Migrant Labor Markets and the Welfare of Rural Households in the Developing World: Evidence from China

Migrant Labor Markets and the Welfare of Rural Households in the Developing World: Evidence from... Abstract Increased ability to migrate from China's rural villages contributed to significant increases in the consumption per capita of both non-durable and durable goods, and these effects were larger in magnitude for households that were relatively poor before the easing of restrictions to migration. With increased out-migration, poorer households invested more in housing and durable goods than rich households, while richer households invested significantly more in non-agricultural production assets. As migration became easier, increased participation in migrant employment was greater among poorer households on both the extensive and intensive margins, and poorer households reduced labor days in agriculture. migration, migrant networks, consumption growth, inequality In developing countries, barriers to the movement of labor are a common institutional feature that may contribute to geographic poverty traps. Whether maintained by formal institutions, by cultural or linguistic differences across regions, or simply by high transaction costs associated with finding migrant employment, constraints on the movement of labor within developing countries may reinforce an inefficient allocation of resources across regions and influence investment levels in poor areas, potentially hindering growth (Jalan and Ravallion 2002). When barriers to cross-regional labor mobility are removed, the resulting improved efficiency of resource allocation may have important consequences for rural living standards (Yang 2008). Remittances to household or family members remaining in rural areas may supplement income earned locally and directly reduce exposure to poverty. Migration may also have indirect effects on household and individual welfare within their home communities, either in the form of increased wages with the depletion of the local labor force, or through remittances from migrant employment that are invested in local production (e.g., Woodruff and Zenteño 2007). A growing body of research examines the impact of both international and internal migration on investment, growth, and other outcomes. Relevant for thinking about the identification strategy employed in this paper, considerable research has focused on the role of networks in facilitating migration (e.g., Munshi 2003) and influencing migration decisions (Munshi 2011; Munshi and Rosenzweig 2016), and the role that risk and liquidity constraints may play in shaping migration decisions (Bryan et al. 2014). Within the China literature, recent research has focused on the relationship between migration and poverty (Du, Park, and Wang 2005); migration, remittances, and household income (Taylor, Rozelle, and de Brauw 2003); migration and risk-coping (Giles 2006; Giles and Yoo 2007); and the prospects of migration and life satisfaction (Frijters, Liu, and Meng 2012), but identification of migration in each of these papers requires uncomfortable assumptions. To date, the published literature on the effects of out-migration has considered neither the distributional effects of migration within villages, nor the effects of out-migration on the sectoral distribution of household activities across the within-village wealth distribution. This paper addresses these gaps in the literature and makes four important contributions to our understanding of the impact of rural-urban migration on both consumption and sectoral choice in rural China. The main specification first highlights the positive effect of migration on annual changes in log consumption between 1988 and 2002: out-migration explains between 2 and 2.9 percent of the annual increase in consumption per capita, or between 65 and 93 percent of the annual consumption growth recorded in the sample villages between 1988 and 2002. As earlier work suggests that income from work outside the village (by both migrants and commuters) contributes to reducing inequality (Benjamin, Brandt, and Giles 2005), this paper exploits the opportunity to use matching village surveys to examine the local general equilibrium effects of migration on the per capita consumption of households across the initial consumption distribution. Whether households actually participate in migration or not, consumption grows more rapidly among households in the poor and middle terciles of the initial average consumption distribution. Maintaining a focus on within-village differences across the initial consumption distribution, the paper examines possible mechanisms through which migration influenced earnings, labor supply, and investment decisions. Where others have found that migration contributed to increased labor supply of residents remaining behind (e.g., Mu and van de Walle 2011), this paper finds that out-migration from the village leads to labor reallocation across activities, and labor allocation changes differ by wealth tercile. Agricultural labor days decline among poorer households, who also provide a relatively larger increase in days spent working outside the home township. More affluent households, by contrast, increase labor supply to local non-agricultural activities, potentially reflecting general equilibrium effects of increased local opportunity as migrants remit earnings. In terms of investment decisions, poorer households invest more in housing and durable goods, while more affluent households invest more in non-agricultural productive assets. If focusing on the period of rapid expansion in migration alone, from 1995 to 2002, the effects of migration on both labor supply and investment decisions are even more pronounced. A fourth contribution is the relatively unique instrumental variables (IV) strategy used to identify village-level effects of migration on household consumption and other outcomes. The approach takes advantage of an early reform to China's residential registration (hukou) system making it easier for rural migrants with national identification cards (IDs) to legally reside in cities after 1988. National IDs, first available to urban residents in 1984, were not available in all rural counties as of 1988. While allowing that the timing of ID distribution may be related to fixed unobserved characteristics of villages, the annual change in the share of the village population working as migrants outside the village is shown to be a non-linear function of the time since residents of a county received IDs. As migrant networks in cities take time to build up, this non-linearity should not be surprising. After controlling for household fixed effects, village-specific trends, and province-wide macroeconomic shocks, the change in the cost of migrating is identified by exploiting differences in the timing of access to IDs and the non-linearity in the relationship between the annual change in the village migrant share and the time since IDs were distributed. As IDs were not randomly distributed, the paper demonstrates that the timing of ID distribution was unrelated to other policy and economic shocks that might be correlated with both migration and well-being. 1. Background Over the period covered by the primary data source (1986 to 2002), rapid growth in the number of rural migrants moving to urban areas signaled a dramatic change in the nature of China's labor market. Estimates using the one percent sample from the 1990 and 2000 rounds of the Population Census and the 1995 one percent population survey suggest that the inter-county migrant population grew from just over 20 million in 1990 to 45 million in 1995 and 79 million by 2000 (Liang and Ma 2004). Surveys conducted by the National Bureau of Statistics (NBS) and the Ministry of Agriculture include more detailed retrospective information on past short-term migration, and suggest even higher levels of labor migration (Cai, Park, and Zhao 2008). Rural-Urban Migration in China, Migrant Networks, and Well-Being The use of migrant networks and employment referral in urban areas are important dimensions of China's rural-urban migration experience. Rozelle et al. (1999) emphasize that villages with more migrants in 1988 experienced more rapid migration growth by 1995. Zhao (2003) shows that the number of early migrants from a village is correlated with the probability that an individual with no prior migration experience will choose to migrate. Meng (2000) further suggests that variation in the size of migrant flows to different destinations can be partially explained by the size of the existing migrant population in potential destinations. Finally, appendix table A.1 of de Brauw and Giles (2017) notes that a 2001 survey of migrants in five large Chinese cities shows that more than 90 percent of migrants knew an acquaintance from their home village before migrating. Before labor mobility restrictions were relaxed, households in remote regions of rural China faced low returns to local economic activity, reinforcing geographic poverty traps (Jalan and Ravallion 2002). Over the period of study, household income per capita grew in both urban and rural areas, but at significantly higher rates in urban areas. After spatially deflating for cost-of-living differences, the urban-rural differential in per capita income increased from 38 percent to 71 percent by 2001. Thus, as documented by a considerable body of descriptive evidence, the growth of migration in China raises the possibility that migrant opportunity may be an important mechanism for poverty reduction in rural areas. Recent research in China suggests that migration is associated with higher incomes (Taylor, Rozelle, and de Brauw 2003; Du, Park, and Wang 2005), facilitates risk-coping and risk management (Giles 2006; Giles and Yoo 2007), is associated with higher levels of local investment in productive activities (Zhao 2002), and may alter household labor allocation decisions (Mu and Van de Walle 2011). Earlier research on rural-urban migration in China has suggested large impacts on the earnings of households with migrants, but some of the identification strategies employed likely produced estimates of migration impacts that suffer from significant bias. For example, Taylor, Rozelle, and de Brauw (2003) find that per capita incomes of households with migrant family members were 22, 26, and 29 percent higher at the 25th, 50th, and 75th percentiles of the sample (not within villages). However, the authors used a retrospective report of the village migration network size from seven years earlier as an instrument for household migration, and therefore cannot control for village fixed effects, such as proximity to cities (and markets) and other factors affecting the local economy, which will lead to an upward bias to estimates of the effect of migration on earnings. Another study examines both migrant participation and earnings across the earnings distribution (Du, Park, and Wang 2005), finding that poor households are unlikely to have migrants, but the main barrier to migrant employment is lack of capable laborers rather than other obstacles to working in urban areas. The authors estimate models in first differences, and identify migration through the one-year lag of the village migrant network, finding that households with migrants have per capita incomes 8.5 to 13.1 percent higher than those without migrants. A concern with this approach, however, is that shocks to the village economy, which affect migration decisions, have persistent effects over time. Both current migration and a household's position within the earnings distribution are likely to be affected by shocks reflected in prior migration from the village. The RCRE Household and Village Surveys The primary data sources used for analysis are the village and household surveys conducted by the Research Center for Rural Economy (RCRE) at China's Ministry of Agriculture from 1986 through 2002. The paper uses data from 88 villages in eight provinces (Anhui, Jilin, Jiangsu, Henan, Hunan, Shanxi, Sichuan, and Zhejiang) surveyed over the 16-year period, with an average of 6,305 households surveyed per year. Depending on village size, between 40 and 120 households were randomly surveyed in each village. Each village in the sample is in a different county, so county-level policies affect each village in this sample differently. The RCRE household survey collected detailed household-level information on incomes and expenditures, education, labor supply, asset ownership, land holdings, savings, formal and informal access to credit, and remittances. In common with the National Bureau of Statistics (NBS) Rural Household Survey, respondent households keep daily diaries of income and expenditures, and a resident administrator living in the county seat collects the diaries monthly. As in the NBS Rural Household Survey, estimates of the income and expenditures of household members currently living and working as migrants outside the household were included among household earnings and expenditures. This unusual feature of the survey is beneficial because household size does not change with out-migration. Details on the calculation of consumption from expenditure values and information on housing and durable goods are provided in supplementary online appendix S1, and in the analysis below, both income and consumption are deflated to 1986 values. The paper also makes use of an annual village survey enumerated from meetings with village accountants and other key informants that provides important village-level information on the local economy and on the land, production, and the work locations (migrant, local, and in the village) and occupations of registered residents. Time-varying information from this survey is important for characterizing connection to migrant networks, and when controlling for other time-varying village demographic and economic factors. The village-level survey includes all information on current registered residents who are working outside the village, and does not miss entire families who move (which was a very rare occurrence in rural China during this period). Finally, the identification strategy makes use of retrospective information, specifically the timing of ID availability, from a Village Governance Survey conducted by the authors in 2004. Trends in Migration, Consumption Growth, and Poverty The accompanying village survey includes questions asked annually of village leaders about the number of registered village residents working and living outside the village. For the analysis in this paper, all registered village residents who work outside the home county are considered migrants. Both the tremendous increase in migration from 1987 onward and heterogeneity across villages are evident in fig. 1. In 1987, an average of 3 percent of working-age laborers in RCRE villages worked outside their home counties, and this share rose steadily to 23 percent by 2003. Moreover, the share of working-age laborers working as migrants varies considerably by village. Whereas for some villages, only a small share of legal residents are employed as migrants, from other villages more than 50 percent of working-age adults are employed outside their home county by 2003. Figure 1. View largeDownload slide Share of Village Labor Force Employed as Migrants, by Year Source: RCRE Village Surveys 1986 to 2002. Note: The figure shows the share of registered village residents living and working outside the village and home county. Figure 1. View largeDownload slide Share of Village Labor Force Employed as Migrants, by Year Source: RCRE Village Surveys 1986 to 2002. Note: The figure shows the share of registered village residents living and working outside the village and home county. The relationship between migration and consumption is of central concern for the analysis. The linear fit of the relationship between annual changes in share of the village workforce employed as migrants (village migrant share) and growth in village average consumption suggest a positive relationship (fig. 2). The local polynomial fit, however, suggests the presence of non-linearities. If out-migration is driven by negative shocks or return migration by positive shocks, and both are correlated with movements in consumption, one should be concerned that migration and consumption are endogenous. Even if consumption grows with an increase in the number of residents employed as migrants, it is of particular policy interest to understand which residents within villages are experiencing increases in consumption. Figure 2. View largeDownload slide Village Average Consumption Growth and Change in Migrant Share of Village Population Source: RCRE Village and Household Surveys, 1986–2003. Note:Figure 2 shows the linear and local polynomial fits of the relationship between annual village average consumption growth and annual changes in the share of migrants from the village. Migrants are registered residents of the village who live and work outside the village and home county. Figure 2. View largeDownload slide Village Average Consumption Growth and Change in Migrant Share of Village Population Source: RCRE Village and Household Surveys, 1986–2003. Note:Figure 2 shows the linear and local polynomial fits of the relationship between annual village average consumption growth and annual changes in the share of migrants from the village. Migrants are registered residents of the village who live and work outside the village and home county. 2. Empirical Methodology Behind the estimation approach lies a simple model highlighting the direct and indirect mechanisms through which expanded migrant employment opportunities may affect household consumption (see de Brauw and Giles 2008). The migrant network, which facilitates job referral in distant destinations, may directly and indirectly affect permanent household income and thus also consumption. First, a larger village migrant network may facilitate higher family income from earnings in migrant destinations. The wealth effect eases credit constraints associated with accumulating assets for productive activities (both agricultural and non-agricultural) and “non-productive” uses (e.g., investments in housing and durable goods). Household consumption may also increase by relaxing a credit constraint that led households to consume less and save more in each period as a precaution against potential future production shocks. Second, an increase in the village migrant network size may affect the shadow price of household labor time. If leisure is a normal good, the net effect on family labor supply is indeterminate. A substitution effect will lead families to supply more labor to productive activities, and perhaps shift from farming into other higher-return activities, and an income effect may lead to a reduction in family labor supply. To flesh out the productive mechanisms through which increases in migration may lead to higher consumption, the impacts of migration on household labor supply are also examined, and disaggregated into three broad categories: agricultural, local non-agricultural, and non-agricultural self-employed work. Estimating the Effect of Migration on Consumption An empirical model consistent with the conceptual framework outlined above suggests the following linearized specification for household consumption, cijt (in logs):   \begin{equation} c_{ijt}= \beta {M_{jt}} + {\boldsymbol{X}}_{{\boldsymbol{ijt}}}^{\prime}\alpha + {\boldsymbol{Z}}_{{\boldsymbol{jt}}}^{\prime}\gamma + {u_i} + {v_j} + {{\boldsymbol{t}}_{\boldsymbol{j}}} + {\varepsilon _{ijt}} \end{equation} (1) The logarithm of per capita consumption for household i in village j during period t is a function of the share of the registered working-age village residents working outside the home county as migrants, Mjt. Household characteristics, Xijt, influence consumption through endowments, such as human capital, which affect household permanent income, and through demographic characteristics that influence consumption preferences. Time-varying village variables account for heterogeneity across villages in policies and economic conditions, Zjt, that may influence consumption through productivity. Village-specific trends, tj, account for underlying endowments and initial conditions in the village that may contribute to differences in consumption growth across villages. Finally, village- and household-level unobservables, vj and ui, respectively, related to location (village), to consumption preferences (household), and to the ease of household participation in the migrant labor market (household).1 Equation (1) is initially first-differenced to control for fixed effects at the household and village level:   \begin{equation} {\rm \Delta }{c_{it}} = \beta {{\rm \Delta }}{M_{jt}} + {{\rm \Delta }}{\boldsymbol{X}}_{{\boldsymbol{ijt}}}^{\prime}\alpha + {\rm{\Delta }}{\boldsymbol{Z}}_{{\boldsymbol{jt}}}^{\prime}\gamma + {{\boldsymbol{d}}_{\boldsymbol{j}}} + {{\rm \Delta }}{\varepsilon _{ijt}} \end{equation} (2)Differencing the village-specific trend leaves a vector of village dummy variables, dj, that control for differences in consumption growth trends across villages, which may be related to differences in proximity markets for labor or goods in townships or cities. Finally, province-wide macroeconomic shocks, driven by policy or market effects, might affect the relationship between household consumption and employment opportunities. Province-year interactions, p⊗t, are added in (3) to control for the presence of these shocks:   \begin{equation}{\rm{\Delta }}{c_{it}} = \beta {\rm{\Delta }}{M_{jt}} + {\rm{\Delta }}{\boldsymbol{X}}_{{\boldsymbol{it}}}^{\prime}\alpha + {\rm{\Delta }}{\boldsymbol{Z}}_{{\boldsymbol{jt}}}^{\prime}\gamma + {{\boldsymbol{d}}_{\boldsymbol{j}}} + p \otimes t + {\rm{\Delta }}{\varepsilon _{ijt}}\end{equation} (3)In equation (3), the coefficient of primary interest, β, measures the effect of the migrant labor market on consumption. Given descriptive evidence from the literature suggesting that the effect of migration may differ across the within-village wealth distribution (e.g., Benjamin, Brandt, and Giles 2005), the sample is next split into low, middle, and upper initial wealth terciles within each village. The tercile rank (low, middle, and high) is assigned to households based on average household consumption per capita for each household over the 1987–1989 period. Estimating models by this initial 1987–1989 consumption tercile facilitates examining the different impacts of migration across the initial wealth distribution. Endogeneity Concerns In equation (3), ΔMjt suffers from a well-known endogeneity problem. The village migrant share reflects factors affecting both changes in demand for migrant labor and changes in the labor supply decisions of migrants and potential migrants. Local shocks, for example, decrease household consumption per capita while increasing the relative return to migrant employment in more distant locations, potentially leading to an observed negative relationship between increases in migration and consumption growth. To identify the effect of migration on consumption, it is necessary to find an instrument correlated with the share of village residents working as migrants, but otherwise unrelated to factors affecting growth or negative shocks experienced by the village. The identification strategy employed makes use of two policy changes that, working together, affect the strength of migrant networks outside home counties but are plausibly unrelated to average village consumption growth. First, a new national ID card (shenfen zheng) was introduced in 1984. While urban residents received IDs in 1984, residents of most rural counties did not receive them immediately. In 1988, a reform of the residential registration system made it easier for migrants to gain legal temporary residence for work in cities under a “guest worker” system, but a national ID card was necessary to obtain a temporary residence permit (zanzu zheng) (Mallee 1995). While some counties made national IDs available to rural residents as early as 1984, others distributed them in 1988, and still others did not issue IDs until several years later. In a follow-up survey conducted with RCRE in 2004, local officials were asked when IDs had actually been issued to rural residents of the county. In the analysis sample, 41 of the 88 counties issued ID cards in 1988, but cards were issued as early as 1984 in three counties and as late as 1997 in one county. It is important to note that IDs were not necessary for migration, and large numbers of migrants live in cities without legal temporary residence cards. With temporary residence cards, however, migrants were under a “guest worker” system providing a more secure position in the destination community, and thus held better jobs and made up part of a longer-term migrant network in migrant destinations. Thus, ID distribution had two effects after the 1988 hukou reform. First, the costs of migrating to a city should fall after IDs become available. Second, as the quality of the potential migrant network improves with the years since IDs are available, the costs of finding migrant employment should continue to fall over time. The relative size of the migrant network should therefore be a function of both whether or not cards have been issued and the time since cards have been available to village residents. As we use the share of the village workforce working as migrants to proxy for the migrant network, the size of the potential network has an upper bound. Thus, we expect years-since-IDs-issued to have a non-linear relationship with the share of the village labor force working as migrants, as growth in the migrant network should decline after initially increasing with distribution of IDs. In fig. 3, a polynomial smoother shows the relationship between years since IDs were distributed and the share of working-age village residents working as migrants in year t. Within a couple of years after IDs are distributed, the share of the village labor force working as migrants grows sharply, and then slows after seven years. This pattern suggests non-linearity in the relationship between ID distribution and new participants in the village migrant labor force. Therefore, the primary instrument is specified as a dummy variable indicating that IDs had been issued, interacted with quartic functions of years since IDs were issued.2 Figure 3. View largeDownload slide Share of Village Out-Migrants, versus Years Since IDs Were Distributed Source: RCRE Village Surveys, 1986–2003, and Supplementary Village Governance Survey (2004). Note:Figure 3 shows the relationship between change in annual share of migrants from the village and number of years since ID cards became available in the county. Figure 3. View largeDownload slide Share of Village Out-Migrants, versus Years Since IDs Were Distributed Source: RCRE Village Surveys, 1986–2003, and Supplementary Village Governance Survey (2004). Note:Figure 3 shows the relationship between change in annual share of migrants from the village and number of years since ID cards became available in the county. As IDs were not randomly distributed, it is important to consider the plausible exogeneity of the identification strategy. More detail on the distribution of IDs and the conditional exogeneity of the timing is provided in supplementary online appendix S2, and additional discussion based on a four-province sample can be found in de Brauw and Giles (2017). A second, mechanical source of endogeneity that is well known in the panel data literature may arise, as the regressors included in ΔXit and ΔZjt might not be strictly exogenous. For example, income shocks that affect household consumption decisions may also have an impact on household composition, land characteristics, or village policy. To examine the robustness of our result, initial estimates of equation (3) exclude ΔXit and ΔZjt. If the years-since-IDs quartic is a valid instrument, meaning that it is uncorrelated with the changes in household and village characteristics that are now part of the unobservable, then this model will be identified and potentially endogenous regressors are not a concern. The simplified model is included for estimates over both the full sample and the models estimated by initial consumption tercile. When models include the village and household regressors, ΔXit and Zjt, both likely to explain some variation in consumption, they are treated in successive models as exogenous and then as predetermined but not strictly exogenous. For models in which regressors are treated as predetermined, first-differenced predetermined variables are instrumented with their t − 2 lagged levels [Xit − 2, Zit − 2], a standard panel data approach. Xit − 2 and Zit − 2 will be valid instruments if they are correlated with ΔXit and ΔZjt, but uncorrelated with any time-varying household unobservables remaining in the differenced error term, Δεit. Most importantly, these lagged values, Xit − 2 and Zit − 2, are uncorrelated with shocks that affected household demographic composition or the education composition of the household, and thus reduce one potential source of endogeneity. Use of lagged values does not resolve all endogeneity problems related to forward-looking household behavior. For example, a husband and wife making plans to migrate in the distant future may choose to co-reside with parents and in-laws who could farm land and watch children. In this sense, household structure is endogenous to future migration plans. These unobserved plans are considered a fixed preference that are differenced out in estimation. Further, precise unbiased estimates of the coefficients on these variables are not our central focus. Rather, these controls are included to allow for a more precise estimate of the effects of migration on household consumption, and later on labor supply and investment decisions. 3. Results The First Stage Before estimating equation (3), it is important to first establish that the instruments, period t − 2 values of a polynomial function of the years since ID cards were issued, are significantly related to the change in the share of working-age village residents working as migrants between period t − 1 and t. First-stage models were initially specified with years since IDs were issued as a quadratic, cubic, and quartic function, along with village and province-year dummies (table 1, columns 1 through 3). The quartic is preferred to the quadratic and the cubic due to the added flexibility in the functional form for the effects of ID card distribution on the migrant network. Table 1. Developing the First Stage: Timing of ID Card Distribution and Change in Share of Migrants in Village Population   (1)  (2)  (3)  (4)  (5)  (Years since IDs issued)t-2  −0.000  0.029***  0.044***  0.027***  0.026***    (0.001)  (0.002)  (0.004)  (0.004)  (0.004)  [(Years since IDs issued)t-2]2/10  −0.007***  −0.065***  −0.121***  −0.098***  −0.095***    (0.001)  (0.004)  (0.012)  (0.013)  (0.013)  [(Years since IDs issued)t-2]3/100    0.029***  0.092***  0.073***  0.069***      (0.002)  (0.013)  (0.014)  (0.014)  [(Years since IDs issued)t-2]4/1000      −0.022***  −0.017***  −0.015***        (0.005)  (0.005)  (0.005)  Two-period lag village controls included?  No  No  No  Yes  Yes  Two-period lag household controls included?  No  No  No  No  Yes  Number of observations  69,878  69,878  69,878  67,529  66,487  F-statistic, instruments  43.7  83.2  74.0  75.0  50.6    (1)  (2)  (3)  (4)  (5)  (Years since IDs issued)t-2  −0.000  0.029***  0.044***  0.027***  0.026***    (0.001)  (0.002)  (0.004)  (0.004)  (0.004)  [(Years since IDs issued)t-2]2/10  −0.007***  −0.065***  −0.121***  −0.098***  −0.095***    (0.001)  (0.004)  (0.012)  (0.013)  (0.013)  [(Years since IDs issued)t-2]3/100    0.029***  0.092***  0.073***  0.069***      (0.002)  (0.013)  (0.014)  (0.014)  [(Years since IDs issued)t-2]4/1000      −0.022***  −0.017***  −0.015***        (0.005)  (0.005)  (0.005)  Two-period lag village controls included?  No  No  No  Yes  Yes  Two-period lag household controls included?  No  No  No  No  Yes  Number of observations  69,878  69,878  69,878  67,529  66,487  F-statistic, instruments  43.7  83.2  74.0  75.0  50.6  Source: RCRE Household and Village Surveys, 1986–2002, and the RCRE 2004 Supplementary Village Governance Survey. Note: (1) All models include jointly significant controls for village and province* year effects, as well as other included instruments. (2) Dependent variable is change in share of working-age residents from the village living and working outside their home county between year t-1 and t. (3) Robust standard errors are cluster-corrected at the village, and there are 88 village clusters. (4) Two-period lag village controls include: total number of working-age laborers in registered village labor force, total village land, share of land in village in orchards, and share of total assets owned by the village collective. (5) Two-period lag household controls include: number of working-age laborers in the household, male working-age laborer share of household population, female working-age laborer share of household population, household land per capita, value of household productive assets, and average years of education of working-age laborers. ***indicates significance at the 1 percent level. View Large To anticipate models controlling for endogenous changes in village or household variables, two-period lags of village and household controls are added in columns 4 and 5, respectively. At the village level, the size of the village labor force is included to control for local returns to labor, the cultivable share of village land, total village land, and the share of land planted in orchards, which control for village land endowment and specialization in high-value crops, and the share of village assets held by collectives, which controls for the returns to capital outside agriculture as well as local government involvement in the economy. To control for household-level human and physical capital endowments, the number of working-age members of the household, the share of household members that are working-age males and females, respectively, land per capita, and the average education level of household adults are included. In both columns, the relationship between the migrant network variable and instruments for migration remains strong, and the F-statistic suggests that the complete set of instruments continues to have sufficient power to ease concerns over weak instrument bias after adding the full set of controls. While sufficiently strong, one might be concerned that the timing of ID distribution is also correlated with changes in a range of different village policies. These possibilities are examined in supplementary online appendix S3, and again there is no direct evidence that ID distribution is systematically related to policy changes that may affect consumption through channels other than migration from the village. The Effect of Migration on Household Consumption To begin the examination of the effects of migration on consumption, OLS models of the effects of migration on consumption in both levels and first differences are estimated across all three terciles of the initial consumption distribution. As one might expect if unobserved local shocks are an important factor driving initial migration decisions, the coefficient on migration is positive and insignificant in the OLS levels model, whether or not household- and village-level covariates are included (supplementary online appendix table S3.2). Similarly, when estimated by OLS in first differences, the coefficients on the change in migration are small and not statistically significant from zero. Next, the analysis uses IV-GMM models, which control for simultaneity bias and other unobservables potentially related to the migration measure. The weighting matrix used in the GMM estimator accounts for arbitrary heteroskedasticity and intra-cluster correlation, and it is asymptotically efficient in the presence of heteroskedasticity (Wooldridge 2010). Three first-differenced models are estimated (table 2). In column 1, estimates exclude village and household controls. Village and household controls are then added, treated first as exogenous (column 2) and then as predetermined (column 3), using t − 2 levels of the household and village controls as instruments. The Cragg-Donald F-statistic indicates that the bias in the IV coefficient is less than 5 percent of the OLS bias. Evidence from all three models suggests that the growth of the village migrant share in the labor force has a positive, statistically significant effect on consumption among all households. Table 2. Migration and Household Consumption in Migrant-Sending Villages (All Models in First Differences)   Ln(consumption per capita)  Ln(non-durable consumption per capita)    (1)  (2)  (3)  (4)  (5)  (6)  Share of migrants in village population  3.563***  3.602***  2.507**  3.285***  3.434***  2.308**    (1.188)  (1.103)  (1.132)  (1.268)  (1.184)  (1.131)  Village-level control variables  Village labor force    −0.001  0.003    −0.003  0.000      (0.005)  (0.014)    (0.005)  (0.015)  Cultivable share of village land    0.269***  0.819**    0.270***  0.796**      (0.092)  (0.382)    (0.093)  (0.380)  Total village land    0.016*  0.127***    0.015  0.129***      (0.009)  (0.046)    (0.010)  (0.047)  Share of assets owned by village collective    −0.014  −0.157    −0.015  −0.169      (0.028)  (0.165)    (0.032)  (0.175)  Share of village land in orchards    −0.163  0.420    −0.100  0.443      (0.180)  (0.777)    (0.181)  (0.780)  Household-level control variables  Working-age male share of household population    −0.070**  0.427***    −0.064*  0.479***      (0.032)  (0.127)    (0.036)  (0.151)  Working-age female share of household population    −0.076**  0.131    −0.066*  0.152      (0.031)  (0.127)    (0.035)  (0.149)  Number of working-age laborers in the household    −0.038***  −0.003    −0.026***  0.023**      (0.004)  (0.009)    (0.005)  (0.010)  Cultivable land per capita    0.085***  0.114***    0.085***  0.125***      (0.008)  (0.021)    (0.009)  (0.023)  Household average years of education    0.004**  −0.012**    0.005***  −0.013**      (0.002)  (0.005)    (0.002)  (0.005)  Village, HH controls predetermined?    No  Yes    No  Yes  Regression statistics  Hansen J-statistic  1.860  2.235  4.130  1.244  1.564  4.569  p-value, J-statistic  0.602  0.525  0.248  0.742  0.668  0.206  Cragg-Donald F-statistic  87.23  95.66  32.68  87.56  95.76  32.66  Number of clusters  88  88  88  88  88  88  Number of observations  69,878  68,110  65,607  69,834  68,072  65,570    Ln(consumption per capita)  Ln(non-durable consumption per capita)    (1)  (2)  (3)  (4)  (5)  (6)  Share of migrants in village population  3.563***  3.602***  2.507**  3.285***  3.434***  2.308**    (1.188)  (1.103)  (1.132)  (1.268)  (1.184)  (1.131)  Village-level control variables  Village labor force    −0.001  0.003    −0.003  0.000      (0.005)  (0.014)    (0.005)  (0.015)  Cultivable share of village land    0.269***  0.819**    0.270***  0.796**      (0.092)  (0.382)    (0.093)  (0.380)  Total village land    0.016*  0.127***    0.015  0.129***      (0.009)  (0.046)    (0.010)  (0.047)  Share of assets owned by village collective    −0.014  −0.157    −0.015  −0.169      (0.028)  (0.165)    (0.032)  (0.175)  Share of village land in orchards    −0.163  0.420    −0.100  0.443      (0.180)  (0.777)    (0.181)  (0.780)  Household-level control variables  Working-age male share of household population    −0.070**  0.427***    −0.064*  0.479***      (0.032)  (0.127)    (0.036)  (0.151)  Working-age female share of household population    −0.076**  0.131    −0.066*  0.152      (0.031)  (0.127)    (0.035)  (0.149)  Number of working-age laborers in the household    −0.038***  −0.003    −0.026***  0.023**      (0.004)  (0.009)    (0.005)  (0.010)  Cultivable land per capita    0.085***  0.114***    0.085***  0.125***      (0.008)  (0.021)    (0.009)  (0.023)  Household average years of education    0.004**  −0.012**    0.005***  −0.013**      (0.002)  (0.005)    (0.002)  (0.005)  Village, HH controls predetermined?    No  Yes    No  Yes  Regression statistics  Hansen J-statistic  1.860  2.235  4.130  1.244  1.564  4.569  p-value, J-statistic  0.602  0.525  0.248  0.742  0.668  0.206  Cragg-Donald F-statistic  87.23  95.66  32.68  87.56  95.76  32.66  Number of clusters  88  88  88  88  88  88  Number of observations  69,878  68,110  65,607  69,834  68,072  65,570  Source: RCRE Household and Village Surveys, 1986–2002, and the RCRE 2004 Supplementary Village Governance Survey. Note: All models are run in first differences and include jointly significant village fixed effects to control for village-specific trends, and province-year effects to control for province-wide macroeconomic shocks. Standard errors clustered at the village. *indicates significance at the 10 percent level; **indicates significance at the 5 percent level; ***indicates significance at the 1 percent level. View Large In all three specifications, the coefficients on the village migrant share are relatively close to one another. Coefficients in columns 1 and 2 suggest that a one-percentage-point increase in migration at the village level is associated with a 3.6 percent increase in household consumption per capita, while column 3 (controlling for the dynamic endogeneity of control variables) indicates a 2.5 percent increase. Average village migration increased from 0.012 to 0.126 of the village workforce from 1988 to 2002, so a one-percentage-point rise is slightly higher than the average annual increase in the village migrant share, which is 0.8 percent. At the average migrant share, these estimates imply that migration is associated with a 2 percent annual increase in household per capita consumption in those models treating controls as predetermined, or 2.9 percent when treated as exogenous. With an average of 3.1 percent consumption growth per annum in the RCRE villages, increased ability to migrate thus explains from 65 to 93 percent of the average per capita consumption growth in this sample. Recall that the consumption measure includes the flow of services from housing and durable goods. Given that the value of this flow may be imputed incorrectly for migrant family members living outside the household, the effect of migration on non-durable consumption growth is shown in columns 4 through 6, with education expenses and the flow from housing and durable goods excluded.3 The coefficient on the village migrant share is slightly smaller (suggesting a 2.3 to 3.4 percent increase), but remains statistically significant at the 5 percent level or better. An additional concern might be that the results are driven by comparisons between villages with and without migration. At the beginning of the study period, some villages had no migrants, though by 1996 all 88 villages had some migrants. This possibility is explored by dropping the early years of villages without migrants (supplementary online appendix table S3.3, columns 1 through 3), and there is little difference from the main specifications shown in table 2. In any long panel, one should also be concerned over whether attrition might be biasing results. The average annual household attrition rate is 5.8 percent; by 2002, the average probability that original households are still in the sample is 62 percent. In this paper, as in two others using different samples from this data source (Benjamin, Brandt, and Giles 2011; Giles 2006), there is no evidence of a statistically significant relationship between the village migrant share (or changes in the village migrant share) and attrition in the current or following year, nor are consumption per capita and attrition correlated. To check further for potential attrition bias, we weight each observation by the inverse of the cumulative probability of survival to the current year, with probabilities for survival in each year calculated using household and village characteristics and a province fixed effect (Wooldridge 2010, chapter 19). After controlling for attrition, there is little change in the coefficients estimated for the three main models (columns 4 through 6 of supplementary online appendix table S3.3), suggesting that the main results do not suffer from attrition bias. Who Benefits from Migration? By measuring effects of new migration from the village on household consumption, and with consumption of migrants included in the measure of household expenditures, both made possible through interesting features of the data source, we capture the general equilibrium effects of increasing out-migration. Starting in 1995, the household-level questionnaires began to enumerate whether or not a household has a migrant, so direct household-level participation in migration can be examined, also by terciles of the initial distribution of consumption per capita. For the period beginning in 1995, fig. 4A shows the share of households with at least one household member working and living outside their home township. For each year through 2001, households in the lowest tercile were roughly five percentage points more likely than upper-tercile households to have a registered member living and working outside the village. Further, the lower tercile averages about 12 more days of migrant employment per capita than the upper tercile (fig. 4B). These differences are statistically significant, and remain so after controlling for village fixed effects (supplementary online appendix table S4.2). Figure 4. View largeDownload slide A. Share of Households with Member Working Outside the Township, by Initial Consumption Tercile, 1995–2002 B. Average Days Employed Outside the Township Source: RCRE Household Surveys, 1995–2002. Figure 4. View largeDownload slide A. Share of Households with Member Working Outside the Township, by Initial Consumption Tercile, 1995–2002 B. Average Days Employed Outside the Township Source: RCRE Household Surveys, 1995–2002. Splitting the sample by the initial consumption terciles, equation (3) is re-estimated by tercile, with consumption and non-durable consumption as dependent variables (table 3).4 Coefficients are estimated by tercile for the full 1988–2002 period and for a shorter 1995–2002 period, after migration had begun to grow more rapidly. Households that were initially in the lowest within-village tercile experience more rapid growth in consumption per capita and non-durable consumption per capita than households in the upper tercile. In fact, estimates for the upper tercile for the entire 1988–2002 period show that neither total nor non-durable consumption per capita grows significantly as a result of village-level out-migration. For the lowest and middle within-village tercile, however, a one-percentage-point increase in migration over the 1988–2002 period is associated with a 3 to 4 percent increase in both consumption per capita and non-durable consumption per capita. Over the period when out-migration was growing rapidly (1995–2002), coefficient estimates suggest more rapid consumption growth for both the lowest and the middle tercile (10 percent and 7 percent, respectively, with a one-percentage-point increase in migration), and also suggest positive consumption growth for the upper tercile (4 percent), though the latter increase appears attributable to the accumulation of durable goods and housing. Table 3. Village-Level Migration and Consumption across the Initial Consumption Distribution (All Models in First Differences) Dependent variable  With controls?  Lowest tercile  Middle tercile  Upper tercile  Log (consumption per capita)  No  3.979**  3.936**  1.985  1988–2002    (1.420)  (1.447)  (1.471)    Yes  3.484***  3.718**  2.536**      (1.260)  (1.350)  (1.353)    Pre-det.  2.783*  1.541  1.305      (1.450)  (1.344)  (1.252)  Log (consumption per capita)  No  10.262***  7.703***  3.856**  1995–2002    (2.630)  (2.037)  (1.898)    Yes  8.742***  7.242***  4.476**      (2.815)  (2.156)  (2.184)    Pre-det.  9.069**  5.776*  3.459      (4.393)  (3.323)  (2.451)  Log (non-durable consumption per capita)  No  3.732**  3.413**  1.516  1988–2002    (1.536)  (1.467)  (1.554)    Yes  3.318**  3.330**  1.999      (1.385)  (1.390)  (1.412)    Pre-det.  2.646  1.022  0.936      (1.557)  (1.444)  (1.354)  Log (non-durable consumption per capita)  No  9.253***  7.694***  2.864  1995–2002    (2.535)  (2.123)  (1.789)    Yes  7.99***  7.459***  3.371      (2.774)  (2.270)  (2.012)    Pre-det.  8.452*  5.991*  1.864      (4.421)  (3.469)  (2.244)  Dependent variable  With controls?  Lowest tercile  Middle tercile  Upper tercile  Log (consumption per capita)  No  3.979**  3.936**  1.985  1988–2002    (1.420)  (1.447)  (1.471)    Yes  3.484***  3.718**  2.536**      (1.260)  (1.350)  (1.353)    Pre-det.  2.783*  1.541  1.305      (1.450)  (1.344)  (1.252)  Log (consumption per capita)  No  10.262***  7.703***  3.856**  1995–2002    (2.630)  (2.037)  (1.898)    Yes  8.742***  7.242***  4.476**      (2.815)  (2.156)  (2.184)    Pre-det.  9.069**  5.776*  3.459      (4.393)  (3.323)  (2.451)  Log (non-durable consumption per capita)  No  3.732**  3.413**  1.516  1988–2002    (1.536)  (1.467)  (1.554)    Yes  3.318**  3.330**  1.999      (1.385)  (1.390)  (1.412)    Pre-det.  2.646  1.022  0.936      (1.557)  (1.444)  (1.354)  Log (non-durable consumption per capita)  No  9.253***  7.694***  2.864  1995–2002    (2.535)  (2.123)  (1.789)    Yes  7.99***  7.459***  3.371      (2.774)  (2.270)  (2.012)    Pre-det.  8.452*  5.991*  1.864      (4.421)  (3.469)  (2.244)  Source: RCRE Household and Village Surveys, 1986–2002, and the RCRE 2004 Supplementary Village Governance Survey. Note: Standard errors in parentheses clustered at the village level. * indicates significance at the 10 percent level; ** indicates significant at the 5 percent level; *** indicates significance at the 1 percent level. Each cell represents a separate regression estimated in first differences with IV-GMM, and includes jointly significant village fixed effects and province-year effects to control for province-wide macroeconomic shocks. The share of migrants from the village is treated as endogenous, as are control variables, which follow the specification in columns 3 and 6 of table 2. View Large Migrant Networks, Investment, and Specialization Taken together, the results in tables 2 and 3 demonstrate that increases in household consumption per capita are associated with increasing out-migration, and that these effects are stronger among the poor. However, they do not shed light on how migration may have affected the investments of village residents, and the extent to which out-migration may contribute to productive activity within the local economy. The migrant labor market may relax local credit constraints through remittances, resulting in higher productive investment either in agriculture or non-agricultural self-employment, which contributes to increased earnings. Alternatively, households may respond to the relaxation of credit constraints by investing proceeds from migration in housing or consumer durables. Third, households may shift their labor to more productive activities, either directly as employees in migrant destinations, or through local employment as out-migration reduces the local labor supply. Understanding these mechanisms may have significant implications for rural policy in China. For example, if labor market policies relaxing restrictions on living in urban areas increase agricultural investment, policymakers charged with designing agricultural policy should take these increases into account. Alternatively, if loosening labor market restrictions does not affect agricultural investment, and agricultural policymakers have reason to believe there are still important credit market failures leading to low investment in high-return activities, then these failures should be approached more directly. To shed light on these mechanisms, we next examine the relationship between migration and both investment and labor supply. Investment To observe whether credit constraints are relaxed by migration, dependent variables measuring productive investment, or investment in housing and durables, are estimated using the following specification:   \begin{equation}{\rm{\Delta }}{K_{it}} = \beta {\rm{\Delta }}{M_{jt}} + {{\boldsymbol{d}}_{\boldsymbol{j}}} + p \otimes t + {\rm{\Delta }}{\varepsilon _{ijt}}\end{equation} (4)In alternate models, ΔKit is the change in log value of productive assets, the change in ln(1+value of productive assets related to agriculture), the change in ln(1+value of productive assets for non-agricultural activities), and the change in log of the imputed value of housing and durable goods. The coefficient β measures how each type of investment changes with change in the share of the village labor force employed as migrants; and these models are also estimated by the initial per capita consumption tercile. Results from this exercise make it clear that all groups of village residents increase investment in durable goods and housing with out-migration; and that such investment growth is larger among poorer households (table 4, panel A). Upgrades to housing with out-migration are consistent with other work; Mu and de Brauw (2015) find suggestive evidence that investment in housing, with improved access to tap water, may have contributed to improved nutritional status of children with migrant parents. With the exception of the model for non-agricultural productive assets among households in the upper tercile, coefficients on change in village migrant share are not significant at better than the 5 percent level. This association suggests that remitted earnings from migrants contribute sufficiently to the local economy to increase returns to local activity, perhaps related to home renovation or construction or other activities. An indirect benefit of out-migration may be expansion of non-agricultural productive activities among households in the upper tercile of the initial consumption distribution. Table 4. The Impact of Migration from the Village on Household Investment and Labor Allocation (All Models in First Differences) Dependent variable  Lowest tercile  Middle tercile  Upper tercile  A. Investment in physical assets, housing, and consumer durables  Log (productive assets per capita)  2.768  2.630  4.689*    (3.655)  (3.268)  (3.289)  Log (agricultural assets per capita + 1)  3.296  1.239  2.302    (4.346)  (3.186)  (2.967)  Log (non-agricultural assets per capita + 1)  4.626  4.794  11.067**    (3.377)  (3.882)  (5.018)  Log (durables plus housing per capita + 1)  6.76**  5.493**  3.421*    (2.383)  (2.184)  (1.882)  B. Share of migrants from village and household labor allocation  Total labor days per capita  −1.152  −1.086  −0.173    (1.716)  (1.998)  (2.055)  Agricultural (days per capita)  −7.407*  −2.790  0.471    (3.630)  (2.761)  (3.019)  Local off-farm (days per capita)  4.043  17.299**  16.506**    (6.847)  (7.726)  (7.967)  Self-employment (days per capita)  −1.439  −10.042  −7.592    (6.341)  (8.314)  (8.073)  Dependent variable  Lowest tercile  Middle tercile  Upper tercile  A. Investment in physical assets, housing, and consumer durables  Log (productive assets per capita)  2.768  2.630  4.689*    (3.655)  (3.268)  (3.289)  Log (agricultural assets per capita + 1)  3.296  1.239  2.302    (4.346)  (3.186)  (2.967)  Log (non-agricultural assets per capita + 1)  4.626  4.794  11.067**    (3.377)  (3.882)  (5.018)  Log (durables plus housing per capita + 1)  6.76**  5.493**  3.421*    (2.383)  (2.184)  (1.882)  B. Share of migrants from village and household labor allocation  Total labor days per capita  −1.152  −1.086  −0.173    (1.716)  (1.998)  (2.055)  Agricultural (days per capita)  −7.407*  −2.790  0.471    (3.630)  (2.761)  (3.019)  Local off-farm (days per capita)  4.043  17.299**  16.506**    (6.847)  (7.726)  (7.967)  Self-employment (days per capita)  −1.439  −10.042  −7.592    (6.341)  (8.314)  (8.073)  Source: RCRE Household and Village Surveys, 1986–2002, and the RCRE 2004 Supplementary Village Governance Survey. Note: Standard errors in parentheses clustered at the village level. * indicates significance at the 10 percent level; ** indicates significant at the 5 percent level; *** indicates significance at the 1 percent level. Each cell represents a separate regression estimated in first differences with IV-GMM, and includes jointly significant village fixed effects and province-year effects to control for province-wide macroeconomic shocks. The share of migrants from the village is treated as endogenous, but control variables are treated as exogenous, following the specification in columns 2 and 5 of table 2. View Large Labor Supply Increases in the ability to earn income from the migrant labor market may have negative effects on household labor supply if the wealth effect dominates the substitution effect. Households may have initially faced constraints in their ability to supply labor to the market, and if so, the expansion of migrant opportunity may allow them to increase income through expanded employment. Direct effects on labor supply through work in the migrant labor market may be complemented by indirect effects through depletion of the local labor force or demand for labor in the local construction and service sectors. To investigate this hypothesis, we modify equation (5) and use four measures of the logarithm of the number of labor days supplied (+1) as the dependent variable (table 4, panel B). These measures include the total labor days per capita, and the days per capita worked in agriculture, local wage labor, and non-agricultural self-employment. For the total labor days worked, no significant coefficients are found for any of the three terciles. Based on these results, one might assume that changes in village-level migration did not affect rural household labor supply. The absence of an increase in overall labor supply masks changes in labor allocation across activities, which demonstrate some interesting changes. Specifically, among the lowest tercile, the number of agricultural days worked per capita decline with increased migration; the estimated coefficient is significant at the 10 percent level and suggests a 7.4 percent decline with a one-percentage-point increase in migrant share at the sample mean. Meanwhile, the number of days allocated to the local off-farm market increases among the middle and upper terciles; these coefficients are fairly large and significant at the 5 percent level, and suggest that initially better-off households expand work in the local labor market, most likely as a result of more local activity with investment of remittances in productive activities. Summary The main results suggest that consumption per capita increases with migration, and impacts are stronger among poorer households than among households in the upper tercile of the initial consumption distribution. Later in the study period (after 1995), incomes also increase among all households, though the increase is faster among households that were initially poorer. Increases in consumption and income among poorer households appear to come directly through migration with reduced labor allocation to farming, whereas increases among households in the upper tercile appear to come more from stronger investments in non-agricultural business and increased labor allocation to non-agricultural activities. 4. Conclusion This paper shows the positive effect that internal migration in China has had on the consumption per capita of households remaining in migrant-sending communities, and also demonstrates that these effects are stronger for poorer households within villages. In common with McKenzie and Rapoport (2007) for Mexico, the increased ease of migration from villages of rural China is associated with decreasing inequality within communities. Further, increases in migration from rural China are associated with increased accumulation of housing wealth and consumer durables, with more pronounced growth among poorer households within rural communities. Consistent with both research on return migration in China and the international literature (Woodruff and Zenteño 2007; Yang 2008), this paper shows that migration is associated with more investment in non-agricultural production activities, but these investments are concentrated among more affluent households within villages. From the perspective of raising consumption, shifting labor allocation of the poor, and promoting non-agricultural investment, allowing the expansion of migration played a significant, positive role as a development strategy for China's rural areas. As this research was conducted using data from the period when rural-urban migration in China first accelerated, it is worth thinking about the relevance of these findings for China's ongoing process of urbanization over the past fifteen years. First, while there are announced plans to eliminate hukou, or household registration, these reforms have not been enacted to date. The hukou system does not prevent mobility, but does prevent full integration of rural migrants into China's urban economy. Although recent joint work conducted by the World Bank and China's Development Research Center has concluded that more complete integration of rural migrants into the urban economy will be important for sustaining growth, most migrants do not plan on remaining in cities (Meng 2012). In addition to discrimination in access to housing and education, migrants also found themselves without any employment protection in the face of sharp dislocations, as witnessed in the wake of the Global Financial Crisis (Huang et al. 2011). Indeed, Kong, Meng, and Zhang (2009) report evidence suggesting strong reluctance among those migrants laid off in the wake of the crisis to return to urban centers once the demand for migrant labor picked up again. In the face of continued discrimination in cities, the advent of e-commerce in China may offer a means for more productive investments of migrant earnings in home townships or counties. The e-commerce giant, Alibaba, which operates Taobao (a Chinese equivalent of eBay and Amazon), is explicitly attempting to expand its services to rural communities. There is also explicit interest in promoting the use of Taobao for rural residents to sell both processed agricultural outputs and non-agricultural goods such as handicrafts. Indeed, some observers have raised the question of whether Taobao might spur a “rural revolution” (e.g., Feng 2016). While the opening of markets through Taobao is unlikely to draw the vast majority of migrants back to rural areas, it does raise the prospect that the return to investments in small-scale activities in migrant-sending communities may rise over the next decade. Alan de Brauw is a Senior Research Fellow at the International Food Policy Research Institute (IFPRI); his e-mail address is a.debrauw@cgiar.org. John Giles (corresponding author) is a Lead Economist in the Development Research Group at the World Bank and a Research Fellow at IZA; his e-mail address is jgiles@worldbank.org. The research for this article was supported by the Knowledge for Change Program at the World Bank and the CGIAR research program on Policies, Institutions, and Markets. The authors are grateful to three anonymous referees and a host of participants in seminars and conferences where earlier versions of the paper were presented. A supplementary online appendix for this article can be found at The World Bank Economic Review website, and at https://sites.google.com/site/decrgjohngiles/publications one may find a “director's cut” version. 1 Supplementary online appendix table S1.1 provides summary statistics for selected years on key outcomes, and household- and village-level control variables. 2 de Brauw and Giles (2017) review various ways of using the years since ID distribution to identify the village migrant share, and fig. 5 of that paper shows that the quartic function predicts the same pattern observed from a fully non-parametric set of 19 dummy variables indicating years since ID distribution. As using the quartic requires fewer regressors than a fully non-parametric specification, the F-statistic on the first stage is higher, yielding reduced weak instrument bias. 3 During the period studied, it was rare for rural residents to be able to legally purchase housing or have the funds to do so, so the value of housing imputed from any rental payments is not problematic. A larger problem is imputing the consumption value of employer-provided housing, as it is certain that this dimension of consumption was not considered by enumerators. Education expenses are also excluded, as that consumption may only increase because migrants face higher education expenses for their children. 4 In supplementary online appendix S4, similar results are estimated and discussed for income per capita. References Benjamin D., Brandt L., Giles J.. 2005. “The Evolution of Income Inequality in Rural China.” Economic Development and Cultural Change  53 ( 4): 769– 824. Google Scholar CrossRef Search ADS   Benjamin D., Brandt L., Giles J.. 2011. “Did Higher Inequality Impede Growth in Rural China?” Economic Journal  121 ( 557): 1281– 309. Bryan G., Chowdhury S., Mobarak A. M.. 2014. “Under-Investment in a Profitable Technology: The Case of Seasonal Migration in Bangladesh.” Econometrica  82 ( 5): 1671– 748. Google Scholar CrossRef Search ADS   Cai F., Park A., Zhao Y.. 2008. “The Chinese Labor Market in the Reform Era.” In China's Great Economic Transformation , edited by Brandt L., Rawski T., 167– 214. Cambridge: Cambridge University Press. Google Scholar CrossRef Search ADS   de Brauw A., Giles J.. 2008. “Migrant Labor Markets and the Welfare of Households in the Developing World: Evidence from China.” Policy Research Working Paper No. 4585, World Bank, Washington, DC. de Brauw A., Giles J.. 2017. “Migrant Opportunity and the Educational Attainment of Youth in Rural China.” Journal of Human Resources  52 ( 1): 272– 311. Google Scholar CrossRef Search ADS   Du Y., Park A., Wang S.. 2005. “Is Migration Helping China's Poor?” Journal of Comparative Economics  33 ( 4): 688– 709. Google Scholar CrossRef Search ADS   Feng H. 2016. “Will ‘Taobao’ Villages Spur a Rural Revolution?” China Dialogue , May 24. Frijters P., Liu A., Meng X.. 2012. “Are Optimistic Expectations Keeping the Chinese Happy?” Journal of Economic Behavior and Organization  81 (1): 159– 71. Google Scholar CrossRef Search ADS   Giles J. 2006. “Is Life More Risky in the Open? Household Risk-Coping and the Opening of China's Labor Markets.” Journal of Development Economics  81 ( 1): 25– 60. Google Scholar CrossRef Search ADS   Giles J., Yoo K.. 2007. “Precautionary Behavior, Migrant Networks and Household Consumption Decisions: An Empirical Analysis Using Household Panel Data from Rural China.” Review of Economics and Statistics  89 ( 3): 534– 51. Google Scholar CrossRef Search ADS   Huang J., Zhi H., Huang Z., Rozelle S., Giles J.. 2011. “The Impact of the Global Financial Crisis on Off-Farm Employment and Earnings in Rural China.” World Development  39 ( 5): 797– 807. Google Scholar CrossRef Search ADS   Jalan J., Ravallion M.. 2002. “Geographic Poverty Traps? A Micro Model of Consumption Growth in Rural China.” Journal of Applied Econometrics  17 ( 4): 329– 46. Google Scholar CrossRef Search ADS   Kong S. T., Meng X., Zhang D. 2009. “The Impact of Economic Slowdown on Migrant Workers.” In China's New Place in a World in Crisis: Economic, Geopolitical and Environmental Dimensions , edited by Garnaut R., Song L., Woo W. T., 233– 60. Canberra: Asia Pacific Press. Liang Z., Ma Z.. 2004. “China's Floating Population: New Evidence from the 2000 Census.” Population and Development Review  30 ( 3): 467– 88. Google Scholar CrossRef Search ADS   Mallee H. 1995. “China's Household Registration System under Reform.” Development and Change  26 ( 1): 1– 29. Google Scholar CrossRef Search ADS   McKenzie D., Rapoport H.. 2007. “Network Effects and the Dynamics of Migration and Inequality: Theory and Evidence from Mexico.” Journal of Development Economics  84 ( 1): 1– 24. Google Scholar CrossRef Search ADS   Meng X. 2000. “Regional Wage Gap, Information Flow, and Rural-Urban Migration.” In Rural Labor Flows in China , edited by Zhao Y., West L., 251– 77. Berkeley: University of California Press. Meng X. 2012. “Labor Market Outcomes and Reforms in China.” Journal of Economic Perspectives  26 ( 4): 75– 102. Google Scholar CrossRef Search ADS   Mu R., de Brauw A.. 2015. “Migration and Young Child Nutrition: Evidence from Rural China.” Journal of Population Economics  28 ( 3): 631– 57. Google Scholar CrossRef Search ADS   Mu R., van de Walle D.. 2011. “Left Behind to Farm? Women's Labor Reallocation in Rural China.” Labour Economics  18 (S1): S83– 97. Google Scholar CrossRef Search ADS   Munshi K. 2003. “Networks in the Modern Economy: Mexican Migrants in the U.S. Labor Market.” Quarterly Journal of Economics  118 ( 2): 549– 99. Google Scholar CrossRef Search ADS   Munshi K. 2011. “Strength in Numbers: Networks as a Solution to Occupational Traps.” Review of Economic Studies  78 (3): 1069– 1101. Google Scholar CrossRef Search ADS   Munshi K., Rosenzweig M.. 2016. “Networks and Misallocation: Insurance, Migration and the Rural-Urban Wage Gap.” American Economic Review  106 ( 1): 46– 98. Google Scholar CrossRef Search ADS   Rozelle S., Guo L., Shen M., Hughart A., Giles J.. 1999. “Leaving China's Farms: Survey Results of New Paths and Remaining Hurdles to Rural Migration.” China Quarterly  158: 367– 93. Google Scholar CrossRef Search ADS   Taylor J. E., Rozelle S., de Brauw A.. 2003. “Migration and Incomes in Source Communities: A New Economics of Migration Perspective from China.” Economic Development and Cultural Change  52 ( 1): 75– 101. Google Scholar CrossRef Search ADS   Woodruff C., Zenteño R.. 2007. “Migration Networks and Microenterprises in Mexico.” Journal of Development Economics  82 ( 2): 509– 28. Google Scholar CrossRef Search ADS   Wooldridge J. M. 2010. Econometric Analysis of Cross Section and Panel Data , 2nd ed. Cambridge, MA: MIT Press. Yang D. 2008. “International Migration, Remittances and Household Investment: Evidence from Philippine Migrants’ Exchange Rate Shocks.” Economic Journal  118 ( March): 1– 40. Zhao Y. 2002. “Causes and Consequences of Return Migration: Recent Evidence from China.” Journal of Comparative Economics  30 ( 2): 376– 94. Google Scholar CrossRef Search ADS   Zhao Y. 2003. “The Role of Migrant Networks in Labor Migration: The Case of China.” Contemporary Economic Policy  21 ( 4): 500– 511. Google Scholar CrossRef Search ADS   © The Author 2018. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png The World Bank Economic Review Oxford University Press

Migrant Labor Markets and the Welfare of Rural Households in the Developing World: Evidence from China

Loading next page...
 
/lp/ou_press/migrant-labor-markets-and-the-welfare-of-rural-households-in-the-z608Ka7V2j
Publisher
Oxford University Press
Copyright
© The Author 2018. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com
ISSN
0258-6770
eISSN
1564-698X
D.O.I.
10.1093/wber/lhx023
Publisher site
See Article on Publisher Site

Abstract

Abstract Increased ability to migrate from China's rural villages contributed to significant increases in the consumption per capita of both non-durable and durable goods, and these effects were larger in magnitude for households that were relatively poor before the easing of restrictions to migration. With increased out-migration, poorer households invested more in housing and durable goods than rich households, while richer households invested significantly more in non-agricultural production assets. As migration became easier, increased participation in migrant employment was greater among poorer households on both the extensive and intensive margins, and poorer households reduced labor days in agriculture. migration, migrant networks, consumption growth, inequality In developing countries, barriers to the movement of labor are a common institutional feature that may contribute to geographic poverty traps. Whether maintained by formal institutions, by cultural or linguistic differences across regions, or simply by high transaction costs associated with finding migrant employment, constraints on the movement of labor within developing countries may reinforce an inefficient allocation of resources across regions and influence investment levels in poor areas, potentially hindering growth (Jalan and Ravallion 2002). When barriers to cross-regional labor mobility are removed, the resulting improved efficiency of resource allocation may have important consequences for rural living standards (Yang 2008). Remittances to household or family members remaining in rural areas may supplement income earned locally and directly reduce exposure to poverty. Migration may also have indirect effects on household and individual welfare within their home communities, either in the form of increased wages with the depletion of the local labor force, or through remittances from migrant employment that are invested in local production (e.g., Woodruff and Zenteño 2007). A growing body of research examines the impact of both international and internal migration on investment, growth, and other outcomes. Relevant for thinking about the identification strategy employed in this paper, considerable research has focused on the role of networks in facilitating migration (e.g., Munshi 2003) and influencing migration decisions (Munshi 2011; Munshi and Rosenzweig 2016), and the role that risk and liquidity constraints may play in shaping migration decisions (Bryan et al. 2014). Within the China literature, recent research has focused on the relationship between migration and poverty (Du, Park, and Wang 2005); migration, remittances, and household income (Taylor, Rozelle, and de Brauw 2003); migration and risk-coping (Giles 2006; Giles and Yoo 2007); and the prospects of migration and life satisfaction (Frijters, Liu, and Meng 2012), but identification of migration in each of these papers requires uncomfortable assumptions. To date, the published literature on the effects of out-migration has considered neither the distributional effects of migration within villages, nor the effects of out-migration on the sectoral distribution of household activities across the within-village wealth distribution. This paper addresses these gaps in the literature and makes four important contributions to our understanding of the impact of rural-urban migration on both consumption and sectoral choice in rural China. The main specification first highlights the positive effect of migration on annual changes in log consumption between 1988 and 2002: out-migration explains between 2 and 2.9 percent of the annual increase in consumption per capita, or between 65 and 93 percent of the annual consumption growth recorded in the sample villages between 1988 and 2002. As earlier work suggests that income from work outside the village (by both migrants and commuters) contributes to reducing inequality (Benjamin, Brandt, and Giles 2005), this paper exploits the opportunity to use matching village surveys to examine the local general equilibrium effects of migration on the per capita consumption of households across the initial consumption distribution. Whether households actually participate in migration or not, consumption grows more rapidly among households in the poor and middle terciles of the initial average consumption distribution. Maintaining a focus on within-village differences across the initial consumption distribution, the paper examines possible mechanisms through which migration influenced earnings, labor supply, and investment decisions. Where others have found that migration contributed to increased labor supply of residents remaining behind (e.g., Mu and van de Walle 2011), this paper finds that out-migration from the village leads to labor reallocation across activities, and labor allocation changes differ by wealth tercile. Agricultural labor days decline among poorer households, who also provide a relatively larger increase in days spent working outside the home township. More affluent households, by contrast, increase labor supply to local non-agricultural activities, potentially reflecting general equilibrium effects of increased local opportunity as migrants remit earnings. In terms of investment decisions, poorer households invest more in housing and durable goods, while more affluent households invest more in non-agricultural productive assets. If focusing on the period of rapid expansion in migration alone, from 1995 to 2002, the effects of migration on both labor supply and investment decisions are even more pronounced. A fourth contribution is the relatively unique instrumental variables (IV) strategy used to identify village-level effects of migration on household consumption and other outcomes. The approach takes advantage of an early reform to China's residential registration (hukou) system making it easier for rural migrants with national identification cards (IDs) to legally reside in cities after 1988. National IDs, first available to urban residents in 1984, were not available in all rural counties as of 1988. While allowing that the timing of ID distribution may be related to fixed unobserved characteristics of villages, the annual change in the share of the village population working as migrants outside the village is shown to be a non-linear function of the time since residents of a county received IDs. As migrant networks in cities take time to build up, this non-linearity should not be surprising. After controlling for household fixed effects, village-specific trends, and province-wide macroeconomic shocks, the change in the cost of migrating is identified by exploiting differences in the timing of access to IDs and the non-linearity in the relationship between the annual change in the village migrant share and the time since IDs were distributed. As IDs were not randomly distributed, the paper demonstrates that the timing of ID distribution was unrelated to other policy and economic shocks that might be correlated with both migration and well-being. 1. Background Over the period covered by the primary data source (1986 to 2002), rapid growth in the number of rural migrants moving to urban areas signaled a dramatic change in the nature of China's labor market. Estimates using the one percent sample from the 1990 and 2000 rounds of the Population Census and the 1995 one percent population survey suggest that the inter-county migrant population grew from just over 20 million in 1990 to 45 million in 1995 and 79 million by 2000 (Liang and Ma 2004). Surveys conducted by the National Bureau of Statistics (NBS) and the Ministry of Agriculture include more detailed retrospective information on past short-term migration, and suggest even higher levels of labor migration (Cai, Park, and Zhao 2008). Rural-Urban Migration in China, Migrant Networks, and Well-Being The use of migrant networks and employment referral in urban areas are important dimensions of China's rural-urban migration experience. Rozelle et al. (1999) emphasize that villages with more migrants in 1988 experienced more rapid migration growth by 1995. Zhao (2003) shows that the number of early migrants from a village is correlated with the probability that an individual with no prior migration experience will choose to migrate. Meng (2000) further suggests that variation in the size of migrant flows to different destinations can be partially explained by the size of the existing migrant population in potential destinations. Finally, appendix table A.1 of de Brauw and Giles (2017) notes that a 2001 survey of migrants in five large Chinese cities shows that more than 90 percent of migrants knew an acquaintance from their home village before migrating. Before labor mobility restrictions were relaxed, households in remote regions of rural China faced low returns to local economic activity, reinforcing geographic poverty traps (Jalan and Ravallion 2002). Over the period of study, household income per capita grew in both urban and rural areas, but at significantly higher rates in urban areas. After spatially deflating for cost-of-living differences, the urban-rural differential in per capita income increased from 38 percent to 71 percent by 2001. Thus, as documented by a considerable body of descriptive evidence, the growth of migration in China raises the possibility that migrant opportunity may be an important mechanism for poverty reduction in rural areas. Recent research in China suggests that migration is associated with higher incomes (Taylor, Rozelle, and de Brauw 2003; Du, Park, and Wang 2005), facilitates risk-coping and risk management (Giles 2006; Giles and Yoo 2007), is associated with higher levels of local investment in productive activities (Zhao 2002), and may alter household labor allocation decisions (Mu and Van de Walle 2011). Earlier research on rural-urban migration in China has suggested large impacts on the earnings of households with migrants, but some of the identification strategies employed likely produced estimates of migration impacts that suffer from significant bias. For example, Taylor, Rozelle, and de Brauw (2003) find that per capita incomes of households with migrant family members were 22, 26, and 29 percent higher at the 25th, 50th, and 75th percentiles of the sample (not within villages). However, the authors used a retrospective report of the village migration network size from seven years earlier as an instrument for household migration, and therefore cannot control for village fixed effects, such as proximity to cities (and markets) and other factors affecting the local economy, which will lead to an upward bias to estimates of the effect of migration on earnings. Another study examines both migrant participation and earnings across the earnings distribution (Du, Park, and Wang 2005), finding that poor households are unlikely to have migrants, but the main barrier to migrant employment is lack of capable laborers rather than other obstacles to working in urban areas. The authors estimate models in first differences, and identify migration through the one-year lag of the village migrant network, finding that households with migrants have per capita incomes 8.5 to 13.1 percent higher than those without migrants. A concern with this approach, however, is that shocks to the village economy, which affect migration decisions, have persistent effects over time. Both current migration and a household's position within the earnings distribution are likely to be affected by shocks reflected in prior migration from the village. The RCRE Household and Village Surveys The primary data sources used for analysis are the village and household surveys conducted by the Research Center for Rural Economy (RCRE) at China's Ministry of Agriculture from 1986 through 2002. The paper uses data from 88 villages in eight provinces (Anhui, Jilin, Jiangsu, Henan, Hunan, Shanxi, Sichuan, and Zhejiang) surveyed over the 16-year period, with an average of 6,305 households surveyed per year. Depending on village size, between 40 and 120 households were randomly surveyed in each village. Each village in the sample is in a different county, so county-level policies affect each village in this sample differently. The RCRE household survey collected detailed household-level information on incomes and expenditures, education, labor supply, asset ownership, land holdings, savings, formal and informal access to credit, and remittances. In common with the National Bureau of Statistics (NBS) Rural Household Survey, respondent households keep daily diaries of income and expenditures, and a resident administrator living in the county seat collects the diaries monthly. As in the NBS Rural Household Survey, estimates of the income and expenditures of household members currently living and working as migrants outside the household were included among household earnings and expenditures. This unusual feature of the survey is beneficial because household size does not change with out-migration. Details on the calculation of consumption from expenditure values and information on housing and durable goods are provided in supplementary online appendix S1, and in the analysis below, both income and consumption are deflated to 1986 values. The paper also makes use of an annual village survey enumerated from meetings with village accountants and other key informants that provides important village-level information on the local economy and on the land, production, and the work locations (migrant, local, and in the village) and occupations of registered residents. Time-varying information from this survey is important for characterizing connection to migrant networks, and when controlling for other time-varying village demographic and economic factors. The village-level survey includes all information on current registered residents who are working outside the village, and does not miss entire families who move (which was a very rare occurrence in rural China during this period). Finally, the identification strategy makes use of retrospective information, specifically the timing of ID availability, from a Village Governance Survey conducted by the authors in 2004. Trends in Migration, Consumption Growth, and Poverty The accompanying village survey includes questions asked annually of village leaders about the number of registered village residents working and living outside the village. For the analysis in this paper, all registered village residents who work outside the home county are considered migrants. Both the tremendous increase in migration from 1987 onward and heterogeneity across villages are evident in fig. 1. In 1987, an average of 3 percent of working-age laborers in RCRE villages worked outside their home counties, and this share rose steadily to 23 percent by 2003. Moreover, the share of working-age laborers working as migrants varies considerably by village. Whereas for some villages, only a small share of legal residents are employed as migrants, from other villages more than 50 percent of working-age adults are employed outside their home county by 2003. Figure 1. View largeDownload slide Share of Village Labor Force Employed as Migrants, by Year Source: RCRE Village Surveys 1986 to 2002. Note: The figure shows the share of registered village residents living and working outside the village and home county. Figure 1. View largeDownload slide Share of Village Labor Force Employed as Migrants, by Year Source: RCRE Village Surveys 1986 to 2002. Note: The figure shows the share of registered village residents living and working outside the village and home county. The relationship between migration and consumption is of central concern for the analysis. The linear fit of the relationship between annual changes in share of the village workforce employed as migrants (village migrant share) and growth in village average consumption suggest a positive relationship (fig. 2). The local polynomial fit, however, suggests the presence of non-linearities. If out-migration is driven by negative shocks or return migration by positive shocks, and both are correlated with movements in consumption, one should be concerned that migration and consumption are endogenous. Even if consumption grows with an increase in the number of residents employed as migrants, it is of particular policy interest to understand which residents within villages are experiencing increases in consumption. Figure 2. View largeDownload slide Village Average Consumption Growth and Change in Migrant Share of Village Population Source: RCRE Village and Household Surveys, 1986–2003. Note:Figure 2 shows the linear and local polynomial fits of the relationship between annual village average consumption growth and annual changes in the share of migrants from the village. Migrants are registered residents of the village who live and work outside the village and home county. Figure 2. View largeDownload slide Village Average Consumption Growth and Change in Migrant Share of Village Population Source: RCRE Village and Household Surveys, 1986–2003. Note:Figure 2 shows the linear and local polynomial fits of the relationship between annual village average consumption growth and annual changes in the share of migrants from the village. Migrants are registered residents of the village who live and work outside the village and home county. 2. Empirical Methodology Behind the estimation approach lies a simple model highlighting the direct and indirect mechanisms through which expanded migrant employment opportunities may affect household consumption (see de Brauw and Giles 2008). The migrant network, which facilitates job referral in distant destinations, may directly and indirectly affect permanent household income and thus also consumption. First, a larger village migrant network may facilitate higher family income from earnings in migrant destinations. The wealth effect eases credit constraints associated with accumulating assets for productive activities (both agricultural and non-agricultural) and “non-productive” uses (e.g., investments in housing and durable goods). Household consumption may also increase by relaxing a credit constraint that led households to consume less and save more in each period as a precaution against potential future production shocks. Second, an increase in the village migrant network size may affect the shadow price of household labor time. If leisure is a normal good, the net effect on family labor supply is indeterminate. A substitution effect will lead families to supply more labor to productive activities, and perhaps shift from farming into other higher-return activities, and an income effect may lead to a reduction in family labor supply. To flesh out the productive mechanisms through which increases in migration may lead to higher consumption, the impacts of migration on household labor supply are also examined, and disaggregated into three broad categories: agricultural, local non-agricultural, and non-agricultural self-employed work. Estimating the Effect of Migration on Consumption An empirical model consistent with the conceptual framework outlined above suggests the following linearized specification for household consumption, cijt (in logs):   \begin{equation} c_{ijt}= \beta {M_{jt}} + {\boldsymbol{X}}_{{\boldsymbol{ijt}}}^{\prime}\alpha + {\boldsymbol{Z}}_{{\boldsymbol{jt}}}^{\prime}\gamma + {u_i} + {v_j} + {{\boldsymbol{t}}_{\boldsymbol{j}}} + {\varepsilon _{ijt}} \end{equation} (1) The logarithm of per capita consumption for household i in village j during period t is a function of the share of the registered working-age village residents working outside the home county as migrants, Mjt. Household characteristics, Xijt, influence consumption through endowments, such as human capital, which affect household permanent income, and through demographic characteristics that influence consumption preferences. Time-varying village variables account for heterogeneity across villages in policies and economic conditions, Zjt, that may influence consumption through productivity. Village-specific trends, tj, account for underlying endowments and initial conditions in the village that may contribute to differences in consumption growth across villages. Finally, village- and household-level unobservables, vj and ui, respectively, related to location (village), to consumption preferences (household), and to the ease of household participation in the migrant labor market (household).1 Equation (1) is initially first-differenced to control for fixed effects at the household and village level:   \begin{equation} {\rm \Delta }{c_{it}} = \beta {{\rm \Delta }}{M_{jt}} + {{\rm \Delta }}{\boldsymbol{X}}_{{\boldsymbol{ijt}}}^{\prime}\alpha + {\rm{\Delta }}{\boldsymbol{Z}}_{{\boldsymbol{jt}}}^{\prime}\gamma + {{\boldsymbol{d}}_{\boldsymbol{j}}} + {{\rm \Delta }}{\varepsilon _{ijt}} \end{equation} (2)Differencing the village-specific trend leaves a vector of village dummy variables, dj, that control for differences in consumption growth trends across villages, which may be related to differences in proximity markets for labor or goods in townships or cities. Finally, province-wide macroeconomic shocks, driven by policy or market effects, might affect the relationship between household consumption and employment opportunities. Province-year interactions, p⊗t, are added in (3) to control for the presence of these shocks:   \begin{equation}{\rm{\Delta }}{c_{it}} = \beta {\rm{\Delta }}{M_{jt}} + {\rm{\Delta }}{\boldsymbol{X}}_{{\boldsymbol{it}}}^{\prime}\alpha + {\rm{\Delta }}{\boldsymbol{Z}}_{{\boldsymbol{jt}}}^{\prime}\gamma + {{\boldsymbol{d}}_{\boldsymbol{j}}} + p \otimes t + {\rm{\Delta }}{\varepsilon _{ijt}}\end{equation} (3)In equation (3), the coefficient of primary interest, β, measures the effect of the migrant labor market on consumption. Given descriptive evidence from the literature suggesting that the effect of migration may differ across the within-village wealth distribution (e.g., Benjamin, Brandt, and Giles 2005), the sample is next split into low, middle, and upper initial wealth terciles within each village. The tercile rank (low, middle, and high) is assigned to households based on average household consumption per capita for each household over the 1987–1989 period. Estimating models by this initial 1987–1989 consumption tercile facilitates examining the different impacts of migration across the initial wealth distribution. Endogeneity Concerns In equation (3), ΔMjt suffers from a well-known endogeneity problem. The village migrant share reflects factors affecting both changes in demand for migrant labor and changes in the labor supply decisions of migrants and potential migrants. Local shocks, for example, decrease household consumption per capita while increasing the relative return to migrant employment in more distant locations, potentially leading to an observed negative relationship between increases in migration and consumption growth. To identify the effect of migration on consumption, it is necessary to find an instrument correlated with the share of village residents working as migrants, but otherwise unrelated to factors affecting growth or negative shocks experienced by the village. The identification strategy employed makes use of two policy changes that, working together, affect the strength of migrant networks outside home counties but are plausibly unrelated to average village consumption growth. First, a new national ID card (shenfen zheng) was introduced in 1984. While urban residents received IDs in 1984, residents of most rural counties did not receive them immediately. In 1988, a reform of the residential registration system made it easier for migrants to gain legal temporary residence for work in cities under a “guest worker” system, but a national ID card was necessary to obtain a temporary residence permit (zanzu zheng) (Mallee 1995). While some counties made national IDs available to rural residents as early as 1984, others distributed them in 1988, and still others did not issue IDs until several years later. In a follow-up survey conducted with RCRE in 2004, local officials were asked when IDs had actually been issued to rural residents of the county. In the analysis sample, 41 of the 88 counties issued ID cards in 1988, but cards were issued as early as 1984 in three counties and as late as 1997 in one county. It is important to note that IDs were not necessary for migration, and large numbers of migrants live in cities without legal temporary residence cards. With temporary residence cards, however, migrants were under a “guest worker” system providing a more secure position in the destination community, and thus held better jobs and made up part of a longer-term migrant network in migrant destinations. Thus, ID distribution had two effects after the 1988 hukou reform. First, the costs of migrating to a city should fall after IDs become available. Second, as the quality of the potential migrant network improves with the years since IDs are available, the costs of finding migrant employment should continue to fall over time. The relative size of the migrant network should therefore be a function of both whether or not cards have been issued and the time since cards have been available to village residents. As we use the share of the village workforce working as migrants to proxy for the migrant network, the size of the potential network has an upper bound. Thus, we expect years-since-IDs-issued to have a non-linear relationship with the share of the village labor force working as migrants, as growth in the migrant network should decline after initially increasing with distribution of IDs. In fig. 3, a polynomial smoother shows the relationship between years since IDs were distributed and the share of working-age village residents working as migrants in year t. Within a couple of years after IDs are distributed, the share of the village labor force working as migrants grows sharply, and then slows after seven years. This pattern suggests non-linearity in the relationship between ID distribution and new participants in the village migrant labor force. Therefore, the primary instrument is specified as a dummy variable indicating that IDs had been issued, interacted with quartic functions of years since IDs were issued.2 Figure 3. View largeDownload slide Share of Village Out-Migrants, versus Years Since IDs Were Distributed Source: RCRE Village Surveys, 1986–2003, and Supplementary Village Governance Survey (2004). Note:Figure 3 shows the relationship between change in annual share of migrants from the village and number of years since ID cards became available in the county. Figure 3. View largeDownload slide Share of Village Out-Migrants, versus Years Since IDs Were Distributed Source: RCRE Village Surveys, 1986–2003, and Supplementary Village Governance Survey (2004). Note:Figure 3 shows the relationship between change in annual share of migrants from the village and number of years since ID cards became available in the county. As IDs were not randomly distributed, it is important to consider the plausible exogeneity of the identification strategy. More detail on the distribution of IDs and the conditional exogeneity of the timing is provided in supplementary online appendix S2, and additional discussion based on a four-province sample can be found in de Brauw and Giles (2017). A second, mechanical source of endogeneity that is well known in the panel data literature may arise, as the regressors included in ΔXit and ΔZjt might not be strictly exogenous. For example, income shocks that affect household consumption decisions may also have an impact on household composition, land characteristics, or village policy. To examine the robustness of our result, initial estimates of equation (3) exclude ΔXit and ΔZjt. If the years-since-IDs quartic is a valid instrument, meaning that it is uncorrelated with the changes in household and village characteristics that are now part of the unobservable, then this model will be identified and potentially endogenous regressors are not a concern. The simplified model is included for estimates over both the full sample and the models estimated by initial consumption tercile. When models include the village and household regressors, ΔXit and Zjt, both likely to explain some variation in consumption, they are treated in successive models as exogenous and then as predetermined but not strictly exogenous. For models in which regressors are treated as predetermined, first-differenced predetermined variables are instrumented with their t − 2 lagged levels [Xit − 2, Zit − 2], a standard panel data approach. Xit − 2 and Zit − 2 will be valid instruments if they are correlated with ΔXit and ΔZjt, but uncorrelated with any time-varying household unobservables remaining in the differenced error term, Δεit. Most importantly, these lagged values, Xit − 2 and Zit − 2, are uncorrelated with shocks that affected household demographic composition or the education composition of the household, and thus reduce one potential source of endogeneity. Use of lagged values does not resolve all endogeneity problems related to forward-looking household behavior. For example, a husband and wife making plans to migrate in the distant future may choose to co-reside with parents and in-laws who could farm land and watch children. In this sense, household structure is endogenous to future migration plans. These unobserved plans are considered a fixed preference that are differenced out in estimation. Further, precise unbiased estimates of the coefficients on these variables are not our central focus. Rather, these controls are included to allow for a more precise estimate of the effects of migration on household consumption, and later on labor supply and investment decisions. 3. Results The First Stage Before estimating equation (3), it is important to first establish that the instruments, period t − 2 values of a polynomial function of the years since ID cards were issued, are significantly related to the change in the share of working-age village residents working as migrants between period t − 1 and t. First-stage models were initially specified with years since IDs were issued as a quadratic, cubic, and quartic function, along with village and province-year dummies (table 1, columns 1 through 3). The quartic is preferred to the quadratic and the cubic due to the added flexibility in the functional form for the effects of ID card distribution on the migrant network. Table 1. Developing the First Stage: Timing of ID Card Distribution and Change in Share of Migrants in Village Population   (1)  (2)  (3)  (4)  (5)  (Years since IDs issued)t-2  −0.000  0.029***  0.044***  0.027***  0.026***    (0.001)  (0.002)  (0.004)  (0.004)  (0.004)  [(Years since IDs issued)t-2]2/10  −0.007***  −0.065***  −0.121***  −0.098***  −0.095***    (0.001)  (0.004)  (0.012)  (0.013)  (0.013)  [(Years since IDs issued)t-2]3/100    0.029***  0.092***  0.073***  0.069***      (0.002)  (0.013)  (0.014)  (0.014)  [(Years since IDs issued)t-2]4/1000      −0.022***  −0.017***  −0.015***        (0.005)  (0.005)  (0.005)  Two-period lag village controls included?  No  No  No  Yes  Yes  Two-period lag household controls included?  No  No  No  No  Yes  Number of observations  69,878  69,878  69,878  67,529  66,487  F-statistic, instruments  43.7  83.2  74.0  75.0  50.6    (1)  (2)  (3)  (4)  (5)  (Years since IDs issued)t-2  −0.000  0.029***  0.044***  0.027***  0.026***    (0.001)  (0.002)  (0.004)  (0.004)  (0.004)  [(Years since IDs issued)t-2]2/10  −0.007***  −0.065***  −0.121***  −0.098***  −0.095***    (0.001)  (0.004)  (0.012)  (0.013)  (0.013)  [(Years since IDs issued)t-2]3/100    0.029***  0.092***  0.073***  0.069***      (0.002)  (0.013)  (0.014)  (0.014)  [(Years since IDs issued)t-2]4/1000      −0.022***  −0.017***  −0.015***        (0.005)  (0.005)  (0.005)  Two-period lag village controls included?  No  No  No  Yes  Yes  Two-period lag household controls included?  No  No  No  No  Yes  Number of observations  69,878  69,878  69,878  67,529  66,487  F-statistic, instruments  43.7  83.2  74.0  75.0  50.6  Source: RCRE Household and Village Surveys, 1986–2002, and the RCRE 2004 Supplementary Village Governance Survey. Note: (1) All models include jointly significant controls for village and province* year effects, as well as other included instruments. (2) Dependent variable is change in share of working-age residents from the village living and working outside their home county between year t-1 and t. (3) Robust standard errors are cluster-corrected at the village, and there are 88 village clusters. (4) Two-period lag village controls include: total number of working-age laborers in registered village labor force, total village land, share of land in village in orchards, and share of total assets owned by the village collective. (5) Two-period lag household controls include: number of working-age laborers in the household, male working-age laborer share of household population, female working-age laborer share of household population, household land per capita, value of household productive assets, and average years of education of working-age laborers. ***indicates significance at the 1 percent level. View Large To anticipate models controlling for endogenous changes in village or household variables, two-period lags of village and household controls are added in columns 4 and 5, respectively. At the village level, the size of the village labor force is included to control for local returns to labor, the cultivable share of village land, total village land, and the share of land planted in orchards, which control for village land endowment and specialization in high-value crops, and the share of village assets held by collectives, which controls for the returns to capital outside agriculture as well as local government involvement in the economy. To control for household-level human and physical capital endowments, the number of working-age members of the household, the share of household members that are working-age males and females, respectively, land per capita, and the average education level of household adults are included. In both columns, the relationship between the migrant network variable and instruments for migration remains strong, and the F-statistic suggests that the complete set of instruments continues to have sufficient power to ease concerns over weak instrument bias after adding the full set of controls. While sufficiently strong, one might be concerned that the timing of ID distribution is also correlated with changes in a range of different village policies. These possibilities are examined in supplementary online appendix S3, and again there is no direct evidence that ID distribution is systematically related to policy changes that may affect consumption through channels other than migration from the village. The Effect of Migration on Household Consumption To begin the examination of the effects of migration on consumption, OLS models of the effects of migration on consumption in both levels and first differences are estimated across all three terciles of the initial consumption distribution. As one might expect if unobserved local shocks are an important factor driving initial migration decisions, the coefficient on migration is positive and insignificant in the OLS levels model, whether or not household- and village-level covariates are included (supplementary online appendix table S3.2). Similarly, when estimated by OLS in first differences, the coefficients on the change in migration are small and not statistically significant from zero. Next, the analysis uses IV-GMM models, which control for simultaneity bias and other unobservables potentially related to the migration measure. The weighting matrix used in the GMM estimator accounts for arbitrary heteroskedasticity and intra-cluster correlation, and it is asymptotically efficient in the presence of heteroskedasticity (Wooldridge 2010). Three first-differenced models are estimated (table 2). In column 1, estimates exclude village and household controls. Village and household controls are then added, treated first as exogenous (column 2) and then as predetermined (column 3), using t − 2 levels of the household and village controls as instruments. The Cragg-Donald F-statistic indicates that the bias in the IV coefficient is less than 5 percent of the OLS bias. Evidence from all three models suggests that the growth of the village migrant share in the labor force has a positive, statistically significant effect on consumption among all households. Table 2. Migration and Household Consumption in Migrant-Sending Villages (All Models in First Differences)   Ln(consumption per capita)  Ln(non-durable consumption per capita)    (1)  (2)  (3)  (4)  (5)  (6)  Share of migrants in village population  3.563***  3.602***  2.507**  3.285***  3.434***  2.308**    (1.188)  (1.103)  (1.132)  (1.268)  (1.184)  (1.131)  Village-level control variables  Village labor force    −0.001  0.003    −0.003  0.000      (0.005)  (0.014)    (0.005)  (0.015)  Cultivable share of village land    0.269***  0.819**    0.270***  0.796**      (0.092)  (0.382)    (0.093)  (0.380)  Total village land    0.016*  0.127***    0.015  0.129***      (0.009)  (0.046)    (0.010)  (0.047)  Share of assets owned by village collective    −0.014  −0.157    −0.015  −0.169      (0.028)  (0.165)    (0.032)  (0.175)  Share of village land in orchards    −0.163  0.420    −0.100  0.443      (0.180)  (0.777)    (0.181)  (0.780)  Household-level control variables  Working-age male share of household population    −0.070**  0.427***    −0.064*  0.479***      (0.032)  (0.127)    (0.036)  (0.151)  Working-age female share of household population    −0.076**  0.131    −0.066*  0.152      (0.031)  (0.127)    (0.035)  (0.149)  Number of working-age laborers in the household    −0.038***  −0.003    −0.026***  0.023**      (0.004)  (0.009)    (0.005)  (0.010)  Cultivable land per capita    0.085***  0.114***    0.085***  0.125***      (0.008)  (0.021)    (0.009)  (0.023)  Household average years of education    0.004**  −0.012**    0.005***  −0.013**      (0.002)  (0.005)    (0.002)  (0.005)  Village, HH controls predetermined?    No  Yes    No  Yes  Regression statistics  Hansen J-statistic  1.860  2.235  4.130  1.244  1.564  4.569  p-value, J-statistic  0.602  0.525  0.248  0.742  0.668  0.206  Cragg-Donald F-statistic  87.23  95.66  32.68  87.56  95.76  32.66  Number of clusters  88  88  88  88  88  88  Number of observations  69,878  68,110  65,607  69,834  68,072  65,570    Ln(consumption per capita)  Ln(non-durable consumption per capita)    (1)  (2)  (3)  (4)  (5)  (6)  Share of migrants in village population  3.563***  3.602***  2.507**  3.285***  3.434***  2.308**    (1.188)  (1.103)  (1.132)  (1.268)  (1.184)  (1.131)  Village-level control variables  Village labor force    −0.001  0.003    −0.003  0.000      (0.005)  (0.014)    (0.005)  (0.015)  Cultivable share of village land    0.269***  0.819**    0.270***  0.796**      (0.092)  (0.382)    (0.093)  (0.380)  Total village land    0.016*  0.127***    0.015  0.129***      (0.009)  (0.046)    (0.010)  (0.047)  Share of assets owned by village collective    −0.014  −0.157    −0.015  −0.169      (0.028)  (0.165)    (0.032)  (0.175)  Share of village land in orchards    −0.163  0.420    −0.100  0.443      (0.180)  (0.777)    (0.181)  (0.780)  Household-level control variables  Working-age male share of household population    −0.070**  0.427***    −0.064*  0.479***      (0.032)  (0.127)    (0.036)  (0.151)  Working-age female share of household population    −0.076**  0.131    −0.066*  0.152      (0.031)  (0.127)    (0.035)  (0.149)  Number of working-age laborers in the household    −0.038***  −0.003    −0.026***  0.023**      (0.004)  (0.009)    (0.005)  (0.010)  Cultivable land per capita    0.085***  0.114***    0.085***  0.125***      (0.008)  (0.021)    (0.009)  (0.023)  Household average years of education    0.004**  −0.012**    0.005***  −0.013**      (0.002)  (0.005)    (0.002)  (0.005)  Village, HH controls predetermined?    No  Yes    No  Yes  Regression statistics  Hansen J-statistic  1.860  2.235  4.130  1.244  1.564  4.569  p-value, J-statistic  0.602  0.525  0.248  0.742  0.668  0.206  Cragg-Donald F-statistic  87.23  95.66  32.68  87.56  95.76  32.66  Number of clusters  88  88  88  88  88  88  Number of observations  69,878  68,110  65,607  69,834  68,072  65,570  Source: RCRE Household and Village Surveys, 1986–2002, and the RCRE 2004 Supplementary Village Governance Survey. Note: All models are run in first differences and include jointly significant village fixed effects to control for village-specific trends, and province-year effects to control for province-wide macroeconomic shocks. Standard errors clustered at the village. *indicates significance at the 10 percent level; **indicates significance at the 5 percent level; ***indicates significance at the 1 percent level. View Large In all three specifications, the coefficients on the village migrant share are relatively close to one another. Coefficients in columns 1 and 2 suggest that a one-percentage-point increase in migration at the village level is associated with a 3.6 percent increase in household consumption per capita, while column 3 (controlling for the dynamic endogeneity of control variables) indicates a 2.5 percent increase. Average village migration increased from 0.012 to 0.126 of the village workforce from 1988 to 2002, so a one-percentage-point rise is slightly higher than the average annual increase in the village migrant share, which is 0.8 percent. At the average migrant share, these estimates imply that migration is associated with a 2 percent annual increase in household per capita consumption in those models treating controls as predetermined, or 2.9 percent when treated as exogenous. With an average of 3.1 percent consumption growth per annum in the RCRE villages, increased ability to migrate thus explains from 65 to 93 percent of the average per capita consumption growth in this sample. Recall that the consumption measure includes the flow of services from housing and durable goods. Given that the value of this flow may be imputed incorrectly for migrant family members living outside the household, the effect of migration on non-durable consumption growth is shown in columns 4 through 6, with education expenses and the flow from housing and durable goods excluded.3 The coefficient on the village migrant share is slightly smaller (suggesting a 2.3 to 3.4 percent increase), but remains statistically significant at the 5 percent level or better. An additional concern might be that the results are driven by comparisons between villages with and without migration. At the beginning of the study period, some villages had no migrants, though by 1996 all 88 villages had some migrants. This possibility is explored by dropping the early years of villages without migrants (supplementary online appendix table S3.3, columns 1 through 3), and there is little difference from the main specifications shown in table 2. In any long panel, one should also be concerned over whether attrition might be biasing results. The average annual household attrition rate is 5.8 percent; by 2002, the average probability that original households are still in the sample is 62 percent. In this paper, as in two others using different samples from this data source (Benjamin, Brandt, and Giles 2011; Giles 2006), there is no evidence of a statistically significant relationship between the village migrant share (or changes in the village migrant share) and attrition in the current or following year, nor are consumption per capita and attrition correlated. To check further for potential attrition bias, we weight each observation by the inverse of the cumulative probability of survival to the current year, with probabilities for survival in each year calculated using household and village characteristics and a province fixed effect (Wooldridge 2010, chapter 19). After controlling for attrition, there is little change in the coefficients estimated for the three main models (columns 4 through 6 of supplementary online appendix table S3.3), suggesting that the main results do not suffer from attrition bias. Who Benefits from Migration? By measuring effects of new migration from the village on household consumption, and with consumption of migrants included in the measure of household expenditures, both made possible through interesting features of the data source, we capture the general equilibrium effects of increasing out-migration. Starting in 1995, the household-level questionnaires began to enumerate whether or not a household has a migrant, so direct household-level participation in migration can be examined, also by terciles of the initial distribution of consumption per capita. For the period beginning in 1995, fig. 4A shows the share of households with at least one household member working and living outside their home township. For each year through 2001, households in the lowest tercile were roughly five percentage points more likely than upper-tercile households to have a registered member living and working outside the village. Further, the lower tercile averages about 12 more days of migrant employment per capita than the upper tercile (fig. 4B). These differences are statistically significant, and remain so after controlling for village fixed effects (supplementary online appendix table S4.2). Figure 4. View largeDownload slide A. Share of Households with Member Working Outside the Township, by Initial Consumption Tercile, 1995–2002 B. Average Days Employed Outside the Township Source: RCRE Household Surveys, 1995–2002. Figure 4. View largeDownload slide A. Share of Households with Member Working Outside the Township, by Initial Consumption Tercile, 1995–2002 B. Average Days Employed Outside the Township Source: RCRE Household Surveys, 1995–2002. Splitting the sample by the initial consumption terciles, equation (3) is re-estimated by tercile, with consumption and non-durable consumption as dependent variables (table 3).4 Coefficients are estimated by tercile for the full 1988–2002 period and for a shorter 1995–2002 period, after migration had begun to grow more rapidly. Households that were initially in the lowest within-village tercile experience more rapid growth in consumption per capita and non-durable consumption per capita than households in the upper tercile. In fact, estimates for the upper tercile for the entire 1988–2002 period show that neither total nor non-durable consumption per capita grows significantly as a result of village-level out-migration. For the lowest and middle within-village tercile, however, a one-percentage-point increase in migration over the 1988–2002 period is associated with a 3 to 4 percent increase in both consumption per capita and non-durable consumption per capita. Over the period when out-migration was growing rapidly (1995–2002), coefficient estimates suggest more rapid consumption growth for both the lowest and the middle tercile (10 percent and 7 percent, respectively, with a one-percentage-point increase in migration), and also suggest positive consumption growth for the upper tercile (4 percent), though the latter increase appears attributable to the accumulation of durable goods and housing. Table 3. Village-Level Migration and Consumption across the Initial Consumption Distribution (All Models in First Differences) Dependent variable  With controls?  Lowest tercile  Middle tercile  Upper tercile  Log (consumption per capita)  No  3.979**  3.936**  1.985  1988–2002    (1.420)  (1.447)  (1.471)    Yes  3.484***  3.718**  2.536**      (1.260)  (1.350)  (1.353)    Pre-det.  2.783*  1.541  1.305      (1.450)  (1.344)  (1.252)  Log (consumption per capita)  No  10.262***  7.703***  3.856**  1995–2002    (2.630)  (2.037)  (1.898)    Yes  8.742***  7.242***  4.476**      (2.815)  (2.156)  (2.184)    Pre-det.  9.069**  5.776*  3.459      (4.393)  (3.323)  (2.451)  Log (non-durable consumption per capita)  No  3.732**  3.413**  1.516  1988–2002    (1.536)  (1.467)  (1.554)    Yes  3.318**  3.330**  1.999      (1.385)  (1.390)  (1.412)    Pre-det.  2.646  1.022  0.936      (1.557)  (1.444)  (1.354)  Log (non-durable consumption per capita)  No  9.253***  7.694***  2.864  1995–2002    (2.535)  (2.123)  (1.789)    Yes  7.99***  7.459***  3.371      (2.774)  (2.270)  (2.012)    Pre-det.  8.452*  5.991*  1.864      (4.421)  (3.469)  (2.244)  Dependent variable  With controls?  Lowest tercile  Middle tercile  Upper tercile  Log (consumption per capita)  No  3.979**  3.936**  1.985  1988–2002    (1.420)  (1.447)  (1.471)    Yes  3.484***  3.718**  2.536**      (1.260)  (1.350)  (1.353)    Pre-det.  2.783*  1.541  1.305      (1.450)  (1.344)  (1.252)  Log (consumption per capita)  No  10.262***  7.703***  3.856**  1995–2002    (2.630)  (2.037)  (1.898)    Yes  8.742***  7.242***  4.476**      (2.815)  (2.156)  (2.184)    Pre-det.  9.069**  5.776*  3.459      (4.393)  (3.323)  (2.451)  Log (non-durable consumption per capita)  No  3.732**  3.413**  1.516  1988–2002    (1.536)  (1.467)  (1.554)    Yes  3.318**  3.330**  1.999      (1.385)  (1.390)  (1.412)    Pre-det.  2.646  1.022  0.936      (1.557)  (1.444)  (1.354)  Log (non-durable consumption per capita)  No  9.253***  7.694***  2.864  1995–2002    (2.535)  (2.123)  (1.789)    Yes  7.99***  7.459***  3.371      (2.774)  (2.270)  (2.012)    Pre-det.  8.452*  5.991*  1.864      (4.421)  (3.469)  (2.244)  Source: RCRE Household and Village Surveys, 1986–2002, and the RCRE 2004 Supplementary Village Governance Survey. Note: Standard errors in parentheses clustered at the village level. * indicates significance at the 10 percent level; ** indicates significant at the 5 percent level; *** indicates significance at the 1 percent level. Each cell represents a separate regression estimated in first differences with IV-GMM, and includes jointly significant village fixed effects and province-year effects to control for province-wide macroeconomic shocks. The share of migrants from the village is treated as endogenous, as are control variables, which follow the specification in columns 3 and 6 of table 2. View Large Migrant Networks, Investment, and Specialization Taken together, the results in tables 2 and 3 demonstrate that increases in household consumption per capita are associated with increasing out-migration, and that these effects are stronger among the poor. However, they do not shed light on how migration may have affected the investments of village residents, and the extent to which out-migration may contribute to productive activity within the local economy. The migrant labor market may relax local credit constraints through remittances, resulting in higher productive investment either in agriculture or non-agricultural self-employment, which contributes to increased earnings. Alternatively, households may respond to the relaxation of credit constraints by investing proceeds from migration in housing or consumer durables. Third, households may shift their labor to more productive activities, either directly as employees in migrant destinations, or through local employment as out-migration reduces the local labor supply. Understanding these mechanisms may have significant implications for rural policy in China. For example, if labor market policies relaxing restrictions on living in urban areas increase agricultural investment, policymakers charged with designing agricultural policy should take these increases into account. Alternatively, if loosening labor market restrictions does not affect agricultural investment, and agricultural policymakers have reason to believe there are still important credit market failures leading to low investment in high-return activities, then these failures should be approached more directly. To shed light on these mechanisms, we next examine the relationship between migration and both investment and labor supply. Investment To observe whether credit constraints are relaxed by migration, dependent variables measuring productive investment, or investment in housing and durables, are estimated using the following specification:   \begin{equation}{\rm{\Delta }}{K_{it}} = \beta {\rm{\Delta }}{M_{jt}} + {{\boldsymbol{d}}_{\boldsymbol{j}}} + p \otimes t + {\rm{\Delta }}{\varepsilon _{ijt}}\end{equation} (4)In alternate models, ΔKit is the change in log value of productive assets, the change in ln(1+value of productive assets related to agriculture), the change in ln(1+value of productive assets for non-agricultural activities), and the change in log of the imputed value of housing and durable goods. The coefficient β measures how each type of investment changes with change in the share of the village labor force employed as migrants; and these models are also estimated by the initial per capita consumption tercile. Results from this exercise make it clear that all groups of village residents increase investment in durable goods and housing with out-migration; and that such investment growth is larger among poorer households (table 4, panel A). Upgrades to housing with out-migration are consistent with other work; Mu and de Brauw (2015) find suggestive evidence that investment in housing, with improved access to tap water, may have contributed to improved nutritional status of children with migrant parents. With the exception of the model for non-agricultural productive assets among households in the upper tercile, coefficients on change in village migrant share are not significant at better than the 5 percent level. This association suggests that remitted earnings from migrants contribute sufficiently to the local economy to increase returns to local activity, perhaps related to home renovation or construction or other activities. An indirect benefit of out-migration may be expansion of non-agricultural productive activities among households in the upper tercile of the initial consumption distribution. Table 4. The Impact of Migration from the Village on Household Investment and Labor Allocation (All Models in First Differences) Dependent variable  Lowest tercile  Middle tercile  Upper tercile  A. Investment in physical assets, housing, and consumer durables  Log (productive assets per capita)  2.768  2.630  4.689*    (3.655)  (3.268)  (3.289)  Log (agricultural assets per capita + 1)  3.296  1.239  2.302    (4.346)  (3.186)  (2.967)  Log (non-agricultural assets per capita + 1)  4.626  4.794  11.067**    (3.377)  (3.882)  (5.018)  Log (durables plus housing per capita + 1)  6.76**  5.493**  3.421*    (2.383)  (2.184)  (1.882)  B. Share of migrants from village and household labor allocation  Total labor days per capita  −1.152  −1.086  −0.173    (1.716)  (1.998)  (2.055)  Agricultural (days per capita)  −7.407*  −2.790  0.471    (3.630)  (2.761)  (3.019)  Local off-farm (days per capita)  4.043  17.299**  16.506**    (6.847)  (7.726)  (7.967)  Self-employment (days per capita)  −1.439  −10.042  −7.592    (6.341)  (8.314)  (8.073)  Dependent variable  Lowest tercile  Middle tercile  Upper tercile  A. Investment in physical assets, housing, and consumer durables  Log (productive assets per capita)  2.768  2.630  4.689*    (3.655)  (3.268)  (3.289)  Log (agricultural assets per capita + 1)  3.296  1.239  2.302    (4.346)  (3.186)  (2.967)  Log (non-agricultural assets per capita + 1)  4.626  4.794  11.067**    (3.377)  (3.882)  (5.018)  Log (durables plus housing per capita + 1)  6.76**  5.493**  3.421*    (2.383)  (2.184)  (1.882)  B. Share of migrants from village and household labor allocation  Total labor days per capita  −1.152  −1.086  −0.173    (1.716)  (1.998)  (2.055)  Agricultural (days per capita)  −7.407*  −2.790  0.471    (3.630)  (2.761)  (3.019)  Local off-farm (days per capita)  4.043  17.299**  16.506**    (6.847)  (7.726)  (7.967)  Self-employment (days per capita)  −1.439  −10.042  −7.592    (6.341)  (8.314)  (8.073)  Source: RCRE Household and Village Surveys, 1986–2002, and the RCRE 2004 Supplementary Village Governance Survey. Note: Standard errors in parentheses clustered at the village level. * indicates significance at the 10 percent level; ** indicates significant at the 5 percent level; *** indicates significance at the 1 percent level. Each cell represents a separate regression estimated in first differences with IV-GMM, and includes jointly significant village fixed effects and province-year effects to control for province-wide macroeconomic shocks. The share of migrants from the village is treated as endogenous, but control variables are treated as exogenous, following the specification in columns 2 and 5 of table 2. View Large Labor Supply Increases in the ability to earn income from the migrant labor market may have negative effects on household labor supply if the wealth effect dominates the substitution effect. Households may have initially faced constraints in their ability to supply labor to the market, and if so, the expansion of migrant opportunity may allow them to increase income through expanded employment. Direct effects on labor supply through work in the migrant labor market may be complemented by indirect effects through depletion of the local labor force or demand for labor in the local construction and service sectors. To investigate this hypothesis, we modify equation (5) and use four measures of the logarithm of the number of labor days supplied (+1) as the dependent variable (table 4, panel B). These measures include the total labor days per capita, and the days per capita worked in agriculture, local wage labor, and non-agricultural self-employment. For the total labor days worked, no significant coefficients are found for any of the three terciles. Based on these results, one might assume that changes in village-level migration did not affect rural household labor supply. The absence of an increase in overall labor supply masks changes in labor allocation across activities, which demonstrate some interesting changes. Specifically, among the lowest tercile, the number of agricultural days worked per capita decline with increased migration; the estimated coefficient is significant at the 10 percent level and suggests a 7.4 percent decline with a one-percentage-point increase in migrant share at the sample mean. Meanwhile, the number of days allocated to the local off-farm market increases among the middle and upper terciles; these coefficients are fairly large and significant at the 5 percent level, and suggest that initially better-off households expand work in the local labor market, most likely as a result of more local activity with investment of remittances in productive activities. Summary The main results suggest that consumption per capita increases with migration, and impacts are stronger among poorer households than among households in the upper tercile of the initial consumption distribution. Later in the study period (after 1995), incomes also increase among all households, though the increase is faster among households that were initially poorer. Increases in consumption and income among poorer households appear to come directly through migration with reduced labor allocation to farming, whereas increases among households in the upper tercile appear to come more from stronger investments in non-agricultural business and increased labor allocation to non-agricultural activities. 4. Conclusion This paper shows the positive effect that internal migration in China has had on the consumption per capita of households remaining in migrant-sending communities, and also demonstrates that these effects are stronger for poorer households within villages. In common with McKenzie and Rapoport (2007) for Mexico, the increased ease of migration from villages of rural China is associated with decreasing inequality within communities. Further, increases in migration from rural China are associated with increased accumulation of housing wealth and consumer durables, with more pronounced growth among poorer households within rural communities. Consistent with both research on return migration in China and the international literature (Woodruff and Zenteño 2007; Yang 2008), this paper shows that migration is associated with more investment in non-agricultural production activities, but these investments are concentrated among more affluent households within villages. From the perspective of raising consumption, shifting labor allocation of the poor, and promoting non-agricultural investment, allowing the expansion of migration played a significant, positive role as a development strategy for China's rural areas. As this research was conducted using data from the period when rural-urban migration in China first accelerated, it is worth thinking about the relevance of these findings for China's ongoing process of urbanization over the past fifteen years. First, while there are announced plans to eliminate hukou, or household registration, these reforms have not been enacted to date. The hukou system does not prevent mobility, but does prevent full integration of rural migrants into China's urban economy. Although recent joint work conducted by the World Bank and China's Development Research Center has concluded that more complete integration of rural migrants into the urban economy will be important for sustaining growth, most migrants do not plan on remaining in cities (Meng 2012). In addition to discrimination in access to housing and education, migrants also found themselves without any employment protection in the face of sharp dislocations, as witnessed in the wake of the Global Financial Crisis (Huang et al. 2011). Indeed, Kong, Meng, and Zhang (2009) report evidence suggesting strong reluctance among those migrants laid off in the wake of the crisis to return to urban centers once the demand for migrant labor picked up again. In the face of continued discrimination in cities, the advent of e-commerce in China may offer a means for more productive investments of migrant earnings in home townships or counties. The e-commerce giant, Alibaba, which operates Taobao (a Chinese equivalent of eBay and Amazon), is explicitly attempting to expand its services to rural communities. There is also explicit interest in promoting the use of Taobao for rural residents to sell both processed agricultural outputs and non-agricultural goods such as handicrafts. Indeed, some observers have raised the question of whether Taobao might spur a “rural revolution” (e.g., Feng 2016). While the opening of markets through Taobao is unlikely to draw the vast majority of migrants back to rural areas, it does raise the prospect that the return to investments in small-scale activities in migrant-sending communities may rise over the next decade. Alan de Brauw is a Senior Research Fellow at the International Food Policy Research Institute (IFPRI); his e-mail address is a.debrauw@cgiar.org. John Giles (corresponding author) is a Lead Economist in the Development Research Group at the World Bank and a Research Fellow at IZA; his e-mail address is jgiles@worldbank.org. The research for this article was supported by the Knowledge for Change Program at the World Bank and the CGIAR research program on Policies, Institutions, and Markets. The authors are grateful to three anonymous referees and a host of participants in seminars and conferences where earlier versions of the paper were presented. A supplementary online appendix for this article can be found at The World Bank Economic Review website, and at https://sites.google.com/site/decrgjohngiles/publications one may find a “director's cut” version. 1 Supplementary online appendix table S1.1 provides summary statistics for selected years on key outcomes, and household- and village-level control variables. 2 de Brauw and Giles (2017) review various ways of using the years since ID distribution to identify the village migrant share, and fig. 5 of that paper shows that the quartic function predicts the same pattern observed from a fully non-parametric set of 19 dummy variables indicating years since ID distribution. As using the quartic requires fewer regressors than a fully non-parametric specification, the F-statistic on the first stage is higher, yielding reduced weak instrument bias. 3 During the period studied, it was rare for rural residents to be able to legally purchase housing or have the funds to do so, so the value of housing imputed from any rental payments is not problematic. A larger problem is imputing the consumption value of employer-provided housing, as it is certain that this dimension of consumption was not considered by enumerators. Education expenses are also excluded, as that consumption may only increase because migrants face higher education expenses for their children. 4 In supplementary online appendix S4, similar results are estimated and discussed for income per capita. References Benjamin D., Brandt L., Giles J.. 2005. “The Evolution of Income Inequality in Rural China.” Economic Development and Cultural Change  53 ( 4): 769– 824. Google Scholar CrossRef Search ADS   Benjamin D., Brandt L., Giles J.. 2011. “Did Higher Inequality Impede Growth in Rural China?” Economic Journal  121 ( 557): 1281– 309. Bryan G., Chowdhury S., Mobarak A. M.. 2014. “Under-Investment in a Profitable Technology: The Case of Seasonal Migration in Bangladesh.” Econometrica  82 ( 5): 1671– 748. Google Scholar CrossRef Search ADS   Cai F., Park A., Zhao Y.. 2008. “The Chinese Labor Market in the Reform Era.” In China's Great Economic Transformation , edited by Brandt L., Rawski T., 167– 214. Cambridge: Cambridge University Press. Google Scholar CrossRef Search ADS   de Brauw A., Giles J.. 2008. “Migrant Labor Markets and the Welfare of Households in the Developing World: Evidence from China.” Policy Research Working Paper No. 4585, World Bank, Washington, DC. de Brauw A., Giles J.. 2017. “Migrant Opportunity and the Educational Attainment of Youth in Rural China.” Journal of Human Resources  52 ( 1): 272– 311. Google Scholar CrossRef Search ADS   Du Y., Park A., Wang S.. 2005. “Is Migration Helping China's Poor?” Journal of Comparative Economics  33 ( 4): 688– 709. Google Scholar CrossRef Search ADS   Feng H. 2016. “Will ‘Taobao’ Villages Spur a Rural Revolution?” China Dialogue , May 24. Frijters P., Liu A., Meng X.. 2012. “Are Optimistic Expectations Keeping the Chinese Happy?” Journal of Economic Behavior and Organization  81 (1): 159– 71. Google Scholar CrossRef Search ADS   Giles J. 2006. “Is Life More Risky in the Open? Household Risk-Coping and the Opening of China's Labor Markets.” Journal of Development Economics  81 ( 1): 25– 60. Google Scholar CrossRef Search ADS   Giles J., Yoo K.. 2007. “Precautionary Behavior, Migrant Networks and Household Consumption Decisions: An Empirical Analysis Using Household Panel Data from Rural China.” Review of Economics and Statistics  89 ( 3): 534– 51. Google Scholar CrossRef Search ADS   Huang J., Zhi H., Huang Z., Rozelle S., Giles J.. 2011. “The Impact of the Global Financial Crisis on Off-Farm Employment and Earnings in Rural China.” World Development  39 ( 5): 797– 807. Google Scholar CrossRef Search ADS   Jalan J., Ravallion M.. 2002. “Geographic Poverty Traps? A Micro Model of Consumption Growth in Rural China.” Journal of Applied Econometrics  17 ( 4): 329– 46. Google Scholar CrossRef Search ADS   Kong S. T., Meng X., Zhang D. 2009. “The Impact of Economic Slowdown on Migrant Workers.” In China's New Place in a World in Crisis: Economic, Geopolitical and Environmental Dimensions , edited by Garnaut R., Song L., Woo W. T., 233– 60. Canberra: Asia Pacific Press. Liang Z., Ma Z.. 2004. “China's Floating Population: New Evidence from the 2000 Census.” Population and Development Review  30 ( 3): 467– 88. Google Scholar CrossRef Search ADS   Mallee H. 1995. “China's Household Registration System under Reform.” Development and Change  26 ( 1): 1– 29. Google Scholar CrossRef Search ADS   McKenzie D., Rapoport H.. 2007. “Network Effects and the Dynamics of Migration and Inequality: Theory and Evidence from Mexico.” Journal of Development Economics  84 ( 1): 1– 24. Google Scholar CrossRef Search ADS   Meng X. 2000. “Regional Wage Gap, Information Flow, and Rural-Urban Migration.” In Rural Labor Flows in China , edited by Zhao Y., West L., 251– 77. Berkeley: University of California Press. Meng X. 2012. “Labor Market Outcomes and Reforms in China.” Journal of Economic Perspectives  26 ( 4): 75– 102. Google Scholar CrossRef Search ADS   Mu R., de Brauw A.. 2015. “Migration and Young Child Nutrition: Evidence from Rural China.” Journal of Population Economics  28 ( 3): 631– 57. Google Scholar CrossRef Search ADS   Mu R., van de Walle D.. 2011. “Left Behind to Farm? Women's Labor Reallocation in Rural China.” Labour Economics  18 (S1): S83– 97. Google Scholar CrossRef Search ADS   Munshi K. 2003. “Networks in the Modern Economy: Mexican Migrants in the U.S. Labor Market.” Quarterly Journal of Economics  118 ( 2): 549– 99. Google Scholar CrossRef Search ADS   Munshi K. 2011. “Strength in Numbers: Networks as a Solution to Occupational Traps.” Review of Economic Studies  78 (3): 1069– 1101. Google Scholar CrossRef Search ADS   Munshi K., Rosenzweig M.. 2016. “Networks and Misallocation: Insurance, Migration and the Rural-Urban Wage Gap.” American Economic Review  106 ( 1): 46– 98. Google Scholar CrossRef Search ADS   Rozelle S., Guo L., Shen M., Hughart A., Giles J.. 1999. “Leaving China's Farms: Survey Results of New Paths and Remaining Hurdles to Rural Migration.” China Quarterly  158: 367– 93. Google Scholar CrossRef Search ADS   Taylor J. E., Rozelle S., de Brauw A.. 2003. “Migration and Incomes in Source Communities: A New Economics of Migration Perspective from China.” Economic Development and Cultural Change  52 ( 1): 75– 101. Google Scholar CrossRef Search ADS   Woodruff C., Zenteño R.. 2007. “Migration Networks and Microenterprises in Mexico.” Journal of Development Economics  82 ( 2): 509– 28. Google Scholar CrossRef Search ADS   Wooldridge J. M. 2010. Econometric Analysis of Cross Section and Panel Data , 2nd ed. Cambridge, MA: MIT Press. Yang D. 2008. “International Migration, Remittances and Household Investment: Evidence from Philippine Migrants’ Exchange Rate Shocks.” Economic Journal  118 ( March): 1– 40. Zhao Y. 2002. “Causes and Consequences of Return Migration: Recent Evidence from China.” Journal of Comparative Economics  30 ( 2): 376– 94. Google Scholar CrossRef Search ADS   Zhao Y. 2003. “The Role of Migrant Networks in Labor Migration: The Case of China.” Contemporary Economic Policy  21 ( 4): 500– 511. Google Scholar CrossRef Search ADS   © The Author 2018. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com

Journal

The World Bank Economic ReviewOxford University Press

Published: Feb 1, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 12 million articles from more than
10,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Unlimited reading

Read as many articles as you need. Full articles with original layout, charts and figures. Read online, from anywhere.

Stay up to date

Keep up with your field with Personalized Recommendations and Follow Journals to get automatic updates.

Organize your research

It’s easy to organize your research with our built-in tools.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve Freelancer

DeepDyve Pro

Price
FREE
$49/month

$360/year
Save searches from Google Scholar, PubMed
Create lists to organize your research
Export lists, citations
Access to DeepDyve database
Abstract access only
Unlimited access to over
18 million full-text articles
Print
20 pages/month
PDF Discount
20% off