Structural change and long memory in the GARCH(1,1)-model

Abstract

It has long been known that the estimated persistence parameter in the GARCH(1,1) - model is biased upwards when the parameters of the model are not constant throughout the sample. The present paper explains the mechanics of this behavior for a particular class of estimates of the model parameters. It gives sufficient conditions for the estimated persistence to tend to one when the mean of the process changes, both for a given sample size (as the size of the structural change increases), and as sample size increases, extending previous results that were concerned with changes in the volatility parameters.