Under-investment in Profitable Technologies when Experimenting is Risky: Evidence from a Migration Experiment in Bangladesh

Abstract

The rural north-western districts of Bangladesh, home to 10 million people, experience a pre-harvest seasonal famine, locally known as Monga, with disturbing regularity. Inspired by the observations that wages are higher, jobs are more plentiful in nearby urban areas than in the monga-prone region, and that there are no official restrictions on mobility, we provide monetary incentives in 100 study villages to encourage people to seasonally migrate out in search of employment. We employ a randomized intervention design to study the patterns of responsiveness to our incentives using a 1900 household sample, which illuminates the current constraints to seasonal out-migration. The randomization also allows us to cleanly estimate the (causal) returns to migration in terms of household expenditures, savings and earnings, and caloric intake. The propensity to seasonally out-migrate is very responsive to a 600-800 Taka cash/credit incentive, raising the migration rate to 57%, relative to 34% in a set of control villages. Simply providing information on job availability and wages at destination has no effect on the migration rate. Comparing the characteristics if migrants in the treatment and control areas, the incentives appear to induce people who otherwise feel less comfortable migrating – i.e. those without job leads or social network presence at the destination, and new migrants. Households closer to subsistence are less likely to migrate, but are much more responsive to the incentive. The migrant experience was very successful on average: monthly consumption increased by at least Tk 300 ($4) per person per month, or Tk 1050 ($15) per household per month due to the induced migration. Earnings, savings and remittances from each migration episode are several multiples of the initial investment (the incentive, or the round-trip cost of moving). Most strikingly, migration rate in the treatment areas remained significantly higher in the treatment areas (47% vs. 35%) a year after the program, even after all incentives are removed. Induced migration in one year increases the propensity to re-migrate by 40 percentage points. There is evidence of learning among induced migrants, such as a greater propensity to remigrate when the initial migration episode was more successful, greater growth in savings/earnings per day in treatment group, and larger consumption effect in 2009 among re-migrants. These results are consistent with a model of a migration-based poverty trap, where individuals are uncertain about their own prospects at the destination, and particularly worried about a bad outcome (e.g. undertaking the costs of moving, but then not finding a job) during a period in which their family is under the threat of famine. The uncertainty associated with migration prevents households from investing, even when the expected returns are positive. Our intervention insures households against the bad outcome, thereby allowing them to invest, learn about their private returns to the investment, and for those with positive realizations, re-migrate the next period even in the absence of the intervention. Many of the empirical results are also consistent with a simple model of a liquidity/credit constraint preventing migration. These results suggest that a grant program, or a credit program with limited liability (which amounts to insuring households against the possibility of a bad outcome) is likely to be welfare enhancing, and can be an effective policy response against the threat of localized seasonal famines such as Monga. More broadly, providing small grants or credit that enable households to search for jobs, and leads to a better spatial and seasonal matching between potential employers and employees may be a useful complement to the muchmore- popular microcredit programs that aim to create new entrepreneurs and new businesses.