THE EFFECT OF DIFFERENT LENGTH TREND VARIABLES ON THE PREDICTION OF FIRM DEFAULT PROBABILITIES.

Abstract

This paper expands on the forward intensity model in Duan et al.(2012) by including a second trend component of firm-specific variables. We performed a modified grid search of moving averages for both trend variables. Altogether, 67 combinations were tried and their performance across prediction horizons from 1 to 36 months were compared. For the in-sample, out-of-sample (Cross-sectional), and out-of-sample (Time) tests, the best models showed average improvements in Accuracy Ratios across all 36 prediction horizons of 0.76%, 0.59%, and 1.34% respectively. However, these models underperformed the original model for the out-of-sample (Cross-Sectional) prediction horizons of 30 months and above. An equal-weighted rank system was employed for model selection. The best models had significantly different durations of trend moving averages. Separately, net-income-to-total-asset and distance-to-default were found to have significantly different coefficients for its 2 trend components. This supports our hypothesis of a bi-directional relationship between firm-specific variables and firm’s financial health.