VALUE-ADDING ACTIVE CREDIT PORTFOLIO MANAGEMENT OPTIMISATION AT ALL LEVELS Dr. Christian Bluhm Head Credit Portfolio Management Credit Suisse, Zurich September 28-29, 2005, Wiesbaden
AGENDA INTRODUCTION AND OVERVIEW EXAMPLE AT SINGLE-NAME LEVEL: THE WAY TOWARDS MORE COMPETITIVE RATINGS EXAMPLE AT PORTFOLIO LEVEL: THE WAY TOWARDS MORE MEANINGFUL VALUATIONS PORTFOLIO STEERING AND OPTIMISATION Disclaimer: This talk reflects the opinion of the author and not necessarily the opinion of Credit Suisse. Folie 1
AGENDA INTRODUCTION AND OVERVIEW EXAMPLE AT SINGLE-NAME LEVEL: THE WAY TOWARDS MORE COMPETITIVE RATINGS EXAMPLE AT PORTFOLIO LEVEL: THE WAY TOWARDS MORE MEANINGFUL VALUATIONS PORTFOLIO STEERING AND OPTIMISATION Disclaimer: This talk reflects the opinion of the author and not necessarily the opinion of Credit Suisse. Folie 2
WHAT IS CREDIT PORTFOLIO MANAGEMENT? Main task Optimization of the risk/return profile of the bank s credit portfolio Main lever of implementation Consequent and up-to-date optimization of measurement and steering instruments at single-name and portfolio level Measurement of risks and performance Management of credit risk (buy & sell) Client and market orientation Super Senior CDS Update Reference Portfolio Originator Swap Premium CDS Protection Payments SPV Proceeds Notes Super Senior Swap AAA.. BBB Junior Swap Interest / Principal Proceeds Junior Swap Collateral Business-challenged Innovative & up-to-date Folie 3
CREDIT PORTFOLIO MANAGEMENT AT ALL LEVELS high Components Level of sophistication moderate Active steering of credit portfolio Portfolio risk Time dynamics of risk drivers Single-name risk Steering of the portfolio decomposition Capital allocation, limits, directives Transfer pricing mechanism Securitisation, hedging, risk transfer Measurement of the portfolio's diversification Time evolution of risk drivers Economic cycle adjustments Ratings & default probabilities Exposures, recovery quotes & LGDs Observation 1: Observation 2: Credit portfolio management plays at all levels of the pyramid. Most major banks try to improve their "status quo" at all levels. Folie 4
CHALLENGES: SOPHISTICATION & EFFICIENCY Improved risk parameter estimates PD LGD EAD Dependencies Returns Consequent Steering of Risk/Return Profile Relationship banking and competition Clear specification of bank's risk appetite Top-down steering and bottomup pricing (consistent) Increase market share Client relationship Pricing pressure Reduction of process costs Justifiable development costs Folie 5
AGENDA INTRODUCTION AND OVERVIEW EXAMPLE AT SINGLE-NAME LEVEL: THE WAY TOWARDS MORE COMPETITIVE RATINGS EXAMPLE AT PORTFOLIO LEVEL: THE WAY TOWARDS MORE MEANINGFUL VALUATIONS PORTFOLIO STEERING AND OPTIMISATION Disclaimer: This talk reflects the opinion of the author and not necessarily the opinion of Credit Suisse. Folie 6
OVERVIEW ON RATING SYSTEMS Standard Corporate Client Lending Rating / Scoring models relying on balance sheet data (financials) and a sound selection of qualitative criteria Specialized Lending Business Balance sheet based scoring models or causal models (modelling underlying risk drivers and the link to the default event) or "hybrids" Private / Retail Lending Business Application of classical scoring methods, based on the most often huge database of client-related data Structured Asset-backed Business Asset-backed securities, secured (depot) lending, collateralized debt obligations, etc. are evaluated by means of causal models Rating approaches depend on the considered client segment and business model. Folie 7
EXAMPLE 1: B/S-SCORING (1/4) Scoring-/rating system development/optimisation Score transformation 1 Score B/S / Financials Assets Liabilities Long list of ratios Structural coverage of: Ratio No. 1 Capital structure Ratio No. 1...... Liquidity Profitability Growth...... Ratio No. N... Score transformation N Score Selection Ratio mix is based on a multivariate optimisation of the system's Ratio No. N discriminatory power* S total = α + N k= 1 w k S k w-σ * Measured in terms of AUROC (area under the receiver operating characteristic) Folie 8
EXAMPLE 1: B/S-SCORING (2/4) Optimisation of discriminatory power is value-adding Optimisation of discriminatory power ("AUROC" increase) reduces the potential rejection of non-defaulters (optimisation of β-error) reduces the potential to approve defaulters (optimisation of α-error) has to be optimised in a "balanced way" - rejection of a non-defaulter generates a "loss" in size of [volume x margin] - default of a customer generates a loss in size of [net exposure x LGD] (illustrative!) Banks with high discriminatory power in their rating systems have a good chance to avoid adverse selection, enable a more competitive lending business in a competitive market, and have a significant advantage compared to banks with average rating/scoring systems Folie 9
Hit Rate [%] EXAMPLE 1: B/S-SCORING (3/4) - illustrative - Higher discriminatory power can dramatically change the PD distribution of a portfolio... Example: 100 obligors PD average = 100 bps Perfect rating ("crystal ball"): AUROC = 100% Realistic model: AUROC = 70-85% Random model: AUROC = 50% False Alarm Rate [%] Extrem case I Extrem case II Increase of discriminatory power when moving from a random model to a perfect model reduces 99 PDs and increases 1 PD Hit Rate [%] Hit Rate [%] Perfect rating ("crystal ball"): AUROC = 100% Realistic model: AUROC = 70-85% Random model Random model: AUROC = 50% False Alarm Rate [%] Perfect rating ("crystal ball"): AUROC = 100% Realistic model: AUROC = 70-85% Perfect model Random model: AUROC = 50% False Alarm Rate [%] Defaulter Defaulter PD distribution 100 PDs á 1% PD average = 1% PD contains no default information 1% 100% PD distribution: 1 PD á 100% 99 PDs á 0% PD average = 1% "Crystal Ball" PDs Folie 10
EXAMPLE 1: B/S-SCORING (4/4) Further aspects in the context of b/s scorings Calibration of default probabilities - Example: B/S Score Score Class Rating Class PD - Calibration "Score Class PD" relies on historic default experience Ratings "point-in-time" versus "through-the-cycle" - Short-term versus long-term loans should be treated differently - Basel II requirements have to be met (regulatory view) - Short-term loans: "point-in-time" aspects via pricing* Time evolution/dynamics of ratings/pds - Calibration of "PD term structures" relies on historic migration experience - Under-year PDs via "generator approaches" or PD interpolation Validation and backtesting - Basel II says: "A bank must demonstrate to its supervisor that the internal validation process enables it to asses the performance of internal rating and risk estimation systems consistently and meaningfully." (Capital Accord, 500) * e.g., via macroeconomic pricing adjustments Folie 11
EXAMPLE 2: CAUSAL RATINGS (1/4) Illustrative example: investment in a CDO tranche (e.g., AA-Tranche) Reference Portfolio Originator Swap Premium CDS Protection Payments Junior Swap Super Senior CDS Interest / Principal SPV Proceeds Notes Proceeds Super Senior Swap AAA.. BBB Junior Swap Amortization "top down" AA-Tranche Performance-linked to the reference portfolio Collateral Loss allocation "bottom up" Folie 12
EXAMPLE 2: CAUSAL RATINGS (2/4) Illustrative example: investment in a CDO tranche (e.g., AA-Tranche) Reference Portfolio Structural Def. Offering Circular Presale Reports Term Sheet AA-Tranche Target: Determination of internal rating for AA-Tranche Approach: Use of the causality: performance of reference pool performance of AA-Tranche Model: Performance at asset side determines performance at liability side AA-Trance is performancelinked to performance of overall reference portfolio CDO modell as causal rating system for tranches Folie 13
EXAMPLE 2: CAUSAL RATINGS (3/4) Further aspect in the context of causal rating systems Typical area of application - Structured credit products, "asset-backed", e.g., secured lending, CDOs, etc. - In general: credit products where the analyst can explicitely model the link between underlying risk drivers and the performance of the considered security Modelling efforts - Typically higher as in the case of standard scoring systems due to higher complexity - Modelling approach typically based on "scenario transformation" (see Slide 16) by means of Monte Carlo simulation Loss severity - Often the LGD of the underlying assets has a significant impact on the LGD of the asset-backed structured securities (e.g., CDO tranches) - The LGD of the security, e.g., for a CDO tranche can be determined via LGD Tranche = EL Tranche / PD Tranche ) Folie 14
EXAMPLE 2: CAUSAL RATINGS (4/4) Illustration of a causal rating model scheme for CDO investments Collateral Pool / Reference Portfolio Structured Notes / Tranched Securities Parameters: DP Distribution Exposure Distrib. Industry Distrib. Country Distrib. Recoveries Maturities Amortization Profiles etc. Structural Mapping Structure: Tranching Cash Flows Coverage Tests Triggers Overcollateralization Credit Enhancements Fees Maturity etc. Performance: Returns IRR, ROIC DP of tranches EL of tranches Loss on Principal Loss on Interest Excess Spread Maturity of Tranche etc. Example: distribution of the return of a CDO equity tranche (FLP) Probability Space Random Variable Distribution ( Ω,( ), P) X r 1 P o X r F t 25% 20% 15% Equity Return Distribution 10% underlying models: factor model, interest rate model, portfolio model, etc. market data: macroeconomic indices, interest rates, indices, etc. 5% 0% -10% -5% 0% 5% 10% 15% 20% 25% 30% Quelle: B / Overbeck / Wagner; An Introduction to Credit Risk Modeling; CRC Press, 2nd Reprint (2003) Folie 15
AGENDA INTRODUCTION AND OVERVIEW EXAMPLE AT SINGLE-NAME LEVEL: THE WAY TOWARDS MORE COMPETITIVE RATINGS EXAMPLE AT PORTFOLIO LEVEL: THE WAY TOWARDS MORE MEANINGFUL VALUATIONS PORTFOLIO STEERING AND OPTIMISATION Disclaimer: This talk reflects the opinion of the author and not necessarily the opinion of Credit Suisse. Folie 16
INTRO: WHAT MEANS DIVERSIFICATION? Two obligors (illustrative): PD 1 = 100bps, LGD 1 = 100% PD 2 = 50bps, LGD 2 = 100% Exposure weights according to w 1 = 0...1, w 2 = 1 - w 1 0.01 0.009 0.008 EL PF small correlation Calculation: Expected Loss (EL) Unexpected Loss (UL) 0.007 0.006 perfect correlation EL UL PF 2 PF = = w w 1 2 1 PD 1 1 + 2ρ w + w 1 2 PD (1 PD w 2 PD 1 2 ) + w 1 2 2 PD 2 PD (1 PD 0.005 0.06 0.07 UL PF 0.08 0.09 0.1 (1 PD 1 ) PD 2 2 ) + (1 PD 2 ) Plot of two-dimensional points (UL PF, EL PF ) for w 1 = 0...1 in the coordinate system above Assumption: Correlation of underlying driving ability-to-pay of the obligors equals ρ = 10% ( default correlation = 74bps) Diversification means reduction of the portfolio variance/ecap Folie 17
THE EARLY TIMES OF CREDIT PORTFOLIO MODELS First "off-the-shelf"-models in the market: Frequency CreditMetrics (RMG) ~ 1997 Portfolio Manager (Moody's KMV) ~ 1987* CreditRisk+ (CSFP) ~ 1997 Credit Portfolio View (McK) ~ 1997 Clear model target setting: Quantification of expected and unexpected loss of a credit portfolio Understanding the portfolio decomposition Loss [%] of Exposure EL = mean value of loss distribution E(R)C = [quantile - EL] or expected shortfall Expected Loss [ACP] Economic (Risk) Capital [ERC] Applications: Capital buffers, loss reserves, capital allocation, portfolio steering, limit setting, etc. * 1987 was the release date of seminal work by O. A. Vasicek; however, KMV Corporation has its hour of birth several years earlier than 1987... Folie 18
TODAY: TIME-DYNAMIC APPROACHES ARE COMMON Today credit risk modelling* is more "dynamic": Li (default baskets) ~ 2000 Finger (stochastic default rate models) ~ 2000 Schmidt & Ward (default baskets) ~ 2002 single-name τ Rating time (semi-annually) B & Overbeck (semi-analytic DTs) ~ 2003 Clear model target setting: Still quantification of expected and unexpected loss of a credit portfolio, but also Understanding of the timing of defaults Evaluation of structured credit products Financial engineering of credit-linked products portfolio view Recovery Single credit risks are combined by a "copula function" (Gaussian or Student-t or...) time (semi-annually) Implementation: [term structure of default probabilities] [one or more copula functions] * The four mentioned contributions are just a very small subset of the rich literature on credit risk modelling and default times of recent years. Folie 19
EXAMPLE: DEFAULT RISK ADJUSTMENT Portfolio loss is based on default time vector of assets Default time stochastic also impacts stochastic of cash flows Risk management: risk measures in the "classical" sense Pricing, structuring: dt-adjusted stochastic cash flows Default time stochastic drives both approaches in a consistent way Folie 20
AGENDA INTRODUCTION AND OVERVIEW EXAMPLE AT SINGLE-NAME LEVEL: THE WAY TOWARDS MORE COMPETITIVE RATINGS EXAMPLE AT PORTFOLIO LEVEL: THE WAY TOWARDS MORE MEANINGFUL VALUATIONS PORTFOLIO STEERING AND OPTIMISATION Disclaimer: This talk reflects the opinion of the author and not necessarily the opinion of Credit Suisse. Folie 21
STEERING OF THE CREDIT PORTFOLIO Source: B and Mussil; Basket Kreditderivate und Collateralized Debt Obligations als Instrumente des Portfoliomanagements; in: Handbuch Kreditderivate, zweite Auflage, edited by Burghof et al.; Schäffer-Poeschel Verlag Stuttgart (2005) Folie 22
EXAMPLE: SYNTHETIC TRANSFER PRICING Simplified scheme (illustrative) Definition of transfer price: TP has to cover EL and ECC* TP should provide flexibility to CPM Market / Clients Loan... LoC Bank Origination Central unit / units, e.g., front office CPM, ALM for either booking the risk or selling/hedging the risk STP incorporates the option to give more freedom to origination units, but TP always has to be paid, e.g. Pricing Fair price Margin- Decrease Achieved price Margin Front office benefit Margin Margin Risik premium Transfer credit risk Credit portfolio mgmt. Risk premium Risk premium Unit cost Administration Unit cost Unit cost Refinancing Transfer interest risk Asset liability mgmt. Refinancing Refinancing * Economic Capital Costs (ECC) Folie 23
VALUE LEVERS: WRAP-UP At single-name level - Optimisation of discriminatory power of rating/scoring systems - Migration from scoring to causal models where possible (alternative: hybrid models) - Sound calibration of EAD and LGD (reflecting the bank's workout process) - Reasonable incorporation of the time dynamics of risk drivers At portfolio level - Regular re-calibration of all involved parameters - Economic hedging of "tail events" via CSO or default basket - Consequent economic capital based steering of the portfolio's risk/return profile At bank level - Consequent risk/return steering (top-down) and overall pricing (bottom-up) - Measurement and management of the overall customer profitability (no simple cross selling argument without "proof") - Regular adaption of the bank's risk appetite (e.g., "cycle adjustments") - Clean data warehousing and historisation of all risk quantities Folie 24