In 2009, the Thai government implemented a price insurance scheme for rice, cassava, and maize farmers. The program, which was abandoned after only two years, added to the incomes of registered farmers a non-negative but stochastic amount that was decoupled from farmers’ agricultural activities. We apply a simple machine learning algorithm based on rich panel data to control for self-selection into the program and study its impact on small-scale rice farmers in relatively poor Northeastern Thailand. Program participation increases rice production but also leads to shifts in investment behavior and the composition of income generating activities away from agriculture, which may be beneficial for rural development. Moreover, farmers who are initially less poor experience a substantial increase in overall income. Decreasing risk-aversion and relieved credit constraints may be possible channels for these effects.

Keywords: Cash Transfers, Counter-Cyclical Payments, Decoupling, Agricultural Subsidies, Rural Development, Machine Learning, Thailand.