Variable selection in competing risks model Alessandra Amendola and Marialuisa Restaino Abstract Our aim is to investigate the performance of different variable selection methods, focusing on a statistical procedure suitable for the competing risks model. In this setting, same variables might have different degrees of influence on the risks due to multiple causes and this effect has to be taken into account in the choice of the ”best” subset. The proposed procedure, based on shrinkage techniques, has been evaluated by means of empirical analysis on default risk predictions. Key words: competing risks model, variable selection, shrinkage techniques. 1 Introduction Since the seminal works of [1], the analysis of firms’ survival are increasingly in- vestigated in the corporate failure literature. The main interest is in building a model that is able to predict the firms’ potential ending up in financial distress. However, fi- nancially distressed companies may exit the market for several reasons (bankruptcy, liquidation, merge, acquisition, etc.) and a challenging task is to identify which fi- nancial indicators may influence each reason of exit. Different variable selection techniques, such as information criterion, stepwise procedure and penalized regres- sion, have been used for selecting predictors in different statistical frameworks (dis- criminant analysis, logistic regression, neural networks and survival analysis). How- ever, only few of them have been considered in competing risks models. The aim of this paper is to investigate the determinants of the probability of different types of firms’ market exit through a competing-risks hazard model, focusing in partic- ular on the variable selection problem. We propose to fit a competing risk model Alessandra Amendola Department of Economics and Statistics, University of Salerno, Via Ponte don Melillo, 84084 Fisciano, Salerno (Italy) e-mail: alamendola@unisa.it Marialuisa Restaino Department of Economics and Statistics, University of Salerno, Via Ponte don Melillo, 84084 Fisciano, Salerno (Italy) e-mail: mlrestaino@unisa.it 1