Loan Portfolio Selection and Risk Measurement Chapters 10 and 11 The Paradox of Credit • Lending is not a “buy and hold”process. • To move to the efficient frontier, maximize return for any given level of risk or equivalently, minimize risk for any given level of return. • This may entail the selling of loans from the portfolio. “Paradox of Credit” – Fig. 10.1. Saunders & Allen Chapters 10 & 11 2 Figure 10.1 The paradox of credit. B The Efficient Frontier Return C 0 A Risk Saunders & Allen Chapters 10 & 11 3 Managing the Loan Portfolio According to the Tenets of Modern Portfolio Theory • Improve the risk-return tradeoff by: – Calculating default correlations across assets. – Trade the loans in the portfolio (as conditions change) rather than hold the loans to maturity. – This requires the existence of a low transaction cost, liquid loan market. – Inputs to MPT model: Expected return, Risk (standard deviation) and correlations Saunders & Allen Chapters 10 & 11 4 The Optimum Risky Loan Portfolio – Fig. 10.2 • Choose the point on the efficient frontier with the highest Sharpe ratio: – The Sharpe ratio is the excess return to risk ratio calculated as: r R p f p Saunders & Allen Chapters 10 & 11 5 Figure 10.2 The optimum risky loan portfolio B D Return (Rp ) rf A C Risk (p ) Saunders & Allen Chapters 10 & 11 6 Problems in Applying MPT to Untraded Loan Portfolios • Mean-variance world only relevant if security returns are normal or if investors have quadratic utility functions. – Need 3rd moment (skewness) and 4th moment (kurtosis) to represent loan return distributions. • Unobservable returns – No historical price data. • Unobservable correlations Saunders & Allen Chapters 10 & 11 7 KMV’s Portfolio Manager • Returns for each loan I: – Rit = Spreadi + Feesi – (EDFi x LGDi) – rf • Loan Risks=variability around EL=EGF x LGD = UL – LGD assumed fixed: ULi = EDF (1 EDF ) – LGD variable, but independent across borrowers: ULi = EDFi(1 EDFi) LGDi2 EDFiVOL2i – VOL is the standard deviation of LGD. VVOL is valuation volatility of loan value under MTM model. – MTM model with variable, indep LGD (mean LGD): ULi = EDFi(1 EDFi) LGDi2 EDFiVVOL2i (1 EDFi)VVOL2i Saunders & Allen Chapters 10 & 11 8 Valuation Under KMV PM • Depends on the relationship between the loan’s maturity and the credit horizon date: • Figure 11.1: DM if loan’s maturity is less than or equal to the credit horizon date (maturities M1 or M2). • MTM if loan’s maturity is greater than credit horizon date (maturity M3). See Appendix 11.1 for valuation. Saunders & Allen Chapters 10 & 11 9 Figure 11.1 Loan maturity ( 0 M1 M) versus loan horizon ( M2 H Saunders & Allen Chapters 10 & 11 H). M3 Date 10 Correlations • Figure 11.2 – joint PD is the shaded area. • GF = GF/GF • GF = JDFGF ( EDFG EDFF ) EDFG (1 EDFG ) EDFF (1 EDFF ) • Correlations higher (lower) if isocircles are more elliptical (circular). • If JDFGF = EDFGEDFF then correlation=0. Saunders & Allen Chapters 10 & 11 11 Figure 11.2 Value correlation. Market Value of Assets - Firm G Firm G Market Value of Assets - Firm F Firm F 100(1-LGD) 100 Face Value of Debt Firm F ’s Debt Payoff Saunders & Allen Chapters 10 & 11 12 Role of Correlations • Barnhill & Maxwell (2001): diversification can reduce bond portfolio’s standard deviation from $23,433 to $8,102. • KMV diversifies 54% of risk using 5 different BBB rated bonds. • KMV uses asset (de-levered equity) correlations, CreditMetrics uses equity correlations. • Correlation ranges: – KMV: .002 to .15 – Credit Risk Plus: .01 to .05 – CreditMetrics: .0013 to .033 Saunders & Allen Chapters 10 & 11 13 Calculating Correlations using KMV PM • Construct asset returns using OPM. • Estimate 3-level multifactor model. Estimate coefficients and then evaluated asset variance and correlation coefficients using: • First level decomposition: – Single index model – composite market factor constructed for each firm. • Second level decomposition: – Two factors: country and industry indices. • Third level decomposition: – Three sets of factors: (1) 2 global factors (market-weighted index of returns for all firms and return index weighted by the log of MV); (2) 5 regional factors (Europe, No. America, Japan, SE Asia, Australia/NZ); (3) 7 sector factors (interest sensitive, extraction, consumer durables, consumer nondurables, technology, medical services, other). Saunders & Allen Chapters 10 & 11 14 CreditMetrics Portfolio VAR • Two approaches: – Assuming normally distributed asset values. – Using actual (fat-tailed and negatively skewed) asset distributions. • For the 2 Loan Case, Calculate: – Joint migration probabilities – Joint payoffs or loan values – To obtain portfolio value distribution. Saunders & Allen Chapters 10 & 11 15 The 2-Loan Case Under the Normal Distribution • Joint Migration Probabilities = the product of each loan’s migration probability only if the correlation coefficient=0. – From Table 10.1, the probability that obligor 1 retains its BBB rating and obligor 2 retains it’s a rating would be 0.8693 x 0.9105 = 79.15% if the loans were uncorrelated. The entry of 79.69% suggests a positive correlation of 0.3. Saunders & Allen Chapters 10 & 11 16 Mapping Ratings Transitions to Asset Value Distributions • Assume that assets are normally distributed. • Compute historic transition matrix. Figure 11.3 uses the matrix for a BB rated loan. • Suppose that historically, there is a 1.06% probability of transition to default. This corresponds to 2.3 standard deviations below the mean on the standard normal distribution. • Similarly, if there is a 8.84% probability of downgrade from BB to B, this corresponds to 1.23 standard deviations below the mean. Saunders & Allen Chapters 10 & 11 17 Joint Transition Matrix • Can draw a figure like Fig. 11.3 for the A rated obligor. There is a 0.06% PD, corresponding to 3.24 standard deviations below the mean; a 5.52% probability of downgrade from A to BBB, corresponding to 1.51 std dev below the mean. • The joint probability of both borrowers retaining their BBB and A ratings is: the probability that obligor 1’s assets fluctuate between –1.23 to +1.37 and obligor 2’s assets between –1.51 to +1.98 with a correlation coefficient=0.2. Calculated to equal 73.65%. Saunders & Allen Chapters 10 & 11 18 Figure 11.3 The link betw een asset value volatility ( and rating transition for a BB rated borrow er. Class: Def CCC B BB BBB Transition Prob. (%): 1.06 1.00 8.84 80.53 7.73 Asset (): 2.30 2.04 1.23 1.37 Saunders & Allen Chapters 10 & 11 ) A AA AAA 0.67 0.14 0.03 2.39 2.93 3.43 19 Calculating Correlation Coefficients • Estimate systematic risk of each loan – the relationship between equity returns and returns on market/industry indices. • Estimate the correlation between each pair of market/industry indices. • Calculate the correlation coefficient as the weighted average of the systematic risk factors x the index correlations. Saunders & Allen Chapters 10 & 11 20 Two Loan Example of Correlation Calculation • Estimate the systematic risk of each company by regressing the stock returns for each company on the relevant market/industry indices. • RA = .9RCHEM + UA • RZ = .74RINS + .15RBANK + UZ • A,Z=(.9)(.74)CHEM,INS + (.9)(.15)CHEM,BANK • Estimate the correlation between the indices. • If CHEM,INS =.16 and CHEM,BANK =.08, then AZ=0.1174. Saunders & Allen Chapters 10 & 11 21 Joint Loan Values • Table 11.1 shows the joint migration probabilities. • Calculate the portfolio’s value under each of the 64 possible credit migration possibilities (using methodology in Chap.6) to obtain the values in Table 11.3. • Can draw the portfolio value distribution using the probabilities in Table 11.1 and the values in Table 11.3. Saunders & Allen Chapters 10 & 11 22 Credit VAR Measures • Calculate the mean using the values in Table 11.3 and the probabilities in Tab 11.1. – Mean = 64 pV i 1 – Variance = i i 64 p i 1 i (Vi Mean ) 2 – Mean=$213.63 million – Standard deviation= $3.35 million Saunders & Allen Chapters 10 & 11 23 th 99 Calculating the percentile credit VAR under normal distribution • 2.33 x $3.35 = $7.81 million • Benefits of diversification. The BBB loan’s credit VAR (alone) was $6.97million. Combining 2 loans with correlations=0.3, reduces portfolio risk considerably. Saunders & Allen Chapters 10 & 11 24 Calculating the Credit VAR Under the Actual Distribution • Adding up the probabilities (from Table 11.1) in the lowest valuation region in Table 11.3, the 99th percentile credit VAR using the actual (not normal) distribution is $204.4 million. • Unexpected Losses=$213.63m - $204.4m = $9.23 million (>$7.81m). • If the current value of the portfolio = $215m, then Expected Losses=$215m - $213.63m = $1.37m. Saunders & Allen Chapters 10 & 11 25 CreditMetrics with More Than 2 Loans in the Portfolio • Cannot calculate joint transition matrices for more than 2 loans because of computational difficulties: A 5 loan portfolio has over 32,000 joint transitions. • Instead, calculate risk of each pair of loans, as well as standalone risk of each loan. • Use Monte Carlo simulation to obtain 20,000 (or more) possible asset values. Saunders & Allen Chapters 10 & 11 26 Monte Carlo Simulation • First obtain correlation matrix (for each pair of loans) using the systematic risk component of equity prices. Table 11.5 • Randomly draw a rating for each loan from that loan’s distribution (historic rating migration) using the asset correlations. • Value the portfolio for each draw. • Repeat 20,000 times! New algorithms reduce some of the computational requirements. • The 99th% VAR based on the actual distribution is the 200th worst value out of the 20,000 portfolio values. Saunders & Allen Chapters 10 & 11 27 MPT Using CreditMetrics • Calculate each loan’s marginal risk contribution = the change in the portfolio’s standard deviation due to the addition of the asset into the portfolio. • Table 11.6 shows the marginal risk contribution of 20 loans – quite different from standalone risk. • Calculate the total risk of a loan using the marginal contribution to risk = Marginal standard deviation x Credit Exposure. Shown in column (5) of Table 11.6. Saunders & Allen Chapters 10 & 11 28 Figure 11.4 • Plot total risk exposure using marginal risk contributions (column 6 of Table 11.6) against the credit exposure (column 5 of Table 11.4). • Draw total risk isoquants using column 5 of Table 11.6. • Find risk outliers such as asset 15 which have too much portfolio risk ($270,000) for the loan’s size ($3.3 million). • This analysis is not a risk-return tradeoff. No returns. Saunders & Allen Chapters 10 & 11 29 Figure 11.4 Credit limits and loan selection in CreditMetrics. 9 15 8 7 7 6 5 14 4 3 “Isoquant” Curve of Equal Total Risk $70,000 16 13 6 2 9 5 1 12 10 20 1 0 0 2 4 6 8 8 18 10 12 14 16 Credit Exposure ($ Millions) Saunders & Allen Chapters 10 & 11 30 Default Correlations Using Reduced Form Models • Events induce simultaneous jumps in default intensities. • Duffie & Singleton (1998): Mean reverting correlated Poisson arrivals of randomly sized jumps in default intensities. • Each asset’s conditional PD is a function of 4 parameters: h (intensity of default process); (constant arrival prob.); k (mean reversion rate); (steady state constant default intensity). • The jumps in intensity follow an exponential distribution with mean size of jump=J. • So: probability of survival from time t to s: p(t,s) = exp{(s-t)+(s-t)h(t)} where (t) = -(1 – e-kt)/k (t) = -[t + (t)] – [/(J+k)][Jt – ln(1 - (t)J)] Saunders & Allen Chapters 10 & 11 31 Numerical Example • Suppose that =.002, k=.5, =.001, J=5, h(0)=.001 (corresponds to an initial rating of AA). • Correlations across loan default probabilities: = vVc + V • Vc=common factor; V=idiosyncratic factor. As v0, corr0 As v1, corr1. • If v=.02, V=.001, Vc=.05: the probability that loani intensity jumps given that loanj has experienced a jump is = vVc/(Vc+V) = 2%. If v= .05 (instead of .02), then the probability increases to 5%. • Figure 11.5 shows correlated jumps in default intensities. • Figure 11.6 shows the impact of correlations on the portfolio’s risk. Saunders & Allen Chapters 10 & 11 32 Figure 11.5 Correlated default intensities. Source: Duf f e and Singleton (1998), p.25. The f igure shows a portion of a simulated sample path of total def ault arriv al intensity (exactly 1,000 f irms). An X denotes a def ault ev ent. 150 100 50 0 2.2 2.4 2.6 2.8 3 Year 3.2 3.4 Marketw ide Credit Event Saunders & Allen Chapters 10 & 11 3.6 3.8 4 Calendar Time 33 Figure 11.6 Portfolio default intended. Source: Duf f e and Singleton (1998), p.27. The f igure shows the probabilty of anm-day interv al within 10 y ears hav ing f our or more def aults (base case). 0.7 0.6 0.5 0.4 0.3 High Correlation Medium Correlation Low Correlation 0.2 0.1 0 0 10 20 30 40 50 60 70 80 90 100 Time Window m (Days) Saunders & Allen Chapters 10 & 11 34 Appendix 11.1: Valuing a Loan that Matures after the Credit Horizon – KMV PM • Maturity=M3 in Figure 11.1. • Four Step Process: Use MTM to value loans. – 1. Valuation of an individual firm’s assets using random sampling of risk factors. – 2. Loan valuation based on the EDFs implied by the firm’s asset valuation. – 3. Aggregation of individual loan values to construct portfolio value. – 4. Calculation of excess returns and losses for portfolio. • Yields a single estimate for expected returns (losses) for each loan in the portfolio. Use Monte Carlo simulation (repeated 50,000 to 200,000 times) to trace out distribution Saunders & Allen Chapters 10 & 11 35 Step 1: Valuation of Firm Assets at 3 Time Horizons – Fig. 11.7 • A0 , AH , AM valuations. Stochastic process generating AH, AM: ln AH = ln A0 + (-.52)tH + HtH (11.21) where AH = the asset value at the credit horizon date H, = the expected return (drift term) on the asset valuation, = the volatility of asset returns, tH = the credit horizon time period, H = a random risk term (assumed to follow a standard normal distribution). • The random component = systematic portion f + firm-specific portion u. Each simulation draws another risk factor. • Using AH and AM can calculate EDFH and EDFM Saunders & Allen Chapters 10 & 11 36 Step 2: Loan Valuation Using Term Structure of EDFs • Convert EDF into QDF by removing risk-adjusted ROR. V0 = PV0(1 – LGD) + PV0(1-QDF)LGD (11.22) where V0 = the loan’s present value, PV0 = the present value factor using the riskfree rate to discount the loan’s cash flows to time t=0, QDF = the (cumulative) risk neutral quasi-EDF, LGD = the loss given default • Also value loan as of credit horizon date H: VH|ND = CH + PVH(1 – LGD) + PVH(1-QDF)LGD (11.23) where VH|ND = the loan’s expected value as of the credit horizon date given that default has not occurred, CH = the cash flow on the credit horizon date, PVH = the present value factor using the riskfree rate as the discount factor to discount the loan’s cash flows to time t=H. However, there is a possibility that the loan will default on or before the credit horizon date. The expected value of the loan given default is: VH|D = (CH + PVH)LGD (11.24) VH = (EDF) VH|D + (1-EDF) VH|ND Saunders & Allen Chapters 10 & 11 (11.25) 37 Step 3: Aggregation to Construct Portfolio • Sum the expected values VH for all loans in the portfolio. Vt P = Vt i (11.26) i where Vt P = the value of the loan portfolio at date t=0,H, Vt i = the value of each loan i at date t=0,H. Saunders & Allen Chapters 10 & 11 38 Step 4: Calculation of Excess Returns/Losses • Excess Returns on the Portfolio: RH = V HP V0P RF V0P (11.27) where RH = the excess return on the loan portfolio from time period 0 to the credit horizon date H, V HP = the expected value of the loan portfolio at the credit horizon date, V0P = the present value of the loan portfolio, RF = the riskfree rate. • Expected Loss on the Portfolio: ELH VH | ND VH V0 (11.28) • Repeat steps 1 through 4 from 50,000 to 200,000 times. Saunders & Allen Chapters 10 & 11 39 A Case Study: KMV PM valuation of 5 yr maturity $1 loan paying a fixed rate of 10% p.a. • Using Table 11.8: V0 = PV0(1 – LGD) + PV0(1-QDF)LGD = 1.2103(.50) + (1.0675)(.50) = $ 1.1389 Table 11.8 Valuing the Loan’s Present Value Time Period Cash flows per period (1) 1 2 3 4 5 Totals (2) .10 .10 .10 .10 1.10 Discount Factor e tRF Risk-free Present Value of Cashflows EDFi QDFi cumulative cumulative (3) (2) x (3) = (4) (5) (6) .9512 .9048 .8607 .8187 .7788 .0951 .0905 .0861 .0819 .8567 1.2103 .0100 .0199 .0297 .0394 .0490 .0203 .0471 .0770 .1088 .1414 Saunders & Allen Chapters 10 & 11 Risky Present Value of Cashflows (7) .0932 .0862 .0795 .0730 .7356 1.0675 40 Valuing the Loan at the Credit Horizon Date =1 • Using Table 11.9: VH|ND = CH + PVH(1 – LGD) + PVH(1-QDF)LGD = 0.10 + 1.1723(.50) + (1.0615)(.50) = $ 1.2169 VH|D = (CH + PVH)LGD = (0.10 + 1.1723)(.50) = $ 0.63615 VH = (EDF) VH|D + (1-EDF) VH|ND = (.01)(.63615) + (.99)(1.2169) = $ 1.2111 Time Period Cash flows per period (1) 1 2 3 4 5 Totals (2) .10 .10 .10 .10 1.10 Discount Factor e tRF Risk-free Present Value of Cashflows EDFi QDFi cumulative cumulative (3) (2) x (3) = (4) (5) (6) 1 .9512 .9048 .8607 .8187 0 .0951 .0905 .0861 .9006 1.1723 .0100 .0199 .0297 .0394 .0203 .0471 .0770 .1088 Saunders & Allen Chapters 10 & 11 Risky Present Value of Cashflows (7) .0932 .0862 .0795 .8026 1.0615 41 KMV’s Private Firm Model • Calculate EBITDA for private firm j in industryj. • Calculate the average equity mulitple for industryi by dividing the industry average MV of equity by the industry average EBITDA. • Obtain an estimate of the MV of equity for firm j by multiplying the industry equity multiple by firm j’s EBITDA. • Firm j’s assets = MV of equity + BV of debt • Then use valuation steps as in public firm model. Saunders & Allen Chapters 10 & 11 42 Credit Risk Plus Model 2 - Incorporating Systematic Linkages in Mean Default rates • Mean default rate is a function of factor sensitivities to different independent sectors (industries or countries). N AB = (mAmB)1/2 AkBk(k/mk)2 (11.20) k 1 where AB = default correlation between obligor A and B, mA = mean default rate for type A obligor, mB = mean default rate for type B obligor, A = allocation of obligor A's default rate volatility across N sectors, B = allocation of obligor B's default rate volatility across N sectors, (k/mk)2 = proportional default rate volatility in sector k. • Table 11.7 shows as example of 2 loans sensitive to a single factor (parameters reflect US national default rates). As credit quality declines (m gets larger), correlations get larger. Saunders & Allen Chapters 10 & 11 43