Portfolio approach

While the asset-by-asset approach is a critical component to managing credit risk, it
does not provide a complete view of portfolio credit risk, where the term ‘risk’ refers to the possibility that actual losses exceed expected losses. Therefore, to gain greater
insights into credit risk, banks increasingly look to complement the asset-by-asset
approach with a quantitative portfolio review using a credit model.
A primary problem with the asset-by-asset approach is that it does not identify or
quantify the probability and severity of unexpected losses. Historical migration
analysis and problem loan allocations are two different methods of measuring the
same variable; i.e. expected losses. The ALLL absorbs expected losses. However, the
nature of credit risk is that there is a small probability of very large losses.

The practical consequence of these two return distributions is that, while the mean
and variance fully describe (i.e. they define all the relevant characteristics of) the
distributions of market returns, they do not fully describe the distribution of credit
risk returns. For a normal distribution, one can say that the portfolio with the larger
variance has greater risk. With a credit risk portfolio, a portfolio with a larger variance
need not automatically have greater risk than one with a smaller variance, because
the skewed distribution of credit returns does not allow the mean and variance to
describe the distribution fully. Credit returns are skewed to the left and exhibit ‘fat
tails’; i.e. a probability, albeit very small, of very large losses. While banks extending
credit face a high probability of a small gain (payment of interest and return of
principal), they face a very low probability of large losses. Depending upon risk
tolerance, an investor may consider a credit portfolio with a larger variance less risky
than one with a smaller variance, if the smaller variance portfolio has some probability
of an unacceptably large loss.
Credit risk, in a statistical sense, refers to deviations from expected losses, or
unexpected losses. Capital covers unexpected losses, regardless of the source; therefore,
the measurement of unexpected losses is an important concern.

Banks increasingly attempt to address the inability of the asset-by-asset approach
to measure unexpected losses sufficiently by pursuing a ‘portfolio’ approach. One
weakness with the asset-by-asset approach is that it has difficulty identifying and
measuring concentration risk. Concentration risk refers to additional portfolio risk
resulting from increased exposure to a borrower, or to a group of correlated borrowers.
For example, the high correlation between energy and real estate prices precipitated
a large number of failures of banks that had credit concentrations in those sectors
in the mid-1980s.
Traditionally, banks have relied upon arbitrary concentration limits to manage
concentration risk. For example, banks often set limits on credit exposure to a given
industry, or to a geographic area. A portfolio approach helps frame concentration
risk in a quantitative context, by considering correlations. Even though two credit
exposures may not come from the same industry, they could be highly correlated
because of dependence upon common economic factors. An arbitrary industry limit
may not be sufficient to protect a bank from unwarranted risk, given these correlations.
A model can help portfolio managers set limits in a more risk-focused manner,
allocate capital more effectively, and price credit consistent with the portfolio risks
entailed.14
It is important to understand what diversification can and cannot do for a portfolio.
The goal of diversification in a credit portfolio is to shorten the ‘tail’ of the loss
distribution; i.e. to reduce the probability of large, unexpected, credit losses. Diversification
cannot transform a portfolio of poor quality assets, with a high level of
expected losses, into a higher quality portfolio. Diversification efforts can reduce the
uncertainty of losses around the expectation (i.e. credit ‘risk’), but it cannot change
the level of expected loss, which is a function of the quality of the constituent assets.

A low-quality portfolio will have higher expected losses than a high quality portfolio.
Because all credit risky assets have exposure to macro-economic conditions, it is
impossible to diversify a portfolio completely. Diversification efforts focus on eliminating
the issuer specific, or ‘unsystematic’, risk of the portfolio. Credit managers
attempt to do this by spreading the risk out over a large number of obligors, with
cross correlations as small as possible.

While the portfolio management of credit risk has intuitive appeal, its implementation
in practice is difficult because of the absence of historical data and measurement
problems (e.g. correlations). Currently, while most large banks have initiatives to
measure credit risk more quantitatively, few currently manage credit risk based upon
the results of such measurements. Fundamental differences between credit and
market risks make application of MPT problematic when applied to credit portfolios.
Two important assumptions of portfolio credit risk models are: (1) the holding
period or planning horizon over which losses are predicted (e.g. one year) and (2)
how credit losses will be reported by the model. Models generally report either a
default or market value distribution. If a model reports a default distribution, then
the model would report no loss unless a default is predicted. If the model uses a
market value distribution, then a decline in the market value of the asset would be
reflected even if a default did not occur.15
To employ a portfolio model successfully the bank must have a reliable credit risk
rating system. Within most credit risk models, the internal risk rating is a critical
statistic for summarizing a facility’s probability of defaulting within the planning
horizon. The models use the credit risk rating to predict the probability of default by
comparing this data against: (1) publicly available historical default rates of similarly
rated corporate bonds; (2) the bank’s own historical internal default data; and (3)
default data experienced by other banks. A sufficiently stratified (‘granular’) credit
risk rating system is also important to credit risk modeling. The more stratified the
system, the more precise the model’s predictive capability can be. Moreover, greater
granularity in risk ratings assists banks in risk-based pricing and can offer a
competitive advantage.
The objective of credit risk modeling is to identify exposures that create an unacceptable risk/reward profile, such as might arise from credit concentrations.
Credit risk management seeks to reduce the unsystematic risk of a portfolio by
diversifying risks. As banks gain greater confidence in their portfolio modeling
capabilities, it is likely that credit derivatives will become a more significant vehicle
to manage portfolio credit risk. While some banks currently use credit derivatives to
hedge undesired exposures, much of that activity only involves a desire to reduce
regulatory capital requirements, particularly for higher-grade corporate exposures
that incur a high regulatory capital tax. But as credit derivatives are used to allow
banks to customize their risk exposures, and separate the customer relationship and
exposure functions, banks will increasingly find them helpful in applying MPT. 338