|
During the 1950s, Harry Markowitz championed the notion
that market risk should be managed at the portfolio level. It took almost
50 years for researchers to develop effective models for doing the same
thing with corporate credit risk. Certainly, retail credit risk—auto
loans, mortgages, credit card debt—has been managed at a portfolio level
for decades. This has been possible due to the relative homogeneity of
retail obligors. Corporate obligations—loans,
bonds,
derivatives, etc.—are
far less homogenous. Also, they represent greater concentrations of risk.
We are more comfortable modeling 50,000 consumer loans of
USD 10,000 each
as homogenous than we are doing so with 10 corporate obligations of USD
50MM each!
Investors and counterparties have generally managed the
portfolio-level implications of credit risk by
diversifying
their exposures and avoiding
concentrations.
Several factors have contributed to the new interest in
more systematically measuring and managing the sum credit risk of an
entire portfolio of obligations—what is called
portfolio credit risk. These
include:
bank industry interest in having the
Basel II accordpermit
the use of internal models for calculating credit risk
capital charges;
efforts to improve the application of
risk limits and
capital allocation for portfolio credit risk;
efforts to price
credit derivatives linked
to baskets of defaultable obligations;
the emergence of collateralized debt
obligations, which represent customized interests in portfolios of
defaultable bonds.
Portfolio credit
risk models are financial models that assess portfolio credit
risk. Output takes various forms. Many models can be run in either of two
modes. In a mark-to-market mode
fluctuations in a portfolio's market value resulting from defaults or
changes in credit ratings are modeled. Output might be the standard
deviation or some quantile of market value. This form of analysis differs
from that of value-at-risk
measures of market risk in that they consider changes in market value due
only to obligor-specific credit events. It does not consider, for example,
changes in credit spreads due to general market sentiment or changes in
liquidity. In a default mode, portfolio
losses due to actual defaults are modeled. Output generally includes what
are known as:
expected loss:
the expected value over some specified
horizon of portfolio losses due to default;
unexpected
loss: some metric related to the second moment of portfolio losses
due to default over the same horizon. Metrics may be the
standard deviation or
some quantile of portfolio loss.
Both CreditMetrics and KMV offer mark-to-market and
default modes.
Portfolio credit risk models are constructed by
associating some sort of correlation
model with a default model.
The default model specifies unconditional probabilities of default for
individual obligations. The correlation model assigns default correlations
to pairs of obligations. A simple correlation model might assign all pairs
of obligations the same correlation. This may be reasonable if obligations
are fairly homogenous—perhaps all bank loans. More sophisticated are
factor models (also called sector models). These split each obligation's
default probability into two components. One is a function of some factor,
such as the performance of the stock market. The other is obligor
specific. For example, if the the factor is the stock market's performance
and all obligors are publicly traded firms, the allocation might be made
based upon the betas of the obligors'
stocks. Having a common factor impact all of their default probabilities
to a greater or lesser extent imposes a correlation on the obligations'
defaults. Generalizations allow for multiple factors.
The literature on portfolio credit risk models is
extensive. Crouhy et al. (2000)
and Gordy (2000) survey the
literature and compare standard models.
|
|