Information and Prediction Criteria in Selecting the Forecasting Model

Abstract

The purpose of the paper it to compare the performance of both information and prediction criteria in selecting the forecasting model on empirical data for Poland when the data generating model is unknown. The attention will especially focus on the evolution of information criteria (AIC, BIC) and accumulated prediction error (APE) for increasing sample sizes and rolling windows of different size, and also the impact of initial sample and rolling window sizes on the selection of forecasting model. The best forecasting model will be chosen from the set including three models: autoregressive model, AR (with or without a deterministic trend), ARIMA model and random walk (RW) model.