Variance Estimates and Model Selection

Abstract:     
The large majority of the criteria for model selection are functions of the usual variance estimate for a regression model. The validity of the usual variance estimate depends on some assumptions, most critically the validity of the model being estimated. This is often violated in model selection contexts, where model search takes place over invalid models. A cross validated variance estimate is more robust to specification errors (see, for example, Efron, 1983). We consider the effects of replacing the usual variance estimate by a cross validated variance estimate, namely, the Prediction Sum of Squares (PRESS) in the functions of several model selection criteria. Such replacements improve the probability of finding the true model, at least in large samples.

Number of Pages in PDF File: 17

Keywords: Autoregressive Process, Lag Order Determination, Model Selection Criteria, Cross Validation

JEL Classification: C13, C15, C22, C52

Accepted Paper Series