Recovering Your Money: Insights Into Losses From Defaults

by Karen Van de Castle of Standard & Poor's, and
David Keisman of Portfolio Data Management, LLC

June 16, 1999

Beginning Paragraphs: Recovery data has always been a weak link of quantitative credit loss models and has long lagged research done on default and migration. Modeling, whether it be to estimate return on capital, return on risk-adjusted capital, value at risk, or pricing, utilizes increasingly sophisticated methods of predicting default based on credit rating equity, price volatility and financial statistics. This default analysis has tended to be paired with static loss assumptions with, at best, a single average used for all secured loans and another single average used for all unsecured loans. Unsurprisingly, the result has been loss given default assumptions marred by high standard deviations and fat tails (excess data points at both ends of the distribution). An unfortunate result of this high standard deviation is an inability to fine-tune spreads, capital allocation and ratings based on historical loss experience.

Portfolio Management Data LLC (PMD), working with Standard & Poor's, has assembled an empirical credit loss database that will enhance loss assumptions and tighten standard deviations. This permits improved capital allocation, more precise pricing and better portfolio management.

In order to achieve this goal, the new database focuses not only on tranche of debt (e.g., bank loans, senior secured debt, senior subordinated debt, etc.) and collateral type, but also on capital structure--how much debt is above or below a particular instrument in the balance sheet. By analyzing loss data by the amount of debt subordinated to the loans, we have been able to significantly tighten the distribution around the mean in compiling loss statistics. When this subordination data is combined with a specific type of collateral, the standard deviation is further reduced.

Published in: Credit Week, Vol. 16. (June 1999), pp. 29-34.

Download paper (60K PDF) 6 pages