The loss function is a mathematical representation of the costs experienced by a forecaster when observed outcomes differ from what was predicted. Prior studies suggest that USDA forecasts are not optimal based on an assumed mean-zero quadratic loss function. This study proposes an alternative view of forecast evaluation, which assumes all USDA forecasts are produced to minimize the forecasters’ costs, and searches for the dimensions of the loss function under which optimality holds. We illustrate the degree to which USDA loss functions vary across a series of WASDE price forecasts. A better understanding of USDA forecasters’ costs will benefit forecasters and forecast users.
