1 / 37

Class 5 Credit risk models

Class 5 Credit risk models. Up to now. Why credit risk is harder to model than market risk What affects credit losses Probability of default Loss given default (LGD) / Recovery rate (RR=1–LGD) Credit exposure How to measure credit risk

renate
Télécharger la présentation

Class 5 Credit risk models

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Class 5Credit riskmodels

  2. Up to now • Why credit risk is harder to model than market risk • What affects credit losses • Probability of default • Loss given default (LGD) / Recovery rate (RR=1–LGD) • Credit exposure • How to measure credit risk • Internal approach: analyzing company’s characteristics • External approach: using market data on bonds and stocks

  3. What if there is no external rating?

  4. Internal rating systems • Criteriа • Probability of default • Recovery rate • Horizon • Usually, 1 year • Scale • According to S&P/Moody’s or its own • Factors: what are the 5C?

  5. What can we learn (and cannot learn) from financial reports? • Different financial ratios • Liquidity, leverage, profitability, turnover,… • Backward-looking • Extrapolation of the past into the future gives imprecise forecast • Often reports are corrupted • To misguide tax authorities, minority shareholders, banks, etc. • In Russia: little trust to reports • RAS is clearly inferior to IAS • IAS has become compulsory for the exchange-listed companies • Role of human factor • Expert’s opinion: visit to the company, talks with top managers • But: possibility of corruption and subjectivity

  6. Credit scoring models: a statistical approach to predicting default • Statistical model: PD = function of the borrower’s characteristics • How to evaluate the model? • Type-1 error: default by the borrower who received a loan • Type-2 error: predicting default for the good borrower • Typical strategies: • RusskiStandart vs. Sberbank in 2000s Bad borrowers Good borrowers Income Credit history

  7. Which factors can be used in retail lending scoring models? • Gender, age, family • Registration • Car, apartment, dacha • Income • Debt • Credit history • Conviction • …

  8. Altman (1968): scoring model for firms Z = 1.2X1 + 1.4X2 + 3.3X3 + 0.6X4+ 0.999X5 • Dependent variable: Z-score • Sample: 66 companies, half of which defaulted • Determinants of PD • X1: Working Capital to Assets • X2: Retained Earnings to Assets • X3: EBIT to Assets • X4: Market Value of Equity to Book Value of Liabilities • X5: Sales to Assets • Interpretation • Z > 3: default is unlikely • 2.7 < Z ≤ 3: closer to “dangerous zone” • 1.8 < Z ≤ 2.7: likely default • Z ≤ 1.8: very high PD • Results: • More than 90% firms correctly classified

  9. What are pros and cons of scoring models? • ‘Optimal’ treatment of factors • …without subjectivity • Automated decisions • Quick • Economy of scale • Flexibility: calibrate models • …for behavior, collectors, fraud • …with non-linear or step-wise functional forms • Required by regulators (Basel II) • Need large and clean data base • Ideally, including rejected applicants • Some data are hard to verify (quickly and cheaply) • Model risk • How “average” results fit each specific segment • How often the model should be reevaluated

  10. What can we learn from bond spread? • Bond spread: • Difference in yields of risky bond and risk-free bond • For a one-period zero-coupon bond: BS ≡ R-Rf≈ PD*LGD • LGD: relative to face value • PD: risk-neutralprobability • Which factors influence bond spread besides credit risk? • Macro factors (market liquidity) • Bond-specific factors • Is bond spread a better measure of credit risk than credit rating? • Ratings often react with a lag to the change in bond spread • Bond spreads may change due to other factors or noise

  11. Are spreads of high-yield bonds in the US too low? • The average yield ≈ 5.5% • Cf. long-term average 10% over the past 20 years • Close to long-term average yield of investment-grade corporate bonds • Low defaults: 2.6% in 2012, 2.3% in 2013 • Below the long-term historical average of 4.8% • Due to low leverage, lots of cash, little need in refinancing • How will the Fed’s tapering affect the bond market?

  12. European sovereign bond spreads

  13. What can we learn from CDS spread? • Credit default swap (CDS): agreement between two parties to exchange the credit risk of a reference entity • Similar to buying insurance against default • What is the difference? • Why did it become so popular? • CDS started in early 1990’s, was standardized in 1999 • CDS spread = Premium paid by the protection buyer to the protection seller • Quoted in basis points p.a. of the contract’s notional amount • Paid quarterly • In case of default, the protection seller covers occurred losses • Usually, buyer delivers the asset (bonds, loans) to seller and receives 100% of notional • For a one-year bond: CDS spread ≈ PD*LGD • Just like bond spread, CDS spread depends on other factors: • Liquidity, counterparty risk, equity risk premia, market sentiment (speculation),… • The basket (first-to-default) swap is harder to value

  14. Пример: Tyco

  15. European sovereign CDS spreads

  16. What is a better measure of credit risk: CDS spread or bond spread? • No arbitrage conditions should ensure that CDS spread ≈ bond spread • Indeed, credit risk prices for bonds and CDS are equal over the long term • What could be the reason for short-term price differentials? • Factors unrelated to the credit risk (especially liquidity) • Contractual arrangements in CDS contract • Restructuring clauses • Delivery options • Quotation differences

  17. Greece: CDS spread vs. bond spread

  18. CreditMetrics (1997) • Credit risk is modeled as a change in credit rating • Using migration probabilities • Compute a probability distribution of the future value of the instrument using • Forward rates for each rating • Thus, account for duration and spread effects • Recovery rate in case of default (depending on seniority) • Estimate VaR as a difference between given quantileand expected value • Bottom up approach: • First estimate VaR for each instrument • Then aggregate to portfolio VaR accounting for correlations • See computations in the Excel file

  19. Migration of credit ratings

  20. CreditMetrics: critique and potential for improvement • Relying on credit ratings • Assuming homogeneity within the same rating class • Using discrete migration matrix based on average historical frequencies • Credit Portfolio View: adjust transition probabilities according to the current stage of the business cycle • Downgrade probabilities are higher during the recession • Ignoring market risk • It is better to use stochastic interest rates • Easy to account for this via Monte Carlo

  21. Merton (1974) • The company’s capital structure includes • Equity, with value E • Debt (zero-coupon), with face value F and maturity T • Default occurs at maturity if VT<F • Stockholders receive at T: max(VT-F,0) • Creditors receive at T: min(VT, F) • Stockholders: call option on the value of the company V • Exercise date: T • Price of the underlying asset: V • Volatility: σV

  22. Merton model: illustration

  23. Derivation of the parameters • V and σV are unobservable, derived from two equations: • The value of equity by Black-Scholes: E=V*N(d1)-Fe-rTN(d2) • r: risk-free rate • d1=[ln(V/F) + T(r+σ2/2)] / [σV√T], d2=d1-σV√T • The equation for stock volatility: σEE = N(d1) σVV • Risk-neutral probability of default: PD = 1-N(d2) = N(-d2)

  24. Merton’s model: Implicit assumptions • Stockholders’ behavior – as given • Though they are interested in raising risk • Lognormal distribution • Underestimate PD at short horizon • Bankruptcy when V is below the face value of debt • Default may be different from bankruptcy

  25. KMV Portfolio Manager (1998) • Estimate V and σV based on equity prices • Compute distance to default • Empirically estimated default point: DPT = short-term liabilities + ½ long-term liabilities • Distance to default: DD = ln[E(V)-DPT]/σV • in %, in σ

  26. Estimating EDF (cont.)

  27. Estimating EDF • The historical frequency of defaults with given distance to default: EDF = # defaults / # firms (with given DD) • EDFis a continuous company-specific risk measure • Credit rating is an ordinal measure

  28. Example: EDF for FedEx • Why did EDF change? • Increase in asset volatility • Increase in financial leverage

  29. EDF vs. credit ratings

  30. EDF vs. credit ratings • EDFgoes up 1-2 years before the default • Ratings agencies react with a lag

  31. EDF vs. credit ratings • The probability of keeping (changing) the rating seems overstated (understated) by the rating agencies

  32. What are pros and cons of EDF? • Company-specific • Continuous • Not biased by periods of high or low defaults • Quick reaction • EDF rises sharply 1-2 years before the default • Best applied to publicly traded firms • Relies on market liquidity and efficiency • Ignores more complicated features of the debt • Seniority, collateral, etc.

  33. Credit derivatives as risk management tool • May be linked to different types of credit risk: • Credit event, value of the underlying asset, recovery rate, maturity • Completing the market • Risk managers hedge credit risk • Investors find interesting instruments to speculate • Arbitrageurs can short-sell credit risk • Separating credit riskmgt from the underlying asset • Banks can clean the balance without selling loans • Tax considerations / underpricing / client • Hedge funds can invest in credit risk, with leverage • Avoid transfer of property rights and administrative costs • Rapid growth since the end of 1990s until the crisis

  34. Risks of credit derivatives • Correlation • Simultaneous default of the underlying asset and protection seller • Basis • Legal • 1999, 2003: ISDA adopted standard terms and documentation • Liquidity • Credit derivatives are usually traded OTC • Protection: • Bilateral netting • Option for premature abortion of the contract in the case of the counterparty’s financial distress

  35. Do credit derivatives reduce or raise systemic risk? • It is argued that CDS contributed to the global financial crisis of 2008 • Many traders speculated on CDS linked to the mortgage market • …and investment banks exposed to mortgages: Bear Stearns, Lehman Brothers • …contributing to their decline • What were the problems? • How it was resolved: • ISDA: standardization of contracts and settlement • CCP: centralized clearing party • More transparency (central repository) • More than 90% contracts are covered by collateral • (Banning naked CDS)

More Related