Spatial Panel Models of Crop Yield Response to Weather: Econometric Specification Strategies and Prediction Performance

This study scrutinizes spatial econometric models and specifications of crop yield response functions to provide a robust evaluation of empirical alternatives available to researchers. We specify 14 competing panel regression models of crop yield response to weather and site characteristics. Using county corn yields in the US, this study implements in-sample, out-of-sample, and bootstrapped out-of-sample prediction performance comparisons. Descriptive propositions and empirical results demonstrate the importance of spatial correlation and empirically support the fixed effects model with spatially dependent error structures. This study also emphasizes the importance of extensive model specification testing and evaluation of selection criteria for prediction.