CAPITAL REQUIREMENTS

Credit Measurement: Avoiding Unintended Results Part 1

by Peter O. Davis his article—the first in a series—provides an overview of measurement terminology and a basic risk meas- Turement framework. The series focuses on credit risk measurement concepts, how they are implemented, and potential inconsistencies between theory and application. Subsequent articles will present common implementation options for a given concept and examine how each affects the credit risk measurement result. he basic concepts of Trend Toward Credit Risk scorecards to lower underwriting credit risk measure- Quantification costs and improve portfolio man- T ment— probabili- As credit risk modeling agement. Quantification of com- ty, recovery rate, exposure at methodologies have improved mercial credit risk has moved for- default, and unexpected loss—are over time, have incorporat- ward at a slower pace, with a sig- easy enough to describe. But even ed models into risk-grading, pric- nificant acceleration in the past for people who agree on the con- ing, portfolio management, and few years. The relative infre- cepts, it’s not so easy to imple- decision processes. As the role of quency of defaults and limited ment an approach that is fully credit risk models has grown in historical data have constrained consistent with the starting con- significance, it is important to model development for commer- cept. Small differences in how understand the different options cial credit risk. credit risk is measured can result for measuring individual credit Although vendors offer in big swings in estimates of cred- risk components and relating default models for larger (typically it risk—with potentially far-reach- them for a complete measure of public) firms, quantifying the risk ing effects on risk assessments credit risk. of small business credits remains a and business decisions. Consumer lenders, in particu- challenge for banks where there is lar, rely heavily on borrower a void between scorecard-driven © 2004 by RMA. Peter Davis is director of Credit Risk Services in Ernst & Young LLP’s Global Financial Services Advisory practice.

88 The RMA Journal April 2004 Credit Risk Measurement: Avoiding Unintended Results Part 1 approaches to retail credits and Credit Risk Measurement having dual ratings systems (one formally graded large facilities. Conceptual Framework rating for PD and the other for The general framework for LGD), most banks had a one- Basel II as a driver. The measuring credit risk is simple. dimensional rating system. introduction of dual rating sys- Credit risk can be divided into Borrower and facility were consid- tems that would meet the criteria two components: expected loss ered together in assigning a rating, proposed in Basel II has acceler- and unexpected loss. As formal- which was the product of the ated the interest in credit models ized in Pillar I of Basel II, expect- (PD) and to support risk-grading frame- ed losses are calculated as: the fraction of the loan’s value works. Credit models have been Exposure likely to be lost in the event of introduced to explicitly measure Probability of Default (PD) default (LGD). Typically, strong borrower default risk and the risk (LGD) collateral supporting a facility of loss given default on individual (EAD) could notch up an obligor risk rat- loan facilities. Basel II has been a Expected Loss (EL) ing. These rating systems were catalyst for banks to develop and Historically, banks have com- commonly used to rank-order risk, or strengthen measures of loss bined the risk of PD and LGD rather than to link ratings to given default. Significant into a single measure. Rather than explicit measures of credit loss. improvements in LGD modeling approaches can be expected over the next few years as more data Credit Risk Measurement Terminology becomes available and institu- A sign of the advancement of credit risk measurement has been the emergence of tions validate recently imple- common terminology. This reflects, in part, the maturity of measurement methodolo- mented modeling/grading gies and the initial success of Basel II in creating a common language. Several years methodologies. ago, the discussion of PDs, LGDs, and EADs would have produced blank stares in most The development of robust circles. These acronyms have now worked their way into everyday discussions among frameworks has credit and noncredit professionals. allowed banks to measure credit Probability of Default (PD). This is the risk that an obligor, or borrower, will fail to risk concentrations and the mar- make full and timely payment on its financial obligations over a given time horizon. ginal contribution of new credit This is a frequency measure. Another common name for this concept is expected exposures. As institutions have default frequency (EDF). become more comfortable with their economic capital frame- Loss Given Default (LGD). This is a less mature measurement concept. It measures the works, risk-adjusted performance risk that a loss will be incurred given that there has been a default event. This same concept may be called loss in the event of default (LIED) or loss severity. The inverse measurement has become a key of this concept is the recovery rate. LGD is typically expressed on a net present value driver for assessing customer basis. It is expressed as a percent of the outstanding balance at default. profitability, measuring business line performance and setting sen- Exposure at Default (EAD). This concept only applies to non-term exposures, such as ior management compensation. lines of credit and is also known as usage given default (UGD). This is the measurement Basel II also will be a catalyst for of the expected drawn exposure at the time of default. For consumer portfolios, the con- the introduction and enhance- cept of exposure at default is often captured in the dollar-weighted default probability. ment of economic capital frame- Expected Loss (EL). This concept is simply the product of the above three metrics mul- works. To qualify for the tiplied by the exposure and is a measure of the probability of credit loss over a given advanced approaches under Basel time horizon. For a loan portfolio, over an extended period of time, expected loss II, banks must be able to meas- should be comparable to the net charge-off rate. ure economic capital and use this information to support business Unexpected Loss (UL). This is a measure of the volatility of credit risk and captures decisions. portfolio concentration risk. Unexpected losses may be calculated using a credit value- at-risk methodology (as popularized by JPMorgan's CreditMetrics methodology) or by using stress testing and scenario analysis.

89 Credit Risk Measurement: Avoiding Unintended Results Part 1

The last term in the formula As much by convention as by tomer or industry could lead to for expected loss—exposure at theory, expected loss is typically greater volatility in credit losses. default (EAD)—is necessary for measured over a one-year period. commitments and revolving cred- Under Basel II, this one-year hori- Theory Versus Practice: Getting its, which may be undrawn, fully zon is formalized under Pillar I. A the Details Right drawn, or anywhere in between. one-year view is also often taken Theorists can deal in con- (Depending on how an institution for economic capital frameworks. cepts, but practitioners need con- manages its lines, a borrower may In contrast, many banks have used crete (and consistent) results. not be able to draw as it nears a life-of-the-loan concept for set- Given the costs of being wrong, default or the borrower may fully ting loan loss reserves. Rather than the process of quantifying credit draw down on the line.) assessing risk over a one-year peri- risk is like walking through a If exposure-weighted default od, credit risk is measured over the minefield: Every step requires rates are used, then EAD is remaining life of the portfolio. judgment and experience, and a already implicitly captured in the While expected loss is a misstep can have disastrous default risk component. (An expo- measure of average losses over a results. Each element in the sure-weighted default rate meas- given risk horizon, unexpected process—data selection, risk driv- ures the default amount as per- loss (UL) is a measure of what ers, measurement methodology, cent of the outstanding portfolio could go wrong in some stressed, and model application—drives balance.) If incidence-weighted or unexpected, scenario over this differences in the final estimate. default rates are used (as required same time period. Institutions A sound understanding of the for Pillar I internal ratings calcula- typically use a credit value-at-risk nuances of credit risk measure- tions under Basel II), then out- (CVaR) approach to measure at a ment will help bankers choose standing exposure at the time of given confidence interval the risk which lending opportunities to default must be measured sepa- of unexpected losses over a given pursue and then price and moni- rately in an EAD calculation. (An period. Measuring UL gives an tor the credits appropriately. ❒ incidence-weighted default rate institution a measure of the Contact Davis by e-mail at measures the number of defaults potential volatility in its credit [email protected]. as a percent of the number of portfolio. For example, an active borrowers.) increased exposure to a given cus-

Practical Implementation Issues: Future Article Topics While basic credit risk measurement concepts are easily understood, the way these concepts are implemented can yield a range of results—some useful, some potentially misleading. Future articles in this series will examine the following topics: Default probabilities—use of incidence versus exposure-weighted default rates. This article will examine the challenges and potential pitfalls in moving from an incidence-based measure of default risk to a measure of expected credit losses, which implicitly relies upon an exposure-weighted default measure. The article will show how using unadjusted incidence-based default probabilities could result in understating or overstating potential defaults. Loss given default—selection of the discount rate. This article will focus on the selection of the discount rate for the net present value calculation of loss given default (LGD). The U.S. supervisor’s guidance for Basel II implementation proposes that banks use a current market rate for discounting post-default cash flows. To calculate reserves for specific impaired loans, institutions are required to use the rate on the loan at the time of default. This article shows how these alternatives—current market rate versus bor- rower rate—affect the LGD estimate. It also examines how LGDs need to be applied and interpreted differently under the two dis- counting approaches. Basel II Pillar I and loan loss reserves—similarities and differences. As institutions move toward quantifying the credit risk of commercial lending activities, there will be an inevitable convergence and/or comparison of metrics used across the institution. Basel II’s Pillar I and loan loss reserves for non-impaired loans is an obvious starting point. This article will focus on the integration of Basel II expected loss calculations and loan loss reserve frameworks. The quantification of credit risk will drive an understanding of the sim- ilarities and differences considering a variety of factors such as time horizons, lookback periods, and the incorporation of unexpected losses. This article will focus on the similarities and differences between these two measures of expected losses.

90 The RMA Journal April 2004