests for differences between cohorts of time trends in the evolution of household income inequality 1930s 1940s 1950s 1940s 1.11 (0.2967).. 1950s 0.23 (0.6370) 1960s 2.19 (0.1453) 0.33 (0.5668) 0.44 (0.5096). 1.18 (0.2833) F-statistics distributed with (1, 51) degrees of freedom; p-values in parentheses.
Appendix 3: Cohort Year Cell Sizes for Monthly Wage Earners Cohort age in 1997: 10s 20s 30s 40s 50s 60s 70s 80s 90s 1997 0 0 0 0 0 0 0 0 0 1998 0 1 1 9 14 48 57 74 16 1999 0 1 2 6 13 46 53 84 21 2000 0 0 2 4 15 45 62 75 18 2001 0 0 3 5 10 45 59 86 23 2002 0 0 2 5 12 41 58 81 39 2003 0 0 0 3 9 37 61 65 31 2004 0 0 1 5 8 33 62 71 46 2005 0 0 1 6 7 34 61 78 37 2006 0 0 1 5 6 32 45 69 46 2007 0 0 1 4 6 29 33 49 37 2008 0 0 0 4 8 26 38 54 55 2009 0 0 0 3 8 23 38 51 65 2010 0 0 0 3 5 25 40 61 70 2011 0 0 0 3 2 23 36 52 81
Appendix 4: Cohort Year Cell Sizes for Daily Wage Earners Cohort age in 1997: 93 83 73 63 53 43 33 23 13 1997 0 0 0 0 0 0 0 0 0 1998 0 1 6 29 49 72 106 105 37 1999 0 0 5 29 61 86 105 130 45 2000 0 0 7 23 62 79 101 129 65 2001 0 0 4 17 62 64 101 134 83 2002 0 0 4 22 55 53 93 115 92 2003 0 0 4 17 56 57 112 111 109 2004 0 0 4 17 52 66 101 123 126 2005 0 0 6 19 56 84 107 101 126 2006 0 0 1 15 50 72 100 99 121 2007 0 0 1 16 50 89 129 129 141 2008 0 0 1 23 61 101 138 136 140 2009 0 0 1 7 41 81 120 118 120 2010 0 0 1 6 35 74 103 94 133 2011 0 0 1 5 41 76 101 93 142
Important Factors that We Do Not Observe The reason for migrating, i.e. marriage, to set up an independent household elsewhere, to remit resources to the household of origin. The earnings of the migrant at the destination.