CAPITAL REQUIREMENTS

Estimating Probability of via External Data Sources: A Step Toward Basel II

by Nadeem A. Siddiqi and Shelley B. Klein xternal data will likely be necessary for most to achieve the foundation or advanced IRB approach. Although Ethe decision to use the data may be relatively simple, effec- tive use of it is more complicated and can be critical to success. Careful analysis of the definitions and characteristics implicit in the external data is essential, and banks that do this will likely be reward- ed with more appropriate -rating results and an easier time defending their processes to the regulators. anks considering their methods of calculating risk capital probability rates available from strategies for compliance as required by Basel II, can, and major external ratings agencies Bwith the Basel II Capital are, attempting to employ the such as Standard and Poor’s Accord will likely use external foundation internal ratings-based (S&P) and Moody’s Financial data sources for the estimation of approach proposed in Basel II. To Services. probability of default as part of calculate regulatory capital, banks However, banks need to be the solution. Definitions and must determine the probabilities careful when using this data as characteristics must be consistent of default associated with their proxies for specific loan portfolios. or adjusted to be so if the exter- portfolios, and then apply regula- 1. They need to understand the nal data is to be an effective and tor-determined method of calculating default statistically useful supplement to and rates. to ensure that it matches both internal data. Banks that do not One seemingly simple approach the ’s philosophy and have the most sophisticated banks can take is to use default processes. © 2002 by RMA. Siddiqi and Klein are senior consultant and director, respectively, with the Credit Analytics and Data Management Group of the Financial Services business unit of BearingPoint, Inc. The authors thank Charles Hill and Savitha Sagar for valuable research assistance. All information provided is of a general nature and is not intended to address the circumstances of any particular individual or entity. Although we endeavor to provide accurate and timely information, there can be no guarantee that such information is accurate as of the date it is received or that it will con- tinue to be accurate in the future. No one should act upon such information without appropriate professional advice after a thorough examination of the particular situation. The views and opinions are those of the authors and do not necessarily represent the views and opinions of BearingPoint, Inc.

68 The RMA Journal December 2002 - January 2003 Estimating Probability of Default Via External Data Sources: A Step Toward Basel II

2. They must be careful in 3. The advanced IRB approach mately $116 billion of debt. matching the characteristics allows banks to fully estimate These defaults represented a and structures of the loans the risk and regulatory capital record 4.09% of all rated issuers at with those of the underlying requirements internally. The the beginning of the year. This bonds from the published obvious advantage of using default ratio surpassed the previ- default studies, especially for the advanced IRB approach is ous record of 4.01%, set in 1991.1 loans below investment that if a well-managed bank In addition to a generally weak grade, which may require spe- calculates its capital require- global economy, an abundance of cific adjustments. ments for itself, and its risk recently issued speculative-grade 3. Some loan grades have little profile requires less capital debt, particularly in the leisure published default data avail- than the conservative regula- time/media and telecom sectors, able, and banks may need to tor-determined charges, more contributed to the very high extrapolate their default rates capital will be freed for other default rate. To reflect the chang- from available data. use, thus improving overall ing economic environment and Capitalizing on continuous profitability. the associated increases in advances in technology and ana- defaults, yield spreads on highly lytical methods, Basel II is Recent Default Activity speculative debt widened to a prompting banks to learn more The poor global and domestic point where costs associated with about quantitative economic environment that pre- issuing debt at such levels were measurement. Many banks’ vailed in 2001 resulted in record prohibitive to most prospective response will include using exter- defaults. During that year, S&P issuers. To lower their cost of capi- nal data to calculate risk capital. observed 216 defaults on approxi- tal, many issuers have been forced Basel II offers three different ways of determining risk capital: Table 1 1. The standardized approach, Corporate Defaults Rates—All Ratings, 1982 through 2001 which applies industry-based Year Default Rate (%) Number of Total Debt Defaulting averages to different asset Defaults ($ Billions) classes and is most suitable 1982 1.26 18 0.9 for very small banks that have 1983 0.68 10 0.4 a relatively homogenous asset 1984 0.83 13 0.4 base and do not have the 1985 1.09 18 0.3 resources to meet regulatory 1986 1.69 32 0.5 requirements to internally cal- 1987 0.93 19 1.6 culate risk. 1988 1.49 32 3.3 2. The foundation internal-rat- 1989 1.75 39 7.3 ings-based (IRB) approach, 1990 2.86 64 21.2 whereby banks need to inter- 1991 4.01 88 23.6 nally estimate part of the risk 1992 1.35 31 5.4 calculation formula, with the 1993 0.86 22 2.4 1994 0.62 18 2.3 rest being determined by 1995 0.97 32 9.0 (conservative) regulator-deter- 1996 0.56 20 2.7 mined rules. This approach 1997 0.59 23 4.9 may be suitable for small and 1998 1.26 56 11.3 medium-sized banks looking 1999 2.19 108 37.8 to get a foothold in the 2000 2.56 132 42.3 advanced approach, without 2001 4.09 216 116.1 having the resources to com- Source: Standard and Poor's, Special Report - Record Defaults in 2001 the Result of Poor mit to it fully. Credit Quality and a Weak Economy.

69 Estimating Probability of Default Via External Data Sources: A Step Toward Basel II to use alternative debt structures. ping is complete, any loan can pools. As used by S&P, the static In some cases, issuers provide col- then be given an equivalent S&P pool assesses the default rates of lateral when issuing debt, or issu- rating, for example, based on the all bonds of a given rating, ing , to receive high- internal bank rating. Then, based regardless of age.3 er credit ratings. A number of on the applicant’s , his- The S&P default rates do not issuers have also issued debt with torical default performance obser- take into account the age of an structures, including interest vations from S&P may be used to issuance. Default curves obtained reserves, which ensure interest estimate the probability and from S&P’s CreditPro database do payments to investors for a num- potential timing of default. Thus not reflect a new issue bias. New ber of years after issuance. Table 1 starting from existing internal bank issue bias refers to the expected provides historical corporate ratings of commercial loans, proba- results of the hypothesis that an default rates for the past 20 years. bilities of default can be attached. issuer that has just received a sub- stantial cash inflow from a bond The Basic Process A Few Kinks offering or loan is not likely to The risk calculation process While this process of obtain- default in the near term, regard- outlined in Basel II for those ing default probabilities from less of the issuer’s credit rating. A institutions seeking to apply one external data vendors seems new issue bias would be expected of the IRB approaches is also straightforward, several caveats to reduce marginal default rates in composed of three parts. First, the are in order. First, most of the the first three years subsequent to probability of default (PD) must empirical work on corporate the issuance of new debt, with the be obtained. Second, the loss defaults thus far has concentrated effect getting more pronounced as given default (LGD) must be on publicly traded bonds. Due to the credit quality declines. established. Finally, the exposure the private nature of the loan mar- To address this concern, at default (EAD) must be estimat- ket, there is limited publicly avail- Professor Ed Altman from New ed. The foundation IRB approach able loan default data. Second, York University developed a requires banks to estimate only since loan portfolios vary from series of default curves based the PD, while the LGD and EAD bank to bank, even if a reasonable upon this new issue bias hypothe- are determined by regulatory default database were available, it sis. As stated in Managing Credit guidance. In this article, we focus would be inappropriate to gener- Risk: The Next Great Financial on one methodology that can be alize these results for the entire Challenge, “The aging effect is used to estimate the probabilities market.2 intuitively sound, since most com- of default without committing What this implies is that panies have a great deal of cash substantial internal resources, to external default data must be just after they issue a bond. Even utilize the foundation IRB carefully scrutinized for suitability if their operating cash flow is neg- approach to regulatory capital cal- of application to any particular ative, they are usually able to culation, as well as to check any bank’s portfolio. This scrutiny meet several periods of interest internal calculations under the should center around at least two payments.”4 S&P also notes that advanced IRB approach. dimensions: default calculation “relatively few issuers default In processes methodology and characteristics early in their rated history.”5 used by virtually all banks, com- of the instruments. Similar to S&P’s default mercial loans are assigned a credit analysis, Altman grouped his rating. These credit ratings can be Default Calculation analysis according to credit-rating letter grades or numerical grades Methodology cohorts. However, Altman’s analy- covering multiple states. A map- At least two different sis differed in that his cohorts ping of these bank credit ratings methodologies have been used to were organized by issuers with can then be made to published rat- estimate default probabilities in the same original rating at ings of bonds by agencies such as the commercial sector. S&P’s issuance, as opposed to issuers S&P and Moody’s. Once this map- standard default curves use static with the same ratings as of a ran-

70 Estimating Probability of Default Via External Data Sources: A Step Toward Basel II

Table 2 Table 3 indicates that, histori- Cumulative Default Rates cally, issuers that received funds at speculative ratings tended to Year 1 2 3 4 5 6 7 avoid default for longer periods compared with issuers who may S&P Credit Pro have issued debt at higher ratings but subsequently slid down the A 0.04 0.12 0.18 .027 0.43 0.28 .072 BBB 0.29 .068 .098 1.52 2.04 2.42 2.84 rating scale because of financial BB 1.07 2.97 5.27 7.26 8.94 10.73 11.82 difficulties. For example, Table 3 B 9.29 18.21 24.22 27.71 30.23 32.47 33.99 indicates that issuers who CCC 24.72 33.06 38.40 42.60 46.87 48.48 49.62 received an initial rating of CCC and eventually defaulted have Altman’s MMR taken, on average, 3.1 years to A 0.00 0.00 0.03 0.15 0.21 0.23 .025 default (CCC; “Average Years BBB 0.02 0.31 0.58 1.25 1.49 1.89 1.99 from Original Rating”). BB 0.38 1.13 3.78 5.26 7.56 8.49 10.50 Conversely, issuers who had a last B 1.16 4.15 9.75 15.30 19.21 21.62 23.82 rating of CCC and eventually CCC 2.06 15.6 28.51 34.53 36.52 42.71 44.91 defaulted (but may have had a dom observation date (that is, ning of the measurement period. higher rating when they issued static pools). Dr. Altman devel- Altman’s data tends to be more debt) typically defaulted within oped his default probability volatile from year to year, espe- five months of receiving the CCC curves by measuring defaults in a cially when significant bonds rating (CCC; “Average Years from particular period relative to the default, such as the multi-billion Last Rating”). base population in the same peri- dollar Texaco default in 1987 or Problems with both method- od. Therefore, depending on the the Enron default in 2001. The ologies begin to emerge for the number of observed issuers with- seven-year cumulative default highly speculative issuances below in each cohort at the beginning of rates implied by the two method- CCC (CCC-, CC, and C). Whether the year that could possibly ologies are shown in Table 2. The using static pools or first rating default during the year, the difference in marginal default cohorts, there is no data available denominator used to calculate the rates over the first few years for ratings below CCC. While probabilities of default may between the two methodologies cumulative and marginal default change every year. This method- is clearly visible, and the differ- probability data are available from ology was based on the actuarial ence gets more pronounced as the S&P for CCC-rated debt, no methodologies used by insurance credit quality declines. default rates are provided by S&P companies and was named the for any ratings below CCC. Default Marginal Mortality Rate Table 3 Methodology (MMR). This Time to Default by Rating Category methodology also differs signifi- Average Years Average Years cantly from S&P’s analysis in that Original Rating from Original Rating Last Rating from Last Rating Altman measured the magnitude of default as the dollar value of AAA 8.0 AAA N.A. defaults as a percent of the total AA 11.9 AA N.A. dollar value of debt rated at the A 10.7 A N.A. BBB 7.4 BBB 1.3 beginning of the period. S&P dif- BB 5.1 BB 1.5 fers in its methodology by meas- B 3.8 B 1.3 uring the number of issuers that CCC 3.1 CCC 0.4 have defaulted as a percentage of N.R. N.A. N.R. 3.3 all the issuers rated at the begin- Total 4.7 Total 1.1 71 Estimating Probability of Default Via External Data Sources: A Step Toward Basel II

Table 4 Syndicated Bank Loans versus —At-Issuance Mortality Rates

Year 1 Year 2 Year 3 Year 4 Year 5 Bank Bond Bank Bond Bank Bond Bank Bond Bank Bond Baa 0.04% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 054% 0.00% 0.00% Ba 0.17% 0.00% 0.60% 0.38% 0.60% 2.30% 0.97% 1.80% 4.89% 0.00% B 2.30% 0.81% 1.86% 1.97% 2.59% 4.99% 1.79% 1.76% 1.86% 0.00% Caa 15.24% 2.65% 7.44% 3.09% 13.03% 4.55% 0.00% 21.72% 0.00% 0.00%

Source: Caouette, John B., Altman, Edward I., and Narayanan, Paul. Managing Credit Risk: The Next Great Financial Challenge. rates will therefore need to be ments. These factors might con- dicated bank loans, making it very extrapolated by banks for credit tribute to lower initial default rates difficult to verify any of the find- ratings below CCC. Hence, it (and in some cases higher credit ings presented in Table 4. Dr. should be noted that any compar- ratings). These structural character- Altman notes that, “The marginal isons made and conclusions drawn istics need to be compared to the mortality rate results and its infor- at the CCC level need to be tem- loan structures the bank is issuing. mation content concerning the pered by a consideration of the lack aging effect of corporate loan of data points at this rating level. Corporate Bonds versus default rates is not conclusive.”6 Syndicated Bank Loans As the liquidity of the sec- Instrument/structure char- As noted above, due to the ondary whole-loan market increas- acteristics. The second dimen- lack of available data on the histori- es in the future, so will the quan- sion that needs to be scrutinized cal performance of syndicated bank tity and quality of public informa- before default probabilities can be loans, historical corporate bond per- tion available. Until then, it will applied to bank portfolios is the formance data has often been used remain difficult to accurately draw characteristics of the loans in the to predict the probability and tim- meaningful conclusions, particu- bank portfolio relative to those of ing of default that may occur. larly at the speculative end of the the bonds in the published Table 4 indicates that while credit-rating range. ❐ default studies. This is especially cumulative default rates for bonds the case for speculative-rated and loans may even out after a Siddiqi and Klein may be contacted at debt, whose specific structures number of years, syndicated loans [email protected] and greatly affect the cash flow. This tend to have much higher initial [email protected] is probably one reason why, in default rates, again especially at Table 2, we note that the differ- speculative ratings. These rates Notes ence in default rates between the 1 Standard and Poor’s, 2002, Special Report— indicate that bond structures may Ratings Performance 2001. static pool and at-issuance have been substantially different 2 Altman, E., Suggitt, H., 2000, Default Rates in the methodologies increases as we Syndicated Bank Loan Market: A Mortality Analysis, from loan structures. These differ- Journal of Banking & Finance, 24, 229-253 move down the credit rating. ences could be due to the use of 3 Standard and Poor’s, 2001, CreditPro 5.0 User In addition to the added liq- Guide. According to S&P, “A Static Pool consists of all more complex structures for bond of the rated obligors on the first day of the year or a uidity provided by the proceeds of quarter and these obligors are followed from that issuances that are designed to pro- a bond issuance, recently issued point on. Thus, a Static Pool is a grouping of obligors tect investors from default during whose members remain constant. The results pre- bonds, especially speculative sented in tables with all Static Pools represent a the years immediately following bonds, have had structural weighted average based on the number of obligors issuances well as the typically in each pool and rating category, at the beginning of enhancements such as interest pay- each period, over a specific time period. shorter average maturities for ment guarantees for the first years 4 Caouette, John B., Altman, Edward I., and bank debt compared to corporate Narayanan, Paul., 1998, Managing Credit Risk: The of the bond’s life, or in some cases Next Great Financial Challenge. New York: John bonds. Unfortunately, very little Wiley and Sons, Inc. they have been structured with data is available specifically on 5 Standard and Poor’s, 2001, CreditPro 5.0 User escrow accounts containing a por- Guide. the historical performances of syn- 6 Altman, E., Suggitt, H., 2000. tion of the bond’s interest pay-

72 The RMA Journal December 2002 - January 2003