Covariate Effects in Periodic Hazard Rate Models

Ulrich Poetter and Kai Kopperschmidt

Abstract

Labour market participation, consumer behaviour, and many other phenomena exhibit strong periodic patterns that result from cyclic behaviour, constraints on the timing of events, or seasonal variation. While these periodicities can generally be neglected when dealing with small data sets or coarsely grouped event times, they pose challenges to the analysis of large data sets with precise recordings. It seems natural to require that statistical models used in the analysis of such data sets reproduce any underlying periodicities. In particular, the conditional hazard rate given covariates should be periodic for all possible values of the covariates. We show that this requirement severely restricts the class of covariate effects models.