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...