3.7 MEASURING CREDIT RISK FOR CREDIT PORTFOLIOS
3.7.3 Credit Loss Measurement Definition .1 DM Versus MTM Models
Both the default-mode (DM) and mark-to-market (MTM) approaches esti- mate the credit losses from adverse changes in credit quality. The quality of credit models is primarily influenced by the fit between the model out- put and the model application. The model choice should be made based on the circumstances of the use and application. An institution that uses a portfolio of liquid credits and exposure to credit spreads (i.e., the hedge transactions for credit portfolios using credit spreads between different term structures) may require a credit loss measurement definition that in- corporates potential shifts in credit spreads and thus opt for the (more complicated) MTM model with the multistate nature.
3.7.3.2 DCCF Versus RNV Approaches
In practice, the difference between the approaches is smaller than in theory, because the credit value is priced as a discounted present value of its future cash flows in both approaches. The dichotomy is sharper in theory because the discount factors are calculated differently. The discounted contractual cash flow (DCCF) approach assumes a nonparametric approach to esti- mating these discount factors. The public debt issuers (issues) are grouped into rating categories, and the credit spreads on the issuers are then aver- aged within each rating “bucket.” On the other hand, the risk-neutral val- uation (RNV) approach is more complex. In a structural process, each credit is simultaneously modeled in an individual framework. This means that the modeling of the market risk premium for each credit in the RNV model is typically referenced to credit spreads from the debt market.
The empirical evidence of the econometric theory shows that highly structural estimators make efficient use of available data but are vulnera- ble to model misspecification. Nonparametric estimators make minimal use of modeling assumptions but underperform where data integrity is a problem. The two approaches will, in general, assign different credit losses to any given loan. Under normal market conditions (liquidity and information efficiency) and the stable assumptions of the RNV model,
Credit Risk 147
both approaches should deliver reasonable output for a well-diversified credit portfolio.
The probability that a particular credit contract will default during the time horizon is a critical input. The bank’s credit staff has to assign internal credit risk ratings for all credit contracts. This is done for most corporate customers. The trend is that all customers, corporate or retail, are assigned risk ratings in order to obtain the overall risk profile of the credit portfolio, including the correlations between the different segments of the portfolio as they are exposed to different business cycles, macroeconomic factors, etc.
There are basically three approaches (which can be combined) for as- signing a credit rating to a customer or the contract:
• The traditional approach, based on financial, accounting, and other characteristics of the customer. This approach is very subjective and is based on the reliability and availability of specific information.
• Credit-scoring models supplied by commercial vendors, which also deliver a database that reflects the best-practice standards in the market.
• Credit-scoring models developed internally, which reflect the structure and the processes of the credit department.
External ratings are frequently combined with internal rating cate- gories, which allows combination of the internal credit authorization process and external rating data. The expected default frequency (EDF) can be interpreted as a credit’s probability of migrating from its current inter- nal rating grade to default within the credit model’s time horizon. The like- lihood of such a migration from its current risk rating category to another category is defined in the transition matrix, as illustrated in Table 3-3.
Given the contract’s actual rating (defined by each row), the proba- bility to migrate to another category (defined by the columns) is defined within the intersecting cell of the transition matrix. Thus, in Table 3-3, the likelihood of a credit contract rated A migrating to BBB within one year would be 7.4 percent. The likelihood of a credit contract rated CCC mi- grating to default within one year would be 18.6 percent.
3.7.3.3 Unconditional Versus Conditional Models
In a broad sense, all models are conditional, as they process input infor- mation on the credit quality of the borrower and the credit contracts. In a narrower sense, it is possible to distinguish between unconditional and conditional models. The unconditional models process information lim- ited to the borrower and the credit contracts. The transition matrix and the correlations are modeled to capture the long-run values of these parame- ters. But such long-run averages may misrepresent the short-term condi- tion, as correlations and default frequency tend to vary systematically
with the course of the business cycle. The conditional models process in- formation on the borrower and the credit contracts and, in addition, are linked to macroeconomic information such as gross domestic product, current levels and trends in domestic and international employment, in- flation rates, indicators specific for particular sectors, etc.
CreditMetrics and CreditRisk are examples of unconditional risk mod- els. They estimate the expected default frequency and correlations between historical default data and borrower- or contract-specific information, such as the rating. These models are based on data collected and estimated over many credit cycles to reflect the averages of these parameters. They should predict reasonable credit loss probabilities based on the transition matrix, if the credit portfolio is composed of similar credit contracts. This type of model has drawbacks; if the borrowers are upgraded or downgraded over time, their expected default rates will be revised downward or upward.
Such a portfolio will not have a similar standard deviation over time and will be more complex to manage relative to given risk levels. The uncondi- tional models are not able to incorporate macroeconomic parameters such as business cycle effects. The tendency for rating improvement or deteriora- tion is positive during cyclical upturns or downturns, respectively.
The drawbacks of the unconditional models are avoided with the con- ditional models. Examples of conditional credit risk models include McKin- sey and Company’s CreditPortfolioView12 and KMV’s PortfolioManager.
Credit Risk 149
T A B L E 3-3
Sample Credit Rating Transition Matrix: Probability of Migrating to Another Rating Within 1 Year
Rating at Year End, % Initial
Rating AAA AA A BBB BB B CCC Default
AAA 87.74 10.93 0.45 0.63 0.12 0.10 0.02 0.02
AA 0.84 88.23 7.47 2.16 1.11 0.13 0.05 0.02
A 0.27 1.59 89.05 7.40 1.48 0.13 0.06 0.03
BBB 1.84 1.89 5.00 84.21 6.51 0.32 0.16 0.07
BB 0.08 2.91 3.29 5.53 74.68 8.05 4.14 1.32
B 0.21 0.36 9.25 8.29 2.31 63.89 10.13 5.58
CCC 0.06 0.25 1.85 2.06 12.34 24.86 39.97 18.60
SOURCE: Greg M. Gupton, Christopher C. Finger, and Mickey Bhatia, CreditMetrics Technical Document,New York: Morgan Guaranty Trust Co., April 1997, 76. Copyright © 1997 by J. P. Morgan & Co., Inc., all rights reserved. Reproduced with permission of RiskMetrics Group, Inc.
Within its conceptual modeling, the transition matrices of CreditPortfo- lioView are related to the state of the economy, as the matrices of the covari- ance matrix are modified to give an increased likelihood of an upgrade (and decreased likelihood of a downgrade) during an upswing (or downswing) in a credit cycle. KMV’s PortfolioManager links the process of estimating the asset values, rates of return, and volatility to current equity prices, which are information-efficient and incorporate all information available in the market. This approach is comparable to the arbitrage price theory (APT) and the multifactor models. Empirical research has generated empirical ev- idence. The drawbacks of these models are timing and parameterization.
They might underestimate losses as the credit cycle enters a downturn and overestimate losses as the cycle bottoms out.