By using cohorts defined by narrow age groups, I am able to analyze how wages of such overall similar workers change as their cohort faces a higher rural out-migration shock. My investigation is challenged by the fact that information on the rural/urban status of a migrant’s locality of origin is not available in most population censuses and in the Brazilian censuses in particular. While the administrative region of a migrant’s origin may be known, these regions are often too large to meaningfully use data on population size or density to construct a proxy for the rural/urban status of the locality. In this paper I use the size of a cohort relative to a baseline year to measure rural net out-migration rates. More specifically a cohort’s rural out-migration in 1991 is measured as the change in the size of the cohort between 1980 and 1991. This net-migration method has been shown to perform well in the Mexico-US immigration by Martinez and Woodruff and Hanson and McInstosh. My empirical strategy is implemented using population census data from the Brazilian census of 1980, 1991 and 2000. The basic identification strategy is based on the fact that cohorts of workers have differential rural net out-migration rates and that these differences vary over time. I also address two key concerns about the potential sources of bias in the estimates. First, rural out-migration and wages may be simultaneously determined by, for instance,cut flower bucket pull factors such as urban employment growth or wages. Following the literature I apply an instrumental variables strategy using past rural out-migration rates as an instrument for current out-migration rates.
The results show that the instrument is strongly related to rural out-migration rates and reinforce the OLS estimates. Second, I address the issue of selectivity bias which arises because the measured wages in rural areas are obtained from selected workers who decided not to migrate. This is because, in the presence of self-selection into migration, rural out-migration may both reduce overall rural labor supply and affect the composition of rural labor supply. I show that the OLS and IV results are robust to controlling for key worker characteristics that may affect wages. Although addressing the issue only partially, this approach provides evidence that the OLS and IV results are robust to controlling for key characteristics that determine the composition of rural labor. My results show that a 10% increase in rural out-migration rate in a cohort raises wages in that cohort by 1 to 5%. This result suggests that rural out-migration flows in Brazil between 1991 and 2000 have increased wages by 3 to 6.5%. Moreover the impact on wages differs substantially by cohort. I find that rural out-migration between 1991 and 2000 led to a 4.5 to 5.6% increase in wages for the older cohorts of workers and between 7 to 9% for the younger cohorts. The remainder of this chapter is organized as follows: Section 2.2 provides an overview of the theoretical literature and presents the empirical framework. Section 2.3 describes the data, the basic results and some robustness checks. Section 2.4 discusses the issue of selection bias and changes in the composition of rural labor. Section 2.5 concludes.In this section I present the implications of existing models that could explain the effect of rural emigration on rural wages.
As discussed by Mishra, partial equilibrium models with closed economies suggest that, under certain hypotheses on the production function and uniform inputs, rural out-migration will raise real wages in rural areas leading to a welfare gain for workers, a welfare loss for owners of capital and an aggregate welfare loss. Models such as those in Findlay and Davis and Einstein extend the Ricardian model of international trade to allow for migration. In these models, under free trade, rural out-migration from will increase the real wage in rural areas in terms of the importable goods and remain unchanged in terms of the exportable good, leading to an increase of the real wage. However in the presence of a non-traded labor-intensive good, the real wages in rural areas could decrease with the degree of rural out-migration as shown in Quibria and Rivera-Batiz. This arises if migrants possess a large fraction of the capital or when there are increasing returns to the production of the non-traded good. Beyond the labor supply push, rural out-migration may also affect rural wages by increasing the rate of capital inflow. Remittances may increase consumption and investment of households at origin thus potentially raising local demand for goods and services and pushing prices and wages upward. In this paper I examine empirically how rural out-migration measured by changes in cohort size residing in a rural area affects the wage of rural workers in the same cohort. In light of the discussion above such analysis will measure, in general, the combined effect of changes in labor supply, capital inflow, and local demands for goods and services following a higher out-migration shock.
An important question is whether one can disentangle these effects or provide evidence on the main channels. The empirical model developed below presents a framework to estimate the effect of rural out-migration on rural wages. Due to the absence of data on rural capital stock, I assume that the stock of capital is fixed. As such, the results presented here provide long-run estimates of the effect of rural out-migration on rural wages in a general equilibrium setting. Where cohortsize is the size of cohort c living in rural areas of state s at time t. The variable shockemigrationcst measures the number of individuals which belong to cohort c of state s that have migrated out of rural areas as a share of the size of that cohort. The first issue is related to data limitations. While census data often includes information related to an individual’s migration status and administrative region of origin, they do not generally record whether the migrant was originally from a rural or urban area. Moreover, the administrative regions of origin are often too large to, meaningfully,flower display buckets use information on population size or density to construct measures of the rural/urban status of the locality of origin. The second issue is related to the identification of the effect of rural out-migration on rural wages. With information on individual’s migration status and rural/urban status of the locality of origin, one could measure the out-migration shock as the size of the rural locality residing elsewhere. However this will raise the concern that higher rural out-migration in a state could increase rural in-migration of individuals from other areas . Nonetheless while the approach in this paper deals with such concern, it doesn’t allow to separately estimate the effect of rural out-migration of native population and rural in-migration. The last issue is related to the definition of a rural area and whether this definition changes over time. In Brazil, the status of a locality as an urban or rural area is defined by municipal law and doesn’t adhere to any specific thresholds based on population size or density for instance. Moreover, the rural/urban status of localities did not change between the 1980 and 2000 censuses limiting concerns that rural/urban classification may be endogenous to living standards or wages. Lastly, the official rural/urban classification captures the differences in occupational structure that characterizes these two areas. In our sample over 80% of individuals residing in areas classified as rural are engaged in agricultural activities against 5% in localities classified as urban. Figure 2.1 shows smoothed densities of the measure of rural net out-migration in 1991 and 2000. Except for three states all states experienced net rural out-migration during this period. Our results are not affected by excluding the three states with positive net rural in-migration. Figures 2.2 to 2.8 show the distribution of out-migration by cohort. Both the theoretical and the empirical literature suggest that migration decisions are endogenous to wage differential between source and destination regions. As a consequence falling rural wages and job prospects in urban areas may lead to differential propensity to migrate across cohorts. To control for this reverse causality I instrument the emigration shock by lagged emigration shock .
This instrument is motivated by evidence on the importance of networks and social in migration decision . Since networks are usually based on blood relationship and friendship, our instrument which varies by cohort and state should be able to pick-up a large fraction of the movements in rural out-migration within the same cohort-by-state group. The data from my analysis are 5% random samples from the 1980, 1991 and 2000 Public Use Microdata Samples of the decennial census of Brazil. Unfortunately the 1980 census doesn’t contain reliable data on wages. However the 1980 census is used to construct the emigration shocks in 1991 and 2000. Individuals are divided in 7 cohorts using their age in 1991. The cohorts are individuals aged 20-24, 25-29, 30-34, 35-39, 40-44, 45-49 and 50-54 in 1991. The sample is restricted to individuals aged 20 to 54 in 1991 who earn between no less than 200 and no more 50,000 Reais per month in real terms which constitutes about 86% of the original sample. Table 2.1 reports basic summary statistics for the measure of rural net emigration shock and the logarithm of real monthly earnings. The degree of rural out-migration is considerable. In 1991, the rural emigrants shock was about 48% of the rural population. By 2000 this number increased to over 95%. In other words for every 100 individuals still living in rural areas in 2000, over 95 individuals had left rural areas since 1980. The emigrant shock is strikingly different across cohorts with the largest shocks appearing for the younger cohorts of individuals. Moreover these shocks are large compared to international immigration figures. By way of comparison Mishra finds that, in the case of MexicoUS migration, the largest immigration shock across skill groups is a little above 50% of the Mexican labor force. The real wages of individuals in the sample increased slightly between 1991 and 2000 despite several periods of hyperinflation. The increase in real wages range between 1.6 to 3.4 percentage points across cohorts with the largest increase occurring for the younger cohorts . As discussed by Mishra , in the presence of self selection of individuals into migration, the parameter estimated in equations and do not provide un-biased estimates of an exogenous change in rural labor supply. This is because, in the presence of self-selection into migration, rural out-migration affects rural labor markets in two distinct ways: a reduction in overall rural labor supply, and a change in the composition of rural labor supply. The instrumental variable strategy used in the analysis described above provides exogenous changes in rural labor supply but may not provide exogenous changes in the composition of rural labor supply. With pooled cross-sections I cannot model or control for individual’s self selection into migration. I provide some evidence that the results are robust to controlling for some of the observable characteristics of changes in workforce composition. I estimate again controlling for the share of a cohort that is male, with some primary education and that have completed primary education. The results from this estimation are reported in Table 2.4. Comparison the results in Table 4 to the coefficient estimates in columns to of Table 2 provides some indication on the robustness the previous results are to changes to workforce composition. I find that the OLS estimates are not affected by the educational attainment and gender composition of the population that remains in rural areas. However the IV estimates are lower by an order of 2 to 3 points of standard errors. This result is consistent with models of positive selection into migration, where out-migrants have a higher educational attainment and tend to be males. The results show that a 10% increase in out-migration increases rural wages by 1.4 to 2.1%. This suggests that after controlling for changes in workforce composition, the wage effect of rural out-migration flows between 1991 and 2000 drops from 6.5% to about 3%. This points to the fact that changes in workforce compositions resulting from higher rural out-migration could explain up to half of the effect of rural out-migration on rural wages.