Estimation non-paramétrique du taux de hasard : application à des durées de chômage censurées à droite

Different hazard rate nonparametric estimators for randomly right-censored data are considered in this paper. Lacking sufficiently homogeneous theoritical results allowing for a comparison between their performances, an empirical procedure is adopted. This procedure is based on an empirical estimation of the MISE criterion ("Mean Integrated Squarred Error") from simulated data. It highlights the fact that an estimator which is based on the nearest-neighbor method, and which we introduce, is more dependable thus more appropriate for unemployment data analysis. Application of this new nonparametric estimator allows us to validate results from a parametric model estimation, and particularly to show the information loss caused by inflexible parametric specification of the hazard rate.