Stress Testing* Mariano Cortés IMF Monetary and Capital Markets Department World Bank/International Monetary Fund/Federal Reserve System Seminar for Senior Bank Supervisors from Emerging Economies October 20, 2009 *This presentation draws on previous presentations by A. Tieman, M. Segoviano, and L. Schumacher. 1 Agenda I. Stress testing in context financial stability analysis II. Basics of stress testing III. Credit risk IV. Other risks V. Concluding remarks 2
Agenda I. Stress testing in context financial stability analysis 3 What Is Financial Stability? A supportive environment in which the financial system facilitates a relatively efficient allocation of resources (positive) Allowing households to engage in optimal consumption and savings decisions Promoting investment in most-productive activities Redistributing risks efficiently across households and business enterprises Financial stability is the absence of financial instability (negative) Market failure or externalities in financial system Substantial enough to threaten aggregate economic activity Consequences: Undesirable outcomes of market imperfections: excess leverage, inadequate risk management, financial panics Accompanied by long-lasting distortions 4
Causes Of Financial Instability Inherent vulnerabilities associated with maturity transformation and banking system vulnerabilities (low capital, credit risk and assets quality problems, market risks, liquidity risk) Shortcomings in regulation and supervision (forbearance, non riskbased supervision, gaps complex groups, non-banks) Financial infrastructure: safety net (risk mitigation), payment and settlement systems (large exposures and contagion), markets (money and T-bills market seizure), legal (property and creditors rights, collateral) Liberalization and globalization: poorly sequenced and weak supervision, increased competition and sudden capital flows reversals 5 Consequences Of Financial Instability Banking crises are common (map), unpredictable (e.g., non-economic triggers, contagion), and surface unrecognized or underestimated vulnerabilities Monetary policy implementation is made more difficult bank responses to interest rate policy changes are less predictable, concerns over the health of banks may limit the scope for policy action Negative fiscal consequences potentially large build-up of debt to support/resolve banks and to recapitalize the central bank build-up of contingent liabilities in the form of guarantees of deposits and credits Real economy consequences Inefficient savings intermediation Weaker corporate governance More pronounced business cycles 6
Banking Problems Worldwide 1980-2002 Banking Crisis Significant Banking Problems No Significant Banking Problems/Insufficient Information 7 Fiscal impact of crises (percent of GDP) Indonesia 1997-present Chile 1981-1983 Thailand 1997-present Turkey 2000-present Korea 1997-present Ecuador 1998-2001 Mexico 1994-1995 Venezuela 1994-1995 Finland 1991-1993 Malaysia 1997-2001 Sweden 1991-1993 United States 1984-1991 Norway 1987-1989 Gross Cost Net Cost 0 10 20 30 40 50 60 Sources: IFS, WEO and national authorities. 8
Assessing financial stability To some extent, market imperfections are unavoidable and pervasive, and therefore financial stability can never be absolute. This in turn results in a series of questions: Can we measure the degree of financial stability? Can we assess when the economy is at the greatest risk of financial instability? How well and how far ahead can risks of instability be predicted? 9 Dimensions of financial stability Macroeconomic Conditions Monetary policy Debt structure Exchange regime Economic growth Regulatory and Supervisory Conditions Regulatory framework Supervisory efficacy Contingency planning and management Market and Infrastructure Conditions Money, FX, and securities markets Payment and settlement systems Accounting and auditing Legal framework 10
A framework for financial stability Assessment analysisof Macro- Prudential Analysis Structure Soundness Vulnerabilities Observance of Standards and Codes Peer review Review of Financial Sector Infrastructure and Governance Framework 1. Overall assessment of financial stability 2. Development needs assessment 3. Formulation of a coordinated reform program 11 Agenda II. Basics of Stress Testing 12
What is stress testing (1)? Stress testing is a range of techniques used to assess the vulnerability of the financial system to exceptional but plausible shocks. Stress tests impose a coherent structure in which to discuss risks and can add rigor to systemic analyses. Stress test were originally developed for use at the level of portfolios and for individual institutions. 13 Stress tests attempt to What is stress testing (2)? combine a forward-looking macroeconomic perspective with an assessment of the sensitivity of financial institutions to major shocks in the economic and financial environment. The system-wide nature does not imply that the tests should be performed only on aggregate data Aggregate data can disguise substantial exposures and risk concentrations at the institutional level that would be netted out through aggregation It is therefore important to perform stress tests on an institution-byinstitution basis as much as possible 14
Stress tests should be tailored to: What is stress testing (3)? country-specific circumstances complexity of the financial system data availability. Stress testing balance-sheet positions only can be misleading because: off-balance-sheet positions can quantitatively and qualitatively alter onbalance-sheet exposures. it may not be clear where market and credit risks ultimately reside credit risk derivatives or contingent liabilities. off-balance-sheet funding vehicles (conduits and SIVs) can also be sources of vulnerabilities. 15 What is stress testing (4)? Three main types of stress tests: Single-factor sensitivity analysis to identify how portfolios respond to changes (one at a time) in relevant economic variables such as interest rates, exchange rates, and equity prices; Scenario analysis (model-based simultaneous moves in a group of risk factors) to assess the resilience of financial institutions and the financial system to exceptional events; Contagion analysis to take account of the transmission of shocks from individual institutions to the financial system as a whole. 16
Stress Testing Five steps 1) Identification of major risks and exposures 2) Define coverage Institutions--all systemically relevant ones; Risks credit, market, liquidity, interbank contagion, cross border contagion 3) Calibrate the shocks and scenarios Exceptional but plausible tail events Movements in assets prices, volatilities, correlations 4) Implement the methodology (e.g., models) and run the tests 5) Interpretation of results More useful if done regularly changes in risk Absolute level of risk could be highly uncertain and the model misspecified 17 Agenda III. Credit Risk 18
Types of Stress Testing Single Factor: Credit Risk Credit risk losses due to counterparty default or changes in credit quality (e.g., rating) Building blocks of credit-risk modeling (obligor i) Default D i, D=0 (no default) or D=1 (default); Exposure at default EAD i; Loss given default LGD i ; Recovery rate: 1-LGD i Probability that obligor i will default within a given horizon PDi The credit loss distribution Loss=Default arrival EAD LGD Portfolio loss at a given horizon is sum of individual losses Loss = I I Loss = = 1 i= i i Di EADi LGDi 1 The portfolio credit loss distribution depends critically on the dependency across individual losses 19 Types of Stress Testing (continued) Single Factor: Credit Risk Expected and Unexpected Losses-- Credit Loss Distribution Function For an individual obligor i, the expected loss is: EL i = PDi EADi LGD i Probability At the portfolio level, the expected loss is the sum of individual expected losses: I EL = = EL i 1 i Unexpected loss (UEL) is usually defined relative to an extreme percentile of the loss distribution i.e., Mean 99 percentile UEL = Loss 99% EL Expected losses Provisions Unexpected losses Economic Capital Losses 20
Types of Stress Testing (continued) Single Factor: Credit Risk Popular portfolio models of credit risk include: Creditmetrics (J.P.Morgan); CreditPortfolioView (McKinsey); KMV (KMV Corp.); CreditRisk+ (Credit Suisse) In the context of credit risk models, stress tests entail shocking the model s parameters, with the shocks linked to macroeconomic factors using judgment/and or models Using a modified transition matrix Using stressed PDs (Examples FSAPs for Austria and Spain ) Increases in LGDs Increases in asset correlations Shock the entire multivariate distribution (Example FSAP for Denmark: application of Segoviano and Padilla methodology) 21 Types of Stress Testing Single Factor: Credit Risk Austria FSAP Update (2008)(1)*: estimate PD model (10 sectors; data: quarterly 1969-2007) The average PD is modeled as a logistic function of a sector specific macroeconomic index : 1 ps, t = y s, t 1+ e where, p s,t : average default probability for sector s at time t y s,t : macroeconomic index for sector s at time t, which depends on the values of the macroeconomic variables under consideration *See http://www.imf.org/external/pubs/ft/scr/2008/cr08204.pdf Spain FSAP (2006)*: estimate NPL model (4 credit categories-- household mortgages; real estate and construction; other sectors; consumer and other credit; data: annual credit institution level, 1992-2004. Logit-transformed NPLs are regressed on lagged NPLs and a set of macroeconomic factors: * See http://www.imf.org/external/pubs/ft/scr/2006/cr06216.pdf 22
Types of Stress Testing Single Factor: Credit Risk--Austria FSAP Update (2008) (2) for each sector the following regression model is estimated: y s, t = β 0, s + β1, s x1, s, t + β 2, s x2, s, t + K + β K, s xk, s, t + ε s, t where y = y y s, t s, t s, t 1 is the change of the macroeconomic index of sector s. The original model was altered to model the change in the macroeconomic index because the index itself contains a unit root no reasonable model for five sectors, for which a model based on the aggregated PD of the Austrian economy was applied. 23 remaining models contained 2-4 macro variables from the following set: Types of Stress Testing Single Factor: Credit Risk--Austria FSAP Update (2008) (3) Macroeconomic Variables included in the Model-Selection Cyclical Indicators Interest rate indicators Gross domestic product (GDP) Nominal short-term interest rate Industrial Production International (IPI) Real short-term interest rate Dom. Industrial production less energy (IPexE) Nominal 5y term interest rate Price stability indicators Real 5y term interest rate Consumer price Index (CPI) Nominal 10y term interest rate Inflation rate (InfR) Real 10y term interest rate Household Indicators External variables Unemployment Rate (seasonal adjusted) (Uersa) Exports Private Consumption (privcon) Export/GDP Corporate Indicators Imports Gross fixed capital formation (Gfctot) FX-Rate USD Gross fixed capital equipment(gfceq) FX-Rate CHF Gfctot/GDP FX-Rate JPY Ifo business climate index FX-Rate GBP Equity Oil price (Brent Crude) in EUR Dow Jones Index Austria FSAP Update Technical Note on Stress Testing and Short-term Vulnerabilities 24
Types of Stress Testing (continued) Single Factor: Credit Risk -- Structural Approach 0.2 0.15 Marginal Distribution Asset Type A 0.1 0.05 0 4 2 0-2 -4-4 -2 0 2 4 Portfolio Multivariate Distribution PLD Marginal Distribution Asset Type B PORTFOLIO DIVERSIFICATION EFFECTS Joint likelihood of changes in credit quality Dependence structure Individual likelihood of changes in credit quality 25 Types of Stress Testing (continued) Single Factor: Credit Risk Problems with the structural approach The estimation of the distributional parameters hinges critically on the availability of data on the evolution of the firms underlying asset values and their correlations (i.e. the firms joint credit risk quality), which is frequently not available. Misspecification consequences: incorrect risk assessments undermining profitability and ultimately viability of the bank; adverse systemic consequences Information restrictions arise due to: Arm s-length relationships with the firms: IMF economists and many regulators. Data not publicly available: SME s and unlisted firms. In most of these cases, it is only possible to get short time series of frequencies of default (e.g. PDs or NPLs) of loans grouped by economic sector or ratings. 26
0. 2 0. 1 5 0. 1 0. 0 5 0 4 2 0-2 -4-4 -2 0 2 4 Types of Stress Testing (continued) Single Factor: Credit Risk Non parametric approach: Segoviano & Padilla Schematic Representation Frequencies Default Macro/Financial variables Step 1 Forecast of PoDs f (Macro/Fin Variables) Step 2 Aggregated Exposure Step 3 Multivariate Density LGD Economic Capital (VaR) Simulation 27 Types of Stress Testing Scenario Analysis Plausible, internally consistent, and challenging macroeconomic scenarios to assess the impact of tail events on individual banks portfolio positions and the adequacy of capital to absorb losses Could be constructed based on historical events Sensitivity tests--replicate the largest change in a risk factor (e.g. interest rates) say in the last 10 years; Extreme historical scenarios--a recent financial crisis; but historical events-based scenarios have problems; Plausibility--the financial system and instruments may have substantially changed Or on hypothetical events may be more realistic and flexible, particularly if the financial structure has changed considerably; encourage risk managers to be forward-looking. 28
Types of Stress Testing Scenario Analysis (concluded) Two approaches for running the stress tests who does it? the bottom-up approach each selected institution Key advantages: runs on internal risk management systems using detailed information on exposures and portfolios; But no two banks employ the exact same measurement approach even if their methodologies are broadly the same the top-down approach--the supervisory agency or the central bank With the model selected, the effects of macro variables on banks provisions, earnings, and balance sheets are estimated using supervisory data; Can also be used as a check on the results of a bottom-up exercise; 29 Agenda IV. Other risks 30
What is market risk? It is the risk that the value of an investment will decline due to move in market factors. The standard market risk factors are: Interest rate risk Foreign exchange rate risk Assets price risks Liquidity risk Derivatives risk Commodity risk 31 Liquidity Risk Liquidity risk is currently receiving much attention in light of the recent crisis. Liquidity stress testing is mostly done as single factor analysis, but could also be done as scenario analysis. 32
Liquidity Risk What happens if funding dries up (funding liquidity); and/or if assets thought to be liquid become illiquid (market liquidity)? Retail funding via deposits Wholesale funding on financial markets (such as CP market) Contingent liabilities (including through back-up lines of credit for off-balance-sheet vehicles; or for purely reputational risk considerations) Illiquidity of liquid assets (such as CDO tranches). 33 Interbank Contagion Tests Two types of tests: Pure interbank stress tests these tests explore the effects of the failure of one bank on the other banks in the system through interbank exposures; Integrated macro interbank stress tests in these tests the initial shocks are the macro shocks of the scenario which may (or may not) lead to bank failures, which in turn may affect other banks that are already weakened by the external shock. The key input for interbank contagion tests is a matrix of bilateral interbank exposures. 34
Agenda--Concluding remarks Presenting stress tests results: What would be losses associated with a change in risk factor/scenario X and the impact on profitability and capital solvency? What are the worst case scenarios or parameters configurations? What could the fiscal impact be if banks need to be recapitalized? What actions can be undertaken to limit the losses associated with the worst case scenario? 35 Thank you. 36