Credit Risk讲稿.ppt

1、CREDIT RISK,Credit Risk Modeling,3,Measuring Credit Risk: Overview,Credit Migration Approach: CreditMetrics (from J.P. Morgan) CreditVaR (CIBC) CreditPortfolioView (McKinsey) The Option Pricing Approach: KMV (from KMV Corp.),What are the current proposed industry sponsored Credit VaR methodologies?,

2、4,What are the current proposed industry sponsored Credit VaR methodologies?,Measuring Credit Risk: Overview,The Actuarial Approach: CreditRisk+ (from Credit Suisse First Boston) The Reduced Form Approach: Jarrow/Turnbull Duffie/Singleton,5,Credit migration approach,Contingent claim,approach,Actuari

3、al approach,Reduced form approach,Software,CreditMetrics,CreditPortfolioView,KMV,CreditRisk+,Kamakura,Definition of,Risk,D,Market Value,Market Value,Default losses,Default losses,Default losses,Credit events,Downgrade/Default,Downgrade/Default,Continuous default,probabilities,Default,Default,Risk dr

4、ivers,Asset Values,Macro-factors,Asset Values,Expected default,rates,Hazard rate,Transition,probabilities,Constant,Driven by Macro,factors,Driven by:,-Individual term,structure of EDF,-Asset value process,N/A,N/A,Correlation of,credit events,Standard,multivariate normal,distribution (equity-,factor

5、model),Conditional default,probabilities as,functions of macro-,factors,Standard,multivariate normal,asset returns (asset,factor model),Conditional,default,probabilities as,functions of,common risk,factors,Conditional default,probabilities as,functions of macro-,factors,Recovery rates,Random (Beta,d

6、istribution,Random (empirical,distribution),Random (Beta,distribution),Loss given default,deterministic,Loss given default,deterministic,Numerical,approach,Simulation/Analytic,Simulation,Analytic/Simulation,Analytic,Tree based /simulation,D,Comparison of Models,6,Measuring Credit Risk: Overview,Cred

7、it risk models should capture: Spread risk Downgrade risk Default risk Recovery rate risk Concentration risk (portfolio diversification and correlation risk),7,Measuring Credit Risk: Overview,Credit risk models generate: Loss distribution (default risk) KMV, CreditRisk + Portfolio value distribution

8、 (migration and default risks) CreditMetrics, CreditVaR, CreditPortfolioView,8,Measuring Credit Risk: Overview,Typical market returns,Typical credit returns,Portfolio Value,Source: CIBC,Comparison of the distributions of credit returns and market returns,Frequency,9,Measuring Credit Risk: Overview,K

9、ey input parameters common to all models obligors information exposures recovery rate (loss given default: LGD) default correlations (concentration risk),IIThe Credit Migration Approach,11,Credit Migration Approach,Key input parameters: Credit data: Credit horizon Credit rating system: Moodys, S而在经济

10、衰退时期,违约率则很高。 2)模型假定资产收益服从正态分布,它是进行模拟的基础,但资 产收益的实际分布有待进一步研究。 3)模型中假定企业资产收益之间的相关度等于公司证券收益 之间的相关度,该假设有待进一步验证,模型计算结果对于这一 假定的敏感性很高。 4)模型中假定无风险利率是固定不变的,影响投资组合价值的 只有各种信用事件,市场风险对于投资组合价值没有影响。,40,对上述模型的评论：KMV,优点：,首先，该模型可充分地利用资本市场信息， 对所有公开上市企业进行信用风险的量化度量和分析。 其次，由于这种方法所获得的数据来自股票市场的资料， 而非企业历史账面价值，因此更能反映企业当前的信用状况

11、。 再次，预期违约频率指标在本质上是一种对风险的基数衡量法， 从而可以反映风险水平差异的程度，因而更准确,41,对上述模型的评论：KMV(2),缺点：,首先，它只适用于上市公司的信用风险评估， 对非上市公司则要借助某些会计信息 或其他能反映借款企业特征值的指标来替代模型中一些变量。 其次，该模型假定借款企业的资产价值呈正态分布。 再次，该模型不能够对长期债务的不同类型进行分辨。但实际上， 可以依据其优先偿还顺序、有否担保、有否契约、能否转换 等来区别不同的长期债务，因而导致违约点DP的不确定。 最后，该模型属于静态模型，因为KMV模型基础的默顿期权模型是假设： 借款企业管理层一旦将企业的债务结

12、构确定下来之后， 该企业的这一结构就不变化了。,42,Internal Ratings-Based approach,43,Changing Regulatory Environment,1988Regulators recognized need for risk-based Capital for Credit Risk (Basel Accord) 1995Capital Regulations for Market Risk Published 1996-98Capital Regulations for Credit Derivatives 1997Discussion of usi

13、ng credit risk models for selected portfolios in the banking books 1999New Credit Risk Recommendations Bucket Approach - External and Possibly Internal Ratings Expected Final Recommendations by Fall 2001 Postpone Internal Models (Portfolio Approach) 2001Revised Basel Guidelines Revised Buckets - Sti

14、ll Same Problems Foundation and Advanced Internal Models Final Guidelines Expected in Fall 2002 - Implemented by 2005,44,Capital Adequacy Risk Weights from Various BIS Accords(Corporate Assets Only),Original 1988 Accord,All Ratings 100% of Minimum Capital (e.g. 8%),1999 (June) Consultative BIS Propo

15、sal,Rating/Weight AAA to AA- A+ to B-Below B-Unrated,20%,100%,150%,100%,2001 (January) Consultative BIS Proposal,AAA to AA-A+ to A-BBB+ to BB-Below BB-Unrated,20%,50%,100%,150%,100%,Altman/Saunders Proposal (2000,2001),AAA to AA-A+ to BBB-BB+ to B-Below B-Unrated,10%,30%,100%,150%,Internally Based A

16、pproach,45,The Importance of Credit Ratings,For Risk Management in General Greater Understanding Between Borrowers and Lenders Trade off between risk and return,46,Rating Systems,Bond Rating Agency Systems US (3) - Moodys, S&P (20+ Notches), Fitch/IBCA Bank Rating Systems 1 9, A F, Ratings since 199

17、5 Office of Controller of Currency System Pass (0%), Substandard (20%), Doubtful (50%), Loss (100%) NAIC (Insurance Agency) 1 6 Local Rating Systems Three (Japan) SERASA (Brazil) RAM (Malaysia) New Zealand (NEW) etc.,47,Debt Ratings,48,S&PS Debt rating process,Request Rating,Assign analytical team C

18、onduct basic research,Meet issuer,Rating Committee Meeting,Issue Rating,Appeals Process,49,Moodys rating analysis of an industrial company,Rating process includes quantitative, qualitative and legal analyses Quantitative analysis is mainly based on the firms financial reports Qualitative analysis is

19、 concerned with management quality, reviews the firms competitive situation as well as an assessment of expected growth within the firms industry plus the vulnerability to technological changes, regulatory changes, labor relations, etc,50,Issue,Company Structure,Operating /Financial Position,Managem

20、ent Quality,Industry / Regulatory Trends,Sovreign / macroeconomic Analysis,51,Scoring Systems,Qualitative (Subjective) Univariate (Accounting/Market Measures) Multivariate (Accounting/Market Measures) Discriminant, Logit, Probit Models (Linear, Quadratic) Non-Linear Models (e.g., RPA, NN) Discrimina

21、nt and Logit Models in Use Consumer Models - Fair Isaacs Z-Score (5) - Manufacturing ZETA Score (7) - Industrials Private Firm Models (eg. Risk Calc (Moodys), Z” Score) EM Score (4) - Emerging Markets, Industrial Other - Bank Specialized Systems,52,Scoring Systems(continued),Artificial Intelligence

22、Systems Expert Systems Neural Networks (eg. Credit Model (S&P), CBI (Italy) Option/Contingent Models Risk of Ruin KMV Credit Monitor Model,53,Basic Architecture of an Internal Ratings-Based (IRB) Approach to Capital,In order to become eligible for the IRB approach, a bank would first need to demonst

23、rate that its internal rating system and processes are in accordance with the minimum standards and sound practice guidelines which will be set forward by the Basel Committee. The bank would furthermore need to provide to supervisors exposure amounts and estimates of some or all of the key loss stat

24、istics associated with these exposures, such as Probability of Default (PD), by internal rating grade (Foundation Approach). Based on the banks estimate of the probability of default, as well as the estimates of the loss given default (LGD) and maturity of loan, a banks exposures would be assigned t

25、o capital “buckets” (Advanced Approach). Each bucket would have an associated risk weight that incorporates the expected (up to 1.25%) and unexpected loss associated with estimates of PD and LGD, and possibly other risk characteristics.,54,Recent (2001) Basel Credit Risk Management Recommendations,M

26、ay establish two-tier system for banks for use of internal rating systems to set regulatory capital. Ones that can set loss given default estimates, OR Banks that can only calculate default probability may do so and have loss (recovery) probability estimates provided by regulators. Revised plan (Jan

27、uary 2001) provides substantial guidance for banks and regulators on what Basel Committee considers as a strong, best practice risk rating system. Preliminary indications are that a large number of banks will attempt to have their internal rating system accepted. Basel Committee working to develop c

28、apital charge for operational risk. May not complete this work in time for revised capital rules. Next round of recommendations to take effect in 2004.,55,Risk Weights for Sovereign and Banks(Based on January 2001 BIS Proposal),Sovereigns,Credit Assessment AAA A+ BBB+BB+Below of Sovereignto AA-to A-

29、to BBB-to B- B- Unrated Sovereign risk weights 0%20% 50%100% 150% 100%,56,Risk Weights for Sovereign and Banks(Based on January 2001 BIS Proposal) (continued),Banks,Credit Assessment AAA A+ BBB+BB+Below of Banksto AA-to A-to BBB-to B- B- Unrated Risk weights 20% 50% 50%100% 150% 50% Risk weights for

30、 short-term claims 20% 20% 20% 50% 150% 20%,57,BIS Collateral Proposals,January 2001 Proposal introduced a W-factor on the extent of risk mitigation achieved by collateral W-factor is a minimum floor beyond which collateral on a loan cannot reduce the risk-weight to zero. Main rationale for the floo

31、r was “legal uncertainty” of collecting on the collateral and its price volatility September 2001 amendment acknowledges that legal uncertainty is already treated in the Operational Risk charge and proposes the the W-factor be retained but moved form the Pillar 1 standard capital adequacy ratio to P

32、illar 2s Supervisory Review Process in a qualitative sense Capital Ratio = Collateral Value (CV) impacts the denominator More CV the lower the RWA. Leads to a higher capital ratio on the freeing up of capital while maintaining an adequate Capital Ratio CV is adjusted based on 3 Haircuts: HE based on

33、 volatility of underlying exposure HC based on volatility of collateral HFX BASED on possible currency mismatch,58,BIS Collateral Proposals (continued),Simple Approach for most Banks (Except Most Sophisticated) Partial collateralization is recognized Collateral needs to be pledged for life of exposu

34、re Collateral must be marked-to-market Collateral must be revalued with a minimum of six months Floor of 20% except in special Repo cases Constraint on Portfolio Approach for setting collateral standards Correlation and risk through Systematic Risk Factors (still uncertain and not established),59,Re

35、lative Capital Allocation of Risk for Banks(Based on Basel II Guidelines Proposed),SAMPLE ECONOMIC CAPITAL ALLOCATION FOR BANKS,CREDIT RISK COMPONENTS,CREDIT RISK PARAMETERS,Default Probability Default Severity Migration Probabilities,Scoring Models Recovery Rates Transition Matrices,60,Expected Los

36、s Can Be Broken Down Into Three Components,EXPECTED LOSS $,=,Probability of Default (PD) %,x,Loss Severity Given Default (Severity) %,Loan Equivalent Exposure (Exposure) $,x,The focus of grading tools is on modeling PD,What is the probability of the counterparty defaulting?,If default occurs, how mu

37、ch of this do we expect to lose?,If default occurs, how much exposure do we expect to have?,Borrower Risk,Facility Risk Related,61,Rating System: An Example,PRIORITY: Map Internal Ratings to Public Rating Agencies,62,The Starting Point is Establishing a Universal Rating Equivalent Scale for the Clas

38、sification of Risk,Performing,Substandard,63,Default Probabilities Typically Increase Exponentially Across Credit Grades,64,Loan scoring / grading is not new, but as part of BIS II it will become much more important for banks to get it right Building the models and tools Factor / Variable selection

39、Model construction Model evaluation From model to decision tool “Field performance” of the models Stratification power Calibration Consistency Robustness Application and use tests Importance of education across the Bank,At the Core of Credit Risk Management Are Credit Scoring/Grading Models,65,There

40、 are four potentially useful criteria for evaluating the field performance of a scoring or grading tool: Stratification: How good are the tools at stratifying the relative risk of borrowers? Calibration: How close are actual vs. predicted defaults, both for the book overall and for individual credit

41、 grades? Consistency: How consistent are the results across the different scorecards? Robustness: How consistent are the results across Industries, over time and across the Bank Stratification is about ordinal ranking (AA grade has fewer defaults than A grade) Calibration is about cardinal ranking (

42、getting the right number of defaults per grade) Consistency concerns the first two criteria across different models: Different industries or countries within Loan Book (LOB) Across LOBs (e.g. large corporate, middle market, small business) Especially for high grades (BBB and above), field performanc

43、e is hard to assess accurately,Now That the Model Has Been in Use, How Can We Tell If Its Any Good?,66,Backtesting la VaR models is very hard, practically: Lopez might need multiple years However: difficult to assume within year independence Macroeconomic conditions affect everybody This will affect

44、 the statistics A test for grading tools: how do they fare through a recession During expansion years: expect “too few” defaults During recession years: expect “too many” defaults Two schools of credit assessment: Unconditional (“Through-the-cycle”): ratings from agencies are sluggish / insensitive

45、Conditional (“Mark-to-market): KMVs stock price-based PDs are sensitive / volatile / timely Z-Scores based PDs are sensitive / less volatile / less timely,Some Comments on Performance “In the Field”,67,Altman (1968) built a linear discriminant model based only on financial ratios, matched sample (by

46、 year, industry, size) Z = 1.2 X1 + 1.4 X2 + 3.3 X3 +0.6 X4 + 1.0 X5 X1 = working capital / total assets X2 = retained earnings / total assets X3 = earning before interest and taxes / total assets X4 = market value of equity / book value of total liabilities X5 = sales / total assets Most credit sco

47、ring models use a combination of financial and non-financial factors Financial Factors Non-financial Factors Debt service coverage Size Leverage Industry Profitability Age / experience of key managers Liquidity ALM Net worth Location,Many Internal Models are Based on Variations of the Altmans Z-Scor

48、e and Zeta Models,68,Decision Points When Building a Model,Sample selection: How far back do you go to collect enough “bads” ? Ratio of “goods” to “bads” ? Factor or variable selection Financial factors Many financial metrics are very similar highly correlated Non-financial factors More subject to m

49、easurement error and subjectivity Model selection Linear discriminant analysis (e.g. Altmans Z-Score, Zeta models) Logistic regression Neural network or other machine learning methods (e.g. CART) Option based (e.g. KMVs CreditMonitor) for publicly traded companies Model evaluation In-sample Out-of-s

50、ample (“field testing”),Decision Points When Building a Model,69,In binary event modeling (“goods” vs. “bads”), the basic idea is correct classification and separation There is a battery of statistical tests which are used to help us with selecting among competing models and to assess performance,Al

51、l Model Evaluation is Done on the Basis of Error Rate Analysis,Predicted Negatives,Predicted Positives,True Negatives,False Positives (type I error),False Negatives (type II error),True Positive,Actual Negatives,Actual Positives,2x2 Confusion / Classification Table,Error Rate = false negatives + fal

52、se positives Note that you may care very differently about the two error types Cost of Type I usually considerably higher (e.g. 15 to 1),70,It is One Thing to Measure Risk & Capital, It is Another to Apply and Use the Output,There are a host of possible applications of a risk and capital measurement

53、 framework: Risk-adjusted pricing Risk-adjusted compensation Limit setting Portfolio management Loss forecasting and reserve planning Relationship profitability Banks and supervisors share similar (but not identical) objectives, but both are best achieved through the use and application of a risk an

54、d capital measurement framework,SUPERVISOR,BANK,Capital Adequacy “Enough Capital”,Capital Efficiency “Capital Deployed Efficiently”,71,Applications Include Risk-Adjusted Pricing, Performance Measurement and Compensation,At a minimum, risk-adjusted pricing means covering expected losses (EL) Price =

55、LIBOR + EL + (fees & profit) If a credit portfolio model is available, i.e.correlations and concentrations are accounted for, we can do contributory risk-based pricing Price = LIBOR + EL + CR + (fees & profit) Basic idea: if marginal loan is diversifying for the portfolio, maybe able to offer a disc

56、ount, if concentrating, charge a premium With the calculation of economic capital, we can compute RAROC (risk-adjusted return to economic capital) - Returns relative to standard measure of risk Used for LOB performance measurement by comparing RAROCs across business lines Capital attribution and con

57、sumption Input to compensation, especially for capital intensive business activities (e.g. lending, not deposits) Capital management at corporate level,72,Risk Based Pricing Framework,=,+,+,Price (Interest Rate),Cost of Funds,Credit Charge,Loan Overhead & Operating Risk,73,Proposed Credit Risk Prici

58、ng Model,Credit Charge,Risk Charge,Overheads,Expected Loss Charge,Capital at Risk,Default Rate,1-Recovery Rate,Hurdle Rate,Capital at Risk,=,+,74,Risk Based Pricing: An Example,Given:5-Year Senior Unsecured Loan Risk Rating = BBB Expected Default Rate = 0.3% per year (30 b.p.) Expected Recovery Rate

59、 = 70% Unexpected Loss () 50 b.p. per year BIS capital Allocation = 8% Cost of Equity Capital = 15% Overhead + Operations Risk Charge = 40 b.p. per year Cost of Funds = 6% Loan Price(1) = 6.0% + (0.3% x 1-.7) + (6 0.5% x 15%) + 0.4% = 6.94% Or Loan Price(2) = 6.0% + (0.3% x 1-.7) + (8.0% x 15%) + 0.4% = 7.69% (1) Internal Model for Capital Allocation (2) BIS Capital Allocation method,75,Four