SEASONAL MIGRATION AND IMPROVING LIVING STANDARDS IN VIETNAM

Transcription

1 SEASONAL MIGRATION AND IMPROVING LIVING STANDARDS IN VIETNAM ALAN DE BRAUW AND TOMOKO HARIGAYA We use panel data methods to explore whether households in Vietnam used seasonal migration to increase their living standards during the 1990s. Using per capita expenditures as our primary measure of living standards and historical and latent network variables as instruments for migration, we can attribute 5.2 percentage points of annualized expenditure growth to increased migration. The results are robust to several alternative measures of living standards. As the estimates suggest migration accounts for a 3 percentage point decrease in the poverty headcount, we conclude migration played an important role in the improvement of living standards observed in Vietnam. Key words: instrumental variables, measurement error, migration, panel data, Vietnam. DOI: /j x The migration of labor out of agriculture is a primary feature of the economic development process. Both historically and in the present, the share of labor working in agriculture within a country declines as per capita GDP increases (Taylor and Martin 2001). In fact, if predominantly agricultural economies are to take full advantage of the geographically concentrated increasing returns to scale in industrial production (Krugman 1991), farmers must migrate to provide the industrial sector with labor. Whereas migration is an important facet of economic development, the effects of migration on the rural areas migrants leave are complex. Migrants typically continue to have economic interactions with the source households and communities they leave behind (Stark and Bloom 1985), and these interactions are particularly important when markets do not function well. A growing literature on the effects of migration on source communities has documented both the use of migration as part of a household development strategy and the importance of source community networks for finding jobs in urban Alan de Brauw is research fellow, International Food Policy Research Institute, Washington DC, and assistant professor, Department of Economics, Williams College, Williamstown, MA. Tomoko Harigaya is research assistant, Innovation for Poverty Action, New Haven, CT. Senior authorship is shared. We are grateful to Steve Boucher, Zhuo Chen, John Gibson, David McKenzie, Steve Sheppard, Lara Shore-Sheppard, three anonymous referees, and seminar participants at the American Agricultural Economics Association meetings, Denver, Colorado, the International Food Policy Research Institute, and Williams College for helpful comments. de Brauw acknowledges the support of the Henry George Fund at Williams College. areas (Stark 1991; Carrington, Detragiache, and Vishnawath 1996). However, the effects of migration and migration networks on households are difficult to identify econometrically, because unobservable factors that affect decisions about migration almost certainly also affect other decisions made by households. Two sources of exogenous variation that have been used to identify migration are weather shocks and historical events which change the cost of migrating. Munshi (2003) used weather shocks to villages in Mexico to identify the size of migrant networks among Mexican laborers in the United States. McKenzie and Rapoport (2004), Hildebrandt and McKenzie (2005), and Woodruff and Zenteno (2007) use varying time to completion of rail lines in the early 1900s, which facilitated the creation of networks to the United States, to identify the effects of migrant networks on inequality, health status among children, and microenterprise development, respectively. Another source of exogenous variation that could be used to identify migration are institutional details unique to countries that had policies either relocating people or severely restricting migration in the past. In this article, we use an approach unique to Vietnam to help identify seasonal migration and its effects on household well-being. Rural Vietnam is organized into communes, yet 23% of people found living in rural communes in were not born there (Lucas 2000). After the Vietnam War ended, there was a brief period of relocation, followed by a period of Amer. J. Agr. Econ. 89(2) (May 2007): Copyright 2007 American Agricultural Economics Association

2 de Brauw and Harigaya Seasonal Migration in Vietnam 431 severely restricted movement. People born in either Hanoi or Ho Chi Minh City (HCMC) prior to 1975 were likely to have contacts in those cities that became useful again after restrictions on movement were lifted. As a result, communes with more members born in either city may have advantages in migration over members of other communes. Vietnam is a particularly salient place to study seasonal migration for several reasons. First, its economy has been growing rapidly and rural urban inequality has increased (Benjamin and Brandt 2004). As the rural urban income gap grows, migration from rural to urban areas should increase (Harris and Todaro 1970). But Vietnam also lacks wellfunctioning markets, so households are likely to hold onto their land, which takes on additional value due to tenuous land rights. In the similar setting of China, researchers have not observed many whole families migrating, as they continue to work the land partially due to fear of expropriation (e.g., Jacoby, Li, and Rozelle 2002). In this article, we document the effects of seasonal migration on household well-being, which we primarily measure using household per capita expenditures. Though Glewwe, Gragnolati, and Zaman (2002) described determinants of household expenditures in the Vietnam Living Standards Survey (VLSS), their analysis did not include variables that could be considered endogenous, such as migration or other variables. Our article extends their analysis to explore the effects of seasonal migration on household expenditures in rural Vietnam. Our study has three primary objectives. First, we document the increase of seasonal migration in Vietnam during the 1990s. Second, we analyze the effects of seasonal migration on household consumption growth using instrumental variables and panel data techniques. We test our primary result for robustness by redefining household expenditures in two different ways, and we also test whether migration affects a nonmonetary measure of well-being, children s height-for-age z-scores. Third, we use our regression results in a counterfactual experiment, similar to the one conducted by Barham and Boucher (1998), to analyze how participation in seasonal migration has affected poverty and inequality measures. We explain the results of our experiment by discussing which part of the expenditure distribution migrants were more likely to come from. The article proceeds as follows. After a short description of Vietnam s economy during the 1990s, we describe the data set and some patterns we observe in household expenditures and migration over time. Next, we provide a brief theoretical model, followed by the empirical model and justification for our instrumentation strategy. The following section presents our primary results, robustness checks, and implications for poverty and inequality. The final section concludes. Economic Reform in Vietnam in the 1990s Although Vietnam s Doi Moi reforms began in 1986, in several ways Vietnam s transition to a market economy accelerated during the 1990s. The collapse of the Soviet Union in 1989 may have been a catalyst for additional reforms, as Vietnam had been fiscally dependent on Soviet aid. After Soviet aid abruptly ended, Vietnam s government opened its economy to foreigners and made several reforms that affected rural areas. One such reform was decollectivization, which allowed individual households to farm their own plots and make decisions about inputs and crops. Rural reforms led to significant increases in household incomes. As markets became increasingly open during the early 1990s, agricultural growth accelerated, and Vietnam became the second largest rice exporter in the world. Output growth in rice and other agricultural products can be attributed to the liberalization of fertilizer markets, which reduced input costs; the liberalization of output markets, which increased prices and exports; and the expansion of the individual household farming system (Benjamin and Brandt 2004). Through fertilizer market liberalization, the price of fertilizer fell throughout Vietnam. As internal trade barriers were removed, the producer price of rice increased in the South relative to the North, encouraging higher production in the South. As rice farmers in the South were also more efficient than farmers in the North, farmers in the North switched out of rice production in order to maintain their incomes. The switch was possible because markets for other products were growing rapidly and decollectivization gave farmers more control over crop choices. Therefore, market liberalization in agriculture benefited farmers in both regions (Benjamin and Brandt 2004). Resolution 5, a new Land Law enacted in 1993, may have further improved rural

3 432 May 2007 Amer. J. Agr. Econ. economic performance. Land tenure was extended to twenty years for annuals and fifty years for perennials, and farmers were given the rights to transfer, lease, inherit, and mortgage land. Although the government continued to own the land and reforms were implemented slowly, land security increased and farmers began to invest in land productivity (Do and Iyer 2005). Improved land rights also enabled enable transfers from inefficient to efficient users and encouraged inefficient farmers to work off-farm. Although Ravallion and van de Walle (2003) estimate that only one-third of the initial inefficiency was eliminated through land use rights transfers between and , Deininger and Jin (2003) suggest that smaller landowners gained greater access to land, as relatively rich people rented or sold land in order to move off-farm and increase their earnings. Therefore, the combination of recently improved land rights and more robust off-farm labor markets contributed to improved rural household welfare. Data To understand whether seasonal migration has helped households in rural Vietnam increase their well-being, we use data obtained from the VLSS, conducted in and in by the World Bank in collaboration with the Vietnam State Planning Committee and the General Statistical Office. The VLSS is a comprehensive nationwide survey consisting of two main parts: a household survey and a commune-level survey. The household survey collected information on multiple aspects of living conditions, including individual-level education, off-farm employment, on-farm labor, and migration, as well as household demographics, housing conditions, assets and expenditures, and income sources. Anthropometric information was also collected from almost all household residents. To investigate migration behavior by households, we used a specific module asked in both surveys about seasonal migration. This module also asked whether people were born outside of the commune they were living in. 1 The commune-level survey provided further information on local living conditions. For this article, we used several parts of the commune-level survey, including information on the proportion of the commune workforce that was migrating in , the distance of the commune from market centers and roads, and measures of general commune economic characteristics. The 1993 and 1998 surveys have quite different sample sizes and geographic compositions. 2 The sample of 4,799 households in the 1993 survey was chosen to be nationally representative and self-weighting, but the 6,000 households in the 1998 survey include over 1,500 households that were added from another survey to increase the sample size (World Bank 2001). For this article, we construct a panel of the 3,492 rural households included in both surveys. Since the 1993 sample is selfweighting, our sample can be considered selfweighting if sample attrition can be considered random. Of the rural households surveyed in 1993, 344 were not resurveyed in 1998, and three other households were dropped because of incomplete records. The attrition rate of 9% is quite low for a panel survey in a developing country (e.g., Thomas, Frankenburg, and Smith 2001). However, since attrition was not randomly distributed across communes or regions, we test whether attrition affects our econometric results. We use per capita expenditures as our primary measure of household well-being. Total household expenditures are calculated by summing consumption expenditures on food, the value of home-produced food, nondurable and nonfood goods, the estimated rental value of durable goods, the estimated rental value of the dwelling, and the value of in-kind transfers from off-farm employers. We then divide by the total number of household members, including seasonal migrants, to come up with per capita household expenditures. To ensure that our calculations are comparable across regions and over time, the values of food and nonfood expenditures were separately adjusted for both regional and timing differences in the survey. Both surveys collected commune level prices, which were aggregated to regional indices based on average budget shares for consumption goods. Expenditures were also adjusted for the differential timing of the survey, using a monthly consumer price index. To make prices comparable in the two surveys, adjusted prices for 1 In our sample, questions regarding individual labor supply, including migration, were answered by the individual themselves 75.3% of the time in When the individual themselves could not be interviewed, the household head provided responses. 2 Both rounds of the VLSS took place over the course of a calendar year. To avoid labeling confusion, we use the year 1993 to refer to the survey and 1998 to refer to the survey.

4 de Brauw and Harigaya Seasonal Migration in Vietnam 433 the 1993 survey were inflated to January 1998 levels. Therefore, all monetary values in this article are expressed in thousands of January 1998 Vietnamese dong. 3 Measuring the Increase in Household Well-Being According to the VLSS, economic growth in Vietnam translated to increased living standards for most rural households. Whereas the median per capita expenditure level was 1,506 thousand dong in 1993, in real terms median household per capita expenditures had increased to 2,015 thousand dong by 1998, or 33%. Although many transition countries have experienced increased inequality along with economic growth, descriptive statistics on relative inequality in rural Vietnam show little change. The Gini coefficient for per capita household expenditures was in 1993, and only increased to 0.28 in Vietnam s rural households, then, seem to have become better off without experiencing the expected increase in inequality for Vietnam s level of development (Kuznets 1955). Households did not necessarily maintain the same position in the income distribution between 1993 and 1998, as some households benefitted from economic growth more than others (Glewwe and Nguyen 2004). To assess differences in household performance over the sample period, we calculate an annualized expenditure growth rate for each household, as the difference in the logarithm of per capita expenditures over the time between the two surveys. The average per capita expenditure growth for sample households is 5.7% per annum. The kernel density of expenditure growth rates shows there is a great deal of variation among household expenditure growth rates (figure 1). Although the sample standard deviation is 0.084, over half the variation in per capita expenditures can be attributed to measurement error (Glewwe and Nguyen 2004). As a result, only a portion of the variation found in the data reflects true differences in expenditure growth. Measurement error in expenditures may have two different effects on our results. First, since expenditure growth is negatively correlated with initial expenditures, the measurement errors are likely to be both serially and negatively correlated, or 3 In 1998, the exchange rate was approximately 13,900 Vietnamese dong to US$1. Density Average Annual Growth in Per Capita Expenditures Figure 1. Average annual growth in household per capita expenditures, VLSS, mean reverting. Mean reverting measurement error that is not otherwise related to other variables affecting seasonal migration and expenditure growth would cause an underestimate of the effect of migration on expenditure growth (Bound and Krueger 1991; Kim and Solon 2005). However, the portion of measurement error attributable to food expenditures is likely correlated with household size (Gibson 2002). Since migrants tend to come from larger households, we must also ensure that our results are robust to measurement error in food expenditures. Measurement error is not the only reason that per capita expenditures are not an ideal measure of household well-being. Per capita expenditures attribute the same amount of consumption to each household member, ignoring that children consume less than adults, and that economies of scale exist in household consumption (e.g., Deaton and Paxson 1998). Furthermore, measures of monetary and nonmonetary poverty do not fully overlap in the VLSS (Baulch and Masset 2003). Therefore, we test whether our results are robust to three alternative measures of household well-being. First, we replace per capita expenditures with nonfood expenditures. Although eliminating food expenditures from the measure removes the potential correlation between measurement error and household size, it could mismeasure the effect of migration on household well-being, as migrant households may choose to consume more nonfood goods than nonmigrant households for other, unobservable reasons. Second, we check whether our results are robust to measuring well-being with expenditures per adult equivalent, counting each child as half an adult and using a scale parameter of 0.9, as recommended by

5 434 May 2007 Amer. J. Agr. Econ. Deaton and Zaidi (2002) for developing countries. Third, we test whether our results are robust to using height-for-age z-scores among children aged one through ten in 1993 as a proxy for well-being, since they are a good indicator of children s well-being over long time periods, and measure the stock of investments made in children s nutrition (WHO 1986). Though we must drop a large portion of the sample to use z-scores, they provide an additional measure of household welfare improvement that does not suffer from potential measurement error bias as do the expenditurebased measures. Seasonal Migration in Vietnam The VLSS show that from a very small base, seasonal migration increased nearly sixfold between 1993 and We define seasonal migrants as household members who left the household for part of the year to work, but are still considered household members. 4 Typically, these migrants indicated that they were away between busy seasons on the farm. The destination for a sizable proportion of seasonal migrants is either Hanoi or HCMC; over one third of the migrants in 1998 migrate to one of the two big cities. We will use this fact in our identification strategy. The aggregate number of households in the panel sending out seasonal migrants increases from 65 in 1993 to 369 in 1998 (table 1). Migrant households are generally from communes in specific geographic areas. In 1998, over 20% of households in coastal areas and hills/midlands had at least one seasonal migrant (rows 4 and 6). In contrast, few migrants left high mountainous areas; only 2.4% of rural households had a migrant in The lack of mobility in high mountainous areas is likely due to underdeveloped transportation networks and limited off-farm employment opportunities. Seasonal migrants in Vietnam share characteristics with migrants from other countries (table 2, panel A). Migrants are typically 4 We choose to analyze seasonal rather than long-term migration for pragmatic reasons; the survey explicitly asked household members about migration over the past twelve months, where household members were defined generally as individuals who normally live and eat their meals in this household (World Bank 2001). To include more permanent migrants in our definition, we would have had to infer additional information about individuals who were not considered household members by the survey protocol. Since we cannot discern whether migrants who were away from the household at the time of the survey will return, we may include some permanent migrants as seasonal migrants in our definition. young, relatively well-educated men when compared with the rest of the rural population (rows 1 through 3). The average migrant in the sample has 6.8 years of schooling, while nonmigrants have an average of 5.9 years of schooling. However, the difference in schooling levels can wholly be attributed to the difference in the average age of the two groups. Migrants in 1998 are also twice as likely as others to have some vocational training. In general, migrants tend to be younger members of households with a relatively large endowment of human capital. When we label households as either migrant households, defined as households that have increased participation in migration between 1993 and 1998, or nonmigrant households, we also find differences in descriptive statistics (table 2, panel B). In general, migrant households have lower per capita expenditure levels than the sample mean (1,740 thousand dong in 1993). Thus, the typical migrant household can be characterized as a relatively poor household residing in a lower lying areas, which may have more developed social networks through which to migrate. However, expenditures among migrant households grew faster than among nonmigrant households (6.3% versus 5.7%). Although this difference is small, it might be important for poor households; more migrant households were below the World Bank s poverty line for Vietnam in Furthermore, these figures do not account for other observable differences between migrant and non migrant households. We control for such differences in our econometric section. Theoretical Model To illustrate how seasonal migration can affect household expenditure growth, consider a household composed of N laborers that produces a farm good using technology f (L; K), where f ( ) is a strictly increasing function, L is the labor input on the farm, and K is the household capital stock, including land. For now, we ignore the household s demographic composition as well as its human capital endowment, and assume that the farm good is the only product of the household. We further assume that f ( ) is concave in labor and in the short run the capital stock does not change. If the household consumes all of its income and is credit constrained, then its base consumption is f (N; K). For illustrative purposes, we assume that

6 de Brauw and Harigaya Seasonal Migration in Vietnam 435 Table 1. Selected Characteristics of Migrant Households, VLSS, and Number of migrant households Median per capita expenditures, ,264 1,437 Median expenditure growth rate Commune geography (proportion of households with migrants) Coastal 5.3% 21.3% Inland delta 2.1% 11.2% Hills/Midlands 0.5% 24.2% Low mountains 1.3% 5.6% High mountains 0.2% 2.4% Note: All descriptive statistics are conditional on migration occurring. Table 2. Characteristics of Individuals and Households in Rural Vietnam, by Migration Status, VLSS Panel A: Demographic Characteristics of Individuals in Rural Vietnam, by Migration Status Migrants Nonmigrants Proportion male (0.456) (0.498) Age (11.9) (18.0) Years of education (3.21) (3.65) Skill training? (1 = yes) (0.304) (0.218) Married? (1 = yes) (0.500) (0.485) Number of observations ,360 Panel B: Differences between Migrant and Nonmigrant Households, Vietnam Migrant HHs Other HHs Mean per capita expenditures, ,693 1,736 (1,155) (976) Proportion below poverty line, (0.48) (0.50) Mean expenditure growth rate 6.3% 5.7% (9.0%) (8.4%) Age of household head, (13.1) (14.9) Years of education, household head (3.76) (3.91) Household size (1.94) (2.14) Number of observations 353 3,139 Note: Standard deviations are in parentheses. Households characterized as migrant households are defined as those that increased their participation in migration between 1993 and f (L) = ln(l), where the effect of capital is absorbed in the constant. As migrant labor markets begin to develop, the household can dedicate a share of its labor endowment m to migration. When households decide whether or not to send out migrants, they consider wages w in distant markets, migration costs c, and the information Z they

7 436 May 2007 Amer. J. Agr. Econ. Food production f(l) (w 0,c 0,Z 0 ) (w 1,c 1,Z 1 ) 0 N N(1-m * ) Figure 2. Illustration of relationship between migration and household production possessed that shapes expectations about the expected net returns to migration, including knowledge about jobs, the probability of employment, and the ease of transition to the urban environment. Informational factors only affect household consumption through their influence on migration. When deciding whether or not to participate in migration, the household considers the net return to migration (w, c, Z), where ( ) is increasing in w and Z and decreasing in c. The model implies straightforward expressions for both participating in migration and for consumption. Since households maximize consumption, they choose a positive value of m if ln(n(1 m)) + Nm (w, c, Z) > ln(n) for some m > 0. If so, total household consumption will be C = ln(n(1 m)) + Nm (w, c, Z). In this case, it is straightforward to show that the household will choose a migration N (w,c,z) level m = 1. If farm productivity is high or the household small, the household will send out a smaller share of migrants; if net returns to migration are high, then the household will send out a larger share of migrants. On the other hand, if the marginal product of labor on the farm exceeds the net return to migration when all N laborers work on the farm, the household will not participate in migration and consumption will be ln(n). Abstracting from the functional form assumption for farm production and assuming that net returns to migration are linear, the two possible equilibria can be illustrated (figure 2). In the extreme case, assume that the expected net returns to migration are zero (e.g., (w, c, Z) = 0). Then the household will not send out migrants (m = 0), as the marginal product of labor in farming always be higher than in migration. If the expected return to migration is positive and higher than the marginal product of labor when the entire household works on the farm, it will send out Nm migrants, and the equilibrium farm production is f (N(1 m )). The latter equilibrium is the point of tangency between the line with slope (w, c, Z) = K and the farm production function. Since our interest is in understanding the relationship between migration and consumption growth, we consider a household that did not perceive net returns to migration in the first period, but did in the second period. For this case, first period consumption is C 1 = ln(n), and second period consumption is C 2 = ln(n(1 m )) + Nm (w, c, Z). The change in consumption between periods can be written: (1) C = ln(1 m ) + Nm (w, c, Z). The first term in equation (1) represents the loss of farm production due to migration, whereas the second term represents the increase in consumption due to migration, which could come either as remittances or money that migrants bring home. Our theoretical model suggests that household consumption may depend upon household participation in migration, the number of household members, wage rates for migrants, the cost of migration, and the household s capital stock. Therefore, in estimating the relationship between migration and consumption growth, we must account for as many of these factors as we can observe. As migration is a choice variable, it is influenced by net returns to migration as well as the informational factors that affect participation in migration. The model implies that we can exclude variables that proxy for informational factors from the consumption equation, as they only affect consumption through migration. Empirical Model and Estimation Strategy In order to explain the effect of migration on household expenditure growth in the spirit of equation (1), we first abstract somewhat from the model. Workers in the household with different human capital attributes may have different marginal products on the farm or in migration. Therefore, instead of including household size in our empirical model, we control for the demographic composition of the household, X. We then specify the relationship

8 de Brauw and Harigaya Seasonal Migration in Vietnam 437 between per capita consumption, migration, and other variables affecting consumption and decisions about migration as log-linear: (2) ( ) C ln N hvt = h + 1 M hvt + 2 X hvt + 1 w hvt + 2 c hvt + 3 K hvt + ε hvt where h, v, and t index households, communes, and time, respectively, M represents the number of migrants sent out by the household. Since we have data for two periods, we include a household dummy variable which accounts for any household or supra-household variables ( h ) that do not vary over time. We include an error term, ε hvt, which is assumed to be correlated within communes but independent between communes. The error term represents both the random component of per capita consumption as well as any unobservable factors that might affect per capita consumption and vary over time. Since M is endogenous variable, we must account for its endogeneity in estimation. Before estimating equation (2), we consolidate some of its terms because unobservables regarding household decision making may be correlated with some of the variables. Any unobservable factors about the household that will affect its expenditures may also be correlated with its propensity to send out migrants. So, if we were to estimate equation (2), even if we used instrumental variables techniques to limit bias in the estimated coefficient of interest, 1, we might introduce bias by including other endogenous variables. Such variables, some of which are unobservable in the VLSS, include measures of migrant wages, migration costs, and the physical capital of the household. To avoid this concern, rather than attempting to measure the coefficients in equation (2), we drop those variables from the model. We, therefore, assume that these variables can be absorbed into either the household dummy variable or a regional growth rate. To remove the household dummy variable, we difference the two time periods and estimate the effect of the difference in migration behavior on the annualized expenditure growth rate. We write our initial estimator as: (3) r hvp = p + 1 M hvp + 2 X hvp + ε hvp where r represents the annualized per capita expenditure growth rate, p indexes regions, and ΔX represents the household demographic profile, which includes the number of elderly men and women, the number of working age men and women, and the number of school age children. 5 To ensure we account for regional differences in economic conditions, we include regional dummies p in all specifications. To understand whether asset holdings or other observable coefficients affect our estimate of 1, we also control for vectors of variables measured in 1993 at the commune (C) and household level (K). The commune level variables include general economic conditions as well as the distance to Hanoi and HCMC, and the household level variables include measures of both human and physical capital. The most general equation we estimate is: (4) r hvp = p + 1 M hvp + 2 X hvp + C vp + K hvp + ε hvp. Identifying Migration Because our theoretical model suggests that unobservables that affect migration will also affect household expenditures, we attempt to identify migration using measures of informational factors that affect migration but would not directly affect household expenditures. The literature suggests that migration networks increase the amount of information that certain households have about opportunities away from the commune, lowering the implicit costs of migration (Stark 1991; Carrington, Detragiache, and Vishnawath 1996). We use two variables measuring different types of networks that rural Vietnamese households may use to find employment away from the village. First, we use a standard measure of migration networks from the commune survey, which is the percentage of the commune that was seasonally migrating in Migrants who left communes soon after restrictions of movement were loosened may have given those communes an inherent advantage in migration. Second, we use a measure of networks unique to Vietnam, the percentage of people in the commune who were born in either Hanoi or HCMC before Households in 5 We omit children under age five from the demographic profile because we are concerned about the endogeneity of fertility. This is consistent with our theory, as children that young do not work.

9 438 May 2007 Amer. J. Agr. Econ. communes with more members born in either Hanoi or HCMC might have an advantage in finding jobs in the city. Neither instrument is necessarily randomly distributed across communes, so we are concerned that the instruments may have a more direct correlation with growth. For example, villages with early migrant networks could have had better economic conditions at the beginning of the 1990s, and therefore they were inherently able to grow faster than other communes. Along similar reasoning, individuals who were born in Hanoi or HCMC may have had access to better education, and therefore these households would be in a better position to grow as the economy liberalized during the 1990s. To ensure our instruments are largely uncorrelated with observables at the commune level, we regress our instruments on a number of observables at the commune level that could be correlated with inherently higher growth rates (table 3). We find no significant correlations between the percentage of the commune population who were seasonal migrants in 1993 and any of the observable commune characteristics (column 1). In fact, we cannot reject the hypothesis that all of the coefficients are jointly zero. The proportion of commune population that was born in Hanoi or HCMC is somewhat correlated with whether or not a commune has electricity and is negatively correlated with both the distance to Hanoi and to HCMC. Other estimated coefficients are statistically insignificant (column 2). As a result, we include a variable that measures the presence or lack of electricity, as well as the distance variables, as part of the commune characteristics vector C. Second, as Vietnam s agricultural sector became more marketized during the 1990s, farmers who were educated in Hanoi or HCMC may have received a higher quality education and were therefore better positioned to grow. To test whether or not farm income was affected by higher quality education, we regressed the logarithm of farm income on the household head s years of schooling, an interaction between schooling and the percent of the commune born in Hanoi/HCMC, a vector of variables that should affect farm income (land, physical capital, and human capital), and a set of commune dummies. If farmers with an urban education were able to benefit more from marketization more than other farmers, the coefficient on the interaction term would be positive. We found no evidence of a positive coefficient, so we feel comfortable that the Hanoi/HCMC instrument is not correlated with growth of a major portion of income. Estimation and Results We first estimate equations (3) and (4) using OLS and an instrumental variables, Generalized Method of Moments (IV-GMM) estimator (table 4). 6 The IV-GMM estimator we use incorporates a weighting matrix that accounts for arbitrary heteroscedasticity and intracluster correlation, and is asymptotically efficient in the presence of heteroscedasticity (Wooldridge 2002; Baum, Schaffer, and Stillman 2003). Estimating the effect of migration on expenditure growth using OLS, we find a small (0.004), statistically insignificant coefficient (model 0). However, unobservables should be correlated with both the migration decision and expenditure growth. For example, households with poorer business skills, or less ability to market crops, might have been more likely to send out migrants, because they could earn a wage away from the commune whereas they would have to farm or run a business within the commune. In that case, the OLS coefficient would both measure the effect of migration and poor business acumen on expenditure growth, and the coefficient estimate would be lower than the true effect of migration on expenditure growth. When we estimate equations (3) and (4) using the IV-GMM estimator, we find that the both instruments have a strong correlation with the migration variable. In all specifications, the cluster corrected F statistic testing the hypothesis that the coefficients on both estimates are jointly zero is larger than 8 (Appendix table B). In our base IV regression (model 1), the estimates coefficient is with a z-statistic of Among households that were likely to respond to migrant networks, our estimate suggests that an additional migrant is associated with an expenditure growth rate that is 6.3 percentage points higher than it would have been without migration participation. 7 The overidentification tests indicate that the results do not differ when each 6 Descriptive statistics for the included variables in 1993 and 1998 in Appendix table A1. 7 As suggested by Hahn and Hausman (2002), we re-estimated the reverse of our model, using migration as the dependent variable and expenditure growth as the endogenous explanatory variable. They show that if instruments are strong, the coefficient estimated by reverse two-stage least squares should be equal to the

10 de Brauw and Harigaya Seasonal Migration in Vietnam 439 Table 3. Effects of Observable Commune Variables on Instrument Candidates Share of Village, Born Share of Village, in Hanoi or HCMC Dependent Variable Migrants in 1993 before 1975 Average education level (0.446) (0.343) Log, commune population (1.155) (0.888) Public transport available? (1 = yes) (1.185) (0.911) Distance to paved road (km) (0.152) (0.117) Electricity in commune? (1 = yes) (1.348) (1.036) Factory present in 1989? (1 = yes) (1.228) (0.944) Log, average expenditures, (2.619) (2.013) Total land, commune, (hectares) (0.238) (0.183) Share of land, high quality (0.023) (0.018) Secondary school present in commune (1 = yes) (1.068) (0.821) Commune in majority kinh (1 = yes) (2.144) (1.649) Distance to Hanoi (km/100) (0.683) (0.525) Distance to HCMC (km/100) (0.796) (0.612) Regional dummies yes yes Number of observations F statistic, excluding distance to Hanoi/HCMC p-value F statistic, including distance to Hanoi/HCMC p-value Note: Robust standard errors are in parentheses. The F statistics test the hypothesis that estimated coefficients are jointly zero. The dependent variables in these regressions are percentages, whereas we use proportions in other regressions, to amplify the magnitude of these coefficients. instrument is used separately, as the Hansen J statistic indicates we cannot reject the overidentifying restrictions. When we add commune characteristics (C) measured in 1993 to the model (model 2), the estimated coefficient drops to 0.053, but remains significantly different from zero at the 5% level. Adding commune characteristics that may have affected commune-specific growth rates to the model, we find that among households affected by the instruments, an additional migrant is associated with 5.3 additional percentage points of annual per capita expenditure growth. Controlling for human capital characteristics of the household head (model 3) and household land holdings and productive assets (measured in 1993, model 4), we estimate a slightly lower coefficient of Since the consumption data exhibit mean reverting measurement error, these estimates should all be considered lower bounds. inverse of coefficient in equation (2). We performed the Hahn Hausman test on our main specification (table 4) and could not reject the hypothesis that the estimate ˆ was identical to the inverse of ˆ for all IV specifications, confirming the findings from our F-tests that the instruments are strong. 8 The land variable and value of productive assets variable proxy for the household s physical capital. We further experimented with more detailed variables describing land holdings and productive assets, and found that the estimated coefficients on the additional variables were never statistically different from zero. Therefore, here we use a parsimonious specification.

11 440 May 2007 Amer. J. Agr. Econ. Table 4. Regressions Explaining Effect of Migration on Household Per Capita Expenditure Growth, Vietnam, (0) (1) (2) (3) (4) Model OLS IV IV IV IV Number of migrants, household (0.004) (0.026) (0.024) (0.024) (0.024) Changes in household demographics Number of women, aged fifty-five and over (0.004) (0.004) (0.004) (0.004) (0.004) Number of men, aged sixty and over (0.005) (0.004) (0.004) (0.004) (0.004) Number of women, aged eighteen to fifty-four (0.002) (0.003) (0.002) (0.003) (0.003) Number of men, aged eighteen to fifty-nine (0.003) (0.003) (0.003) (0.003) (0.003) Number of children, aged six to seventeen (0.001) (0.001) (0.001) (0.002) (0.001) Household characteristics, 1993 Gender of head (1 = male) (0.004) (0.004) Years of schooling, household head (0.001) (0.001) Age of household head (0.001) (0.001) Age of head, squared (0.001) (0.001) Total land holdings, (hectares) (0.003) Log value of productive Assets, 1993 (0.002) Commune characteristics, 1993 Distance to road (km) (0.001) (0.001) (0.001) Log, commune population (0.007) (0.007) (0.007) Total land, commune (heactare) (0.002) (0.002) (0.002) Electricity in commune (1 = yes) (0.007) (0.007) (0.007) Average education level, commune (0.003) (0.003) (0.003) Distance to Hanoi (km/100) (0.004) (0.004) (0.004) Distance to HCMC (km/100) (0.005) (0.005) (0.005) Number of observations 3,492 3,492 3,462 3,462 3,462 Hansen J statistic Hansen J statistic, p-value Hahn Hausman test (t-statistic) F statistic, instruments p-value, F statistic R 2 or centered R Note: Standard errors clustered at the commune level are in parentheses. Regional dummies are also included in all equations. The F statistic is corrected for clustering and tests the hypothesis that the coefficients on instruments in the first stage are jointly zero. The Hansen J statistic tests the overidentifying restrictions in the model.

12 de Brauw and Harigaya Seasonal Migration in Vietnam 441 Table 5. Regressions Explaining Effect of Migration on Various Definitions of Household Expenditure Growth, Vietnam, Expenditures per Per Capita Dependent Variable Nonfood Expenditures Adult Equivalent Expenditures Model (1) (2) (3) (4) (5) (6) Number of migrants, household (0.039) (0.039) (0.025) (0.025) (0.024) (0.024) Changes in household demographics Number of women, aged fifty five and over (0.006) (0.006) (0.004) (0.004) (0.004) (0.004) Number of men, aged sixty and over (0.007) (0.007) (0.004) (0.004) (0.004) (0.004) Number of women, aged eighteen to fifty-four (0.004) (0.003) (0.003) (0.003) (0.003) (0.003) Number of men, aged eighteen to fifty-nine (0.004) (0.004) (0.003) (0.003) (0.003) (0.003) Number of children, aged six to seventeen (0.002) (0.002) (0.001) (0.001) (0.001) (0.001) Other controls Household, Household, Household, Household, Household, Household, commune commune commune commune commune commune Weighting? no yes no yes no yes Number of observations 3,458 3,458 3,462 3,462 3,462 3,462 F statistic, instruments Hansen J statistic Hansen J statistic, p-value Centered R Note: In all columns, specification is the same as in column (4) of table 4. Observations in columns (2), (4), and (6) are weighted by the inverse of the estimated probability that the household stayed in the sample. Other notes are the same as for table 4. Robustness Checks Our results imply that demographics matter when explaining household expenditure growth. Therefore, we might be concerned that either measurement error in consumption is correlated with household size and affects our results, or that we mismeasure the amount of consumption allocated to each individual in the household. Therefore, we re-estimate equation (3) using both nonfood expenditures and expenditures per adult equivalent as the dependent variable (table 5, models 1 through 4). When we use per capita nonfood expenditures as the dependent variable, we find that among households likely to respond to migrant networks, an additional migrant is associated with 7.9% higher nonfood expenditure growth per annum (models 1 and 2). The higher coefficient is consistent with our previous findings, as nonfood expenditures grew much faster than food expenditures over the study period. When expenditures per adult equivalent are used instead of per capita expenditures (models 3 and 4), the estimated coefficient is approximately 0.054, or slightly higher than the same specification using per capita expenditures (0.052). These findings strengthen the view that migration improves the monetary well-being of households; households increasing their participation in migration between the two surveys also experience a large increase in their well-being, measured in terms of consumption. Since attrition did not occur randomly across communes, we also test whether correcting for attrition bias affects the results (table 5, models 2, 4, and 6). Using a procedure described in Wooldridge (2002), we first use a probit model to estimate the probability that each household surveyed in 1993 stays in the data set, using our full set of variables as explanatory variables. We then re-estimate each model, using inverse probability weighting to weight each observation by the estimated probability that it remains in the data set. The weighted estimates are virtually the same as the unweighted estimates, so we can conclude that attrition bias does not affect our results. To this point, the results imply that increasing participation in migration leads to an increase in monetary well-being. However, Baulch and Masset (2003) found that in the VLSS, measures of monetary well-being are not strongly correlated with other nonmonetary measures of well-being. We test

13 442 May 2007 Amer. J. Agr. Econ. Table 6. Regressions Explaining Effect of Migration on z-scores of Children Aged One to Ten in 1993, Vietnam Children Aged Children Aged Children Aged One to Five One to Eight One to Ten (1) (2) (3) (4) (5) (6) Number of migrants, household (0.333) (0.339) (0.229) (0.271) (0.231) (0.252) Gender of child (1 = male) (0.043) (0.043) (0.032) (0.033) (0.028) (0.028) Changes in household demographics Number of women, aged fifty-five and over (0.100) (0.098) (0.068) (0.068) (0.054) (0.054) Number of men, aged sixty and over (0.106) (0.104) (0.081) (0.084) (0.073) (0.075) Number of women, aged eighteen to fifty-four (0.050) (0.052) (0.039) (0.040) (0.033) (0.034) Number of men, aged eighteen to fifty-four (0.046) (0.047) (0.037) (0.037) (0.031) (0.031) Number of children, aged six to seventeen (0.024) (0.026) (0.019) (0.020) (0.016) (0.017) Other controls Household, Household, Household, commune commune commune Number of observations 1,773 1,769 2,974 2,966 3,770 3,755 Hansen J statistic p-value, Hansen J statistic, p-value Centered R Adjusted R Note: All columns include child age dummies (in years). Columns (2), (4), and (6) also include all household and commune controls in column 4 of table 4. Other notes are the same as for table 4. our main result using height-for-age z-scores among children as the dependent variable, calculated using charts published by the Center for Disease Control. Since heights among children in the VLSS increased faster than weights (Glewwe, Koch, and Nguyen 2004), height-forage z-scores are more likely to respond to migration than other measures of improved child nutrition. We found 3,770 children aged one to ten in 1993 with height measures in both surveys, and use the change in their z-scores between surveys as a dependent variable. To control for inherent differences between children in our regressions, we include the child s gender and age dummies (in years). We then tested whether migration affects height-forage z-scores among children aged one to five, one to eight, and one to ten in We find that increasing migration associated with improvements in child height-forage z-scores among children aged one to eight and one to ten, but not aged one to five in 1993 (table 6). Among children one to five, we estimate a positive coefficient, but it is not statistically different than zero (columns 1 and 2). When we include children aged six to eight or six to ten in the regressions, we estimate coefficients between and 0.45 on migration, depending upon the sample and specification (columns 3 to 6). These results are somewhat surprising given that height-for-age z-scores are usually associated with the aggregate stock of health in children, and one would expect effects among younger rather than older children. However, z-scores among children aged six to ten in 1993 were particularly low; they were 2.28 on average, whereas the average was 2.11 among younger children. As Glewwe, Koch, and Nguyen (2004) show a remarkable amount of catch-up in heights among children in Vietnam in general between the two surveys, our results suggest that in households with good access to migrant networks, catch-up among those children is associated with the increase in migration. In summary, we find that when we control for the endogeneity of migration using two variables that measure the availability of