## _wp1246

## Source details

**Canonical URL:** [_wp1246](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp1246.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp1246.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp1246.pdf.json)

---

### Introduction and scope
- Study sample: largest banks in a sample of 18 countries, using data spanning the 2001-2009 period.
- Objective: analyze empirical drivers of tail-risk interdependence (systemic contagion) from an individual institution to the international financial system, with focus on large-scale, complex institutions typically considered too-big-to-fail.
- Methodological foundation: builds on the CoVaR methodology (Adrian and Brunnermeier (2009)) and generalizes it to account for international banks and asymmetric tail dependence.

### Main empirical findings
- Wholesale funding as primary driver
  - Short-term wholesale funding is identified as the most significant balance sheet determinant of individual contributions to global systemic risk.
  - Quantified effect: an increase of one percentage point in short-term wholesale funding leads to an increase in the contribution to systemic risk of 40 basis points of quarterly asset returns.
  - Policy implication referenced: supports regulatory initiatives to increase bank liquidity buffers and address asset-liability maturity mismatches (example preserved): liquidity coverage ratio under Basel III requires banks to maintain sufficient liquid assets to contain a 100% run-off of unsecured wholesale funding provided by financial institutions during a 30-day stress scenario, contrasting with the 5 to 10% run-off assumed for retail deposits during a significant liquidity stress episode.
- Asymmetry and balance-sheet contraction
  - Sensitivity of left-tail global returns to a shock in an institution’s market-valued asset returns is on average about 0.3.
  - Elasticity conditional on an institution having a shrinking balance sheet is almost three times larger than the unconditional sensitivity.
  - Implication: controlling for balance-sheet contraction is crucial for ranking institutions by systemic risk contribution.
- Crisis and recapitalization effects
  - Restricting attention to balance-sheet contraction episodes:
    - The credit crisis added up 0.1 percentage points to the co-movement between individual and global asset returns.
    - Recapitalization during the crisis period dampened co-movement by 0.2 percentage points.
  - Timing of recapitalization matters:
    - Banks receiving prompt recapitalization in Q4 2008 improved their relative position during the crisis period.
    - Banks rescued later in Q4 2009 became relatively more systemic during the crisis period.
  - Marginal contribution dynamics:
    - The marginal contribution of an individual bank to overall systemic risk increases from 0.76 quarterly percent returns in an average quarter to 0.92 in a quarter characterized by money market turbulence.
- Additional notes
  - Within the class of large-scale banks, neither relative size nor leverage provides strong predictive power for future systemic risk once short-term wholesale funding is accounted for.
  - Short-term wholesale funding increases interbank interconnectedness, maturity mismatch exposure, and vulnerability to liquidity risk, amplifying systemic contagion channels.
  - Importance of incorporating asymmetric patterns (expanding vs. contracting balance sheets) when measuring tail-risk dependence and systemic contributions.

### CoVaR framework and motivation (Section 4)
- Builds on the CoVaR methodology proposed by Adrian and Brunnermeier (2009) to generate time-varying estimates of systemic risk contribution for each bank.
- Extensions relative to prior work:
  - Focus on an international sample of large banks.
  - Account for asymmetric responses, recapitalization effects, and structural changes during the global financial crisis.
- Rationale for CoVaR:
  - Captures contagion under balance sheet deleveraging.
  - Informative about dynamics of each bank’s systemic contribution and allows analysis of effects of observable variables on that latent process.
  - Can incorporate crisis indicators and recapitalizations, and generalize to non-linear/asymmetric patterns.
- Empirical claim: marginal effects of individual banks on the global system differ economically and statistically in good and bad times.

### Data, sample and state variables
- Final sample: a total of 54 large firms from 18 countries, starting in July 2001 and ending in December 2009.
- Initial sample began with 93 banks but constrained to 54 due to data limitations.
- Average size of the representative bank: USD 830 billion and accounts for 57.7 percent of domestic GDP.
- Regional grouping and predictor strategies:
  - “2-Regions”: group banks (48 out of the 54) into U.S. + Canada and Europe; excludes 6 banks in Asia, Africa and Australia.
  - “1-Region”: use U.S. state variables as common conditioning variables for all 54 banks.
- State variables (sampled weekly) — U.S. variables include VIX index; liquidity spread (3-month U.S. repo rate minus 3-month U.S. T-bill yield); change in U.S. Treasury bill secondary market 3-month rate; change in slope (10-year minus 3-month); change in credit spread (10-year Moody’s Baa minus 10-year U.S. Treasury); S&P 500 Composite Index return. European counterparts similarly defined.
- Balance-sheet drivers (quarterly/semi-annual from Bloomberg): leverage, market-to-book ratio, short-term wholesale funding, relative size, marketable securities.
- Dummy variables capture timing of recapitalizations (Appendix B).

### Modelling and estimation approach
- Two-stage procedure:
  - Stage 1: estimate time-varying conditional VaR dynamics of individual banks and the global system; construct CoVaR and ΔCoVaR estimates using quantile regression for the 5% quantile on weekly returns.
  - Stage 2: treat CoVaR estimates as proxies for unobservable systemic risk and analyze determinants via panel-data and pooled cross-sectional regressions.
- Portfolio construction:
  - Individual bank returns: weekly simple returns from a portfolio formed by market-valued total assets.
  - Global system portfolio for institution i: weighted average of returns of the remaining banks in the sample (bank i excluded).
  - Weighting variable used in results: lagged total assets.
- Addressing leverage frequency mismatch:
  - Assumption 1: leverage ratio approximately constant within quarter/semester (constant approach) — primary results use this.
  - Assumption 2: cubic spline interpolation to weekly frequency (smoothing approach) — used for sensitivity checks.
- CoVaR specifications:
  - Baseline (symmetric) CoVaR model and Asymmetric CoVaR allowing different responses depending on sign of bank returns.
  - Extended specification includes crisis dummy Dt (equals one after September 2008) and recapitalization indicators.

### Key estimation and empirical results (first-stage and CoVaR outputs)
- First-stage QR fit (1 Region, 438 weekly observations, 54 banks):
  - Baseline pseudo-R2: 0.551; Asym. pseudo-R2: 0.635; Asym. Ext. pseudo-R2: 0.641.
  - Selected median coefficients (Baseline): Constant: -0.023 (-2.41); Volatility: -0.001 (-3.55); Liquidity Spread: -0.056 (-2.72); ΔTbill: -0.087 (-1.59); ΔCredit Spread: -0.102 (-1.92); Market Return: 0.005 (4.32); 1i,t-1 X: 0.270 (2.40).
- First-stage QR fit (2 Regions, 438 weekly observations, 48 banks):
  - Baseline pseudo-R2: 0.570; Asym. pseudo-R2: 0.639; Asym. Ext. pseudo-R2: 0.644.
  - Selected median coefficients (Baseline): Constant: -0.041 (-3.51); Volatility: -0.001 (-2.02); Market Return: 0.003 (2.26); 1i,t-1 X: 0.321 (2.38).
- Asymmetric specification findings:
  - Allowing for asymmetric effects substantially improves model fit as measured by pseudo-R2.
  - Median of estimates: δ_{λ,i}^- is almost three times larger than δ_{λ,i}^+.
- Selected bank-level asymmetric coefficients (asset-weighted system returns, two regions) — preserve exact entries:
  - ING — X_t<0: 1.677; X_t≥0: 0.223; (X_t<0)*Crisis: -1.129; (X_t<0)*Recap: 0.047; Pseudo-R2: 0.737
  - HSBC — 1.575; 0.349; 0.619; —; 0.667
  - SVK — 1.306; 0.160; -0.511; —; 0.673
  - POP — 1.113; 0.208; -0.056; —; 0.699
  - DBK — 1.097; 0.024; 0.070; —; 0.735
  - LLOY — 1.083; 0.034; -0.427; -0.611; 0.579
  - RBS — 1.072; 0.020; -0.421; -0.010; 0.657
  - CSGN — 1.025; 0.187; -0.532; —; 0.648
  - NDA — 1.042; 0.126; -0.254; —; 0.696
  - CBK — 1.040; 0.141; -0.387; 0.195; 0.697
  - DEA — 1.039; 0.208; -0.167; —; 0.717
  - BMPS — 0.983; -0.037; 0.047; -0.321; 0.696
  - (Table continues with exact values for remaining banks; selected final row for JPM: -0.032; 0.277; -1.578; 0.556)
- Systemic risk contribution to quarterly asset returns (ΔCoVaR) — selected overall top entries:
  - STT: -2.224
  - C: -1.667
  - BBT: -1.521
  - BIN: -1.421
  - BK: -1.348
  - Crisis-period top entries include STT: -3.249; BIN: -3.180; PNC: -2.483; TD: -2.033; KBC: -2.004.

### Determinants of systemic risk — panel regressions (Tables 5 and 6)
- Baseline asymmetric panel (Table 5; One-way; 1,280 observations):
  - Dependent dynamics and coefficients (selected, preserve exact numbers/significance):
    - Constant: -0.058
    - ΔCoVaR_it-1: 0.843 ***
    - VaR_it-1: -0.095 ***
    - Leverage_it-1: -0.002
    - WSF_it-1: -0.410 ** / -0.410 ***
    - Size_it-1: 2.440
    - MTB_it-1: -0.000
    - Mktb_it-1: -0.048
  - Bank Dummies: Yes; Time Dummies: Yes; R2 (%): 78.98.
  - Notes: All ΔCoVaRs estimated using percentile 5. Significance: * 10%, ** 5%, *** 1%.
- Asymmetric model with Recapitalization (Table 6; One-way; 1,280 observations):
  - Added Recap_it and interactions; selected coefficients:
    - Constant: -0.081
    - ΔCoVaR_it-1: 0.845 ***
    - VaR_it-1: -0.097 ***
    - Leverage_it-1: -0.002
    - WSF_it-1: -0.416 ** / -0.416 ***
    - Size_it-1: 2.433
    - MTB_it-1: -0.002
    - Mktb_it-1: -0.033
    - Recap_it: 0.383
    - Recap_it x WSF_it-1: 2.508
    - Recap_it x Mktb_it-1: -2.543 *
  - Bank Dummies: Yes; Time Dummies: Yes; R2 (%): 79.10.
- Main regression conclusion: wholesale funding (WSF_it-1) appears as a robust determinant of systemic risk; relative size, leverage and marketable assets are not significant at conventional confidence levels in these specifications.

### Robustness checks (selected)
- Alternative VaR estimation:
  - Parametric GARCH(1,1) on conditionally demeaned returns produced results remarkably similar to the quantile regression (QR) approach for individual VaR dynamics.
- Alternative systemic-risk measures:
  - Using Marginal Expected Shortfall (MES_it) and regressing on lagged accounting ratios again found short-term wholesale funding a significant predictor (significant at the 10% benchmark).
- Alternative ΔCoVaR definitions and estimation/specification checks (two-way clustering, GMM, inclusion of macro variables, symmetric CoVaR, cubic-spline smoothing of balance-sheet data, system weights using liabilities) left the main conclusion intact: short-term wholesale funding is the main driver of systemic risk.

### Policy-relevant implications and recommendations
- Measurement and regulation
  - Risk measurement frameworks that assume symmetric response to banks’ returns can significantly understate systemic risk; regulators should incorporate asymmetric CoVaR considerations.
  - Market volatility indices and market returns are strong predictors of downside risk and useful for early-warning and stress-testing frameworks.
  - CoVaR built on market-valued total assets aligns with regulatory concerns about balance-sheet deleveraging and is sensitive to recapitalization events.
- Supervisory priorities and Basel III relevance
  - For large international complex institutions, CoVaR-based evidence indicates short-term wholesale funding is positively and significantly related to systemic risk, while leverage and relative size do not add incremental information beyond wholesale funding.
  - Basel III systemic importance factors (size, interconnectedness, substitutability, global activity and complexity) find empirical support for penalizing excessive exposures to liquidity risk by focusing on short-term wholesale funding.
  - An optimal capital buffer structure for systemic banks could weight driving factors by relative importance.
- Additional regulatory measures suggested
  - Strengthen regulation by incentivizing disclosure of contingent short-term liabilities, in particular those related to possible margin calls under credit default swap contracts and repo funding.
  - The 2010 revamp of the Basel II capital framework to cover market risk associated with banks’ trading book positions is likely to decrease individual risk and contribute to mitigating systemic risk, especially for banks recapitalized during the crisis.

### Appendices — selected inventory highlights
- Appendix A: List of Financial Institutions — total listed entries: 26 (country-by-country listing preserved in source).
- Appendix B: List of recapitalizations — selected chronological entries (exact wording preserved):
  - ERS — Oct 30 2008 — Injection of €2.7 bn of non-listed, non-voting, non-transferable capital
  - KBC — Oct 27 2008 — Injection of €3.5 bn from the government, and €2.7 bn from the Flemish Regional Government
  - ING — Oct 21 2008 — Government capital injection of €10 bn
  - BAC — Jan 16 2009 — Capital injection of $20 bn from the TARP in exchange for preferred stock with 8% dividend
  - C — Nov 23 2008 — Capital injection of $20 bn from the TARP in exchange for preferred stock with 8% dividend
  - PNC — Oct 30 2009 — Capital injection of $7.6 bn from the TARP in exchange for preferred stock with 8% dividend
  - WFC — Oct 30 2010 — Redemption of $25 bn issued to the government under the TARP
- Selected market-variable summary statistics (Table 1a, U.S. weekly variables — exact figures):
  - Mean: CREDIT SPREAD 0.004; CHANGE TBILL -0.008; LIQ SPREAD 0.246; S&P RETURN -0.020; VIX 21.827; YIELD SPREAD 0.004
  - Median: -0.020; 0.000; 0.170; 0.147; 19.400; -0.010
  - Maximum: 0.830; 0.690; 1.140; 16.889; 72.916; 0.710
  - Minimum: -0.580; -0.790; -0.040; -26.537; 10.185; -0.554
  - Std. Dev.: 0.154; 0.118; 0.223; 2.964; 10.579; 0.161

*Source: _wp1246 - References; Section 4; conclusion; Appendices A and B.*

### References .............................................................................................................

### _wp1246 - References .............................................................................................................

### Introduction and scope
- Study sample: largest banks in a sample of 18 countries, using data spanning the 2001-2009 period.
- Objective: analyze empirical drivers of tail-risk interdependence (systemic contagion) from an individual institution to the international financial system, with focus on large-scale, complex institutions typically considered too-big-to-fail.
- Key methodological foundation: builds on the CoVaR methodology (Adrian and Brunnermeier (2009)) and generalizes it to account for international banks and asymmetric tail dependence.

### Main empirical findings
- Wholesale funding as primary driver
  - Short-term wholesale funding is identified as the most significant balance sheet determinant of individual contributions to global systemic risk.
  - Quantified effect: an increase of one percentage point in short-term wholesale funding leads to an increase in the contribution to systemic risk of 40 basis points of quarterly asset returns.
  - Policy implication: supports regulatory initiatives to increase bank liquidity buffers and address asset-liability maturity mismatches (e.g., the liquidity coverage ratio under Basel III).
    - Liquidity coverage ratio detail preserved: requires banks to maintain sufficient liquid assets to contain a 100% run-off of unsecured wholesale funding provided by financial institutions during a 30-day stress scenario, contrasting with the 5 to 10% run-off assumed for retail deposits during a significant liquidity stress episode.
- Asymmetry and balance-sheet contraction
  - Evidence of asymmetric contribution to systemic risk depending on whether a bank’s balance sheet is expanding or contracting.
  - Sensitivity of left-tail global returns to a shock in an institution’s market-valued asset returns is on average about 0.3.
  - Elasticity conditional on an institution having a shrinking balance sheet is almost three times larger than the unconditional sensitivity.
  - Implication: controlling for balance-sheet contraction is crucial for ranking institutions by systemic risk contribution.
- Crisis and recapitalization effects
  - Restricting attention to balance-sheet contraction episodes:
    - The credit crisis added up 0.1 percentage points to the co-movement between individual and global asset returns.
    - Recapitalization during the crisis period dampened co-movement by 0.2 percentage points.
  - Timing of recapitalization matters:
    - Banks receiving prompt recapitalization in Q4 2008 improved their relative position during the crisis period.
    - Banks rescued later in Q4 2009 became relatively more systemic during the crisis period.
  - Marginal contribution dynamics:
    - The marginal contribution of an individual bank to overall systemic risk increases from 0.76 quarterly percent returns in an average quarter to 0.92 in a quarter characterized by money market turbulence.

### Additional analytical notes and implications
- Within the class of large-scale banks, neither relative size nor leverage provides strong predictive power for future systemic risk once short-term wholesale funding is accounted for.
- The paper emphasizes the role of short-term wholesale funding in increasing interbank interconnectedness, maturity mismatch exposure, and vulnerability to liquidity risk, thereby amplifying systemic contagion channels (including effects on interbank markets and money market mutual funds).
- The analysis underscores the importance of incorporating asymmetric patterns (expanding vs. contracting balance sheets) when measuring tail-risk dependence and systemic contributions.

*Source: _wp1246 - References .............................................................................................................*

### Section 4 lays out our CoVaR framework and estimation framework and shows the

### _wp1246 - Section 4 lays out our CoVaR framework and estimation framework and shows the

### CoVaR framework and motivation
- Builds on the CoVaR methodology proposed by Adrian and Brunnermeier (2009) to generate time-varying estimates of systemic risk contribution for each bank.
- Extensions relative to prior work:
  - Focus on an international sample of large banks.
  - Account for asymmetric responses, recapitalization effects, and structural changes during the global financial crisis.
- Rationale for choosing CoVaR:
  - Captures contagion under balance sheet deleveraging, aligning with regulatory concerns.
  - Informative about dynamics of each bank’s systemic contribution and allows analysis of effects of observable variables on that latent process.
  - Can incorporate crisis indicators and recapitalizations, and generalize to non-linear/asymmetric patterns.
- Empirical claim: marginal effects of individual banks on the global system differ economically and statistically in good and bad times.

### Data and sample
- Population and perimeter:
  - Financial system defined as a network of large, internationally active regulated banking institutions.
  - Analysis restricted to the regulated banking sector.
- Sample construction and characteristics:
  - Final sample: a total of 54 large firms from 18 countries, starting in July 2001 and ending in December 2009.
  - All variables measured in USD.
  - Initial sample began with 93 banks but constrained to 54 due to data limitations.
  - The average size of the representative bank is USD 830 billion and accounts for 57.7 percent of domestic GDP.
  - Among the final sample, 21 out of the 24 banks in a Financial Times list (November 2009) are included.
- Regional grouping and predictor strategies:
  - Two approaches for predictors:
    - “2-Regions”: group banks (48 out of the 54) into U.S. + Canada and Europe and use regional predictive variables; excludes 6 banks in Asia, Africa and Australia due to data limitations.
    - “1-Region”: use U.S. state variables as common conditioning variables for all 54 banks.
- State variables (sampled weekly):
  - U.S. variables: VIX index; liquidity spread (3-month U.S. repo rate minus 3-month U.S. T-bill yield); change in U.S. Treasury bill secondary market 3-month rate; change in slope of yield curve (10-year minus 3-month); change in credit spread (10-year Moody’s Baa minus 10-year U.S. Treasury); S&P 500 Composite Index return.
  - European counterparts: Euronext volatility index; difference between 3-month U.K. repo rate and 3-month U.K. T-bill yield; first difference of French 3-month interest rate; first difference of French yield slope (5-year minus 3-month); difference between Baa corporate bonds and 10-year German government bond; FTSE European stock index.
  - Data sources: Chicago Board Options Exchange, Federal Reserve Board’s H.15 Release, Datastream.
- Balance-sheet drivers (quarterly/semi-annual from Bloomberg):
  - Measures: leverage, market-to-book ratio, short-term wholesale funding, relative size, marketable securities.
  - Dummy variables included to capture timing of recapitalizations (Appendix B has details).

### Modelling and estimation approach
- Two-stage procedure:
  - Stage 1: measure time-varying conditional VaR dynamics of individual banks and the global system; construct CoVaR and ΔCoVaR estimates.
  - Stage 2: treat CoVaR estimates as proxies for unobservable systemic risk and analyze determinants via panel-data and pooled cross-sectional regressions.
- VaR and CoVaR definitions and choices:
  - VaR: λ-quantile of the conditional loss distribution for confidence level λ ∈ (0,1).
  - CoVaR: λ-quantile of system portfolio’s conditional loss distribution given the other portfolio’s VaR (equation (1)).
  - Focus on weekly horizon and the 5% quantile of the conditional loss distribution.
- Portfolio construction:
  - Individual bank returns: weekly simple returns from a portfolio formed by market-valued total assets.
  - Addressing leverage frequency mismatch:
    - Assumption 1: leverage ratio approximately constant within quarter/semester (constant approach).
    - Assumption 2: cubic spline interpolation of low-frequency leverage to weekly frequency (smoothing approach).
    - Primary results presented using the constant approach.
  - Global system portfolio for institution i: weighted average of returns of the remaining banks in the sample (bank i excluded to avoid spurious interdependence).
  - Weighting variables considered: lagged total assets and lagged book value of liabilities; results reported using total assets as the weighting variable.
- Quantile Regression (QR) estimation:
  - QR used to estimate 5% VaR processes, balancing tail focus and finite-sample identification issues.
  - Predictors include macro-financial state variables Zt and a crisis dummy Dt (equals one after September 2008).
  - General predictive QR model for 5% VaR: Yt = Zt-1' βλ + u_{t,λ} (equation (4)).
  - Augmenting QR to obtain CoVaR by including returns of the i-th bank and setting Yt to system returns conditional on bank returns.
- CoVaR specifications estimated:
  - Baseline (symmetric) CoVaR model: equation (5) with contribution captured via δ_{λ,i}; ΔCoVaR formula in (6).
  - Asymmetric CoVaR: allows different responses depending on sign of bank returns (equation (7)); yields forecast in (8).
  - Recapitalization and crisis time-effects included in extended specification (equation (9)) using indicators for negative returns during crisis and on recapitalization dates.

### Key estimation and empirical results
- Estimation samples:
  - “1-Region”: all 54 banks with U.S. state variables.
  - “2-Regions”: U.S. + Canada and Europe with regional predictors; excludes 6 banks in Asia, Africa, Australia.
- QR result highlights (summary reported; full individual results available upon request):
  - Market volatility index and market return exhibit the strongest predictive power for VaR/CoVaR.
  - Other predictors (changes in T-bill rate, liquidity spreads, credit spreads) are generally associated with larger one-period ahead VaR but their statistical significance is specification-sensitive.
  - Lagged returns of a potentially systemic bank are always significant and improve forecasting of the global system’s tail performance.
  - Allowing for asymmetric effects substantially improves model fit as measured by pseudo-R^2.
  - Asymmetric specification implication:
    - Symmetric models tend to underestimate the size of the link between a bank and its system portfolio, yielding conservative systemic risk predictions.
    - Median of estimates: δ_{λ,i}^- is almost three times larger than δ_{λ,i}^+ (text: "the median of the estimates of the coefficient ,i    is almost three times larger than ,i   ").
  - Time-effects related to the crisis provide moderate incremental gains over the asymmetric specification, with heterogeneity across banks.

### Policy-relevant implications (drawn from methodology and findings)
- Importance of accounting for asymmetries:
  - Risk measurement frameworks that assume symmetric response to banks’ returns can significantly understate systemic risk; regulators should incorporate asymmetric CoVaR considerations.
- Relevance of market volatility and market returns:
  - Market volatility indices and market returns are strong predictors of downside risk and can be useful in early-warning and stress-testing frameworks.
- Use of balance-sheet-based systemic measures:
  - CoVaR built on market-valued total assets aligns with regulatory concerns about balance-sheet deleveraging and provides a measure sensitive to recapitalization events.
- Sample and weighting practices:
  - Excluding the institution under analysis from the system portfolio is a useful adjustment to avoid mechanical correlation effects and potential overstatement of tail dependence; weighting by liabilities can be informative where interconnectedness via claims is central.

*Source: _wp1246 - Section 4 lays out our CoVaR framework and estimation framework and shows the (IMF working paper PDF content provided).*

### conclusion. In general terms, most banks in our sample have become more systemic in the

### _wp1246 - conclusion

### Key findings on systemic contributions and crisis dynamics
- In general terms, most banks in our sample have become more systemic in the aftermath of the global financial crisis.  
- On average, the systemic risk of a financial institution tends to decrease after a capital injection, pointing at the success of recapitalization programs in containing systemic risk.  
- Table 4a (as described) ranks banks by the asymmetric sensitivity of the system to negative bank returns and shows pronounced asymmetries:
  - Banks identified as among the most systemic under the asymmetric beta coefficient include ING, HSBC, Lloyds and Royal Bank of Scotland.
  - For ING, the coefficient on negative returns is around 8 times larger than the coefficient on positive returns.
  - The more systemic a bank is, the more asymmetric its contribution to overall systemic risk is; Figure 1 shows a strong relation between ranking position and size of the asymmetry.
  - Some banks are more systemic during crises while others are less so; two Canadian banks—Bank of Montreal and Toronto-Dom Bank—became very systemic during the crisis.
  - Recapitalizations had a very positive effect for several European institutions, particularly for Italian banks, suggesting government intervention helped mitigate systemic risk.
- Table 4b (average ΔCoVaR metric) observations:
  - Some large institutions, such as Citigroup, appear very systemic during the whole period.
  - On average across banks, contributions are 0.2 percentage points higher during the crisis period.
  - The ranking in the crisis period is influenced by the timing of public intervention:
    - Banks that received prompt recapitalization in Q4 2008 (Citigroup, Bank of America, ING, Commerzbank) improved their relative position during the crisis period.
    - Banks rescued later in Q4 2009 (RBS and Lloyds) became relatively more systemic during the crisis period.
- Cross-country comparisons (Tables 4c and 4d):
  - The system is most sensitive to Dutch banks in distress, essentially due to the large effect of ING on the system, especially before the crisis.

### Determinants of systemic risk — regression analysis (model and principal results)
- Baseline predictive regression (quarterly aggregation of weekly ΔCoVaRit estimates) with fixed effects follows equation (10) using a “2-Region” asymmetric specification accounting for crisis and recapitalization effects.
- Explanatory variables included (lagged at t-1):
  - Leverageit-1: total assets to equity ratio (higher leverage expected to relate negatively to solvency and thus negatively to ΔCoVaRit).
  - WSFit-1: total short-term borrowings to total assets ratio (proxy for short-term wholesale funding, interconnectivity, and liquidity risk; expected negative relation with ΔCoVaRit).
  - Sizeit-1: bank i total assets over total assets of all banks in sample (expected larger size → higher contribution).
  - MTB it-1: market-to-book ratio (could imply negative relationship with ΔCoVaRit under mispricing).
  - Mktbit-1: marketable securities to total assets ratio (proxy for available-for-sale/fair-value instruments; expected negative relation).
  - Bank j and Time k dummies: bank and time fixed effects.
- Main empirical conclusions reported in Table 5:
  - Wholesale funding (WSFit-1) appears as a robust determinant of systemic risk.
  - Relative size, leverage and marketable assets are not significant at any standard confidence levels in these specifications.
  - The evidence supports the view that riskier funding (short-term wholesale funding) is a key contributor to systemic risk rather than mere size.
- Economic channels explaining the key role of short-term wholesale funding:
  - Interconnectivity channel: wholesale funding in interbank unsecured markets increases balance-sheet interconnections and provides a direct contagion channel, raising likelihood of global domino effects.
  - Liquidity/maturity-mismatch channel: reliance on short-term funding creates maturity mismatch and vulnerability to liquidity risk, increasing fire-sale likelihood and externalities to other intermediaries holding similar assets.
- Recapitalization effects:
  - Model (10) extended with Recapit and interaction terms Recapit x Vit (Vit ∈ {Leverageit, WSFit, Sizeit, MTB it, Mktbit}) to capture cross-effects.
  - Results (Table 6): wholesale funding remains robust; among recapitalization cross-effects, only the size of marketable securities significantly increases systemic risk when a bank has been recapitalized, with the estimate significant at the 10% nominal size, lending some support to the fire-sales channel.

### Robustness checks (selected)
- Alternative VaR estimation:
  - Parametric GARCH (1,1) on conditionally demeaned returns produced results remarkably similar to the quantile regression (QR) approach for individual VaR dynamics.
- Alternative systemic-risk measures:
  - Using the Marginal Expected Shortfall (MESit) approximating SES (average global market returns during the 5% worst days of bank i per quarter) and regressing on lagged accounting ratios again found short-term wholesale funding a significant predictor (significant at the 10% benchmark, though weaker).
- Alternative ΔCoVaR definitions:
  - Using the Adrian and Brunnermeier (2009) ΔCoVaR definition and other ΔCoVaR definitions yielded similar determinant results.
- Estimation and specification checks:
  - Corrections for standard errors with time effects, two-way clustering with bank and country dummies, GMM estimation (with all independents as instruments), inclusion of macro variables (unemployment and interest rates), symmetric CoVaR specification, cubic-spline smoothing of balance sheet data, and construction of system weights using liabilities all left the main conclusion intact: short-term wholesale funding is the main driver of systemic risk.

### Concluding policy implications and recommendations
- For large international complex institutions (those likely seen as too-big-to-fail), CoVaR-based evidence indicates short-term wholesale funding (linked to interconnectedness and liquidity risk) is positively and significantly related to systemic risk, while leverage and relative size do not add incremental information beyond wholesale funding.
- Asymmetric responses matter: the sensitivity of system returns to individual bank returns is much higher during balance-sheet deleveraging, reinforcing the need to account for asymmetries in systemic-risk assessment.
- Relevance for Basel III framework:
  - Basel III categorizes systemic importance factors as size, interconnectedness, substitutability, global activity and complexity; empirical results formally support penalizing excessive exposures to liquidity risk by focusing on short-term wholesale funding.
  - An optimal capital buffer structure for systemic banks could weight driving factors by relative importance.
- Additional regulatory measures suggested:
  - Strengthen regulation of systemic risk by incentivizing disclosure of contingent short-term liabilities, in particular those related to possible margin calls under credit default swap contracts and repo funding.
  - The 2010 revamp of the Basel II capital framework to cover market risk associated with banks’ trading book positions is likely to decrease individual risk and contribute to mitigating systemic risk, especially for banks recapitalized during the crisis.

*Source: _wp1246 - conclusion (IMF working paper content supplied).*

### Appendix A. List of Financial Institutions

### Appendix A. List of Financial Institutions

### List by country
- AUSTRIA
  - ERSTE GROUP BANK — EBS AV
- AUSTRALIA
  - COMMONW — BK AUSTR CBA AU
  - NATL AUST BANK — NAB AU
  - WESTPAC BANKING — WBC AU
- BELGIUM
  - KBC GROEP — KBCB PZ
- BRITAIN
  - BARCLAYS PLC — BARC LN
  - HSBC HOLDINGS — HSBC LN
  - LLOYDS BANKING — LLOY LN
  - ROYAL BK SCOTLAND — RBS LN
  - STANDARD CHARTER — STAN LN
- CANADA
  - BANK OF MONTREAL — BMO CN
  - BANK OF NOVA SCO — BNS CT
  - CAN IMPL BK COMM — CM CT
  - ROYAL BANK OF CA — RY CT
  - TORONTO-DOM BANK — TD CT
- DENMARK
  - DANSKE BANK A/S — DANSKE DC
- FRANCE
  - BNP PARIBAS — BNP FP
  - SOC GENERALE — GLE FP
- GERMANY
  - COMMERZBANK — CBK GR
  - DEUTSCHE BANK-RG — DBK GR
- IRELAND
  - ALLIED IRISH BK — ALBK ID
- ITALY
  - BANCA MONTE DEI — BMPS IM
  - INTESA SANPAOLO — ISP IM
  - UNICREDIT SPA — UCG IM
- JAPAN
  - DAIWA SECS GRP — 8601 JT
  - NOMURA HOLDINGS — 8604 JT
- NETHERLANDS
  - ING GROEP NV-CVA — INGA NA
- NORWAY
  - DNB NOR ASA — DNBNOR NO
- SOUTH AFRICA
  - STANDARD BANK GR — SBK SJ
- SPAIN
  - BBVA — BBVA SM
  - BANESTO SA — BTO SM
  - BANCO POPULAR — POP SM
  - BANCO SANTANDER — SAN SM
- SWEDEN
  - NORDEA BANK AB — NDA SS
  - SEB AB-A — SEBA SS
  - SVENSKA HAN-A — SHBA SS
  - SWEDBANK AB-A — SWEDA SS
- SWITZERLAND
  - CREDIT SUISS-REG — CSGN VX
  - UBS AG-REG — UBSN VX
- UNITED STATES
  - BANK OF AMERICA — BAC UN
  - BB&T CORP — BBT UN
  - BANK NY MELLON BK — (abbreviation not provided)
  - CITIGROUP INC — C US
  - CAPITAL ONE FINA — COF UN
  - GOLDMAN SACHS GP — GS UN
  - JPMORGAN CHASE — JPM US
  - MORGAN STANLEY — MS UN
  - PNC FINANCIAL SE — PNC UN
  - REGIONS FINANCIA — RF UN
  - SLM CORP — SLM UN
  - SUNTRUST BANKS — STI UN
  - STATE ST CORP — STT UN
  - US BANCORP — USB US
  - WELLS FARGO & CO — WFC UN

### Key statistics and counts
- Total listed entries: 26

*Appendix A. List of Financial Institutions*

### Appendix B: List of recapitalizations

### Appendix B: List of recapitalizations

### List of bank recapitalizations (chronological entries as provided)
- ERS — Oct 30 2008 — Injection of €2.7 bn of non-listed, non-voting, non-transferable capital
- KBC — Oct 27 2008 — Injection of €3.5 bn from the government, and €2.7 bn from the Flemish Regional Government
- BARC — Sept 16 2009 — Sale of $12 bn of risky credit assets to a special purpose vehicle
- Lloyds — Sep 18 2008 — Competition rules waived to allow the merger with HBOS
- Lloyds — Oct 19 2008 — The government injected £4 bn of preference shares
- RBS — Jan 19 2009 — The government swapped preferred shares for ordinary shares worth £5 bn
- RBS — Feb 26 2009 — The bank received £13 bn in additional capital for a participation fee of £6.5 bn
- RBS — Nov 3 2009 — The authorities announced an additional injection of £25.5 bn shoring up the gov stake to 84 %
- BNP — Oct 22 2008 — The bank issued hybrid subordinated debt for €2.55 bn
- BNP — March 1 2009 — The French banking plan purchased €5.1 bn of non-voting shares; the hybrid debt was redeemed
- SGE — Oct 22 2008 — The bank issued hybrid subordinated debt for €1.7 bn
- CBK — Nov 4 2008 — The government announced an injection of €8.2 bn with a further injection of €10 bn
- ALBK — Feb 11 2009 — Injection of €3.5 bn of tier I capital
- UC — March 18 2009 — The bank issued €4.0 bn of government capital instruments
- BIN — March 20 2009 — The bank announced the issuance of €4 bn of subordinated debt subscribed by the government
- BMPS — March 27 2009 — The bank announced the issuance of €1.9 bn of special bonds subscribed by the government
- ING — Oct 21 2008 — Government capital injection of €10 bn
- BAC — Jan 16 2009 — Capital injection of $20 bn from the TARP in exchange for preferred stock with 8% dividend
- C — Nov 23 2008 — Capital injection of $20 bn from the TARP in exchange for preferred stock with 8% dividend
- C — (subsequent) — Further issuance of $7 bn of preferred stock to the Treasury and the FDCI
- COF — Oct 30 2008 — Capital injection of $3.55 bn from the TARP in exchange for preferred stock with 8% dividend
- PNC — Oct 30 2009 — Capital injection of $7.6 bn from the TARP in exchange for preferred stock with 8% dividend
- WFC — Oct 30 2010 — Redemption of $25 bn issued to the government under the TARP

*Source: Bloomberg, authorities' websites, and IMF*

### Key statistics and summary tables (selected)
- Table 1a. U.S. weekly market variables (summary statistics):
  - Variables: CREDIT SPREAD, CHANGE TBILL, LIQ SPREAD, S&P RETURN, VIX, YIELD SPREAD
  - Mean: 0.004, -0.008, 0.246, -0.020, 21.827, 0.004
  - Median: -0.020, 0.000, 0.170, 0.147, 19.400, -0.010
  - Maximum: 0.830, 0.690, 1.140, 16.889, 72.916, 0.710
  - Minimum: -0.580, -0.790, -0.040, -26.537, 10.185, -0.554
  - Std. Dev.: 0.154, 0.118, 0.223, 2.964, 10.579, 0.161
  - Skewness: 1.475, -1.212, 1.731, -1.763, 1.772, 0.640
  - Kurtosis: 9.200, 15.083, 5.984, 22.104, 7.008, 5.331
  - 1st order autocorrelation: 0.284, 0.084, 0.874, -0.150, 0.972, -0.010

- Table 1b. European weekly market variables (summary statistics):
  - Variables: CREDIT SPREAD, CHANGE TBILL, LIQ SPREAD, FTSE RETURN, VOLAT INDEX, YIELD SPREAD
  - Mean: 0.000, 0.010, 0.037, 0.078, 25.890, 0.002
  - Median: 0.001, 0.000, 0.016, 0.232, 22.870, -0.001
  - Maximum: 0.793, 0.299, 0.831, 13.592, 81.030, 0.316
  - Minimum: -0.461, -0.948, -0.041, -25.130, 11.600, -0.294
  - Std. Dev.: 0.099, 0.089, 0.074, 3.417, 12.228, 0.079
  - Skewness: 1.309, -4.963, 4.873, -1.036, 1.421, 0.594
  - Kurtosis: 12.667, 41.238, 36.567, 7.504, 2.209, 3.038
  - 1st order autocorrelation: 0.296, 0.254, 0.863, -0.070, 0.949, 0.011

### First-stage regressions (median coefficients, 5% quantile regressions)
- Table 2. 1 Region (438 weekly observations, 54 banks) — baseline, asymmetric, asymmetric extended:
  - Baseline pseudo-R2: 0.551; Asym. pseudo-R2: 0.635; Asym. Ext. pseudo-R2: 0.641
  - Selected median coefficients (Baseline; t-statistics in parentheses):
    - Constant: -0.023 (-2.41)
    - Volatility: -0.001 (-3.55)
    - Liquidity Spread: -0.056 (-2.72)
    - ΔTbill: -0.087 (-1.59)
    - ΔSlope: 0.029 (0.77)
    - ΔCredit Spread: -0.102 (-1.92)
    - Market Return: 0.005 (4.32)
    - Crisis Dummy: -0.011 (-1.20)
    - 1i,t-1 X: 0.270 (2.40)

- Table 3. 2 Regions (438 weekly observations, 48 banks — Europe, and US+Canada):
  - Baseline pseudo-R2: 0.570; Asym. pseudo-R2: 0.639; Asym. Ext. pseudo-R2: 0.644
  - Selected median coefficients (Baseline):
    - Constant: -0.041 (-3.51)
    - Volatility: -0.001 (-2.02)
    - Liquidity Spread: -0.075 (-1.02)
    - ΔTbill: 0.054 (0.91)
    - ΔSlope: -0.061 (-1.28)
    - ΔCredit Spread: 0.034 (0.79)
    - Market Return: 0.003 (2.26)
    - Crisis Dummy: -0.015 (-1.39)
    - 1i,t-1 X: 0.321 (2.38)

### Systemic risk contribution: bank-level (Table 4a)
- Table shows asymmetric coefficients and related terms for each bank in the most general model (asset weighted system returns, two regions). Banks sorted by asymmetric coefficient on negative returns (X_t<0).
- Selected entries (preserve exact numbers and order as provided):
  - ING — X_t<0: 1.677; X_t≥0: 0.223; (X_t<0)*Crisis: -1.129; (X_t<0)*Recap: 0.047; Pseudo-R2: 0.737
  - HSBC — 1.575; 0.349; 0.619; —; 0.667
  - SVK — 1.306; 0.160; -0.511; —; 0.673
  - POP — 1.113; 0.208; -0.056; —; 0.699
  - DBK — 1.097; 0.024; 0.070; —; 0.735
  - LLOY — 1.083; 0.034; -0.427; -0.611; 0.579
  - RBS — 1.072; 0.020; -0.421; -0.010; 0.657
  - CSGN — 1.025; 0.187; -0.532; —; 0.648
  - NDA — 1.042; 0.126; -0.254; —; 0.696
  - CBK — 1.040; 0.141; -0.387; 0.195; 0.697
  - DEA — 1.039; 0.208; -0.167; —; 0.717
  - BMPS — 0.983; -0.037; 0.047; -0.321; 0.696
  - Remainder of table continues with exact values for other banks including RY, BNP, STI, UBS, SCH, STAN, BTO, KBC, SWED, C, UC, BBVA, TD, BARC, CM, BIN, DNB, BK, PNC, ALBK, SGE, USB, WFC, STT, DAB, BAC, RF, ERS, BMO, COF, MS, BNS, GS, BBT, SLM, JPM (JPM: -0.032; 0.277; -1.578; 0.556)

### Systemic risk contribution to quarterly asset returns (Table 4b)
- Rankings of ΔCoVaR for three periods: Overall Period (Q4-2001 to Q3-2009), Pre-crisis Period (Q4-2001 to Q2-2007), Crisis Period (Q3-2007 to Q1-2009).
- Overall top entries (ΔCoVaR):
  - STT: -2.224
  - C: -1.667
  - BBT: -1.521
  - BIN: -1.421
  - BK: -1.348
  - (table continues with exact ΔCoVaR values for all listed banks)
- Pre-crisis top entries:
  - BBT: -1.957
  - STT: -1.805
  - BK: -1.523
  - BNS: -1.308
  - (table continues)
- Crisis period top entries:
  - STT: -3.249
  - BIN: -3.180
  - PNC: -2.483
  - TD: -2.033
  - KBC: -2.004
  - (table continues)

### Systemic risk contribution by country (Tables 4c and 4d)
- Table 4c (average contribution by country; sorted by asymmetric coefficient on negative returns):
  - Netherlands — X_t<0: 1.677; X_t≥0: 0.223; (X_t<0)*Crisis: -1.129; (X_t<0)*Recap: 0.047; Pseudo-R2: 0.737
  - Germany — 1.069; 0.083; -0.158; 0.195; 0.716
  - Sweden — 1.053; 0.219; -0.308; 0.702
  - Britain — 1.038; 0.004; -0.150; -0.400; 0.621
  - Switzerland — 0.952; -0.019; -0.492; —; 0.658
  - Spain — 0.867; 0.242; -0.006; —; 0.685
  - Belgium — 0.832; 0.319; -0.160; -0.156; 0.737
  - Italy — 0.772; 0.009; 0.447; -0.762; 0.683
  - France — 0.731; 0.188; 0.398; -0.007; 0.741
  - (remaining countries: Norway, Ireland, Canada, Denmark, US, Austria with exact numbers as listed)

- Table 4d (average quarterly ΔCoVaR by country across periods):
  - Overall Period top countries (ΔCoVaR):
    - Netherlands: -1.042
    - Canada: -1.002
    - US: -0.972
    - Italy: -0.932
    - Germany: -0.919
    - Britain: -0.821
    - Belgium: -0.789
    - Sweden: -0.703
    - Switzerland: -0.640
    - Spain: -0.631
    - South Africa: -0.607
    - Norway: -0.574
    - Austria: -0.546
    - Denmark: -0.525
    - Ireland: -0.445
    - France: -0.397
    - Japan: -0.241
    - Australia: -0.190
  - Pre-crisis Period top countries:
    - Netherlands: -1.128
    - US: -0.991
    - Germany: -0.933
    - Canada: -0.848
    - Sweden: -0.753
    - Italy: -0.673
    - Spain: -0.622
    - Switzerland: -0.593
    - Ireland: -0.508
    - Austria: -0.476
    - Norway: -0.459
    - Belgium: -0.343
    - France: -0.337
    - South Africa: -0.292
    - Japan: -0.114
    - Australia: 0.026
  - Crisis Period top countries:
    - Belgium: -2.004
    - Italy: -1.739
    - South Africa: -1.509
    - Canada: -1.343
    - Denmark: -1.316
    - Britain: -1.115
    - US: -0.907
    - Norway: -0.876
    - Germany: -0.838
    - Switzerland: -0.716
    - Australia: -0.674
    - Netherlands: -0.610
    - Austria: -0.606
    - Spain: -0.582
    - Japan: -0.517
    - France: -0.500
    - Sweden: -0.474
    - Ireland: -0.324

### Systemic risk factors regressions (asymmetric model)
- Table 5: Asymmetric model (Panel; One-way)
  - Dependent variable dynamics and significant coefficients:
    - Constant: -0.058
    - ΔCoVaR_it-1: 0.843 ***
    - VaR_it-1: -0.095 *** (in one column) / -0.095 ** (in other column)
    - Leverage_it-1: -0.002
    - WSF_it-1: -0.410 ** / -0.410 ***
    - Size_it-1: 2.440
    - MTB_it-1: -0.000
    - Mktb_it-1: -0.048
  - Bank Dummies: Yes
  - Time Dummies: Yes
  - R2 (%): 78.98
  - Number of observations: 1,280
  - Notes: All ΔCoVaRs estimated using percentile 5. Significance: * 10%, ** 5%, *** 1%.

- Table 6: Asymmetric model controlling for Recapitalization (Panel; One-way)
  - Added variables involving Recap_it (recapitalization dummy) and interactions:
    - Constant: -0.081
    - ΔCoVaR_it-1: 0.845 ***
    - VaR_it-1: -0.097 ***
    - Leverage_it-1: -0.002
    - WSF_it-1: -0.416 ** / -0.416 ***
    - Size_it-1: 2.433
    - MTB_it-1: -0.002
    - Mktb_it-1: -0.033
    - Recap_it: 0.383
    - Recap_it x Leverage_it-1: -0.002
    - Recap_it x WSF_it-1: 2.508
    - Recap_it x Size_it-1: 0.854
    - Recap_it x MTB_it-1: 0.180
    - Recap_it x Mktb_it-1: -2.543 *
  - Bank Dummies: Yes
  - Time Dummies: Yes
  - R2 (%): 79.10
  - Number of observations: 1,280
  - Definition: Recap_it is a dummy equal to one when the bank is recapitalized and zero otherwise.

### Figure: Asymmetric systemic risk model (Figure 1)
- Figure plots the difference between δ– and δ+ from the 1st stage systemic risk model (equation 9) as a function of the coefficient size on individual banks' negative returns reported in Table 4a.
- X-axis: Bank Ranking (Table 4a) — e.g., 1 corresponds to ING (highest δ– coefficient), 10 corresponds to CBK.
- Y-axis: Difference Estimated Coefficients (δ– − δ+), range shown from -0.4 to 1.6 on the provided plot.

*Source: _wp1246 - Appendix B: List of recapitalizations (PDF).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp1246.pdf_