Capital Adequacy and Basel II by Paul H. Kupiec∗ September 2004 FDIC Center for Financial Research Working Paper No. 2004-01 ABSTRACT Using a one common factor Black-Scholes-Merton model, this paper compares unbiased portfolio-invariant capital allocations with Basel II IRB capital allocations for corporate exposures. The analysis identifies substantial biases in the June 2004 IRB framework. For a wide range of portfolio credit risk characteristics considered, the Advanced IRB rules drastically undercapitalize portfolio credit risks. Implied default rates under the Advanced IRB rule exceed 5 percent. In contrast, Foundation IRB capital requirements allocate multiple times the capital necessary to ensure the supervisory target solvency rate of 99.9 percent. The biases that are identified potentially raise a number of important issues including the potential for increased systemic risk as regulatory capital rules promote consolidation in weakly capitalized Advanced IRB banks. Key words: economic capital, credit risk, Basel II, internal models JEL Classification: G12, G20, G21, G28 CFR research programs: risk measurement, bank regulatory policy ∗ Associate Director, Division of Insurance and Research, Federal Deposit Insurance Corporation. The views expressed in this paper are those of the author and do not reflect the views of the FDIC. I am grateful to Mark Flannery, Dilip Madan, Haluk Unal, Dan Nuxoll, Wenying Jiangli, and Andy Jobst for comments on an earlier draft of this paper. Email: [email protected]; phone 202-898-6768. Capital Adequacy and Basel II 1. INTRODUCTION Economic capital fulfills a buffer stock loss-absorbing function. It is defined as the amount of equity financing in a capital structure that is necessary to ensure that the default rate on a bank’s funding debt never exceeds a maximum target rate selected by management. In practice, economic capital allocations are often estimated using value-at-risk (VaR) measures. Given the widespread acceptance of VaR-based capital allocation methods, it may come as a surprise to learn that perfectly accurate VaR models may produce biased estimates of economic capital requirements. Kupiec (2004) derives the general recipe for calculating unbiased economic capital measures. When used for economic capital allocation purposes, VaR must be measured relative to a portfolio’s initial market value and augmented with an estimate of the equilibrium interest payments that must be made on the bank’s funding debt. The second step—estimating equilibrium required interest payments– cannot be accomplished within a traditional VaR framework, but instead requires the use of a formal asset pricing model or an empirical substitute. For the most part, published methodologies for allocating economic capital are biased because they fail to recognize the need to pay interest and a credit spread premium to bank debt holders. Basel II’s Internal Ratings Based (IRB) capital requirements are set using mathematical rules that have been distilled from a well-known class of VaR models that are constructed to generate portfolio-invariant capital requirements. The IRB approaches of Basel II set capital requirements based on exposure type (corporate, sovereign, bank, retail, SME, equity, etc) and the probability of default (PD), loss given default (LGD), and exposure at default (EAD) characteristics of an individual credit. Vasicek (1991), Finger (1999), Schönbucher (2000), and Gordy (2003) develop statistical models of portfolio credit loss distributions that generate portfolio-invariant capital allocation rules. Gordy (2003) explicitly - 2 - argues for using a portfolio-invariant VaR methodology to calibrate IRB capital requirements.1 An important issue that arises regarding the use of portfolio-invariant methods is the bias that arises because these capital allocation models do not properly account for funding debt holder compensation. To the extent that these industry models have guided the Basel II IRB calibration process, IRB capital requirements are biased. This source of potential bias is not widely recognized and its potential magnitude has not been documented in the literature.2 The magnitude of bias in Basel II IRB capital requirements can be analyzed in the context of a portfolio-invariant version of a Black-Scholes-Merton (BSM) equilibrium model of credit risk. The BSM model is the theoretical foundation for virtually all economic capital models of credit risk including portfolio-invariant methods. While the basic BSM model may not accurately predict all features of the historical data on credit spreads and losses, it represents a useful equilibrium benchmark. It is consistently formulated and facilitates the construction and comparison of alternative measures of risk and capital in a controlled environment. If IRB capital measures are significantly biased in the BSM equilibrium setting, IRB model performance must also be suspect in the more complicated and opaque setting that characterizes real world credit risks. The results of the IRB model calibration comparison are surprising and have important regulatory and competitive implications. Compared to a true unbiased economic capital allocation, the June 2004 Basel II Advanced IRB approach requires less than 20 percent of the capital needed to achieve the 0.1 percent regulatory target default rate assuming that all bank capital is Tier 1 (equity) capital. Even if all bank capital is equity, the Advanced IRB capital rules are consistent with a true bank default rate of over 5 percent—or 1 Basel II background papers highlight the influence that the portfolio invariant capital allocation literature has on the IRB calibration process. For example, footnote 26 of the Basel Committee on Banking Supervision (2001) references the working paper that subsequently has been published as Gordy (2003). 2 Kupiec (2004) demonstrates the existence of these biases in VaR-based capital rules in the single credit setting, but does not consider the biases that may arise in a well-diversified portfolio. - 3 - more than 5000 times the regulatory target rate. Moreover, regulatory capital rules allow banks to use subordinated debt and qualified Tier 3 capital to satisfy regulatory requirements. Consequently, an Advanced IRB bank’s true loss absorbing capacity is likely to be more limited than these estimates in this analysis suggest. In contrast to the Advanced IRB approach, Foundation IRB banks will be required to hold many times the capital necessary to achieve a .001 default rate. For high quality portfolios, Foundation IRB capital requirements specify more than 7 times the level of capital needed to achieve the regulatory target rate. As the quality of the credits in a portfolio declines, the Foundation IRB rules are less aggressive in over-capitalizing positions. Estimates suggest that capital for one-year lower quality credits are overstated by more than 160 percent. For short-dated credits, Foundation IRB capital requirements provide capital relief relative the 1988 Basel Accord. As the maturity of credits lengthens, however, the 1988 Basel Accord provides more favorable capital treatment. Overall, the analysis demonstrates that relative to Advanced IRB banks, Foundation IRB banks will be held to a much stricter prudential standard. The substantial differences in the regulatory capital requirements specified by the alternative Basel II approaches potentially raise important prudential and structural issues. To the extent that banks enjoy safety-net engendered subsidies that are attenuated by minimum regulatory capital requirements, Basel II may create a strong incentive for banks to petition their supervisors for Advanced IRB treatment. Banking system assets will be encouraged to migrate toward Advanced IRB banks, either through consolidation or through an increase in the number banks that are granted regulatory approval for the Advanced IRB approach. The capital relief granted under the Advanced IRB approach may raise prudential concerns as well. For example, should FIDICA prompt corrective action (PCA) minimum leverage requirements be relaxed once Basel II is implemented, regulations would allow Advanced IRB banks to operate with substantial reductions in their capital positions raising the potential for a material increase in systemic risk.3 3 Unless PCA requirements are relaxed (12 U.S.C. Section 1831), PCA may become the binding capital constraint on Advanced IRB banks. - 4 - An outline of this paper follows. Section 2 summarizes the portfolio-invariant capital allocation literature and relates it to IRB model calibration. Section 3 presents a general methodology for setting unbiased buffer stock capital requirements. Section 4 revisits unbiased credit risk capital allocation in the context of the Black-Scholes-Merton (BSM) model. Section 5 derives unbiased portfolio-invariant credit risk capital measures in an infinitely granular one-common factor BSM model specification. Section 6 reports calibration results and Section 7 concludes the paper. 2. PORTFOLIO INVARIANT CAPITAL REQUIREMENTS AND THE IRB APPROACH When a new credit is added to a portfolio, the diversification benefits within the portfolio in part determine the amount of additional capital that is required to maintain a given solvency margin, where solvency margin is defined as 1 minus the probability of default on a bank’s funding debt. The marginal capital required for the new credit depends on the characteristics of the credit that is added, as well as the characteristics of the credits in the existing portfolio. VaR techniques recognize diversification benefits in their prescription for capital requirements. Because an IRB approach sets capital based on characteristics of a credit in isolation, such an approach is not generally capable of recognizing diversification benefits. Under certain conditions, however, IRB capital rules can mimic the capital allocations set by a VaR approach. Vasieck (1991), Finger (1999), Schönbucher (2000), Gordy (2003) and others, have established conditions under which a credit’s contribution to a portfolio VaR measure is independent of the composition of the portfolio to which it is added. For example, Gordy (2003) establishes that contributions to VaR are portfolio-invariant if: (a) there is only a single systematic risk factor driving correlations across obligors; and (b) no exposure in a portfolio accounts for more than an arbitrarily small share of total exposure. Under these assumptions, capital allocations can be set using (only) information on the individual credits’ risk characteristics and the resulting portfolio capital allocation can still be equivalent to a VaR-based capital allocation. Because IRB capital requirements can only be accurate in a portfolio-invariant setting, we adopt the assumption of single common factor risk generation and consider only - 5 - well-diversified portfolios that satisfy the assumptions used in the portfolio-invariant capital literature. In contrast to earlier studies, this study derives the implications of these assumptions in the full BSM model setting and uses the unbiased capital allocation methodology prescribed by Kupiec (2004) to estimate the capital requirements for individual IRB credits held in a well-diversified portfolio. 3. UNBIASED BUFFER STOCK CAPITAL FOR CREDIT RISKS The intuition that underlies the construction of an unbiased economic capital allocation is transparent when considering portfolios composed of long positions in traditional financial assets such as bonds or equities because the value of the portfolio cannot go below zero. While it has not been widely recognized in the literature, should losses have the potential to exceed the initial market value of a portfolio as they can for example when a portfolio includes short positions, futures, derivatives, or other structured products, then economic capital calculations must be modified from the techniques described herein. Modifications to the capital structure alone may not be able to ensure that the bank is able to perform on its liabilities with the desired level of confidence. Kupiec (2004) provides further discussion. Defining an Appropriate VaR Measure VaR is commonly defined to be the loss amount that could be exceeded by at most a maximum percentage (1−α) of all potential future asset or portfolio value realizations at the end of a given time horizon. The VaR coverage rate,α, sets the minimum acceptable solvency margin Assume T is the capital allocation horizon of interest for asset A, which has an initial ~ ~ market valueA , and a time T value of,A . A has a cumulative probability density function 0 T T ~ Ψ(A ,A ). An (α) coverage VaR measure, VaRµ(α),α∈[0,1], is often defined as, T T (~ ) (~ ) VaRµ(α)= E A −Ψ−1 A ,1−α (1) T T - 6 - (~ ) ~ (~ ) where E A represents the expected end-of-period value of A and Ψ−1 A ,1−α T T T ~ represents the inverse of its cumulative density function of A evaluated at 1−α. In T expression (1), profit and loss (P&L) are calculated relative to the expected time T value. When using VaR type measures for capital allocation purposes, Kupiec (1999) establishes the importance of measuring P&L relative to the initial market value of the asset. This measure, VaRA0(α), is defined as, (~ ) VaRA0(α)= A −Ψ−1 A ,1−α (2) 0 T In very short horizon calculations (e.g., daily market risk VaR measures) the difference between VaRµ(α) and VaRA0(α) is inconsequential. As the time horizon lengthens, as it does in most interesting capital allocation problems, the difference between VaRµ(α) and VaRA0(α) can be substantial and use VaRµ(α) will bias VaR capital allocation estimates. Unbiased Capital Allocation for Credit Risk Consider the use of a 99.9 percent, 1-year measure, VaRA0(.999), to determine the necessary amount of equity funding for a long bond or loan position. By definition, there is less than 0.1 percent probability the asset’s value will ever post a loss that exceeds its VaRA0(.999) measure. It is tempting (but incorrect) to conclude that should the level of equity financing be set equal to VaRA0(.999), there would be at most a 0.1 percent probability that subsequent portfolio losses could cause the bank to default on its funding debt. If VaR is measured from the asset’s initial market value and that VaR measure is statistically accurate, VaR can never exceedA . If the bank were to set the share of equity 0 funding equal toVaRA0(.999), the amount of debt finance required to fund the asset would beA −VaRA0(.999). If the bank borrows A −VaRA0(.999) it must promise to pay back 0 0 more than A −VaRA0(.999) assuming that interest rates credit risk compensation are 0 positive. An unbiased economic capital allocation rule for 0.1 percent target default rate is to set equity capital equal to VaRA0(.999) and in addition include the interest that will accrue on - 7 - the funding debt issue. Kupiec (2004) establishes the validity of this capital allocation recipe for both market and credit risks. Portfolio Capital While the discussion of unbiased buffer stock capital has thus far been presented in ~ terms of a single credit, the results generalize to the portfolio context. LetA andA i0 iT represent, respectively, the initial and time T value of asset i. The initial and time T value ~ ~ of a portfolio of assets is given by, A = ∑A and A = ∑A respectively. Let the P 0 i0 P T iT ∀i ∀i ~ ( ~ ) cumulative probability density for A be represented byΨ A , A . An unbiased P T P T P T portfolio economic capital allocation with a solvency rate of α can be estimated by first calculating VaRPA0(α), ( ~ ) VaRPA0(α)= A −Ψ−1 A ,1−α (3) P 0 P 0 and adding to it, an estimate of the equilibrium interest cost on the funding debt issued to finance the portfolio. In order to estimate the equilibrium interest cost, one must go beyond the tools of value-at-risk and utilize formal asset pricing models or empirical approximations to price the funding debt. This topic is discussed in detail in the following section. 4. Unbiased Buffer Stock Capital Allocation in a Black-Scholes-Merton Model If the risk-free term structure is flat and a firm issues only pure discount debt, and asset values follow geometric Brownian motion, under certain simplifying assumptions4, Black and Scholes (1973), and independently Merton (1974), (hereafter BSM) established that the market value of a firm's debt issue is equal to the risk free discounted value of the bond’s par value, less the market value of a Black-Scholes put option written on the value of the firm’s assets. The put option has a maturity identical to the debt issue maturity, and a strike price equal to the par value of the debt. More formally, if A represents the initial 0 4 There are no taxes, transactions are costless, short sales are possible, trading takes place continuously, if borrowers and savers have access to the debt market on identical risk- adjusted terms, and investors in asset markets act as perfect competitors. - 8 - value of the firms assets, B the bond’s initial equilibrium market value, and Parthe bond’s 0 promised payment at maturity date M, the BSM model requires, B = Par e−rfM −Put(A ,Par,M,σ), (4) 0 0 ( ) where r represents the risk free rate and Put A ,Par,M,σ represents the value of a Black- f 0 Scholes put option on an asset with an initial value of A , a strike price of Par, maturityM , 0 and an instantaneous return volatility of σ. The default (put) option is a measure of the credit risk of the bond. Merton (1974), Black and Cox (1976), and others show that the model will generalize as to term structure assumptions, coupon payments, default barrier assumptions, and generalized volatility structures, but the capital allocation discussion that follows uses the simplest formulation of the BSM model.5 Incorporating Credit Risk into the BSM Model In the original BSM model, the underlying assets exhibit market risk. To examine portfolio credit risk issues, it is necessary to modify the BSM model so that the underlying assets in the portfolio are themselves risky fixed income claims. Consider the case in which a bank’s only asset is a risky BSM discount debt issued by an unrelated counterparty. Assume that the bank will fund this bond with its own discount debt and equity issues. In this setting, the bank’s funding debt issue is a compound option. ~ Let A and Par represent respectively the time T value of the assets that support the T P purchased discount debt and the par value of the purchased discount bond. Let Par F represent the par value of the discount bond that is used to fund the asset purchase. For purposes of this discussion we restrict attention to the case where the maturity of the bank’s funding debt matches the maturity of the BSM asset (both equal toM).6 The end-of-period cash flows that accrue to the bank’s funding debt holders are, 5 That is, it assumed that the term structure is flat, asset volatility is constant, the underlying asset pays no dividend or convenience yield, and all debt securities are pure discount issues. 6 Kupiec (2004) discusses the case where maturities differ. - 9 - [ (~ ) ] Min Min A ,Par ,Par . (5) M P F Funding Debt Equilibrium Value In the original BSM model, the firm’s underlying assets evolve in value according to geometric Brownian motion. dA =µAdt +σAdW (6) where dW is a standard Weiner process. If A represents the initial value of the firm’s 0 assets, A the value of the firm’s assets at time T , equation (6) implies that the physical T probability distribution for the value of the firm’s assets is, A~ ~ A e⎜⎜⎝⎛µ−σ22⎟⎟⎠⎞T+σ T ~z (7) T 0 ~ where z is a standard normal random variable. Equilibrium absence of arbitrage conditions impose restrictions on these asset’s drift rate, µ= r +λσ, whereλ is the market price of risk. If dAη =(µ−λσ)Aηdt + Aησdz is f defined as the “risk neutralized” process under the equivalent martingale measure, the underlying end-of-period asset value distribution under the equivalent martingale measure, ~ Aη, is, M A~η ~ A e⎜⎜⎝⎛rf−σ22⎟⎟⎠⎞T+σ T ~z (8) T 0 The initial market value of the bank’s funding discount bond is the discounted (at the risk free rate) expected value of the end-of-period funding debt cash flows taken with respect to ~ probability density Aη . In the held-to-maturity (HTM) case, when the maturity of the bank’s M funding debt matches the maturity of the purchased BSM bond (both equal toM), the initial market value of the bank’s funding debt is, [ [ (~ ) ]] Eη Min Min A ,Par ,Par e−rfM (9) M P F - 10 -