We propose a two-stage probit model (TPM) to predict recovery rates. By the ordinal nature of the three categories of recovery rates: total loss, total recovery, and lying between the two extremes, we first use the ordered probit model to predict the category that a given debt belongs to among the three ones. Then, for the debt that is classified as lying between the two extremes, we use the probit transformation regression to predict its recovery rate. We use real data sets to support TPM. Our empirical results show that macroeconomic-, debt-, firm-, and industry-specific variables are all important in determining recovery rates. Using an expanding rolling window approach, our empirical results confirm that TPM has better and more robust out-of-sample performance than its alternatives, in the sense of yielding more accurate predicted recovery rates.
- Expanding rolling window approach
- Ordered probit model
- Probit transformation regression
- Recovery rate
- Two-stage probit model