A framework for assessing credit risk in depository institutions toward regulatory reform

anking am Finance ! [ 1991) 78.5-8 Robert S Chirinko Harris Graduafr School of Public PoCc_vSrudies The i’niverria~ q/ Chicago Chicago, IL 60637-278s USA Gene D Guill* Bankers Trust Company, New York NY f&11 7, i’S.4 Final version received January 1991 This paper develops and illustrates a new method for evaluating the credit risk borne by depository institutions that may prove to be an important element in a regulatory mxhanism sensitive to risk In our framework, depository institutions are forced to bear the costs of their risk-exposure 2nd hence to internalize this cost when extending loans Using various loan portfolios and alternative macroeconomic scenarios, we present numerical calculations demonstrating the ability of our framework to capture variattons in risk-exposure and highlighting the significance of portfolio concentrations These calculations are based on an important new source of information that permits an econometric analysis of loan losses by industry Introduction The amount of risk faced by depository institutions concern for policymakers because of the perceived link and the performance of the economy There is broad agreement that a substantial number of failures would (Dls) is of substantial bet-seen their stability (though not universal) destatxiize the system *The authors would like to acknowledge the helpful comments from Richard Aspinwall, George French, Gerard Gennotte, Charles Goodhart, Joseph Hotz, Mervyn King, Robert Taggart, an anonymous referee, and seminar participants at the Federal Reserve Bank of Chicago, the London School of Economics, the University of Chicago, and the West Coast Academic-Federal Reserve Economic Research Seminar Excellent research assistance has been provided by Paul Hebert The Shared National Credits data have been kindly distributed by the Office of the Comptroller of the Currency Partial financial support from ?hi: Federal Home Loan Bank Board under Grant No C88066 is gratefully acknowledged All errors, omissions, and conclusions remain the sole responsibility of the authors, and not necessarily reflect the views of the organizations with which they are associated nor the Federal Home Loan Bank Board 03784266/91/$03SO 1991 -Elsevier Science Publishers B.V (North- 786 R.S Chirinko and G.D Guill, Assessing creditrisk in depository instirurions of monetar;r payments and monetary control and impair the flow of funds to borrowers lacking access to capita1 markets Assessing the vulnerability of K)Is has taken on prominent importance following their recent troubles and the dramatic increase in realized U.S government insurance liabilities While it has been recognized for some time that the flat-rate structure of the deposit insurance system substantially lowers the price of incremental risk, tnis explicit incentive for risk-taking had been balanced historically by implicit prices imposed by regulators in the form of operating restrictions.2 owever, the exceptional changes occurring in the financial services industry have lowered, if not effectively eliminated, these implicit regulatory prices Thus, rather than attenuating financial instability as initially intended, the deposit insurance system may have been the primary culprit in disrupting the U.S financial system in recent years In light of these problems, risk-based deposit insurance and capital standards have emerged as important policy options under discussion This study will develop and illustrate a new method for evaluating the risks borne by DIs that may prove important in structuring a regulatory mechanism sens;tive to risk* \vv b begin OUi analysis i-ILsection by reviewing previously proposed methods for assessing DI risk-exposure These proposals have some important limitations avoided with the approach illustrated in this paper The proposed framework relates the credit risk associated with a loan portfolio to variables, such as exchange rates, the prices of primary commodities, Federal Reserve decisions, and federal tax and spending policies These exogenous risk variables will be linked to loan losses by estimated, nonlinear econometric quations relating aggregate variables to the performance of industries from which loan portfolios are constructed The risk borne by a DI will be determined by the proportions of its loan portfolio held in these various industries and by the characteristics of the loan loss distribution reflecting industry means, variances, and covariances In cof- ;tructing loan portfolios, DIs will face a schedule of risk-based premiums, and thus are forced to internalize the costs of their risk-exposure when extending loans Section develops our framework for assessing risks borne by DIs and for setting risk-based regulations A key element is the link between industry loan losses and macroeconomic and industry variables, and this link is forged with an important new source of information from the Shared National Credits database To the best of our knowledge, these data have ‘Se Eisenbeis (1987), Kane (19871, and Gerticr (1988) for recent discussions and citations Apart from macroeconomic go& there is additiona! concern for protecting small depositors who are unlikely to possess sufiicient information to evaluate DIs with a reasonable degree of accuracy This information asymmetry cou!d itself be responsible for bank runs and, hence, also contribute to macroeconOmic instability ‘T’he dual price interpretation of deposit insurance has been deveioped by Buser, Chen and Kane ( 1981) and F!annery ( 1982) estimated loan loss omit scenarios, we present i the ability of our ~ro~o~d met The variety of pr osed methods for assessi~l forecasting, and form te their projections with in financial markets or regulators This section reviews extan general problem of assessing DI risk but does not discuss t important issues concerning adjustments in deposit insuf or capital standards or other aspects of regulatory reforrm3 One set of proposals would turn to financial markets for assistance in assessing risk and, a~ with much of economic analysis, the central idea is to rely on private, self-interested agents to perform the complex risk calculations An innovative proposal introdueed by Merton (1977) views deposit insurance as a put option, and uses modem option valuation techniques to Other market valuation methods determine the value of the guarantee.’ involve assigning risk-premiums based on the spread of uninsured deposits over the safe rate of interest [Peltzman (19X!)] or allowing private parties to co-insure deposits in conjunction with the government [Baer (1985, Boyd and Rolnick (1988)] While these methods portfolio diversification, are sensitive to ex-ante a fundamental difficulty risk and, in principle, with all of these market- based proposals is that they assume sufficient information is publicly available to financial market participants to allow them to form reasonable judgments Yet, this possibility is precluded by the special role in the financial intermediation process held by Dls - using private, client-specific information, they create illiquid assets in the loan market for which no close ‘See, among many studies, Benstcn and Kaufman (1988), Brookings Task Force (1989), Kane (1986), Shadow Financial Regulatory Committee (1989), and White (1989) for comprehensive reform proposals: Avery and Belton ( 1987) Chan Greenbaum and Thakor ( 1988) and Flannery (1989) for a comparison of insurance premiums and capital standards; and Pennacchi (1987a) and Pyle ( 1986) for a consideration of closure procedures ‘Marcus and Shaked (1984) Pennacchi (1987b) and Ronn and Verma (1986) have also applied this technique to valuing deposit insurance The results were shown to be sensitive to assumptions about the distribution of asset returns, the term of the deposit guarantee, and the regulator’s closure rule, respectively 78R R.S Chirinko ond G.B Guill Assessing rreBir risk in depository institutions substitutes exist.’ The potential success of market-based proposals is largely inconsistent with the raison d’etre for overnment regulation of DLL~ is concern is supported by the empirical results of Avery, B&o (1988), who found no relation between the risk-premium on related long-term debt and measures of bank ~~~OIIIUIILX, risk, and bond ratings by private agencies Analyzing bank holding companies, Randall (1989) concluded that the stock market and bond rating unable to anticipate problems before substantial damage had even after identification, neither the market nor agencies were able to uate the seriousness of credit problems In regard to the options approach, there is the further difficulty that 95% of banks are not traded actively enough to be evaluated by this technique Coupled with current questions about the efficiency of financial markets and their substantial volatility, financial markets are unlikely to prove useful in assessing DI risks.’ The aiternative set of proposals provides a more direct role for regulators Relying less on unbridled market forces, these alternatives have received the most serious consideration The Federal Reserve System, ahg with 11 other central banks, has implemented a plan that assesses credit risk by activities in which a DI is engaged Capital standards are determined by allocations among broad asset groups, with commercial and industrial loans taken as a homogenous aggregate requiring the greatest amount of capital [Federal Reserve System (1QSS)] However, it is doubtful that the proportion of assets in these broad categories will serve as a reliable indicator of future prohlems Furthermore, the important role of diversification among loans in the portfolio is not considered systematically ‘The experience of institutions with a heavy exposure to the oil industry - as well as the calculations to be presented in this study - indicates that igrloring the covariance relations between loans in the portfolio can lead to a substantial misstatement of risk-exposure ’ The Federal Reserve plan is likely to have little effect in curtailing risk-exposure by adventurous institutions Another non-market proposal would determine risk-based deposit insure premiums with an objective risk index and subjective evaluations rschhorn ( 1986)J The index is a predictive device for determining ‘See Diamorldand Dybvig (1956) and Goodhart (1987) for further discussion, and Fama (1985) and James (1987) for supporting evidence *Apart from this information problem, the market will fail to value any aggregate externalities arising from bank failures ‘On the volatility issue, see Shiller (1990) Using ootion valuation techniques, Marcus and Shaked (1984 section 3.D) found that their ris” e&mates were sensitive to the standard deviation of the asset rate of return, which is derived from the standard deviation of the equity rate of return *In his study of bank holding companies that experienced difficulties, Randall (1989) concluded that the concentration of loans with common risk characteristics ‘appearsto have played a significant role in nearly all of the cases studied’ (p 15) whether or not variables describi faces two dificuft pro standards are costs t make it difficult for reg en face requirements, costs - due to increased insurance premiums, capital market rates - may we11 find it optimal to undertake policies As with the distribution proposals of loans, are rot adven ral Reserve plan, no acestint is taken of the the deleterious effects of excessive portfolio fully adequate for setting risk-b,>sed premiums or standards An alternative framework The framework developed in this section utilizes portfolid characteristics and other confidential information to force greater sensitivity to ex-ante risk-exposure.10 Our analysis of DI risk quantifies the tsral credit risk arising from outstanding loans From a rtgulatory perspective, total risk is the appropriate concern because it creates demands for regulatory reserves While there are other risks faced by DIs, credit risk has been and is likely to continue to be the primary cause oi failures.” Credit risk is measured by the distribution of loan losses associated with the Dl’s loan portfolio This distribution is calculated as a weighted average of industry loan loss distributions, where the weights equal the proportion of 9A recent example of substantiai classification bias is the accounting practice of restructuring, which ‘lets banks transform non-accruing loans into accruing loans by granting more favorabie terms to the borrower, er,en though those terms may involve below-market interest rates that actually lose money for the bank’ [Suskind (199011 Banks benefit from this reriassification because the amount of non-accruing loans in the portfolio, inter alia determines loss reserves ‘The additional information required for our approach is modest, and could be collected within the existing regulatory apparatus “Credit risk was the primary cause for three-quarters of bank failures [FDIC (1983, p IM)] There are additional sources of risk alfecting Ms Risks arising from unscrupulous management practices are likely to be checked only with supervision, and fall outside of the present study While playing a major role in past problems, interest rate and exchange rate risks are of less current concern because of the emergence of hedging instruments and the relaxation of thrift regulations on deposits Nonetheless, with a detailed description of a Dl s income statement and balance sheet, the framework developed in this paper can assess these additional risks and their relation to credit risk 790 R.S Chirinko and G.D Guill, Assessing credit risk in deposirory institutions the loan portfolio extended to firms in each industry Industry loan losses are determined by industry and economywide variables that affect profitability A distribution of loan losses is constructed with different assumptions a possible states of the world, indexed by s = 1,2, ,S.” These considerations lead to the following set of equations for the I industries, where li,, is loan losses in industry i tor state s per dollar of outstanding loans, ,d[.] is an econometrically eq+imated function (discussed in section 41, yi,s is a vector of variables affecting loan losses, and Ei is an industry-specific shock A DI is characterized by the proportion of its loans (~0~)extended to firms in different industries Fundamental to the evaluation of the DI is that the risks associated with the assets constituting its portfolio may be correlated In terms of (I), this implies that the determinants of loan losses in industry i @ias) may be correlated with the loan loss determinants for industry i’ In our framework, this correlation will be induced by fluctuations in macroeconomic variables, such as interest rates and the components of Gh;P, to which the industries are more or less responsive Critical to making our method operational is linking the industry-specific variables to macroeconomic outcomes, and this is achieved by means of an input/output model for the U.S economy that relates final expenditures to industry production, as well as capturing production flows among industries The final step in our analysis is to identify the sources of risk in the macroeconomy impacting industries through the input/output model We ,issume that the macroeconomy can be adequately represented by a iargez-tale, nonlinear econometric model and that risk arises from the diyersity of possible outcomes of a subset of exogenous variables Exogenoub variabies are divided between those that remain fixed (Z) and those that vdry between states, such as primary commodity prices (including energy), exchange rates and monetary and fiscal policies For each latter variable, wz specify a set of possible outcomes (Sj,.) and the probability that each wil! occur (pi,“), where j refers to an exogenous risk variable and TVa possible outcome The permutations of the _Yj,u’ s {with the associated pi,u’S) are formed, and the probability of any given combination is represented by the probability weights, x,, that sum to unity The relations between industry variables, the macroeconomic and input/output models f Y[.]) and the exogenous variables are epresented by the foiiowing equation, ‘*It may be easiest to think of the model as atemporal from as yet to be specified probability distribution, framework to multiple periods and the states determined but it is straightforward by draws to extend the The Y~,~‘senter e ee vector of loan industries the macroeconometric an industry-specific variables compute a vector of e ed loan losses We have chosen this approach for generating inforuxxtion about industry loan losses because it provides a tractabIe means for using the scarce available information and is forward-looking.‘4 As discussed in section 2, all methods for assessing risk involve forecasting, and our method allows us to enter information about the risks expected to be faced in the future While the usefulness of our framework is enhanced the more accurately the set of possible outcomes can be described, it is no more restrictive than existing or proposed methods, and much more flexi41e in incorporating information about risk variables and the probability of their occurrence Furthermore, it does not restrict the number of possible outcomes (nor their interactions with the fixed exogenous variables) that may be considered.* Being forwardlooking and not trapped by historical happenstance, the framework presented here provides a flexible method for quantifying DI r”sk-exposure Deposit insurance premiums or capital standards would be determined by the interaction between the loan portfolio chosen by a DI and three !‘SC the Appendix in Chirinko and Guill (1990) for a mar: mathematical statement of the method for calculating rhe loan loss distribution Note that the covariance calculations will incorporate the covariance of the ENSfrom the sample period “With similar quantitative relations for other countries (e.g., Project LINK), this approach can be extended to international loans, thus providing a meaxre of the effects of loan concentrations emphasized by Bennett (1984) “The covariance matrix of loan losses fC[!]) generated by our framework will be similar, in principle, to the covarianc- matrix of loan losses computed from historical data (CH(!]) if we expect the future to bc affected by the same shocks as in the past Insofar as they difier, CC11 will be a better estimate In practice, estimating C,[fl will be very difficult because of a lack of data Apart from these problems, using the covariance matrix of the macroeconomic risk variables to approximate CC/] would ignore the substantial information in the mpWoutput model and nonlinear relations in the macroeconomic model Neither of these covariance matrices could provide information on p, and hence could not deliver a complete a%SSment of DI ri:k-exposure 792 RX Chirinko and G.D Guill Assessing credit risk in depository institutions constraints selected by regulators The first set of constraints reflects ceived risks affecting the overall economy and industries and, as discussed eve, is represented by M[1] and C’[l].‘” Second, bas on the relation between loan losses and DI failures, regulators would cho the critical loan loss rate, f*; loan losses in excess of this rate undermine DI viability for a grespecified minimt;m level of capital In our framework, risk-exposure is measured by the area under the portfolio’s loan loss distribution to of l*, and this critical region is labeled q The lower I*, ceterid par rester q and hence the more conservative the stance taken by regulatorsi’ Note that q captures changes in both the location and dispersion of the loan loss distribution Assuming that the DI‘s loan losses are distributed normal, the mean and standard deviation of the loan portfolio completely characterize the distribution The third constraint is the statutory premium schedule, p[q] Apart from q, this would depend on a number of additional factors (e.g., the strength of the balance sheet, management practices) affecting DI performance and the !ike!ihood of insuraace claims l8 Since risk-taking is likely to become acute when DIs near financial distress, ,$q] should be nonlinear in its arguments The proper determination of p[q] would also depend on a welfare analysis that is beyond the scope of the present paper and complicated by the difficulty in defining the relevant objective function The subsidy for risktaking enjoyed currently by DIs is largely, if not entire& shifted to borrowers and, if eliminated, would force other sectors in the economy to bear more risk [Goodman and Santomero (1986)] If there are efficiencies in bearing risk by a partnership between Dls and government insurers or macroeconomic externalities arising from bank failures, maintaining an actuarially fair system is unlikely tc be optimal Payments to the insurance fund (or additions to the capital base) would be determined by p[q], and thus would depend on ex-ante risk and (positively) on both the mean and the standard deviation of the loan loss distribution The latter parameter incorporates the effect of loan portfolio Diversification that we believe is critical to any risk-sensitive regulatory system With M[Q, CC/], I*, and p[q] announced in advance, DIs would determine their insurance payments by altering the composition of their loan portfolios Thus, our framework quantifies the cost of risk-exposure, and forces DIs to internalize this cost when extending loans Given the many factors outside of our model that affect DI performance, ‘6That is to say, regulators choose {pj ‘s, X, rS Y’pC.1.A[-]} and generate {M[I], C[fl} “lt would be incorrect to increase I* becaue of a perceived increase in risk in the macroeconomy, which will be reflected in the choice of the Xj.b’s and pj._‘s, hence in the spread of the computed loan loss distribution “SPremiums would depend negatively on the level of DI capital ia excess of the minimum tevr.! K- d in setting I* Equivalently, I* could be adjusted for the level of capital held by a DI The prop xd framework requires atr on loan Iosses by turn, must be related to variables i t institutions, and be at least $20 million These annual data are constructed at a two-digit SIC code classification and mostly pertain to Commercial and Industrial (C & I) Loans The C & I Evans in the sampIe constitute approximately one-third of the vake of outstanding C&I Loans by the supervised by these three regulatory agencies Given our interest in identifying problem institutions, the loan loss variable is defined as the percentage of loans in one of the following four criticized categories: Special Mention, Sub-Standard, Doubtful and Loss In 1988, loans in these four categories amounted to 7.34?, of the value of loans in the SNC database; the Doubtfuf and Loss categories amounted to 1.07”; of the total In estimating a loan loss equation, we are guided by two considerations Statistically, in order to obtain a reasonable amount of variation in the regression, we pool the sample across industries, though the intercepts (Zi) vary across industries In keeping with the estimation method of other equations in the macroeconometric model, the loan loss equation is estiTheoretically, we postulate that loan mated by ordinary least squares.” losses are related systematically to the recent growth of cash flows for that industry We not have sufkient data permitting explicit cal~~lat~o~ of cash flow, but have available two major components: sales revenues (Ri,,) and the costs of intermediate inputs and labor services (C+), where ‘i’ refers to industries and ‘t’ to time periods The growth rates of these series are entered as 4-year moving averages In addition, to capture t “This estimation technique does not preclude the possibility that the estimated li ‘smay be less than zero, but inspection oh the simulations reveals iittle problem from omttting this constraint R.S Chirinkn and G.D Guill Assessingcredit risk in depository institutions Table Pooled I squares, estimates of the loan loss equation (3).” (I) - Fkxed eflects ^ _ Revenue, /I (21 FIxed effects (3) No fixed e@ects (4) Fixed effts - 0.549 (0.120) -0.237 (0.104) -0.319 (0.096) Cost y 0.431 tO.131) 0.302 (0.117) -0.157 (0.097) 0.029 (0 Ial ) Federal funds (5 0.319 (0.093) _ _ 0.160 (0.087) _ 0.282 (0.089) _ 0.614 (0.175) GNP B 8’ 0.276 108 Observations 0.088 168 0.182 138 (51 No fix effects - - - 0.268 (0.178) 0.151 108 0.362 (0.180) -0.105 ZO.8.180) O.Q9? 108 “Estimation is with annual data from 1985-1988 All coeficients have begn stanJar&& Standard errors are in parentheses For all entries but column (2) the industries included in the estimation had positive loan losses; column (2) includes the zero-loss industries An overall constant term has been inc!uded in m!umns (3! and (S), and is statistically insignificartt effects brought about by variations in the cost of funds, the federal funds rate (F,) enters contemporaneously These considerations give rise to the following equation, + iiF, + Ci.tq i = 1.27, or 42; t = 1985, l%S, where Ei,l is an error term, Table contains the pooled least squares estimates of (3), and all of the reported coeffkients have been standardized The preferred fixed-effect estimates are presented in column (I), and all three coefficients are statistically significant and of the correct sign 2o Over the sample period, the variation in revenue has had the largest effect on 1i.r These estimates are based on a sample in which industries with no loan losses for at least one year in the sample are excluded; column (2) expands the sample to include these additional (zero-loss) industries The coefftcient estimates are broadly similar, “Very similar results are obtained when the q’s are treated as random effects and a variance-components mode! estimated We have chosen to use the fixed effects estimates (as well as abstaining from using instrumental variables) because they are more in keeping with the estimation techniques used in the rna~r~~on~~et~~ model loans, and allowing these categories to have different sensitivities to variations in the rea-zssors might prove important Ail of these issues can be addressed wth a richer dataset and the calculations presented in this paper should be viewed as illustrating the potential of our framework for assessing DI risk Specific ¶~antitati~e asomptions framework proposed in this paper links the mean 1~) and standard deviation (6) of loan portfolio losses to aggregate and indUSFry variables To establish this link, we must make specific quantitative assumptions about values of these risk variables, the weights (or probabilities) associated with each of these values5 and the economic models relating these variables to loan portfolios We begin by specifying the number of exogenous risk varia The “The resuk in columns (I) and (2) OFtable suggest that using the latter estrmates would not radically change the simulation results reported below The estimated in column ( I) haw use, in OUF ~~~~~~a~ pd_pent, the iqe concentration of zeros dktorts the ken used loss coefficients in OUT relatively small sample The zero-loss industries are estinkated in the simulations by estimating z, as the difference in the means of the variables in (3) a~~ro~~ately by /?, and 796 R.S Chirinko end G.D Guill, Assessing credit risk in depository ictitutions values For reasons of computational expense, four sets of risk variables (J =4) are examine:J and, for each set, a Basecase and an Alternative set of values are considered The Basecase assumptions, as well as values for the fixed exogenous variables, are taken from the ongoing forecasting project at DRI/McGraw-Hill Alternative values have been chosen to capture important developments affecting DIs, and are described relative to the Basecase forecast First, the trade-weighted dollar exchange rate appreciates by 2.3% per year Second, the federal deficit is lowered by reducing defense and nondefense spending and increasing personal and corporate taxes Third, a contract;onary monetary policy raises the federal funds rate an average of 167 basis points over the 3-year simulation period; relative to the Benchmark, this represents a 19.7% increase in this interest rate Fourth, the prices of primary commodities are assumed to rise by 4.5% per year Ail changes are annual averages relative to the Benchmark over the 3-year simulation period The permutations of Basecase and Alternative values for four sets of exogenous risk variables lead to 16 possible states, a Benchmark (defined as the Basecase value for each of the four risk variables) and 15 Alternatives From an aggregate perspective these Alternatives are contractionary relative to the Benchmark To provide a more balanced view of possible macroeconomic outcomes and to minimize computational expenses, we assume that the loan loss outcomes are symmetric about the BectiJs variables are represented by X, xs, dimensioned by The states must be weighted by probabilities and, for each risk variable (~EJ), we assume probabilities of occurrence (Pi,,) for the Basecase (U= 1) and two Alternatives (u=2,3) that sum to unity (For computational convenience, we assume that the pj,u’Sare independent across the j’s, though it is straightforward to relax this assumption.) These probabilities of occurrence are multiplied together to form the probability for a given state Calculating the permutations of the pj,u’S across 1, we obtain the probabiiities for the possible states (x,, s = 1, ,31; EWE II, x 1) For example, for Scenario listed in column (1) of table 2, we assume that the Basecase probabilities for all four j’s equal 0.80 Thus, Pi, 1=0.80 for j = 1,2,3,4, and the probability of the Scenario Benchmark is 8.41 ( =0.804) To gain a better understanding of the responsiveness of our measure of risk-exposure to variations in the probabilities, additional plausible scenarios are examined, and are represented by the remaining entries in table Scenarios 2-5 are similar in that they maintain the 0.8O/O.IO/O.10weights for all but one of the sets of exogenous variables In the exceptional ca