Migration Frictions, Earnings Differentials and Spatial Misallocation: Evidence from Thailand.

AuthorShenoy, Ashish
  1. Introduction

    Around the world there exist large differences in earnings across space, both within and between countries, that persist even after adjusting for local prices and human capital (Acemoglu and Dell 2010). Such variation highlights the role played by location in both labour supply and demand. On the demand side, if firms were perfectly mobile, they would relocate to where labour was the cheapest. Similarly, on the supply side, a perfectly mobile labour force would migrate towards higher wages. A geographic mismatch between demand and supply gives rise to earnings variation across markets. The mobility frictions that generate this mismatch may lower aggregate productivity through the misallocation of labour and can increase exposure to local shocks by concentrating economic impacts within a smaller geographic area.

    This paper quantifies the economic importance of mobility frictions across provinces in Thailand in the 1980s and 1990s. Like many countries, Thailand has substantial geographic variation in labour earnings, with the standard deviation across its seventy-three provinces at around half of the median. I analyse the labour supply contribution to this variation through the lens of a spatial equilibrium model of worker location choice (see Rosen 1979; Roback 1982; Moretti 2011). In the model, the migration elasticity to differences in earnings across space is mediated by variation in location-specific amenities common to all workers, idiosyncratic worker preferences for place of residence, and migration frictions raising the cost of relocation. The study uses the model to characterize the net contribution of migration to local labour supply elasticity.

    Model parameters are structurally estimated from the revealed preference migration response to changes in local market conditions. Data come from annual cross-sectional labour force surveys conducted by the government of Thailand. From 1985 to 2000, surveys included questions about the history of residence from which a pseudo-panel of annual province-to-province migration flows is constructed. Data on gross bidirectional flows, as opposed to net flows or gross population changes, allow for separate identification of location-specific preferences and mobility costs, both of which may mute the migration elasticity to local earnings. The net migration response to a change in earnings reflects the importance of earnings relative to workers' geographic preferences. Conditional on net flows, high gross flows would indicate location-specific preference heterogeneity independent of location history, while low gross flows would suggest workers perceive a disutility from leaving their current place of residence. (1) I define the latter to be a mobility friction and simulate counterfactual scenarios to investigate how much such frictions sustain spatial earnings gaps and depress aggregate productivity.

    The main empirical innovation of this paper is to estimate model parameters using international commodity prices to isolate exogenous earnings variation caused by shifts in local labour demand. This identification strategy resolves three distinct challenges. First, instrumenting for local labour demand overcomes the standard problem of labour supply endogeneity. In spatial equilibrium, endogeneity arises when a correlated preference shock drives workers to or from a given destination, shifting the local labour supply curve and inducing a spurious negative relationship between migration and earnings. My empirical strategy addresses this concern by instrumenting for local earnings using global commodity prices interacted with local sensitivity to commodities, thereby leveraging exogenous, demand-based variation in the local earnings process.

    Second, this empirical strategy separates changes in the return to migration from the selection of unobservables. Any revealed-preference empirical strategy relies on observed earnings to infer the potential return to migration. However, recent empirical work has established the possibility that much spatial earnings variation can be attributed to worker selection (Young 2013; Hamory et al. 2021) rather than local productivity. If workers sort geographically according to unobserved traits, then observed earnings levels may not reflect the true potential income one might earn from a move. To avoid selection bias, I exploit variation caused by fluctuations in local labour demand and verify that the resulting earnings changes are not driven by differential worker selection. Therefore, estimates from this paper are accurately based on changes in the potential return to migration rather than on unobserved characteristics of the population at the destination.

    Third, the commodity-based instrument enables results to be quantified in currency units. In general, location choice is a long-term decision with benefits realized over time. Current earnings levels do not convey expectations about the future, making it difficult to calculate the expected return to migration in net present value terms. However, the portion of earnings identified by commodity instruments inherits the same time-series properties as the commodity price series themselves. Therefore, estimation leverages earnings variation for which the expected net present value is readily computed, allowing other parameter estimates to be interpreted relative to monetary returns. I verify in the reduced form that for the same size contemporaneous shock to earnings, there is a greater migration response when the shock is generated by a more permanent commodity series. This differential response indicates that labour markets incorporate information about the expected future value of current earnings shocks, allowing other preference parameters to be scaled into meaningful units comparable to a dollar of net-present-value earnings.

    This final feature expands on existing work using commodity price shocks as a source of identifying variation in local labour demand. A number of other studies use commodity prices to estimate the effect of changes in earnings on outcomes such as occupation selection (e.g., Young 2014; James 2015; Allcott and Keniston 2018) and civil conflict (e.g., Dube and Vargas 2013; Berman et al. 2017; McGuirk and Burke 2020). I extend this literature by leveraging differences in the time-series properties of different commodity prices as a proxy for expectations about the future value for shocks of comparable size.

    This paper also introduces a methodological innovation to overcome two common weaknesses found in the data. First, migration rates are measured with noise. Since the sample size is small relative to the number of potential migration channels, there is high sampling variance and zero observed migration along the majority of province-to-province channels. This issue also arises in retail data with fine or uncommon product categories (e.g., Gandhi, Lu, and Shi 2023) as well as trade data with errors in measurement or reporting (Feenstra et al. 2005; UNCTAD 2012). Second, because sampling is stratified by province, sampling frequencies are endogenous to workers' residence and therefore to their migration decision. Such choice-based sampling arises whenever surveys stratify by outcome or oversample rare outcomes for statistical power (Cosslett 1981). I show that a naive model estimation that ignores these factors would generate inaccurate estimates of the importance of mobility frictions relative to earnings.

    To preserve consistency, this study derives a formula for the data-generating process that comprises both the spatial choice model as well as the survey sampling design. This strategy selects parameters to maximize the joint likelihood of each survey respondent having made their observed migration choice from their place of origin and of having been surveyed at their current destination conditional on that choice. Estimation requires imposing the constraint that aggregate migration flows in and out of each province must sum to the actual change in population for that province. The estimator in this paper is a generalization of the choice-based maximum likelihood approach proposed by (Manski and Lerman 1977) that supplements individual choice data with aggregate choice probabilities. While the Thai data do not accurately measure aggregate migration probabilities along any specific province-to-province channel, the net change in each province's population reflects the aggregate sum across all migration channels. I show that this procedure has an equivalent interpretation that the set of survey respondents in each province represents the outcome of a draw from a multinomial distribution of potential origin provinces, with probabilities governed by migration rates and population sizes.

    Results show the perceived disutility of migration to be substantial. The average utility cost of migrating is equivalent to 1.0-1.2 times annual earnings in net-present-value terms. Furthermore, the variation across individuals in their preferences over local non-wage amenities is on the same order of magnitude as the observed spatial variation in earnings. These two factors combine to limit the role of the labour market in an efficient reallocation. The size of the estimated utility penalty to migration relative to annual earnings is consistent with comparable estimates from Brazil (Morten and Oliveira 2023), and smaller but on the same order as the estimated cost in the US (Kennan and Walker 2011). A parallel avenue of research on the effects of trade finds the cost of changing sector, rather than changing location, to be significantly greater (Artuc, Chaudhuri, and McLaren 2010; Dix-Carneiro 2014). It is unclear in general whether worker behaviour is primarily governed by migration frictions with sector choice being a consequence of the location decision, or by sectoral frictions with location choice conforming to the sector of occupation.

    Despite the...

To continue reading

Request your trial

VLEX uses login cookies to provide you with a better browsing experience. If you click on 'Accept' or continue browsing this site we consider that you accept our cookie policy. ACCEPT