Title: Smooth backfitting of additively structured hazard rates for in-sample forecasting
Authors: Stephan Bischofberger - Cass Business School (United Kingdom) [presenting]
Jens Perch Nielsen - Cass Business School (United Kingdom)
Munir Hiabu - Cass Business School (United Kingdom)
Enno Mammen - University of Heidelberg (Germany)
Abstract: Smooth backfitting has been established in nonparametric regression and in density estimation as a very promising alternative to the classic backfitting method. We apply the concept to a survival model with additively structured nonparametric hazard. The model allows for very general censoring and truncation patterns occurring in many forecasting applications such as medical studies or actuarial reserving. A crucial point is that - in contrast to classical backfitting - we do not assume independence between the covariates. Our estimators are shown to be a projection of the data into the space of multivariate hazard functions with additive components. Hence, our hazard estimator is the closest additive fit even if the actual hazard rate is not additive. Another big advantage of our additive model is that our estimators are straight forward to derive in theory including excellent properties as well as their simple implementation in practice even for high dimensional covariates. We provide full asymptotic theory for our estimators as well as a simulation study.