Systemic Risk Assessment Model for Macroprudential Policy (SAMP)

Transcription

1 Systemic Risk Assessment Model for Macroprudential Policy (SAMP) A. Overview of SAMP (1) Motivations Since the global financial crisis, the roles of central banks in macroprudential policy have been strengthened in many countries. This change has arisen from the perception that macroeconomic stability cannot be achieved without financial stability. In order to implement macroprudential policies effectively, central banks must be able to identify systemic risk factors. 120) To this end, central Overall structure of SAMP banks of major advanced economies including the U.K., Canada and Austria have developed their own systemic risk assessment models to support their macroprudential policies (Aikman et al., 2009; Gauthier et al., 2010; Boss et al., 2006). As a part of the BOK s efforts to carry out its new mandate for financial stability since revision of the Bank of Korea Act in September 2011, the Bank of Korea has developed the Systemic Risk Assessment Model for Macroprudential Policy (SAMP), in a process that started with improvement of the macro stress test model (BOKST-07) developed in By integrating systemic risk assessment and macro stress testing under a unified framework, SAMP is expected to serve as a comprehensive analysis framework for macroprudential policy. Appendix Macro-risk factor module Default contagion module Systemic risk measurement module Estimation of joint probability distribution of macro-risk factors Macroeconometric model (BVAR) Time-varying volatility (GARCH) Co-movement & dependence (Copula) Fat tail-risk (EVT) Macro-risk factors Macro shocks Fundamental loss Bank profit and loss module Credit/market losses Interest /non-interest income Loan loss provisions Fundamental default Fire sale losses Credit crunch losses Default costs Defaults due to loss contagion Default contagion Liquidity contagion Default contagion loss Liquidity contagion loss Funding liquidity contagion module Deleveraging/liquidity withdrawals Fire sale losses Credit crunch losses Higher funding costs Defaults due to funding liquidity contagion Estimation of the number of defaulting/distressed banks Estimation of loss distribution of individual banks/banking system Multi-period module Repeat measurement process until t+4 period Dynamic update of banks B/S Expected Loss Expected Shortfall Systemic risk indicators Value at Risk Probability of systemic crisis 120) As in the case of implementing monetary policy based upon a macroeconomic model, it is desirable that macroprudential policy be implemented based upon a systemic risk assessment model (Alessandri et al., 2008). 145

2 SAMP provides the BOK with a foundation for harmonious operation of its monetary policy and macroprudential policy. SAMP is an integrated systemic risk model framework comprising six modules. Designed to detect systemic crisis, SAMP measures not only the 1st round direct effects of macroeconomic shocks on the financial system, but also the 2nd round effects amplified and propagated by interbank contagion, fire sales, credit crunches and deleveraging. SAMP can be used in diverse macroprudential policy analyses, including systemic risk monitoring, macroeconomic stress testing, policy effect analysis, and analysis of domestic systemically important banks (D-SIBs). (2) Main features The main characteristics of SAMP, which differentiate it from other conventional macroeconometric models or stress test models, are as follow: First, unlike equilibrium models which focus on the average relationships among variables in normal times, SAMP can measure tail risks caused by imbalances in the macroeconomy and the financial sector during a crisis. Second, SAMP is a non-linear model, that reflects the influences of amplification and propagation in the financial system of macro risk factors such as fire sales, interbank contagion, deleveraging, and macro-financial feedback effects. 121) Third, SAMP is an integrated model, through which it is possible to carry out measurement of various kinds of risks, probabilistic assessment of systemic crisis, stress testing, and macroprudential policy simulations under a unified framework. Fourth, SAMP is a modeling platform that enables macro stress testing to assess vulnerabilities of the financial system under specific scenarios. Fifth, SAMP is a tool for policy simulation to analyze the effects of macroprudential policies such as financial regulation, liquidity provision and capital injection. Lastly, as risk assessments are conducted on an individual financial institution basis, SAMP can be used for identifying the vulnerabilities of individual financial institutions. B. Structure of SAMP SAMP is composed of six modules: a macro-risk factor module, a bank profit and loss module, a default contagion module, a funding liquidity contagion module, a multi-period module, and a systemic risk measurement module. Details of these modules follow: (1) Macro-risk factor module The macro-risk factor module estimates a joint distribution of macro-risk factors that have significant impacts on banks profits and losses. In order to derive the joint distribution, this module first estimates the marginal probability distributions of the different macro-risk factors at a future point, and 121) The amplification and propagation mechanism introduced in SAMP generates a sharp rise in losses (a threshold effect) in the event of a shock above a certain level. 146

3 finally combines them into the joint probability distribution. In the first step of macro-risk factor selection, twelve variables 122) are selected as macro risk factors affecting banks credit losses, market losses, and interest and non-interest incomes. In the next step, the time-varying fat-tailed marginal probability distributions of these macro risk factors are estimated by applying methods such as Bayesian VAR, GARCH and extreme value theory (EVT). 123) Finally, the marginal probability distributions of the macro risk factors derived from the previous step are transformed into a joint probability distribution using the t-copula 124) as in Daul et al. (2003). Structure of Macro-risk Factor Module Conditional scenarios Restrict variables Unconditional scenarios Plausibility (Maha distance) Stress scenarios The scenarios for systemic risk assessment and macro stress testing are then extracted from the joint probability distribution reflecting the dependence among the risk factors. (2) Bank profit and loss module The bank profit and loss module estimates banks profits and losses under the selected scenario of macro risk factors. Banks profits and losses are estimated by sub-categories including credit loss, market loss, interest income and non-interest income. Credit losses are the core part of bank losses. The direct impacts of macroeconomic shocks on banks loan portfolios are measured as credit losses (Schmieder et al., 2011). In order to measure credit losses 125) stemming from changes in the macroeconomic variables, the module estimates the changes in the probability of default (PD) and loss given default (LGD) of the credit portfolio. The PDs are estimated by five borrower types: large enterprises, SMEs, housing mortgages, households and credit card loans, based on the Basel III. Because of the lack of time series data on PDs and LGDs, LGDs are estimated 126) by modifying Altman et al. (2003) and S&P (2010) models. Market losses on the trading book incurred by changes in interest rates, stock prices and Appendix 122) List of macro-risk factors in SAMP Classification real economy (4) financial market (4) external sector (4) Macro variables GDP growth rate, Unemployment rate, Inflation rate, Housing price F rate, Equity price, Corporate bond spread, Government bond yield Global GDP growth rate, Global equity price, US treasury bond yield, Oil price 123) EVT is a statistical method estimating the shape of distribution of extreme data belonging to the tail part. It has advantages of being able to calculate the probability of the tail part. 124) The Copula is a statistical method that transforms one-dimensional marginal probability distributions into a multivariate joint distribution. It is appropriate for analyzing dependence among variables that have fat tails. 125) Credit loss = Probability of default (PD) Loss given default (LGD) Exposure at default (EAD) 126) LGDß=LGDç (PDß-PDç), where LGDß and LGDç denote the scenario value and current value of each LGD, PDß and PDç refer to the scenario value and current value of each PD, and is the S&P mode s sensitivity coefficient of LGD to PD. 147

4 F rates, are measured by applying the markto-market method. 127) Profits and losses on available-for-sale securities are not calculated as market loss, but affect equity capital because they fall into the capital adjustment category. Interest income is calculated by deducting estimated interest expenses from estimated interest revenues for each repricing period. Non-interest income is calculated by estimating non-interest revenues and expenses using a simple regression model. Structure of Bank Profit and Loss Module Credit loss (Loan loss provision) Market loss Interest income Non-interest income Regulatory capital ratio (3) Default contagion module The default contagion module measures the 2nd round effects of bank defaults. Bank losses spreading through interbank exposures are captured by analysis of the banking system network structure. 128) The default contagion module also measures fire sale losses and credit crunch feedback losses spreading through macro-financial linkages. In order to estimate default contagion effects, the module is constructed following Eisenberg and Noe (2001) and Cifuentes et al. (2005). Fundamental default occurs when net worth falls below the default threshold point. Interbank losses are estimated through analysis of the banking system network structure (Eisenberg and Noe, 2001). Default costs are assumed to be incurred as the liquidation values of defaulting banks fall below book values as in Elsinger et al. (2006). Additional losses incurred by fire sales of liquid assets can be divided into mark-to-market losses and disposal losses. In cases where banks capital ratios decline due to losses, banks may sharply reduce their loans in order to comply with capital regula- 127) Trading profit and loss = Trading position Price 128) Analysis of the banking system network structure is conducted using an interbank exposure matrix compiled based on interbank transaction information as follows: For an individual bank in, the B/S identity can be written as, where b ij is bank i s borrowings from bank j, and a q i are the liquid and illiquid assets operated by bank i outside the banking system, d q and i dz i are the liquid and illiquid liabilities financed by bank i outside the banking system, and e i is the equity capital of bank i. B/S of bank i Assets Liabilities A banking system network model is constructed based on the B/S of the banking system(n={1,...,n}) composed of N number of banks: Matrix structure of banking system 1 0 b i b d q 1 d z 1 e Banks Liquid Illiquid Equity 1 i N liabilities liabilities capital Banks i b i 0 b i d q i d z i e i a q i a z i d q i d z i e i N b b i 0 d q d z e Liquid assets a q 1 a q i a q Illiquid assets a z 1 a z i a z 148

5 tions, bringing about credit crunches. Credit crunch losses in this process can be estimated by a simple model of the relationship between aggregate lending and PD. 129) If the net worth of a bank which is not in fundamental default falls below the default threshold point when interbank loan losses, fire sale losses and credit crunch feedback losses are taken into account, a loss contagious default occurs and the losses incurred subsequently are re-estimated. Structure of Default Contagion Module change of Basel capital ratio. 130) It then estimates losses generated from banks fire sales of assets to obtain cash in the liquidity contagion stage. It also estimates the liquidity shortages caused by the deposit run-off and the subsequent credit crunch feedback losses. The additional funding costs borne by distressed banks are estimated as well. If the net worth of a bank which is not in fundamental or loss contagious default falls below the default threshold point due to losses in the liquidity contagion stage, a liquidity contagious default occurs and the losses incurred subsequently are re-estimated. Bank loss Structure of Funding Liquidity Contagion Module No Appendix New borrowing No No (4) Funding liquidity contagion module *l = LCR requirement The funding liquidity contagion module measures the contagion effects of funding liquidity across the banking system. The module first estimates the withdrawals of interbank loans and deposits using the 129) This model estimates the impacts of changes in the aggregate loan amount on PD. It is a simple reduced form of macro-financial linkages (macroeconomic shock risk aversion loan reduction credit crunches economic downturn increased default bank losses). 130) The deposit run-off rate is set as proportional to the square of the extent of decline in equity capital, and as inversely proportional to the initial equity capital ratio (Lee, 2011; De Haan and Van den End, 2011). 149

6 (5) Multi-period module The multi-period module estimates annual profits and losses of banks by iterating the previous four modules reflecting quarterly changes in their profits and losses and balance sheet items. The time horizon of SAMP is set at one year. In order to calculate annual bank losses, it repeats the process of bank loss estimation and dynamically updates the B/S reflecting the results of estimation on a quarterly basis. Structure of Multi-period Module The systemic risk measurement module produces systemic risk indicators that can assess the probabilities of systemic crisis using the aggregate loss distributions of the banking system. First, the total losses of individual banks are measured as the sums of their fundamental losses, interbank loan losses, fire sale losses, credit crunch losses, and additional funding costs. Using the estimated losses, the module then determines whether the banks concerned are bankrupt or distressed, and counts the number of defaulting banks and the number of distressed banks. Using Monte-Carlo simulation 131) based on the joint probability distribution of the macro risk factors, the module derives aggregate loss distributions of the banking system. From these aggregate loss distributions, the expected losses, VaR (value at risk), ES (expected shortfall), the number of defaulting banks and the number of distressed banks in the banking system can be calculated (Lehar, 2005; Adrian and Brunnermeier, 2009; White et al., 2010; Acharya et al., 2010). Based on the various definitions of systemic crisis, the probability of systemic crisis can be estimated using the aggregate loss distribution. h=h+1 Structure of Systemic Risk Measurement Module No *Initial value of h is 1 Annual bank losses (6) Systemic risk measurement module Define threshold Define systemic crisis 131) Simulations are conducted for 100,000 scenarios randomly chosen from the macro-risk factor joint probability distribution. 150

7 C. Comparison with other systemic risk models In order to reflect the unique features of the Korean domestic economy and financial system, SAMP models the mechanisms behind the amplification and propagation of various risks. Specifically, compared to other models, SAMP has several comparative advantages, including the enhanced accuracy in its tail risk evaluation, its estimation of the contagion effects of funding liquidity risks, its estimation of credit crunch losses from macro-financial feedbacks, and its setting up of multi-period models. In addition, SAMP capitalizes on a vast amount of private data of individual banks such as information on transactions among financial institutions, obtained through the BOK s own surveys. It therefore has advantages in identifying systemic risk factors that are difficult to capture by existing systemic risk indicators dependent mainly on market data. D. Ways forward and future tasks SAMP is expected to be used for macroprudential policy analyses as follows: Comparison of Systemic Risk Models Korea Korea U.K Austria Canada (SAMP) (BOKST-07) (RAMSI) (SRM) (MFRAF) Name of model Systemic Risk Risk Assessment Systemic Risk Macro-Financial Assessment Model for - Model for Systemic Monitor Risk Assessment Macroprudential Policy Institutions Framework Time of development Macro model (BVAR) (Simultaneous equations model) (BVAR) (VAR) (DSGE) Fat tail-risk (EVT, GARCH) (EVT, GARCH) Probability of default (PD) (5 exposures) (3 exposures) (4 exposures) (11 industries) (7 exposures) Loss given default (LGD) (S&P model) Market loss (Mark-to-market) (Mark-to-market) (Mark-to-market) (Mark-to-market) Income (Interest/Non- (Interest/Non- interest income) interest income) Loss contagion (Network model) (Network model) (Network model) (Network model) Funding liquidity (Contagion effect (Contagion effect (Contagion effect reflected) not reflected) not reflected) Macro-financial (Credit crunch feedbacks reflected) (Under development) Multi-period model (Dynamic B/S update) (Dynamic B/S update) (No B/S adjustment) Time horizon 1 year 1 year Longer than 1 year 1 quarter 1 year Appendix Sources: Aikman et al. (2009), Financial Stability Report of Oesterreichische Nationalbank (June, 2006), Bank of Canada Review (2012 Spring) 151

8 First, SAMP can be used as a tool for quantitative analysis and monitoring of financial stability conditions. Second, in the SAMP framework it is possible to carry out macro stress tests to identify financial system vulnerabilities. Third, it can be used to analyze the effects of Basel III regulations such as the countercyclical buffer and LCR regulations, and of macroprudential policies such as liquidity provision and capital injection. Fourth, it is also possible to analyze D-SIBs, including the calculation of individual bank contributions to systemic risk. Finally, SAMP can be used for monitoring the vulnerabilities of individual banks. The Bank of Korea also plans to further improve SAMP and expand its coverage of financial institutions. First, given Korea s unique situation with its high degree of openness to the external sector, it is necessary to develop additional modules for the conduct of stress tests on foreign currency liquidity. Second, it is necessary to develop a structural model incorporating the process of risk amplification caused by macro-financial linkages. Lastly, while the coverage of financial institutions of SAMP is currently limited to 18 domestic banks, it should be expanded to also include nonbank financial institutions. <References> Acharya, V., L. Pedersen, T. Philippon, and M. Richardson (2010), Measuring systemic risk, Working paper, New York University. Adrian, T., and M. K. Brunnermeier (2009), CoVaR, Federal Reserve Bank of New York: Staff Reports No Aikman, D., P. Alessandri, B. Eklund, P. Gai, S. Kapadia, E. Martin, N. Mora, G. Sterne, and M. Willison (2009), Funding Liquidity Risk in a Quantitative Model of Systemic Stability, Bank of England Working Paper No 372. Alessandri, P., P. Gai, S. Kapadia, N. Mora, and C. Puhr (2008), A Framework for Quantifying Systemic Stability, Bank of England. Altman E., B. Brady, A. Resti and A. Sironi (2003), The link between default and recovery rates: theory, empirical evidence and implications, Journal of Business. Boss, M., T. Breuer, H. Elsinger, G. Krenn, A. Lehar, C. Puhr and M. Summer (2006), Systemic Risk Monitor: Risk Assessment and Stress Testing for the Austrian Banking System, OeNB technical document, available upon request from the authors. Cifuentes, R., H. Shin, and G. Ferrucci (2005), Liquidity Risk and Contagion, Journal of European Economic Association 3, Daul, S., E. De Giorgi, F. Lindskog and A. McNeil (2003), The grouped t-copula with an application to credit risk, Risk, 16, De Haan, Leo, and Jan W. van den End (2011), Banks responses to funding liquidity shocks: Lending adjustment, liquidity hoarding, and fire sales, De Nederlandsche Bank Working Paper No Eisenberg, L., and T. Noe (2001), Systemic Risk in Financial Systems, Management Science 47,

9 Elsinger, H., A. Lehar, and M. Summer (2006), Risk assessment for banking systems, Management Science, Vol. 52(9), pages 1, Elsinger, H., A. Lehar, and M. Summer (2006), Systemically Important Banks: An Analysis for the European Banking System, International Economics and Economic Policy 3, Gauthier, C., A. Lehar, and M. Souissi (2010), Macroprudential Regulation and Systemic Capital Requirements, Bank of Canada Working Paper Gauthier, C. and M. Souissi (2012), Understanding Systemic Risk in the Banking Sector: A MacroFinancial Risk Assessment Framework, Bank of Canada Review (2012 Spring), Lee, S. H. (2011), Systemic Risk and Financial Regulation, Quarterly Economics Analysis 17(3), 1-42.(in Korean) Appendix Lehar, A. (2005), Measuring systemic risk: A risk management approach, Journal of Banking and Finance 29, S&P (2010), Default, Transition, and Recovery: 2009 Annual Global Corporate Default Study and Rating Transition Schmieder, C., C. Puhr, and M. Hasan (2011), Next Generation Balance Sheet Stress Testing, IMF Working Paper(No. 11/83) White, H., T.-H. Kim, and S. Manganelli (2010), VAR for VaR: Measuring Systemic Risk Using Multivariate Regression Quantiles, Manuscript, UCSD. 153

10 Calculation of Systemic Risk of Korean Banking System using SAMP The BOK has performed a pilot calculation of systemic risk indicators for the banking sector 1) using SAMP, so as to assess recent financial stability conditions in Korea. The coverage of institutions is 18 domestic banks (7 commercial banks, 6 local banks, and 5 specialized banks), and the period of analysis is from the first quarter of 2006 through the second quarter of The Expected Shortfall (ES) Index of the distressed bank ratio and the Systemic Crisis Probability Index were used as systemic risk indicators. The distressed bank ratio ES Index is an indicator measuring the tail risks of the banking sector. It measures changes in the distressed bank ratio that may occur with a 1% probability within one year. The Systemic Crisis Probability Index shows changes in the probability of simultaneous defaults of two or more banks within one year. Distressed bank ratio 1) ES Index 3) Systemic Crisis 2) Probability Index 3) st round losses + 2nd round losses 1st round losses + 2nd round losses st round losses st round losses Notes: 1) Distressed banks are defined as banks with less than 8% BIS ratios. ES is an indicator that measures tail risks more conservatively than VaR. In this calculation, ES is estimated conditional on the distressed bank ratio(number of distressed banks/total number of banks) occurring with less than 1% probability. 2) A systemic crisis is defined as a situation in which two or more banks go bankrupt simultaneously. 3) The peak during the global financial crisis is set at 100 in the indices. 4) The 1st round losses are generated directly by macro shocks. The 2nd round losses mean bank losses incurred by 2nd round effects such as interbank contagion effects, fire sales, credit crunches and deleveraging. 0 According to the above analysis, systemic risk has declined significantly in Korea in comparison with the global financial crisis period, thanks to banks efforts to enhance their capital adequacy since the crisis. However, systemic risk has recently shown a modest increase due to the mounting uncertainties in the domestic and global economies and the rising credit risk of household debt. 1) The systemic risk assessment models of other central banks mainly target the banking sector, since the banking sector takes up a large share in the financial system and the contagion effects are mostly generated through interbank transactions. 154