Risk: Measurement and Pricing
Clarendon Lectures in Finance were given by Darrell Duffie.
He spoke on the subject of “Default Risk: Its Measurement and
Pricing” and the subject of his three lectures were bankruptcy
probabilities, default risk pricing and default correlation.
In his first lecture, Darrell Duffie examined the historical
patterns of default, the default intensity, some structural models of
default and estimates of the term structure of default probabilities.
He examined the performance of credit ratings in predicting
defaults and derived estimates of survival probabilities based on default
rates. He then reported
estimates of Chapter 11 and Chapter 7 failures in the US over the period
1971 to 2001 based on the distances of firms from default (the number of
standard deviations distance of assets from liabilities) and a
macroeconomic variable – personal income growth.
From this he was able to derive a term structure of default
his second lecture, Darrell Duffie turned to the pricing of default risk.
He described how default probabilities can be determined from
credit market spreads and expected losses given default.
He used this to determine the risk premium on default insurance
provided by credit derivatives. He
showed that there was considerable time variation in the risk premium
associated with credit derivatives with peaks in the premium occurring in
the middle of 2002. He argued
that this may have been due to a shortage of capital in the market at that
his third lecture, Darrell Duffie extended his analysis to multiple
issuers and the effect of correlation in defaults.
He reported that one year average default event correlations within
sectors average around 5%. A commonly employed technique to dealing with
correlations is a copula, which specifies correlations in variables such
as default times whose probability distributions are already determined.
He argued that this approach is unsatisfactory in so far as it does
not include both correlated default and correlated uncertain changes in
spreads. He argued that the
double stochastic model does not capture correlation in default intensity
processes. This may be due to
missing covariates that lead to omitted common factors.
|The event was sponsored by