Does Spatial Correlation Matter in Econometric Models of Crop Yield Response and Weather?

Due to the rapidly growing availability and accessibility of spatially gridded weather data products, significant effort has been devoted to handling weather and climate variables properly in econometric models. It is, however, noteworthy that relatively less econometric attention is paid to how spatial correlation in weather variables and econometric models can be specified and performed. To fill this gap, this study scrutinizes the main source spatial correlation in econometric models of weather and climate variables, and implements in-sample and out-of-sample prediction analyses with spatial panel model specifications of crop yield response function. First, this paper theoretically and empirically demonstrates that the aggregation bias is a main source of spatial correlation rather than omitted weather variables. With soil variables, we specify six competing specifications of crop yield response function with pooled, fixed effects and random effects with spatially robust standard errors. From the results of prediction performances, we demonstrate that the choice of predictor (prediction models) can be motivated from the purpose of models rather than a better prediction performance. In addition, we empirically argue that the omitted socio-economic variables are not a serious econometric concern in crop yield response function of this study.