Model Selection Criteria for Segmented Time Series from a Bayesian Approach to Information Compression

The principle that the simplest model capable of describing observed phenomena should also correspond to the best description has long been a guiding rule of inference. In this paper a Bayesian approach to formally implementing this principle is employed to develop model selection criteria for detecting structural change in financial and economic time series. Model selection criteria which allow for multiple structural breaks and which seek the optimal model order and parameter choices within regimes are derived. Comparative simulations against other popular information based model selection criteria are performed. Application of the derived criteria are also made to example financial and economic time series.