Spatially Smoothed Kernel Densities with Application to Crop Yield Distributions

This study is motivated by the estimation of many crop yield densities, each with a small number of observations. These densities tend to resemble one another if they are spatially proximate. To gain flexibility and improve efficiency, we propose kernel-based estimators refined by empirical likelihood probability weights derived under spatially smoothed moment conditions. We construct spatially smoothed moments based on spline functions, which are robust to outliers and readily customizable. We use these methods to estimate the corn yield distributions of Iowa counties and to predict the premiums of crop insurance programs. Monte Carlo simulations and an empirical application demonstrate the good performance and usefulness of the proposed methods.