Abstract
This study proposes an optimization-based weather-yield model to reduce the basis risk of weather-based index insurance. This weather-yield model helps us capture the growing season's monthly variation as it involves monthly explanatory weather indices. In addition, it can capture additional extreme weather effects by including extreme cooling or heating weather indices. This study presents an innovative machine learning framework incorporating optimization approaches to ensure the parsimony of weather index models and the accuracy of crop yield predictions, which can be integrated into the conventional policy design and pricing process. The advantages of this modeling approach and the effectiveness of weather index-based insurance based on this approach in reducing basis risk and revenue risk are demonstrated by applying county-level yield data for mid-season rice in the Anhui province, China.