Avoiding Model Selection by the Use of Shrinkage Techniques
Pakistan Institute of Development Economics
Journal of Econometrics Journal of Econometrics, Vol. 25, pp. 73-85, 1984
It is argued that, in most cases, model selection procedures, including those based on the popular criteria such as predictive loss or information, lead to inadmissible procedures. In particular, properties of the Akaike Information Criterion (AIC) in a nested sequence of regression models are analyzed by reduction to an appropriate canonical form. It is shown that the AIC has some undesirable features, and an alternative proper Bayes procedure is suggested. Numerical comparisons of risks are presented.
Keywords: model selection, Akaike Information Criterion, Bayesian model averaging, prediction criteria
JEL Classification: C11, C52
Date posted: March 27, 2009
© 2016 Social Science Electronic Publishing, Inc. All Rights Reserved.
This page was processed by apollobot1 in 0.172 seconds