Forecast Evaluation: A Likelihood Scoring Method

While many forecast evaluation techniques are available, most are designed for the end user of forecasts. Most statistical evaluation procedures rely on a particular loss function. Forecast evaluation procedures, such as mean squared error and mean absolute error, that have different underlying loss functions, may provide conflicting results. This paper develops a new approach of evaluating forecasts, a likelihood scoring method, that does not rely on a particular loss function. The method takes a Bayesian approach to forecast evaluation and uses information from forecast prediction intervals. This method is used to evaluate structural econometric and ARIMA forecasting models of quarterly hog price.