Evaluating the forecast densities of linear and non-linear models: applications to output growth and unemployment

In economics density forecasts are rarely available, and as a result attention has traditionally focused on point forecasts of the mean and the use of mean square error statistics to represent the loss function. In this paper we apply recently developed methods of forecast density evaluation to compare model-based density forecasts of US output growth and changes in the unemployment rate. Since one of the models is non-linear and characterized by a changing error variance, density evaluation may offer greater discrimination than evaluation based on the first moment.