Linear pooling of potentially related density forecasts in crop insurance

Accurate pricing of crop insurance policies relies on forecasts probability densities yields. Yield are dynamic, time series data yields often limited, and yield spatially correlated. We examine linear pooling potentially related, but almost surely misspecified, density forecasts. The pooled combine from other spatial units based out-of-sample forecast performance. result in more accurate premium rates which can reduce incentives for adverse selection. approach is applicable to any setting where the statistical model loss variable likely be misspecified underlying data-generating processes related.