On Specification Diagnostics for Econometric Models of Duration

This paper deals with diagnostics for evaluating the appropriateness of distributional assumptions in econometric models of duration. The purpose is to point out certain extensions, equivalences and interpretations. The exposition is in terms of the Weibull distribution, but the framework is applicable more generally. The diagnostics developed are score tests of the null hypothesis of Weibull distributed durations (conditional on covariates) against alternatives which are based on expansions of various orders in a system of orthogonal functions. The unified framework put forward By Tauchen (1985) and Newey (1985) is used to develop robust conditional moment tests which generalize the score tests. The paper also shows that recently developed diagnostics for uncorrected heterogeneity essentially amount to testing whether a particular moment relationship implied by the null is satisfied. Lastly, some Monte Carlo results are presented