1 Systemic Risk Monitoring ToolsProfessor Connel Fullenkamp Duke University IMF—CAPTAC-DR Course on Macroprudential Policy March 2017 This training material is the property of the International Monetary Fund (IMF) and is intended for use in IMF Institute for Capacity Development  (ICD) courses. Any reuse requires the permission of ICD. Systemic Risk Monitoring Tools

2 An Embarrassment of Riches?Bisias et al (2012) find 31 quantitative Systemic Risk monitoring measures in their survey of the literature Challenge: Organizing our understanding of these tools and choosing ones that are applicable, practical, and accurate for your economy Compromise: overview groups of tools and focus on a few that are relevant and practical Systemic Risk Monitoring Tools

3 Systemic Risk Monitoring ToolsOverview: The Four L’s Systemic risk monitoring is based in large part on measuring the “four L’s of financial crisis:” Leverage Losses Linkages Liquidity Approach: measure these directly but also measure their consequences, such as asset price appreciations and default probabilities Systemic Risk Monitoring Tools

4 Many Ways to Organize the ToolsStarting Approach: group the tools by common methodologies Later: group tools by their natural time horizon (ex-ante, concurrent) Systemic Risk Monitoring Tools

5 The Main MethodologiesMacroeconomic Measures Network Analysis Measures Probability Distribution Measures Contingent Claims and Default Measures Illiquidity Measures Qualitative Measures Systemic Risk Monitoring Tools

6 Main Methodologies and Intuitive Ways to Measure Systemic RiskSystemic Risk “from the ground up:” take measurements of individual institutions and aggregate them in different ways Systemic Risk “from the top down:” take aggregate measures that appear to have high correlations with financial distress Most measures work from the ground up Systemic Risk Monitoring Tools

7 Macroeconomic MeasuresEarly warning indicators Stress tests Risk topography Systemic Risk Monitoring Tools

8 Early Warning IndicatorsFinding and using macro time series that are correlated with elevated financial stress or that have served as signals that preceded stress events Aggregate levels of leverage, liquidity Growth rates of credit and asset prices Bubble indicators Systemic Risk Monitoring Tools

9 Systemic Risk Monitoring ToolsAggregate Leverage Approach: collect data on actual usage of leverage by all financial institutions (including hedge funds) and find aggregate leverage or “collateral rate” Monitor for strong swings in the leverage cycle in which leverage increases dramatically, followed by stress after bad news creates uncertainty and disagreement over asset values Systemic Risk Monitoring Tools

10 Growth Rate IndicatorsCredit-GDP Gap: detrended private credit to GDP ratio Property-price Gap: detrended property-price index (both residential and commercial) Equity-price Gap: detrended market capitalizations of listed companies Detrending usually with HP filter, one-sided, λ = 1600 Systemic Risk Monitoring Tools

11 Systemic Risk Monitoring ToolsRules of Thumb for Gaps Credit-GDP Gap: exceeds 6% Equity Gap: exceeds 60% Property Gap: exceeds 15-25% Need both Credit-GDP Gap and either Equity Gap or Property Gap in their “red zones” to signal future distress Systemic Risk Monitoring Tools

12 Systemic Risk Monitoring ToolsBubble Indicators Many indicators from equity markets Some generalizable to all asset prices Nonparametric statistics (duration dependence, Markov chains) can also identify potential departures of prices from “fundamentals” (See Jurasakuldech et al 2006) Systemic Risk Monitoring Tools

13 Dispersion of ValuationsHistorically, equity price bubbles seem to be characterized by pronounced increases in price in one or a few industries This implies that the dispersion of valuations increases during a bubble, as one favored industry’s firms experience dramatic price increases relative to other industries Examples: housing and construction in Japan, 1980s, IT in the US, 1990s

14 Valuation Dispersion measureForm market-to-book ratios for each industry for the past year, as well as a longer benchmark period Form the standard deviation of the one-year mkt/book ratios and divide this by the full-period standard deviation of mkt/book A ratio significantly greater than 1 indicates higher dispersion, which suggests that prices may be tending toward a bubble (WEO, Apr 2003)

15 Simpler Dispersion MeasureClusters of P/E, P/S ratios may be easier to calculate than Market-Book ratios Example: in early 2014, nearly 15% of the Russell 3000 index companies had P/E ratios over 100 Firms were concentrated in biotech, social media and software industries

16 Bubbles and Turnover In addition to high prices, bubbles tend to feature high levels of trading Example: In November 2013, the daily volume of trading in Tesla shares was 10% of shares outstanding For IBM, annual trading volume was about equal to the number of shares outstanding

17 Test for Explosive Price Growth“Sup ADF” test: Run a standard augmented Dickey-Fuller test regression on a series of prices and calculate the ADF test statistic Add one data point at a time and continue to calculate the test statistic Compare ADF to log(log(TS)) for each S; when ADF > to log(log(TS)), the price is growing explosively Can identify “start” and “end” of bubbles

18 Systemic Risk Monitoring ToolsStress Tests Posit “stress scenarios” and simulate the consequences for solvency, liquidity, and ability to recover for individual institutions as well as entire system Systemic Risk Monitoring Tools

19 Systemic Risk Monitoring ToolsRisk Topography Brunnermeier, Gorton and Krishnamurthy (2010) Uses stress-test methodology as a way to constantly monitor the system Require all financial institutions to report sensitivities to a set of risk factors and scenarios Aggregate these response matrices Systemic Risk Monitoring Tools

20 Network Analysis MeasuresIdentifying systemically important institutions Network models Principal components analysis Systemic Risk Monitoring Tools

21 Identifying systemic institutionsCleveland Fed: Use size, contagion, correlation, concentration, and context. Three-tiered system to classify systemically important financial institutions.

22 Cleveland Fed Tier one: high-risk institutionsLarge, highly complex financial institutions Large interstate banks and multi-state insurance companies Tier Two: based on how connected they are, or their involvement in critical market activities, or how their condition may be affected by stress in the economy. Tier Three: low probability that a failure or stress would cause any widespread ripples throughout the financial system.

23 IFI Criteria G-20 Finance Ministers and Central Bank Governors: Guidance to Assess the Systemic Importance of Financial Institutions, Markets and Instruments: Initial Considerations (October, 2009)

24 Identifying systemic institutionsVolume of financial services provided by the individual component of the financial system Interlinkages where individual failure triggers domino effects Degree of complexity of financial institutions Leverage

25 Dodd-Frank’s CriteriaSize: Automatic Systemic Designation – Bank holding companies with $50 billion or more in assets. No opportunity for notice or appeal. By Designation: The Act establishes the Financial Stability Oversight Council, which will designate systemically important institutions Office of Financial Research (OFR) collects, analyzes and disseminates relevant information for anticipating future crises

26 How to assess systemic linkagesThose that draw inference from market data: De Bandt and Hartmann (2000), Hartmann, Straetmans and de Vries (2006) , Huang, Zhou, and Zhu (2009), Adrian and Brunnermeier (2009), Tarahev, Borio and Tsatsaroinis (2009), De Jonghe (2009), Acharya, Pedersen and others (2012- )

27 How to assess systemic linkagesThose relying on firm-specific default data such as default intensity models Gieseke et.al. (2010) Those relying on balance sheet data such as The contingent claims approach Lehar (2005), Gray, Merton, and Bodie ( ), Gray and Jobst (2009) Network analysis

28 Systemic Risk Monitoring ToolsNetwork Analysis Construct a matrix of (gross) inter-institutional exposures Simulate the impact of institutional default(s) on other institutions and track domino effects This is probably the ground-up approach that most people have in mind Systemic Risk Monitoring Tools

29 Extensions of Network AnalysisApplying Granger causality tests to interconnections Treating international subsidiaries as separate institutions for network analysis purposes Systemic Risk Monitoring Tools

30 Probability Distribution MeasuresMahalanobis Distance CoVaR CoRisk Multivariate Density Systemic Risk Monitoring Tools

31 Systemic Risk Monitoring ToolsMahalanobis Distance Standard statistical technique used to identify “uncharacteristic” behavior in data Applied to asset prices to find “financial turbulence” dt = (yt – m)’ Σ-1 (yt – m) where y = vector of returns, m = average returns, Σ = sample covariance matrix of returns “Turbulent” observations are days on which d is above the 75th percentile Systemic Risk Monitoring Tools

32 Using Mahalanobis DistanceConstruct VaRs using only “turbulent” observations of returns to create the “turbulence-adjusted VaR” This appears to give a better estimate of maximum portfolio losses during crises Also, turbulence appears to be highly persistent, and returns to risk-bearing are substantially lower during turbulent periods Systemic Risk Monitoring Tools

33 Systemic Risk Monitoring ToolsCoVaR and Co-Risk Use value at risk ideas to measure impact of one firm’s distress on measures of other firms’ extreme losses (CoVaR) or default risk (Co-Risk) CoVaR tells what happens to one firm’s (or financial system’s) VaR when a particular firm is at its extreme asset return VaR Co-Risk tells what happens to one firm’s (or financial system’s) extreme CDS spread when another firm is at its extreme CDS spread Systemic Risk Monitoring Tools

34 Systemic Risk Monitoring ToolsMultivariate Density Estimate the joint density of financial institutions’ asset returns, using minimum cross-entropy method Enables users to estimate joint probabilities of default, distress dependencies (conditional probabilities of institution i distress given institution j distress) Systemic Risk Monitoring Tools

35 Default and Contingent Claims MeasuresDefault Intensity Modeling Contingent Claims Analysis Option iPoD Systemic Risk Monitoring Tools

36 Default Intensity ModelingAssumes that there is an average default rate, but whenever a default occurs, the default rate jumps up temporarily and decays at some rate Captures the clustering of default behavior during financial stress First step: estimate the parameters of the default intensity equation, given historical data on defaults Systemic Risk Monitoring Tools

37 Output: VaR for DefaultsUse the estimated parameters from the default intensity model in a Monte Carlo simulation of default behavior Generate a distribution of defaults and use this to find the VaR critical values for the number of defaults Systemic Risk Monitoring Tools

38 Contingent Claims Analysis (CCA)Uses the insight that the equity in a leveraged corporation is equivalent to a long call option on the firm’s assets Uses option pricing equations to estimate the value of assets and their standard deviation, which imply the distance to default in terms of standard deviations above default Need publicly-traded financial institutions in order to implement this approach Systemic Risk Monitoring Tools

39 Fannie Mae Distance to Default (Std Dev) Estimated by CCASenior Debt Credit Risk Models

40 Systemic Risk Monitoring ToolsOption iPoD Uses option prices on a firm’s equity, and equity prices, to generate a probability of default Estimates the density of the asset value by minimizing the cross-entropy between the prior density and the posterior density Must have at least 2 option contracts on the equity in order to estimate Systemic Risk Monitoring Tools

41 Fixed Income SecuritiesMeasures of Liquidity Bid-ask spreads Order Book Depth = average quantity of securities available for sale at best bid and ask prices Estimated price impact per $100 million of net order flow obtained from regressions of 5-minute price changes on 5-minute net order flow Trade size = moving average of trade size of on-the-run issues of individual maturities Fixed Income Securities

42 Other Liquidity Measures“Noise,” or dispersion of yields around a smoothed yield curve = average absolute distance between actual yields and predicted yields Idea: large differences suggest unexploited profit opportunities, which may indicate constraints on market making or poor liquidity Fixed Income Securities

43 Liquidity-isolating SpreadsFind two types of bonds that differ only in liquidity due to issue size or trading turnover The spread is arguably the liquidity premium Example: U.S. Treasuries and Refcorp (Resolution Funding Corporation) bonds with similar cash flows Fixed Income Securities

44 Fixed Income Securities

45 Fixed Income Securities

46 Fixed Income Securities

47 Fixed Income Securities

48 Fixed Income Securities

49 Fixed Income Securities

50 Systemic Risk Monitoring ToolsQualitative Measures Based on Chami, Fullenkamp, Sharma (2010) framework for financial market development Views financial instruments as contracts Contracts contain important compromises between the counterparties, but these compromises change over time Significant changes in these compromises can signal increased likelihood of future stress Systemic Risk Monitoring Tools

51 Contracts and the Global Financial CrisisChanges to the Mortgage contract: option ARMs, teaser rates, 40-year maturity Changes to bond contracts: covenant-lite lending, PIK-toggle bonds Changes to MBS contract: CDO-squared, synthetic CDOs Systemic Risk Monitoring Tools

52 Systemic Risk Monitoring ToolsEx-Ante Measures Early Warning Indicators Stress Tests Network Analysis Default Intensity “Forward” CoVaR (to be presented later) Systemic Risk Monitoring Tools

53 Contemporaneous MeasuresCoVaR and Co-Risk Contingent Claims Analysis Option iPoD Multivariate Density Estimates Mahalanobis Distance Liquidity Measures Systemic Risk Monitoring Tools

54 Details on the Network ApproachSystemic Risk Monitoring Tools

55 Network Metrics CentralityOut Degree number of links leaving a node In Degree the number of links into a node Distance the number of links to go from node i to node j Diameter maximum distance in the network Path Length distance from a node to other nodes in the network. Clustering Coefficient probability that 2 nodes that are linked to another node are linked to each other . Conectivity ratio of links to total potential links

56 Examples for Banking NetworksSource: Upper and Worms (2004), Boss et al. (2004)

57

58

59 Implications Cross-border bank network has become more tightly connected (higher connectivity, shorter average path length, higher average degree, and higher clustering coefficient) Crises have not reversed the strong globalization process begun in early 1990s Once crisis occurs in one country, global impact could be more significant because countries are exposed to greater number of countries

60 Network Analysis ExampleTo analyze a hypothetical credit shock, Espinosa and Solé simulate the individual default of each institution’s cross-border interbank claims and then track the domino effects triggered by this event. Specifically, it is assumed that a bank’s losses are fully absorbed by its capital. A bank fails when its capital is not sufficient to fully cover its interbank losses.

61 Network Analysis: AlgorithmsParameters: λ is loss-given default, δ is asset price haircut, ρ fraction of short-term funding not rolled-over

62 Network Analysis: Tracking shocksInstitution i fails Assess the impact of parameter assumption including on LGD & haircut for each institution: is capital ≤0 ? Institution j fails Compile list of all failed institutions up to this point and re-start algorithm for remaining (non-failed) institutions   Institution k Institution m Institution n fails

63 Network Analysis: from Espinosa-Solé (2009)We start from the following stylized balance sheet identity: Where x = interbank lending a = bank i’s other assets k = bank i’s capital b = other borrowing d = deposits

64 Network Analysis: Tracking shocksWe consider two shocks: (i) a credit shock Pre-Shock Balance Sheet Post-Shock Balance Sheet

65 Network Analysis: Tracking shocks(ii) and a funding shock

66 Network Analysis Bank 1 Bank 1 Bank 1 Bank 1 New failure trigger bankfailures . . . Bank 3 New failure . . . . Bank N-1 Bank N-1 Bank N-1 Bank N-1 Bank N Bank N Bank N Bank N Trigger failure (initializes algorithm) Contagion rounds (algorithm internal loop) Final failures (algorithm converges)

67 Estimating Interbank ExposuresResearchers have relied on a number of sources to complete the matrix X: Bilateral exposures from balance sheet data Data from credit registers Data from payment systems Estimation by maximum entropy

68 Network Analysis: Main FindingsAllows tracking of potential contagion paths:

69 Network Analysis: Main FindingsIt is possible to quantify amount of potential capital losses at institutional level (stress-tests’ second round effects).

70 Network Analysis: Main Findings

71 Japan: Contagion Paths December 2008Trigger Country Affected Countries Contribution to Final Capital Impairment (percent of initial capital) Japan's Total Capital Impairment France -9.7 -14.5 Belgium -1.4 Netherland -3.4

72 Japan: Contagion Paths December 2008Trigger Country Affected Countries Contribution to Final Capital Impairment (percent of initial capital) Japan's Total Capital Impairment

73 Japan: Contagion Paths December 2008Trigger Country Affected Countries Contribution to Final Capital Impairment (percent of initial capital) Japan's Total Capital Impairment

74 Japan: Contagion Paths December 2008Trigger Country Affected Countries Contribution to Final Capital Impairment (percent of initial capital) Japan's Total Capital Impairment

75 Contagion/Vulnerability IndexBank i’s contagion index: the sum of all the capital losses in the system (except for the trigger bank) divided by the sum of the capital of all banks. Bank i’s vulnerability index: the simple average of percentage of capital losses suffered by a country in all the simulations.

76 Contagion/vulnerability indexFirst set: on-balance sheet data (immediate exposures) Indirect contagion effects: Include risk transfer data z stands for ultimate exposure of bank j to bank i, x is immediate exposure of bank j to bank i, τ is risk transfer from bank h to j referenced on i.

77 Contagion and Vulnerability IndexesFigure 6. Vulnerability and Contribution to Contagion of Nordic Banks

78 Network Analysis: indirect contagion effectsKey to include off-balance sheet data in analysis, as it can dramatically alter the risk map (March 2008)

79 Evaluation of Network ModelsThe analysis is very clear about Data Parameters such as loss-given default, haircut and other assumptions But abstracts from many potentially important considerations: Even rating agencies include considerations for potential government support Not all institutions have the same likelihood of failure

80 Systemic Risk Monitoring ToolsSummary Dozens of choices exist for monitoring systemic risk Choose a variety of measures that are best suited to your economy and data Don’t overlook the power of simple measures of systemic risk Systemic Risk Monitoring Tools