630 likes | 982 Views
Overview of RAMSI (Risk Assessment Model for Systemic Institutions). Sujit Kapadia, Bank of England ECB-CFS Conference on “Macro-prudential Regulation as an Approach to Contain Systemic Risk: Economic Foundations, Diagnostic Tools and Policy Instruments” Frankfurt, 27 September 2010
E N D
Overview of RAMSI (Risk Assessment Model for Systemic Institutions) Sujit Kapadia, Bank of England ECB-CFS Conference on “Macro-prudential Regulation as an Approach to Contain Systemic Risk: Economic Foundations, Diagnostic Tools and Policy Instruments” Frankfurt, 27 September 2010 This talk represents the views of the speaker and should not be thought to represent those of the Bank of England or Financial Policy Committee members. The numerical results presented should not be construed to be an accurate measure of the systemic risk in the UK banking system
Outline • Motivation: Macroprudential Policy and RAMSI • Overview of RAMSI • key drivers of individual bank P&L • liquidity risk and feedbacks • Sample Simulation Results
The Role of Macroprudential Policy “…macro-prudential policy has two broad components. • first, policymakers must try to identify, understand and monitor systemic risk. • second, when threats to financial stability are identified, policymakers should seek to counter these risks and vulnerabilities using a range of different tools.” ~ UK Treasury Consultation Paper (2010) • Objectives of ESRB • Identify and prioritise systemic risks • Issue early warnings for systemic risks • Issue policy recommendations for mitigating those risks
SIFIs and Structural / Network Risks Network of large exposures(a) between UK banks(b)(c) (from June 2009 Bank of England Financial Stability Report) Source: FSA returns. (a) A large exposure is one that exceeds 10% of a lending bank’s eligible capital during a period. Eligible capital is defined as Tier 1 plus Tier 2 capital, minus regulatory deductions. (b) Each node represents a bank in the United Kingdom. The size of each node is scaled in proportion to the sum of (1) the total value of exposures to a bank, and (2) the total value of exposures of the bank to others in the network. The thickness of a line is proportionate to the value of a single bilateral exposure. (c) Based on 2008 Q1 data.
UK and International Backdrop UK Developments: • Bank of England to be given macroprudential powers • Financial Policy Committee (FPC) likely to be setting policy within months • possible instruments: capital buffers (aggregate and / or sectoral; systemic add-ons); margin requirements; LTV caps etc. • Independent Commission on Banking has been established • focus on structural issues / network risks / how to handle SIFIs International Initiatives: • Basel countercyclical capital buffer proposal and FSB work on SIFIs • Establishment of European Systemic Risk Board (ESRB) Strong analytical toolkit needed to support policy-making process • senior policymakers (FPC; ESRB) are the key end users • need to be able to take action well in advance (eg 2004 / 2005)
Objectives of RAMSI • Aims to deliver a unified, quantitative framework to (i) sharpen risk assessment work; and (ii) help set macroprudential policy • A modular approach is used to integrate different sources of risk to the UK banking system • Individual bank balance sheets at the core • Non-linearities central: incorporates systemic feedbacks, amplifiers and network dynamics (potential role in identifying SIFIs) • Can be used for forecasting, scenario analysis and story-telling • Considers the role of both capital and liquidity • Currently better suited to assessing resilience of banks; link from banking sector to credit supply a key area for development • Potential weaknesses: • Perhaps less suited to analysing non-bank financial intermediaries • Introducing a stronger role for forward-looking behaviour is a key challenge
Complementary Approaches at the BoE • RAMSI is part of a suite of models envisioned for macroprudential policy analysis • Complementary approaches under development: • Market-based (Merton) models • Early warning indicators • DSGE models with banks / financial sector • VAR models with banks • Regression models to identify systemic importance • Calibrated network models
Overview of RAMSI Draws on: • Aikman et al (2009), “Funding Liquidity Risk in a Quantitative Model of Systemic Stability”, BoE WP 372 • Kapadia et al (2010), “Liquidity Risk, Cash Flow Constraints and Systemic Feedbacks” • [See also OeNB, 2006 – ‘SRM’ and Elsinger et al, 2006]
GDP growth UK Banking System Profits (Assets, Liabilities, Equity, …) Equity returns The Logic of RAMSI • detailed balance sheets of 10 major UK banks
Model Architecture Macroeconomic and financial Shocks Credit, market and income risk Feedbacks Asset-side (“market liquidity risk”) Liability-side (“funding liquidity risk”) UK banks’ balance sheets Network model of UK banks and LCFIs System assets / loss distributions Effects on bank lending
Confidence effects Reduction in Interbank Lending (Liquidity Hoarding) Counterparty credit risk (Network) Failure of a bank Failure of one bank Asset sales Feedbacks in the Financial System 4 1 1 4 4 5 1 1 Funding problems at one or more banks Funding problems at other banks Shock 2 3 3 3
Macro-Credit Risk Macroeconomic risk factors modelled using a Bayesian VAR yield curve interpolated using Nelson-Siegel model For credit risk, we split exposures by: Region: UK, US, EA, RoW Type: mortgages, credit cards, other unsecured, corporate For each type of exposure (eg UK Mortgages), we model An aggregate default probability (PD) An aggregate credit loss rate (ACL) A bank-specific credit loss rate (CL) ACL captures variation in recovery rates; CL is introduced (and calibrated) to capture bank heterogeneity.
Net Interest Income (1) Coupon on Assets = Risk-free rate + Endogenous spread to cover expected write-offs
Net Interest Income (2) • Interest rate spread set to cover expected credit losses. • Coupons are sticky: only a subset of the balance sheet can be repriced at any time. • Example: suppose mortgages are repriced annually and there is a permanent, unanticipated increase in the mortgage PD. E(Income – Credit Loss) < 0 for four quarters • Repricing maturity mismatch varies across banks.
% of initial assets Aggregate UK Banks 0.002 0.0015 Net interest income 0.001 0.0005 0 -0.0005 -0.001 Credit Loss -0.0015 -0.002 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 Quarter Net Interest Income (3): Example – a 100bps Increase in UK Mortgage PDs
Net Interest Income (3) Coupon on Assets = Risk-free rate + Endogenous spread to cover expected write-offs + Exogenous spread to match data Coupon on Liabilities = Risk-free rate + Exogenous spread to match data +Endogenous bank-specific spread linked to bank rating
Net Interest Income (4) Modelled in two stages: • Rating index function produces ratings for each bank in each quarter. • Simple mapping from ratings to credit spreads. Endogenous spreads apply to: • interbank and OFC deposits • commercial paper • certificates of deposit • subordinated debt
Other Income and Costs • Simple ‘rule-of-thumb’ reduced-form estimated equations • Non-interest income excluding trading income: • Pro-cyclicality: Based on empirical evidence from US data, fees and commissions are found to be strongly pro-cyclical. • Bank-specific determinant: 2007 levels of non-interest income used as a base for forecasts. • Operating expenses: • Banks target cost / income ratios. • But banks are unable to immediately adjust expenses given a significant drop in operating income (i.e. there is some stickiness to operating expenses).
Net Profits, Taxes and Dividends • Net profits of a bank (before feedbacks): • Tax rates and the ratio of dividends to profits are in line with recent history. • Profits (or losses) are assumed to increase (or erode) core Tier 1 capital directly. • Core Tier 1 capital ratio computed using Basel II standardised risk weights to obtain RWA.
Reinvestment • Credit losses booked against the relevant exposure for the loss. Banks’ balance sheets are rebalanced via a set of reinvestment rules: • Core Tier 1 ratio: Banks have a bank-specific ‘target’ Tier 1 capital ratio which they aim to meet when investing their funds. • Portfolio allocation: Subject to the first rule, banks invest in assets in proportion to their shares on the bank’s initial balance sheet • Liability generation: The first rule determines total assets after reinvestment and hence the amount of new liabilities which need to be raised. These net new liabilities are allocated in proportion to their shares on the bank’s initial balance sheet. • No active disinvestment • No reshuffling of existing exposures
Closure of Funding Markets: A ‘Danger Zone’ Approach • Information on individual institutions – as the information on the bank deteriorates, danger zone points accumulate. • As the score crosses set thresholds, funding markets close to that institution.
Banks’ Cash Flow Constraint, Defensive Actions and Systemic Feedbacks (1) Liabilities(due)+ Assets(new/rolled over) + Off-balance sheet(liquidity need) < Net Income + Liabilities(new/rolled over) + Assets(due) + Off-balance sheet(liquidity supply) + Value of Assets Sold + Cash
Banks’ Cash Flow Constraint, Defensive Actions and Systemic Feedbacks (2) • Suppose that long-term unsecured funding markets close to a bank • Shortens the funding maturities at other banks – worsens their mismatch position • Effect depends on extent of borrowing from distressed bank (network data) WLD + RLD + WAN,R + RAN,R + OFBLD < Net Inc + WLN,R + RLN,R + WAD + RAD + OFBLS + LAS + Spi*ILAiS Becomes increasingly short-term (snowballing) Hoard by shortening maturity of wholesale lending
Banks’ Cash Flow Constraint, Defensive Actions and Systemic Feedbacks (3) • Now suppose that short-term unsecured funding markets close to a bank WLD + RLD + WAN,R + RAN,R + OFBLD < Net Inc + WLN,R + RLN,R + WAD + RAD + OFBLS + LAS + Spi*ILAiS
Banks’ Cash Flow Constraint, Defensive Actions and Systemic Feedbacks (4) • Now suppose that short-term unsecured funding markets close to a bank WLD + RLD + WAN,R + RAN,R+ OFBLD < Net Inc + WLN,R + RLN,R+ WAD + RAD + OFBLS + LAS + Spi*ILAiS Withdraw wholesale lending completely Use profits to pay maturing liabilities Draw on committed lines Sell or repo liquid assets Sell illiquid assets in a fire sale
Asset Fire Sale and Network Feedbacks • Distressed bank sells securities in a fire sale => asset prices may fall as markets are not perfectly liquid => other banks may incur mark-to-market losses. • If banks cannot meet their cash flow constraint after all possible defensive actions, they default (banks can also fail if their core Tier 1 capital ratio falls below 4%). • Bankruptcy costs incurred upon failure. • Banks suffer interbank losses when counterparties fail (network model) and may become distressed as a result.
Sample Simulation Results • Illustrative simulations from relatively old data • Model currently only at development stage • Further validation and testing to be done • Numerical results should not be construed to be an accurate measure of the systemic risk in the UK banking system.
Simulation Method • 3-year horizon with baseline 2007Q4 • 500 random draws for quarter 1; one draw for each subsequent quarter • All draws from N(0, Σ), where Σ is the Bayesian VAR residual covariance matrix for 24 factors: UK, US, Europe, and Global • Results presented as aggregate distributions • will also show a sample stress test • can also analyse bank-specific results (eg to rank vulnerability) and intermediate outputs (eg credit risk; funding costs)
Credit Losses and Net Interest Income in 2008, relative to 2007 Capital
Total System Assets, Q12: With and Without Liquidity Risk and Feedbacks
Introducing Policy • Model could indicate how risks evolve over the economic cycle as balance sheets change • Can consider the role that extra capital and liquidity could play in enhancing resilience • Identification of key amplifiers and feedback mechanisms can facilitate targeting of structural weaknesses
Concluding Comments • RAMSI demonstrates how systemic risk can be modelled quantitatively: • captures various sources of risk, and (some) key correlations amongst them. • generate fat-tailed loss distributions despite Gaussian shocks • highlight the adverse amplifying effects of various feedback mechanisms which have been critical throughout the crisis • Offers the flexibility to build in wide range of other mechanisms. • feedback from banking sector to credit supply particularly important • Can be used to help guide policy decisions • Introducing a stronger role for forward-looking behaviour is a key challenge
Model Dynamics Start of Period Next Period Feedback Loop Macro/ financial shocks PDs Yields Balance Sheet adjustment Loan book (credit losses) Profit and Loss Trading income Net interest income Other income Operating costs Tax & dividends Rating model Funding Liquidity /Bank failure check Liquidity Feedbacks Asset fire sales Funding (confidence) Network Losses Reinvestment
Probability Stress scenario Baseline 7,200 5,600 4,000 2,400 800 800 2,400 4,000 5,600 - + £ Billions Funding Costs and the Closure of Funding Markets
Why Danger Zones? • Liquidity risk is binary or at least non-linear. • Confidence is key. • Liquidity crises are fairly rare events. • Data availability on those that have occurred is limited. • Difficult to build statistical models to predict the onset of liquidity crises.
Integration into RAMSI RAMSI GENERALISED ILLIQUIDITY / LIQUIDITY SHOCKS LIQUIDITY POSITION SOLVENCY CONCERNS CONFIDENCE CLOSURE OF FUNDING MARKETS
Maturity Mismatch and Snowballing • Short-term wholesale maturity mismatch is given by: • Snowballing effect – if banks lose access to long-term wholesale funding markets, their short-term wholesale liabilities increase. • worsens their mismatch position mechanically (data driven) • increases their danger zone points score (subjective judgement based on overall danger zone framework)
Retail Deposits, Secured Funding and ‘Safe’ Banks • Gradual retail deposit outflows; widespread retail run at or beyond the point where short term unsecured markets close. • Secured funding: if cannot repo assets, assume you can sell them at the prevailing market price (may be a fire sale price).
Asset Fire Sale Equation Illiquidity factor common across asset markets Value of assets sold by failed bank Post fire-sale price Pre fire-sale price Shocks to market depth Market depth
Asset Fire Sales: Calibration • Constrained by limited data. • Illiquidity factor (theta) is calibrated for a case study on US convertible bonds – Mitchell et al (2007) find that price impact is approximately 3% when 5% of the market is sold in 2005. • Given theta, derive ‘implied’ market depths (M’s) that generate 2%, 4% and a 5% price fall for equities, corp. bonds and ABS, respectively, when the UK bank with the largest holdings of these assets sells all of its holdings. • Price falls informed by case studies (Coval and Stafford, 2007; Mitchell et al, 2007; Pulvino, 1998).