摘要
Distressed firms in equity markets are like landmines in the battlefields due to their undetectability and devastating effects. This paper is concerned with distressed firms forecasting by the distance-to-default (DTD) and rare event logit (REL) models via public available data. Comparing these two models by cumulative accuracy profiles (CAP) and receiver operating characteristic (ROC) curves, we conclude that the REL model performs better than the DTD model. The data contains US-listed firms on the S&P 500 for the period January 1986 to December 2012, including 2138 companies and 271,912 firm months, with 444 distressed firms. We set the dynamic thresholds as the last 6% of firms based on the historical cross-section distress rates. Upon Bayesian posterior probability examination, the REL model shows about 40–60% affinity with S&P Domestic Long Term Issuer Credit Rating records on average, and the rate increases to 70% in some situations. We conclude that the REL model can be a good warning indicator of distress in firms at least three years ahead.