In this paper, a default prediction method based on the discrete-time varying-coefficient hazard model (DVHM) is proposed. The new model is constructed by replacing the constant coefficients of firm-specific predictors in the discrete-time hazard model (DHM; see Shumway, 2001; and Chava & Jarrow, 2004) with the smooth functions of macroeconomic variables. Thus, it allows the effects of those firm-specific predictors on the default prediction to change with the macroeconomic dynamics (Pesaran, Schuermann, Treutler, & Weiner, 2006). The coefficient functions in the new model are estimated by a local likelihood approach. One real panel dataset is used to illustrate the proposed methodology. Using an expanding rolling window approach, the empirical results confirm that DVHM has a better and more robust performance than the usual DHM, in the sense that it yields more accurate predicted numbers of defaults and predictive intervals through out-of-sample analysis. Thus, the proposed model is a useful alternative for studying default losses on portfolios.
