Reverse Engineering the Rating Agencies

I get odd quant projects.  Calculate the value of an odd CDO.  One of the subprojects involved in that required me to reverse engineer the senior unsecured corporate credit ratings of the rating agencies.  My parsimonious model explained roughly 2/3rds of the variation in ratings. What were my factors?

  • Market capitalization
  • Bond sector (Financial, Industrial, Utility, Cyclical)
  • Equity implied volatility
  • Past total returns, which were significant last year, and not this year.

Note the absence of obvious fundamental factors.  I tried a bunch of factors, but none proved consistently significant.

My regression coefficients were very similar in 2007 and 2008.  I think the model is fairly stable.

Given the fundamental models used by the rating agencies for their ratings, this may mean that the markets reflect the analyses of the rating agencies.

For the average investor, this simply indicates that the values that the markets calculate in the short run are largely consistent with more complex fundamental models on average in this case.

-=-=-=-=-=-=-==–=-=–==-=-

For those that understand Regression, here is the output of the main model at the 2008 year end:

And here is the same output of the main model at the 2007 year end:

output of the main model at the 2008 year end:

Here are the variable definitions:

  • LNMC — natural logarithm of the market capitalization
  • New — Less than one year old
  • Financial, Utility, Cyclical — If not a Industrial firm, which is the default, what is the difference in credit quality?
  • Volatility — 90-day implied volatility
  • OYTR — one-year total return
  • TYTR — three year total return
  • TYTRD — has the firm been around for three years?
  • DebtMC — Debt/Market Capitalization
  • CDebtMC — Change in Debt/Market Capitalization over the past two years

The dependent variable grades from 1 for a AAA company to a 22 if it is in default.