A Comparison of Model Selection Procedures for Predicting Turning Points in Financial Time Series

The aim of this paper is to compare the influence of different model selection criteria on the performance of ARMA- and VAR-models to predict turning points in nine financial time series. As the true data generating process (DGP) in general is unknown, so is the model that mimics the DGP. In order to find the model which fits the data best, we conduct data mining by estimating a multitude of models and selecting the best one optimizing a well-defined model selection criterion. In the focus of interest are two simple in-sample criteria (AIC, SIC) and a more complicated out-of-sample model selection procedure. We apply Analysis of Variance to assess which selection criterion produces the best forecasts. Our results indicate that there are no differences in the predictive quality when alternative model selection criteria are used.