Non-linear effects and aggregation bias in Ricardian models of climate change

Ricardian models predicting the impact of climate change on agriculture are typically estimated on data aggregated across counties and assuming additively separable effects of temperature and precipitation. We investigate the potential bias induced by such assumptions by using a large panel of farm-level data and estimating a semi-parametric specification. Consistent with the agronomic literature, we observe significant non-linear interaction effects, with more abundant precipitation being a mitigating factor for heat stress. This interaction disappears when the same data is aggregated in the conventional manner, leading to predictions of climate change impacts which are severely distorted.