Information and Prediction Criteria for Model Selection

In model selection literature the methods of model selection are not clearly considered depending on the aims of modeling. The econometric and forecasting literature alike have done little to separate the targets of searching ‘true’ models and of optimizing prediction and have a tendency to blur the distinction between these aims. This is evident from the exaggerated concern about ‘model misspecification’ in a forecasting situation  (where misspecified models may yield excellent forecasts) and from the belief that accurate prediction is the main way that the adequacy of a model must be reflected regardless of the goals of an analysis. The purpose of the paper is to explore the performance of predictive accuracy evaluations as model selection criteria, both from the viewpoint of finding true model and of finding the best forecasting model. This performance is illustrated by means of some Monte Carlo experiments and also empirical examples.

Keywords: information criteria, prediction criteria, model selection, forecasting, predictive accuracy.

JEL Classifications: C32, C52, C53