Model Selection Using Information Criteria and Genetic Algorithms

Automated model searches using information criteria are used for the estimation of linear single equation models. Genetic algorithms are described and used for this purpose. These algorithms are shown to be a practical method for model selection when the number of sub-models are very large. Several examples are presented including tests for bivariate Granger causality and seasonal unit roots. Automated selection of an autoregressive distributed lag model for the consumption function in the US is also undertaken. Copyright Springer Science + Business Media, Inc. 2005