There are ample evidences that economics time series are usually non-stationary over a long period of observation time. In many cases, the existence of non-stationarity is apparently due to the presence of structural breaks. Consequently, failure to detect 'shifts' or breaks might lead to a serious problem. Early researches on structural breaks focus on methods which were developed only for a single break by using historical and sequential test such as CUSUM test and the sequential test statistics. Recent development considers cases with multiple structural breaks based on the model selection approach. This paper revisits the method of estimating the number of breaks and the problem of model selection using Minimum Description Length (MDL) with Genetic Algorithm (GA) as an optimization approach to search for the best fitting model. The effectiveness and the implementation of this approach will be illustrated by using Generalized Autoregressive Conditional Heteroskedasticity (GARCH) model.
展开▼
