Measuring and Managing Agricultural Risk at Scale: Index Insurance and Aflatoxin Early-Warning

This dissertation studies how predictive modeling and optimization can improve agricultural risk management in Low- and Middle-Income Countries. These settings usually feature spatially correlated risk, limited data, and costly monitoring. The first chapter develops an optimization-based framework for index insurance contract design. I model premiums as expected payouts plus a CVaR-based cost of capital, allowing prices to incorporate the insurer's exposure to tail risk. The resulting problem is non-convex, so I develop a conservative, convex relaxation that preserves tractability while maintaining a simple, piecewise-linear contract structure. I then extend the framework from a single-zone setting to a multi-zone portfolio design, allowing contracts to be optimized jointly across regions. I then show that the multi-zone design weakly dominates separate zone-by-zone design. I then evaluate this framework empirically using a novel dataset of Thai rice farmers. The evaluation shows our method improves welfare relative to benchmark contract-design methods.