Testing resilience of the financial system
2 Úvod relativně specifická „technologie“ risk managementu využívaná jak ve finančních institucích, tak v centrálních bankám a dohledových autoritách cílem je zhodnotit odolnost sektoru, identifikovat slabá místa a případně naznačit rozsah potřebného zvýšení kapitálu a komunikovat to veřejnosti
3 Podstata techniky využití modelu vývoje bilance finanční instituce nebo sektoru finančních institucí za účelem kvantifikace dopadu extrémních, leč možných ekonomických podmínek v blízké budoucnosti Testovaná rizika: úvěrové, tržní, riziko poklesu výnosů, riziko nákazy mezi institucemi, riziko likvidity (balance sheet liquidity) či riziko dostatku zdrojů pro financování aktiv (funding liquidity) Zátěžové testy jako nástroj hodnocení odolnosti
4 Top-down přístup (makrozátěžové testy) regulátor/centrální banka vlastní model chování institucí, vlastní (v centrální bance dostupná) data analýza dopadu určitých scénářů na kvalitu portfolií aktiv, příjmy bank a jejich kapitál/solvenci agregátní portfolia aktiv bez znalosti detailních charakteristik jednotlivých aktiv využití makroscénářů, ideálně se zachycením některých dalších feedback efektů (interakce banky versus reálná ekonomika) Zátěžové testy jako nástroj hodnocení odolnosti
5 5 Zátěžové testování a finanční stabilita Rizika v ekonomice -pokles HDP -znehodnocení domácí měny -růst úrokových sazeb -pokles cen nemovitostí Zpětná vazba/feedback effect – NENÍ V TESTECH - ZÁMĚR Zátěžové scénáře makroekonomického vývoje Rizika v bance -úvěrové -tržní -likviditní -zdroje příjmů -mezibankovní nákaza Dopad na ekonomiku (dodatečný pokles HDP apod.) Reakce banky: snížení úvěrování Dopad do bilance banky (výsledné zisky, kapitálová přiměřenost apod.)
6 Bottom-up přístup (individuální testy) centrální banka/regulátor každé bance dodá scénáře jednotlivé banky pak pomocí vlastních modelů a analýz odhadují jejich dopad do výkazu zisků a ztrát a do poměrů solventnosti testují se jednotlivá aktiva s využitím jejich detailních charakteristik (nesplacené objemy, kolaterál, opravné položky, rating pro jednotlivé úvěry) často pouze citlivostní analýza (např. zvýšení hodnoty jednoho faktoru – většinou PD, probability of default, i když může být testováno více faktorů) dobrými interními modely disponují převážně velké banky Zátěžové testy jako nástroj hodnocení odolnosti
7 u obou typů přístupů (top-down in bottom-up) testovány bilance jednotlivých institucí výsledky mohou být agregovány ČNB jako centrální banka a integrovaný regulátor finančního trhu využívá obou přístupů top-down: agregátní zátěžové testy (interně v ČNB na datech reportovaných do ČNB, prováděny od roku 2003) bottom-up: společné zátěžové testy ČNB a vybraných bank (prováděny od roku 2009) Zátěžové testy jako nástroj hodnocení odolnosti
8 for assessing resilience of the financial system the CNB conducts regular stress tests of banks since 2003 three (slightly overlapping) stages in development of the stress testing framework for banks simple static stress testing/sensitivity analysis ( ) static stress testing based on (consistent) macroeconomic scenarios, satellite models and interbank contagion ( ) dynamic model-based stress testing (2009++) in paralel, the CNB develops since 2008 a liquidity stress testing model for banks that is going to be integrated with the main stress testing framework since 2007, the CNB conducts stress tests of insurance companies (market risk, insurance-specific risks) and pension funds (market risk) Stress testing in the CNB
9 Static stress testing based on (consistent) macroeconomic scenarios, satellite models and interbank contagion ( ) FSR 2005, FSR 2006, FSR 2007, FSR 2008/2009 Stress testing in the CNB
10 The framework QPM model (or since late 2008 G3 DSGE model) generates both baseline forecast (the official CNB forecast produced quarterly) as well as alternative „adverse“ macroeconomic scenarios satellite models are credit growth model (ECM model of aggregated credit growth) and credit risk models (corporate, households)
11 Transmission Channels of Credit Risk dependent variable of credit risk models: 12M default rate (i.e. new bad loans over initial portfolio) 12M default rate is also used by commercial banks; the Basel II „PD“ used for IRB approach in credit risk should be „a long-run average of default rates“ model and explanatory variables Corporate Sector Merton model Macroeconomic shocks (explanatory variables); GDP growth, exchange rate, inflation, debt Households Merton model + naive econometric models Unemployment rate, real interest rates, GDP
12 Credit Risk Modeling Macroeconomic credit risk model for the Czech and Germany corporates were estimated (Jakub í k and Schmieder 2008) Czech: German:
13 Credit Risk Modeling Macroeconomic credit risk model for the Czech and German households were estimated (Jakubík and Schmieder 2008) Households models: less successful than for corporates, additional (socio-economic) indicators may improve modelling
14 possible to construct scenarios without a macroeconomic model, but to achieve the highest possible consistency, using a macro model (QPM, DSGE, VAR) is of advantage scenarios should be of a type „low probability – high impact“, but plausible and have some „story“ behind should react to risks identified in risk assessment; in case of double- sided risk, opposite scenarios can be built (e.g. appreciation/depreciation, increase/decrease in interest rates) the story can be reflected in the name of the scenario (makes it easier to remember); „sexy“ names are of advantage use baseline scenario (official forecast) as benchmark; however, problems with interpreting the results if the stress testing model/models calibrated conservatively Scenario building
15 Example FSR 2007: stress test scenarios Three alternative model-consistent scenarios in FSR 2007 (scenarios for the year 2008) A - safe haven (appreciation of currency) B - property market crisis (internal shock with direct impact on banks) C - loss of confidence (external shock – increase in risk aversion)
16 FSR 2007: Impact of Alternative Scenarios on the Banking Sector Example of presentation of the results The results were interpreted as follows: The banking sector seems to be resilient to a wide range of risks Only an extreme macroeconomic scenario would necessitate capital injections to maintain sufficient capitalization
17 Example FSR 2008/2009: macroeconomic scenarios Three scenarios reflecting the risks from the global financial crisis Europe in recession (= baseline prediction) Nervousness of the markets (a la „loss of confidence“, i.e. increase in risk aversion) Economic depression (very large decline in GDP)
18 FSR 2008/2009: presentation of the results same style of presentation information about the capital injections needed
19 Dynamic model-based stress testing (2009++) FSR 2008/2009; FSR 2009/2010 Stress testing in the CNB
20 The framework limited as regards its ability to capture the effects of credit, interest and currency shocks over time in a more dynamic way, analyze the impact of shocks in a longer horizon than a one-year horizon (up to two to three years), estimate the pre-provision income as a function of both the macroeconomic development and a bank’s business model, be expressed in the variables used in current regulatory framework (PD, LGD) and thus mimick the stress testing done by individual banks within Pillar II of Basel II integrate the funding liquidity shock within the macroeconomic stress testing framework, Problems with static stress testing
21 difference in time horizon between the effects of market and credit risks impact of a change in interest rates or other market variables (the exchange rate or stock prices) on the balance sheets of financial institutions is virtually immediate (revaluation of securities) credit risk accumulates over a longer time frame (one to three years) as loans gradually shift into the NPL category the „Phase II“ CNB stress testing framework addressed this discrepancy with a compromise assuming an impact horizon of one year macro variables of the projected year were averaged to produce the „shock“ as the difference between initial and average future value = underestimates peaks in possible crisis (Lehman September 2008) Example of the „time“ problem: market vs credit risk
22 scenario „nervousness of markets“ from the FSR 2008/2009 assumed losses due to unfavourable interest rate changes in some quarters, but these losses are fully reversed in the following periods this dynamics of the directional changes in the shocks over time generates stress situations in the financial sector that cannot be captured by the standard stress tests using averages for the entire test period. Example of the evolution of the impact of shocks
23 Current framework of the dynamic stress tests CNB has performed stress tests with every new quarterly macroeconomic forecasts (i.e. 4 times a year – February, May, August and November), will shift to 2 times a year in 2012 alternative macro scenarios: one scenario reflects actual CNB‘s macroeconomist forecast, one or two adverse scenarios run in DSGE model are outlined by the financial stability team together with modelling division experts (14 variables used), the horizon is set to 8 quarters – for example August 2010 stress tests performed on mid-2010 portfolios with August 2010 forecasts focused on horizon 3Q2010 – 2Q2012.
24 Pre-provision incomeReal estate prices The framework G3 model
25 Dynamic features of CNB‘s stress tests Tests are set as dynamic – for every item in assets, liabilities, income and costs there is an initial state to which the impact of shocks is added in one quarter and the results serve as the initial state for following quarter this is repeated in next 8 quarters for which the prediction is generated. Four risks are tested: credit risk, interest rate risk, currency (FX) risk and interbank contagion Conservative calibration of stress test parameters (slight overestimation of risks, slight underestimation of buffers)
26 Pillar I: change in credit risk terminology/risk factors explicit PD (default rates), LGD, EL (expected loss) loan segments very close to Basel II segments (corporate, retail, other) for banks in IRB approach, application of Basel II formula to determine capital requirements Pillar II: exchange of views with banks on stress testing methodology adjustments in interest rate impact (use of derivatives, interest rate sensitivity of current accounts etc.) explicit (expert) modelling of yield curve Bringing the stress tests in line with Basel II
27 Credit risk I the methodology is being continuously improved; the tests work with four separate loan portfolios: non-financial corporations, households – consumer, households – mortgages, other loans Two impacts o credit risk: 1.Expected loss (EL) PDxLGDxEAD PD is a result of satellite models (dependent variable; smoothed default rate df), LGD set expertly (or via simple models) EAD is non-defaulted stock of exposures; total exposure modelled via credit growth model(s) 2.Risk-weighted assets (RWA) IRB formula using PD, LGD and EAD not precise (non-linearity, not all banks have IRB approach for credit risk management), but close to how banks behave
28 NPLs NPL ratio - the ratio of non-performing loans to total loans product of PD/df, existing NPLs, stock of loans (L) and outflow of NPLs outof the balance sheets NPL(2)/L(2) = approx. [NPL(1) + L(1)*df - a*NPL(1)]/L(2) expert judgment/assumptions about NPL outflow (parameter a of around 15% in a quarter): parameter a may change during bad times, very difficult to model
Illustrative example of credit shock impact: expected loss/provisions, NPL and RWA Calculation of credit losses Impact on RWA New NPLs (0,03 x 1000) NPL outflow (assumed 15% each quarter) For simplicity: 0% credit growth assumed Note: quarterly PDs, yearly PDs = 4 x 3% = 12%
CNB‘s LGDs: first expert estimate in July 2009 versus adjustment in 2010 Parameter LGD Since May 2010 (FSR 2009/2010), simple models for „elevated“ LGDs (role of GDP, property prices and unemployment)
31 Predikce parametru LGD Podniková portfolia pokles HDP o jeden p.b. navýší LGD o 5 p.b. nad výchozí hodnotu LGD 45 % Spotřebitelské úvěry růst nezaměstnanosti o jeden p.b. navýší LGD o 5 p.b. nad výchozí hodnotu LGD 55 % Úvěry na bydlení pokles cen nemovitostí o jeden p.b. navýší hodnotu LGD úvěrů na bydlení taktéž o jeden p.b. nad výchozí hodnotu 22 %
Potential „deleveraging“ leads to higher CAR in worse scenario (protracted recession in July 2009 tests). Thus, in bad times, there are two competiting drivers of RWA PD, LGD – push RWA upwards Stock of exposures – push RWA downwards For comparison a scenario with positive credit growth (and higher PD, LGD): negative impact on CAR confirmed (via higher RWA) Credit growth, RWA & capital adequacy (CAR)
33 until June 2010 (FSR 2009/2010), pre-provision income was expertly set at x % of average of past 2 years (x < 100%, thus additional stress applied in the sense of lower intermediation activity) during 1H2010, a simple model of pre-provision income was estimated (the main determinants: nominal GDP, yield curve, NPLs and capital adequacy) profit/loss is generated using the pre-provision income and the impact of shocks regulatory capital is adjusted every 2Q to get back to initial CAR thus, a P/L account and balance sheet of all banks generated every quarter = possible to cross-check with reality later on How to work with pre-provision income, profits and capital
34 comparison of model estimation versus expert setting of pre-provision income conservative estimation – estimate of the model minus 1 stdev of growth Modelling pre-provision income
Net income, P/L and capital adequacy: an example For final evaluation of banks‘ resilience capital adequacy is estimated. Link between shocks impact and capital adequacy must reflect (net) income generated by banks even under stress, asymmetric treatment of profits in calculation of regulatory capital, topping up of regulatory capital (set for 2nd calender quarter every year).
36 regular consultations with commercial banks on stress testing methodology project of „joint stress tests with selected banks“ basically bottom-up stress tests – CNB gives the increase in risk parameter PD, banks themselves calculate the impact since summer 2009 aggregate results published in the FSR 2009/2010 Regular cross-check of the stress testing framework I
37 Mezibankovní nákaza pokročilých testech banky významně postižené ztrátami z kreditního a tržních šoků se mohou dostat do situace, kdy se zvyšuje pravděpodobnost, že přestanou splácet mezibankovní úvěry věřitelské banky na toto zvýšené riziko protistran vytváří opravné položky, což ovšem dále zvyšuje ztráty těchto věřitelských bank a jejich vlastní pravděpodobnost nesplácení mezibankovních úvěrů dynamická simulace probíhající v několika kolech (domino efekt), než je nalezena rovnováha finální ztráty v mld. Kč ve formě opravných položek na nesplácené (nezajištěné) mezibankovní úvěry jsou dopadem rizika mezibankovní nákazy
38 organized since summer 2009 semi-annually 8 (mostly) largest banks (with IRB method, or close to approved application of IRB) bottom-up stress tests (currently focusing on credit risk only) CNB gives the increase in risk parameter PD banks themselves calculate the impact (on EL and RWAs) LGD and EAD not stressed (assumed constant at reported level) individual results used in supervision (discussion with banks) in financial stability (information about the soundess of the largest banks and as a cross-check of aggregate stress tests) aggregate results published in financial stability reports Joint stress tests with selected banks
39 ZFS 2009/2010 Verifikace zátěžových testů jako součást pokročilého rámce zátěžového testování Procykličnost finančního systému a simulace „feedback“ efektu ZFS 2006, Vývoj kreditního rizika a zátěžové testování bankovního sektoru v ČR References