Presentation on theme: "Developing Macro-stress tests"— Presentation transcript:
1Developing Macro-stress tests Session 9Mindaugas Leika
2Macroprudential policy framework I. Macroprudential policy definition, targets, policy transmission channels and relationships with other policies (Monday) II. Institutional structure (Tuesday) III. Policy tools (Tuesday) IV. Risk identification and quantification: stress testing (This lecture)
3AGENDAWhat is macro stress testing? Macro stress testing framework Macro ST process Use of stress tests Did STs fail?
5How can we group risks: Usually arises gradually Arise instantly A) Credit risk B) Market risk C) Liquidity risk D) Operational risk (e.g. failure of SWIFT, software etc. BoE quantifies this)Arise instantly
6What is the purpose of macro stress testing? Provide quantitative and forward looking assessment of the capital adequacy of the banking system*.Accountability to the publicDecision-making supportMeasuring the impact of systemic riskMacro prudential policy tool: addresses banking system vulnerabilities (capital buffers, exposures to particular sector, absence of diversification, capital planning, investor confidence etc.)*Source: Bank of England (2013). A framework for stress testing the UK banking system.
7Why macro stress testing: Two concepts of losses There are two concepts linked to risk mitigation techniques: I Expected losses (loan loss provisions, loan impairment charges); II Unexpected losses (economic capital). Expected losses are mean loss rate, i.e. amount that bank reasonably expects to lose. Expected losses are usually covered by loan loss provisions or loan impairment charges. It is called known part of losses. Unexpected losses represent volatility of losses, i.e. unknown part. Shareholders equity is used to absorb these losses. We have to presume, that banks not only need capital to absorb these losses, but also have to stay above minimum regulatory capital requirements through the full business cycle. The targeted level of unexpected losses depends on two factors: minimum capital requirements and targeted rating.
8Example IExample: Bank has 100 bn in assets with 100% risk weight. Average interest rate is 8%. Shareholders equity is 8 bn (CAR 8%). Liabilities (deposits and bonds) is 92 bn. Average return on liabilities is 4%. At the end of the year bank expects to receive 8 bn in return from its assets and pay to debt holders 3,68 bn. Return to shareholders is 4,32 bn. However, if bank’s expected losses are 3% (PDs), return to shareholders is lower (return to debt holders is fixed!): 4,32-3=1,32 bn. What happens, if during an economic downturn, PDs increase up to 8% ? Bank’s losses are much higher than expected and equal to 8 bn. In this case, bank’s income is equal to its losses: 8bn- 8bn=0. Bank’s payments to debt holders is fixed, hence bank needs to tap its capital base to pay interest rate: 8-3,68=4,32 bn capital left. That’s below minimum CAR of 8%. Bank needs to be closed or recapitalized.
9Example IIHow much additional capital bank needs to hold? Bank provisions 3% for expected losses and needs additional reserves of 5 bn just to have zero profit. In this case return to its shareholders is 0. To come up with the worst case scenario, and calculate additional reserves, we need to perform a stress test and model expected and unexpected losses. Under Basel II IRB approach we have to model PDs, LGDs. EADs are given. Losses are expressed as: Expected losses=PDs x LGDs x EADs Under Basel II STD approach non IFRS and Basel I we model loan loss provisions (LLPs): ∆LLPs=∆NPLs x provisioning rate Unexpected losses can be measured as a number of standard deviations from expected losses (VaR concept).
10From shocks to their outcome: how transmission mechanism works Credit riskNon- financial corporate sectorTransmission channels: exposureShocks originating in Real sectorReal estate sectorHouseholdsPublic sectorFinancial institutions:profit/loss, capitalShocks originatingin Financial sectorBanksInsurance companiesOther institutionsPayment systemsTransmission channels: common ownership, exposure etc.Feedback effectsMarket, liquidity, counterparty, contagion risks
11Macro Credit risk Stress testing model Real estate prices: collateral value for LGD calculationMacroeconomic forecastsCB’s macro modelGDP, Housing prices, interest rates, FX rate, unemploymentBanking sector dataNPLs, provisions, credit growthShort-term equations with AR(1) terms and/or ECM: NPLs depencence on selected macro variables calculated for 1 to 4 quartersLong-term equations:NPLs depencence on selected macro variables calculated up to 3 yearsEquations on a bank-by-bank basisInterest income, expenses, credit growth, doeposits, interest rate etc.FX risk, concentration risk, income/expense, duration gap modelsLoan migration matrixProjected additional provisionsOn a bank-by-bank basisProjected net losses/profitCARUnexpected lossesMonte-carlo simulationNumber of banks that do not meet minimum CAR
13Three types of models for macro stress testing I Portfolio models (Credit Risk plus; Risk Metrics; Credit Portfolio View etc.) II Balance sheet models (Cihak, Boss et all. and modifications). III Market data based models (CCA). First type of models dominate in private sector, second and third type dominate in regulatory institutions.
15Loss calculation and mapping Macro ST processRisk identificationMain risks are identified and scenarios constructed.III types of scenarios:I economic/industry downturnII market risk eventsIII liquidity crisisLoss calculation and mappingLosses are calculated using either sensitivity or scenario analysis or both approachesLosses in terms of CAR are presentedActionsIncrease capital buffersReduce RWAReduce exposureReview concentration limits etc.
16Understanding the incentives There are at least three stakeholders in the stress-testing process: financial institutions, regulators and the public/markets.Usually they have different incentives: regulators want more data, more time, more extreme scenarios; financial institutions want to provide less data, use in-house models, usually less extreme scenarios. Regulators want to find the weakest components of the banking system, whereas institutions want to show resilience. Public wants “blood”- know institutions that fail the test.
17Macro Stress testing steps Determine the objective of the stress test Design scenario Perform stress test Calculate stress losses Report results Determine actions
18Basel II/III Minimum capital Supervisory review Market discipline Pillar IMinimum capitalMinimum capital requirement; point in time assessmentPillar IISupervisory reviewIndividual capital guidance: Pilar I risks+additional (bank specific) risksStress testsPillar IIIMarket disciplineICAAP: Internal capital adequacy assessment process;Calculation of economic capital
19scenario design (1) Shock matrix High probability Low probability High impactMain focusWatchLow impactSummary mentionIgnore
20Risk mapping: from systemic risks to exogenous shocks scenario design (2)Risk mapping: from systemic risks to exogenous shocksShock calibrationScenario designRisk correlationMacroeconomic modelsBaseline scenarioScenario output: macro and financial variablesAdverse scenario
21Historical experience Credit risk depends on the state of economy (business cycle)
22Defining thresholds Expected (mean losses) P(X) E[x]=μ(x)P(X)LossLoan loss provisionsEconomic capitalConfidence interval is identical to default probability:BBB0,1%A0,07%AA0,03%AAA0,01%Expected shortfallPass/Fail criteria and minimumcapital requirements
23Applying shocks: Normal and shocked PDF P(X)X2 represents shocked PDs, and as it has higher variance, it bears more risk than X1X1X2Loan loss provisionsEconomic capitalLossExpected shortfallLoan loss provisionsEconomic capital
24Losses and business cycle X1=f(GDP↑, FX rate, Unemployment↓, Interest rates↓, Concentration↓ etc.) X2=f(GDP↓, FX rate, Unemployment↑, Interest rates↑, Concentration↑ etc.)P(X)X1 represents upward trendX2 represents downward trendX1X2Loan loss provisionsEconomic capitalExpected shortfallLossEconomic capitalLoan loss provisions
26Calculating CARLoan loss provisions; Forecasted from satellite credit loss modelNet income before loan loss provisions; Forecasted from satellite income modelCurrent Tier I and II capital (regulatory capital)Current RWA for: credit, market and operational risksSatellite credit growth modelLoan loss provisions; Forecasted from satellite credit loss modelMigration matrices
28Theoretical use of stress tests What answers stress tests should provide:How much capital a bank needs to support its risk taking activities? (Forward looking)Is the current level of capital adequate? (Present)Lehman Brothers, Bear Stearns, Dexia, JP Morgan…. Did they do it right?Capital that is available vs. Capital that is needed vs. Capital that regulators need.
29Actual use of stress tests during the crisis Stress tests popped out as a tool to address loss in public confidenceConfidence was boosted by disclosing individual banks’ results, scenarios and data about exposuresSCAP (US) vs. EBA (EU).
30STs before and after crisis Very little public disclosure, usually : “All banks are adequately capitalized, however challenges remain, thus we will be vigilant”Comprehensive analysis, data available on a bank by bank basis. Not all banks pass tests, capital shortfalls are publicStatic analysisDynamic analysisUsually single shocks, VaR basedMacro based, multiple scenarios, dependency among various risk factors, CoVaRIn most cases solo, individual entity basedConsolidated at the parent (group) levelSimple models, usually for credit and market risk separatelyComprehensive models: credit, market, liquidity risks and lost incomeNot necessarily linked to CARLinked to CARNo macroprudential measures or capital conservation plansMacroprudential measures (system wide) and capital conservation plans (individual)
31Can Stress tests detect systemic risks? In theory, macroprudential STs should unveil the sources of systemic risk (see IMF (2012)In practice, sources indeed were identified correctly (e.g housing market in the US, contagion from Greece in the EU etc).Magnitude of shocks and subsequently their impact was miscalculated
33Why stress tests can fail? (1) We can find many “wrongs”:Wrong models: too complexWrong (absence of) data: where risks were “parked”?Wrong scenarios: underestimation of tail risk events and contagion effectsWrong incentives: no need to rock the boat, public will not understandWrong scale: “shadow institutions” escapedWrong policy measures
34We calculate economic capital using 2 or three standard deviations….. If we use a normal distribution, two standard deviations from the mean means we still have almost 5 percent of observations outside of our horizon (2,5 percent in each tail). This means, we overestimate earnings and underestimate losses.
35However during this Global Financial Crisis volatility was much higher….. In August 2007, the Chief Financial Officer of Goldman Sachs, David Viniar, commented to the Financial Times: “We are seeing things that were 25-standard deviation moves, several days in a row”. As Andrew Haldane, executive director at the bank of England noticed: “Assuming a normal distribution, a 7.26-sigma daily loss would be expected to occur once every 13.7 billion or so years. That is roughly the estimated age of the universe. A 25-sigma event would be expected to occur once every 6 x lives of the universe.”
36Volatility depends on time period (sample from Haldane (2009))
37A Hypothetical example of a factor whose relationship to default is not clear until a crisis pushes it to new levelsSource: S&P (2010)
38Bimodal nature of rating transitions Source: Moody’s (2013) Stress Testing of Credit Migration. A Macroeconomic approach.
39Why stress tests can fail? (2) Underestimated probability of adverse outcomes (disaster myopia)Reluctance to include severe scenariosWillingness to hold capital under less extreme scenario onlyPostponement of crisisReverse engineering: scenarios are such, that bank never violates minimum CARShort time series in emerging market countriesData quality issuesWe usually never look beyond economic capital!Historical scenarios are based on historical data. We can not test anything new using data from the past only
40Reduced form vs. full-scale stress tests Most of the stress tests banks do are reduced form stress testsReduced form – Monte Carlo simulationFull scale – links with macro variables. CorrelationsWhy reduced form? If probabilities are unknown, we face uncertainty. In this case randomization is an answer.
41Why reduced form is not suitable for macro STs Reduced form does depend on assumptions about distribution. Beta distribution has fatter tails than the normal oneReduced form ST has very little connection with macro variablesIs opaqueIs good, once we deal with random, uncorrelated price movements or volatilities. Is not suitable, once we deal with systemic events or highly correlated movements
42The way aheadHow to incorporate balance sheet adjustments into core models? How to avoid modeling partial equilibrium situations only, i.e. include feedback effects and adjustments in broader sectors of economy? How to model nonlinearities? System’s stability is most vulnerable then nobody anticipates shocks, i.e. risks are underpriced, real estate prices are at their peaks, GDP and credit grows fast. How to model aggressive risk taking and rapid build-up of imbalances? How to model financial innovations and market liberalization (historical data are not available at all or structural breaks emerge)? How to extend stress tests to other (non-bank) financial institutions?
43Importance for macroprudential policy Based on Borio, Drehmann and Tstatsaronis (2012) objective of the stress tests is to support crisis management and resolution. Drilling down we can formulate this objective more precisely: a) calculation of how much capital should be injected into the system to prevent credit crunch; b) Identification of weakest financial institutions; c) signaling to the market about losses and restoring confidence in the banking system; d) Improve risk management practices, models and data collection; e) Story telling: use stress tests to describe shocks, transmission channels and possible impact on financial system and broader economy.
44Seven best practice principles proposed by the IMF Define appropriately the institutional perimeter for the tests.Identify all relevant channels of risk propagation.Include all material risks and buffers.Make use of the investors’ viewpoint in the design of stress tests.Focus on tail risks.When communicating stress test results, speak smarter, not just louder.Beware of the “black swan.”Source: Macrofinancial Stress Testing—Principles and Practices (2012)
45What to do?Do not constrain yourselves with historical experience and scenarios (it is not contrary to the “this time is different” syndrome, i.e. one should think that worst crisis might repeat again or my country is not necessarily too much different from the ones that experienced crisis earlier)Use judgmental adjustments in scenariosUse reverse stress testing more often to find break-even points (especially important in liquidity stress-testing)Make it simple. Last CCAR (2013) emphasized simplicity. Simplicity means no complicated “black box”: executives should be able to understand and supervisors to verifyIn the end, follow the advise by J.M. Keynes: “It is better to be roughly right than precisely wrong”