Detecting the number of breaks in the mean can be challenging when it comes to the long memory framework. Tree-based procedures can be applied to time series when the location and number of mean shifts are unknown and estimate the breaks consistently though with possible overfitting. For pruning the redundant breaks information criteria can be used. An alteration of the BIC, the LWZ, is presented to overcome long-range dependence issues. A Monte Carlo Study shows the superior performance of the LWZ to alternative pruning criteria like the BIC or LIC.
long memory mean shift regression tree ART LWZ LIC.