Standard & Poor's Financial Institutions Ratings analysts have spent a considerable amount of time in the past few months reviewing our internal capital model. This article introduces our work up to this point and spells out future directions.
What we would most like to achieve is a constructive dialogue on capital adequacy with issuers, investors, and other interested parties. Our objective is to develop an analytical methodology to independently assess the capital adequacy of financial institutions. We will determine required capital for various risk profiles and adjusted risk-weighted assets for inclusion in adjusted regulatory ratios. In recognition of the variation in available information across financial institutions globally, whether due to required public disclosures, regulatory requirements, or the institution's willingness to share confidential information, we are prepared to adopt a multipronged approach. We will use both information that is publicly disclosed and information that is disclosed confidentially only to Standard & Poor's. We will request specific information from all financial institutions, which may be the same as what they are required to report to regulators as per regulatory capital requirements. (By "capital," we mean adjusted total equity, which includes common equity and preferred stock, with limitations on preferred, net of intangibles, and other comprehensive income related to available-for-sale securities. Our definition of capital does not include subordinated or other debt instruments.)
The project has been separated into two phases. Phase I, which has so far been largely completed, involves assessing capital requirements for credit and market risk. Phase II, which begins next year, will deal with interest rate risk, operational risks, diversification benefits, intangibles, earnings impact, and other more involved issues that may surface.
Our capital model is calibrated to an 'A' counterparty credit rating. Companies with lower ratings may have less capital. Companies with ratings higher than 'A' may have less capital than that required by the model but retain those ratings due to offsetting strengths.
Caveats
As with any capital model, there are many assumptions that must be made. We also have to rely on statistics, so a word of caution is in order. Statistical analysis of loss or return data has numerous shortcomings that may not be readily apparent. Firstly, an industry's portfolio is more diversified than the portfolio of any one entity. Therefore, an index of losses for a given loan type will exhibit less volatility than what may be experienced by an individual firm. We implicitly adjusted for this in a few cases by looking at the individual FI worst-case experience or by segmenting portfolios into subcategories. There is also survivorship bias, which is readily apparent from reviewing loss statistics on subprime auto lenders (many of those that defaulted were no longer included in the indices).
In addition, reported financial data often fails to reflect all costs associated with the poor performance of a portfolio. For example, as delinquencies mount, the operating costs of running a collection department also rise, as do the losses. There is also the statistical error of continuing to use a model that no longer reflects reality. A good example of this is the "100-year flood" that seems to occur at least once every five years.
It is also very important to note that what we are presenting here are capital charges for credit and market risks, not for operational risks. In general, we feel that companies need a lot more capital for operational risks than their economic capital models are telling them they need.
The "Going Concern" Problem
Before getting into a detailed discussion of our approach to capital requirements for credit and market risks, it is worthwhile to review what we call the "going concern" problem.
Since all financial institutions rely on confidence-sensitive funding, there is a limit to how much of a loss one can absorb before creditors completely shut out the company from further borrowing. We call this the "going concern" problem. Essentially, a financial institution will default before all of its capital is exhausted by credit, market, or operational losses. However, this may indicate that the institution needs capital for other types of risks. It is not clear, however, what type of risk "going concern capital" is providing a cushion against. If this is a capital layer below capital available for credit, market, liquidity, interest rate, and operational risks, is it for business risk?
On the other hand, one may question whether capital for extreme events covers "going concern" status, i.e. sufficient capital for an extreme enough event would be sufficient for the institution to be a going concern after lesser, albeit extreme, events. Also, an extreme credit or market loss may actually be the manifestation of operational losses. For example, in the case of Barings Bank, Nick Leeson's loss of $800 million on options was a failure of controls (operational risk) that appeared as market losses. We believe that very large losses are not likely to occur at a company without a breakdown of controls. Obviously, further research on the interplay of various risks and market perceptions is needed.
Auto Loans And Other Consumer Loans
We used the annual net charge-offs (NCO) for each of the prime and subprime subsectors of the auto lending industry, which have been maintained by structured finance dating back to 1995. The data are for individual companies that are active loan securitizers. The data from structured deals does not differ materially from the losses disclosed in SEC financial reporting by the same companies. The only significant adjustment done to the raw data was to reclassify Hyundai Motor Co. and Mitsubishi Motors as nonprime companies, as their borrower profiles better match those of a subprime borrower. Additionally, in both subsectors, the highest three NCOs were removed as these were considered outliers.
Using the data, the worst-case scenarios were derived by selecting the actual worst-loss year for each of the two subsectors. Using worst-case experience was selected over a parametric approach to compensate for the short history of the data used to calculate the capital charge; the lack of a severe economic downturn during the observed period (1995-2004); the volatility of used car prices; and the advent of longer-term loans. This results in a base case capital charge of 1.9% for prime lenders and 11.9% for subprime lenders, corresponding to the worst-case years of 1999 and 2000, respectively. These base-case capital charges will be scaled upwards in cases where an institution's loss experience materially exceeds industry averages.
Commercial And Industrial Loans
The capital charge for commercial and industrial (C&I) loans is based on the industry's cumulative three-year worst loan loss experience of the past 20 years. We looked at data from the Federal Reserve on the largest 100 banks. We believe the loss experience of this group of financial institutions is highly representative of C&I lending in general. We examined data with and without seasonal adjustments and arrived at the same conclusion. The highest 12 quarter net charge-off (NCO) rate occurred from fourth-quarter 2000 to third-quarter 2003. This 36-month period corresponds with the fall-out of the telecom/technology bust and the deterioration within the banking industry's shared national credit portfolios. The cumulative three-year net charge-off rate was 4.80%. This is then the capital charge to cover a worst case scenario that lasts three years.
We also applied a capital charge to unfunded commercial loan commitments. Some portion of commitments is eventually funded, and a fraction of these do turn into losses. In determining a capital charge, one must recognize that it is very difficult for banks to determine what percentage of commercial loan commitments will ultimately be funded. The capital charge for unfunded commercial loan commitments is based on each bank's projected 12-month C&I loan growth. The dollar amount of loan growth is added to the beginning of the period amount. For example, if a bank has $100 million of C&I loans outstanding at December 31 of Year One and is projecting 5% growth in Year Two, the capital charge would be applied to $105 million.
In recognition of the differences in portfolio risk profiles between different C&I lenders, we will refine our required capital level (either raising or lowering it) depending on information that would indicate risk differences between the particular FI and the industry. There are three dimensions to this: distribution of internal risk ratings, industry concentrations, and single-name concentrations that all analysts request from their companies as part of ongoing surveillance.
Distribution of internal risk ratings
The overall quality of a commercial loan portfolio can be easily assessed by the distribution of internal risk ratings. Most banks use a one- to-10 scale, with the higher numbers reserved for regulatory definitions of criticized and classified credits.
If the bank's internal risk ratings on outstanding commercial loans can be mapped to Standard & Poor's ratings, we would calculate a more accurate capital charge. Those banks with less risky C&I loans would be charged less capital, while those banks with more risky C&I loans would be charged more.
Before doing this, we will determine whether a bank's internal risk ratings can indeed be mapped. While there is the risk that a bank does not accurately risk-rate its C&I loans, the bank examiners provide a backstop. The standard 4.8% capital charge for C&I credits corresponds to the average 'BB' C&I loan portfolio. The mapping would scale up or down from the standard capital charge, based on a simple extrapolation of Standard & Poor's worst three-year default probabilities.
Single-name concentration
Charges for this dimension will be based on internal risk ratings and amounts outstanding of the top 20 corporate relationships. The aggregate of the top 20 C&I loans outstanding is compared to the bank's capital (i.e., equity plus reserves). We have traditionally considered aggregate amounts greater than 100% as having a concentration in single-name credits. We could put this information to work by adding a surcharge for excess single-name concentrations (e.g., > 120%) and a discount (e.g., < 50%) for well-dispersed C&I loan portfolios.
Industry concentrations
If a bank is overly concentrated in a single industry, we may require additional capital. Alternatively, we might charge only those industries that we identify as having extra risk.
Commercial Real Estate Loans
We used FDIC data to find the worst three years of NCOs for the 100 largest banks. We feel this is more representative of the type of lending that financial institutions typically do, as compared to CMBS or REIT data. CMBS and REITS have a very high proportion of loans to large cash-flowing properties. Also, CMBS and REIT data does not include the early 1990s, which provide the basis for our capital charges since this was in our view a period of significant downturn in commercial real (CRE) estate lending.
We use the worst consecutive three-year cumulative loss rate. Because a severe CRE downturn can result in a prolonged period of illiquidity, it can take several years to work out a troubled CRE portfolio. However, we recognize that earnings on performing assets could reduce the required capital, especially when such a long period is considered. We will look more closely at the impact of earnings on capital requirements in Phase II.
In addition to the FDIC data, we also analyzed publicly disclosed data on Other Real Estate (ORE) write-downs for the top 30 banks in the five years beginning 1990 through 1994. The banks did not break out their ORE write-downs by property type, so we apportioned ORE write-downs to property types. In total, ORE write-downs served to increase the loss rate by 50% on average. The large amount of ORE write-downs is due to the banks' practice of aggressively classifying loans as foreclosed, even when that was not the case in a legal sense. As real estate prices deteriorated, further loss recognition was recorded as a noninterest expense rather than as a charge against reserves.
Table 1
Base Case Capital Charges for CRE Lending
(%)
Construction
15
Multi-family
10
Mortgages
10
Another worst-case scenario we examined is that of the Texas and energy-belt banks. This showed total losses on real estate in higher ranges even before ORE expenses. This would suggest that even higher losses are possible for regional banks in an asset deflationary environment than for the large diversified banks.
We also added commitments in the denominator of the loss rates, since it is reasonable to assume that commitments would be drawn on during the three-year period.
Credit Cards
The capital charge for credit cards is based on a review of quarterly NCO data for the period from 1995 through fourth-quarter 2004. Lagged (12 months) NCOs were used instead of coincident NCOs, as rapid growth in the industry can hide credit problems. We separated the data into two groups, one for subprime issuers and the other for prime issuers. From this data, annual four-quarter NCO numbers were calculated. This avoids the impact of cyclicality throughout the year. The worst-loss year for each of the two subsectors, prime and subprime lending, were 7.38% and 12.87%, respectively. These are higher loss levels than the averages for prime and subprime lenders (see Table 2), but are meant to represent worst cases and not averages.
Base case capital charges will be increased in the event that a company's loss experience is meaningfully higher than industry averages.
Table 2
Worst-Case Credit Card Net Charge-Offs
Worst Year
Coincident (%)
Lagged (%)
Base Capital Charge (%)
Prime Lenders
1997
5.97
7.38
7.38
Subprime Lenders
1998
14.52
12.87
12.87
Issues that have yet to be decided on are unfunded commitments (unused credit lines) and future portfolio growth. It is likely that we will have additional capital charges for portfolio growth. However, the use of lagged NCOs does imply an adjustment for portfolio growth. Also, the relationship between a company's earnings and its asset performance has not been considered. Going forward, we will seek to answer the question of whether they are getting paid enough for the risks they are taking.
Derivatives Credit Risk
In assessing a capital charge for the derivative asset class, the optimal approach would be to first determine the potential future exposure (PFE) for each institution's derivative portfolio and then estimate a worst-case loss. The capital charge would then be the number corresponding to the quantile cutoff of the loss distribution. Given that we don't have the actual position information for each institution's derivative portfolio, an estimate of the PFE is required. Quarterly GAAP financial data on derivative exposures was used from first-quarter 1994 to first-quarter 2005 to help predict PFE.
In all cases the credit exposures represented values after netting agreements were applied, and in some cases the credit exposures were further netted for collateral. A simple linear regression of the reported exposures against time was estimated for each financial institution. All regressions were statistically significant with strong r-squares. The predicted exposure from the regression line for the last data point in the series was used as the estimated exposure. To estimate the worst-case loss, a three-standard deviation of the regression residuals was added to this estimated exposure. This was then used as a measure of the PFE for that point in time.
To determine the required capital charge, the estimated PFE was bucketed into relevant credit quality using the percentages disclosed by the company. For each credit bucket, the average probability of default for one year and three years was determined. A zero recovery rate was assumed, which is consistent with having already deducted collateral. The expected loss for each credit bucket was calculated based on the probability of default.
Residential Mortgages
The residential mortgage capital charge is based on our review of FDIC data. The worst three-year cumulative loss experience (1992, 1993, and 1994) adds up to a NCO rate of 92 basis points (bps), which we rounded up to 1%. However, this does not capture the different risk characteristics of the individual banks' residential loan portfolios (FICO score ranges, loan-to-value (LTV), and other underwriting differences). Certain institutions may have mortgage portfolios with substantially different risk profiles than the average for the industry. To adjust for this, we will look at how a financial institution's loss experience deviates from industry norms and adjust the charge upward if there are meaningful differences.
One alternative to adjusting for loss experience above industry historical average is to customize the charge based on the level of the company's nonperforming loans (NPLs). An NPL add-on would reflect the higher default risk inherent in these loans today as compared to the early 1990s or the 1980s. LTVs tend to be higher and debt-to-income standards are much looser today than they have been historically. We would assume a foreclosure rate of 20% for NPLs and that the institution experiences a loss of 40% on foreclosed properties. The 40% loss captures the likely write-down of the property as well as expenses associated with foreclosure and sale of the house. In a worst-case environment, the costs of foreclosure and sale of ORE, which include carrying costs, can rise substantially. The NPL add-on is based on the Texas loss experience in the 1980s, which was a market that saw housing values drop by as much as 30%. However, this does not capture the loss of value in homes that did not sell for lack of interested buyers.
Because they are second mortgages, home equity lines of credit (HELOCs) warrant a higher capital charge. We will use a charge of 3%.
Securities
We look at securities portfolios based on the accounting under U.S. GAAP because it sets the stage for the information that is disclosed by companies and often indicates their uses. The three categories are securities held for trading, available for sale, or held to maturity.
Trading portfolios
For trading portfolios, required capital is a multiple of value-at-risk (VaR) as reported by the company. This is necessitated because we cannot capture the hedging benefits of being long one stock, for example, and short a closely correlated one (e.g. long General Motors, short Ford Motor Co.) without using VaR. It would not be appropriate to allocate capital solely on the assets in the portfolio, as this would ignore these hedging benefits. However, short positions also need capital. Also, there may be hedges through derivatives that have a nominal or zero value, but nonetheless effectively reduce risk within the portfolio. We would overstate capital needed if we assessed capital only on the amounts appearing on the balance sheet.
Another problem arises with derivatives. We cannot assess capital against derivatives for market risk based on GAAP disclosure, since the nature of the derivatives is not disclosed (i.e. we would not know whether the derivative is a call option on an equity versus an interest rate swap versus an Fx swap, etc.) Only VaR encapsulates the correlations between long and short positions, reflects all hedge affects between cash and derivative positions, and accounts for the varieties of risks represented by derivatives on the balance sheet.
Although we believe that VaR has numerous problems, it is the best starting point for assessing capital for trading portfolios. One alternative is to use the volatility of historical net trading revenue for each firm. But this has its own host of problems. Firstly, to have anything meaningful from a statistical viewpoint, we will need 30 or more observations, i.e. at least 7½ years of quarterly data. Firms change dramatically over such long time periods. Mergers and changes in product lineup make such analysis virtually useless. If we were to shorten the period and accept less statistical validity, we would still have problems with changes in risk due to mergers and product changes. Additionally, trading revenue reporting is not consistent across firms. Some institutions include prime brokerage interest and fees, others do not; some include commissions, but not all do.
We scale up 99% VaR to the 99.96% confidence level, which is consistent with the observed one-year default frequency for 'A' rated debt. We will also scale one-day VaR to a one-year time frame by multiplying by the square root of 365 and not 250. (Although there are approximately 250 trading days in the U.S., values change continuously.)
For those banks and other financial institutions that do not disclose VaR, we will assess capital based on other measures, such as worst-case losses for that asset class in the same way we would do for held-to-maturity securities.
Available-for-sale securities portfolios
These portfolios are often used to manage interest rate risk for the entire balance sheet. We will assess capital on a measure of net interest rate risk as opposed to assessing capital on the carrying value of these assets. This would be restricted to assets whose market value is almost exclusively driven by interest rate risk (U.S. government and agency securities and MBS.)
Securities that do not fall into these categories, such as corporate debt and ABS, would be assessed a capital charge based on the history of total rate of return for the asset type. We will consider the worst-case return and average return scaled for volatility.
Held-to-maturity securities portfolios
We will assess capital charges in the same way we would allocate capital to trading portfolios for which VaR is not publicly available. We will assess capital based on other measures, such as worst-case losses.
Repo and securities lending books
We assess capital for securities lending and repo books at a rate of 0.8% (80 bps) of the amount of asset ("securities borrowed" and "securities purchased under agreements to resell").
Acknowledgements
The following Standard & Poor's analysts and research assistants contributed to this article: Tanya Azarchs, Jack Bartko, Anne Cosgrove, David Cresci, Helene De Luca, Arnaud De Toytot, Jayan Dhru, Michael Driscoll, Nick Hill, Robert Hoban, Vikas Jhaveri, Daniel Koelsch, Anita Ma, Dan Martin, Lisa Perard, Rodrigo Quintanilla, Charles Rauch, Kyelim Rhee, Prodyot Samanta, Victoria Wagner, Yuri Yoshida, and Jeffrey Zaun.
