## _wp10153

## Source details

**Canonical URL:** [_wp10153](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2010/_wp10153.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2010/_wp10153.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2010/_wp10153.pdf.json)

---

### Introduction and purpose
- Objective: identify factors affecting credit default risk of selected European large complex financial institutions (LCFIs) to provide a fundamentals-based surveillance tool and potential early warning device.
- Rationale:
  - EWS models aim to detect underlying economic weaknesses and vulnerabilities and to trigger preemptive interventions.
  - LCFIs played a central role in the ongoing global financial crisis, motivating focused analysis of their business, funding models, risks, and spillovers.
- Approach: combine a standard structural credit risk model with financial soundness indicators to estimate determinants of credit default risk for European LCFIs.

### Credit risk measure and empirical motivation
- Dependent variable:
  - Credit Default Swap (CDS) spread (5-year maturity, yearly average of daily last price quotes from Bloomberg).
- Empirical support:
  - Regression reported: Recapitalization = 0.62(s.e. 0.09) CDS; R2 (adj.) = 0.73.
  - Figure evidence: positive correlation between pre-intervention CDS spreads and extent of needed recapitalization (percent of total assets).
- Claimed advantages of CDS spreads over corporate bond spreads:
  - CDS spread is a more direct measure of credit risk.
  - CDS spread does not seem to contain liquidity components or tax distortions that affect corporate bond spreads.
  - CDS spread leads in the credit risk price-discovery process and represents the upper bound of credit risk.

### Theoretical foundation and methodology
- Theoretical basis:
  - Structural approach based on Merton (1974); applicability to CDS markets following Benkert (2004).
  - Extensions include stock returns and slope of the yield curve to address the credit spread puzzle.
- Dynamic panel specification:
  - CDS_{i,t} = ρ CDS_{i,t-1} + X_{i,t}'β + Z_t'γ + u_{i,t}
  - First-differenced: ∆CDS_{i,t} = ρ ∆CDS_{i,t-1} + ∆X_{i,t}'β + ∆Z_t'γ + ∆ε_{i,t}
- Estimator:
  - Arellano-Bond GMM dynamic panel data estimator.
  - One-step estimator uses weighting matrix W1; Two-step uses optimal weighting matrix W2 based on one-step residuals.
  - Windmeijer (2005) finite sample correction applied to two-step standard errors.
- Justification for estimator:
  - Endogeneity of explanatory variables (e.g., ratings, variance of equity).
  - Need to remove bank-specific time-invariant fixed effects.
  - Instrument lagged dependent variable and regressors because financial market variables exhibit random walk behavior.
  - Panel with T=3, N=29 (T/N → 0) supports Arellano-Bond.

### Sample characteristics and descriptive statistics
- Sample: 29 European LCFIs over 2004–2008 (balanced panel); main estimations focus on 2006–2008 with 87 observations / 29 banks reported in model tables.
- Observed CDS evolution and variability:
  - Dependent variable CDS (basis points) by year: 2004: 13.8, 2005: 12.3, 2006: 9.6, 2007: 27.3, 2008: 108
  - Standard deviation of CDS by year: 2004: 3.9, 2005: 3.1, 2006: 2.7, 2007: 17.5, 2008: 47.4
  - Average CDS spread increased by a factor of 11 (from 9.6 in 2006 to 108 in 2008).
  - Variability (standard deviation) increased 18 times.
- Selected cross-section averages by year (highlights):
  - Tier I ratio: 2004: 8.3, 2005: 8.1, 2006: 8.3, 2007: 7.9, 2008: 8.4
  - Leverage (multiple of equity): 2004: 25.8, 2005: 27.1, 2006: 27.2, 2007: 27.8, 2008: 31.2
  - Z-score: 2004: 3.5, 2005: 3.5, 2006: 3.8, 2007: 3.7, 2008: 1.9
  - Efficiency Ratio (operating costs to revenues): 2004: 61.6, 2005: 58.9, 2006: 56.2, 2007: 60.6, 2008: 104.4
  - Trading income as percent of revenues: 2004: 13.98, 2005: 18.6, 2006: 19.38, 2007: 12.98, 2008: -41.24
  - ROA by year: 2004: 0.7, 2005: 0.7, 2006: 0.8, 2007: 0.7, 2008: 0.2
  - ROE by year: 2004: 16.32, 2005: 18.11, 2006: 19.86, 2007: 16.49, 2008: 0.19
  - Loans to Deposits Ratio: 2004: 128, 2005: 134, 2006: 138, 2007: 139, 2008: 150
  - Liquid Assets to Total Assets: 2004: 13, 2005: 11.9, 2006: 11.3, 2007: 10.4, 2008: 9.8
  - Options Implied Variance of Equity: 2004: 16.8, 2005: 18.3, 2006: 21.6, 2007: 25.6, 2008: 52.5
  - Slope of the Yield Curve: 2004: 1.85, 2005: 1.09, 2006: 0.83, 2007: 0.29, 2008: 0.26
  - Variance of Bank's Equity Price by year (as presented): 2004: 99, 2005: 41, 2006: 02, 2007: 114, 2008: 973345
  - GDP growth by year: 2004: 2.5, 2005: 2.8, 2006: 3.1, 2007: 3.1, 2008: -1.9
  - Volatility of GDP growth by year: 2004: 0.41, 2005: 0.55, 2006: 0.45, 2007: 0.5, 2008: 0.15
  - Moodys' KMV EDF by year: 2004: 50.0, 2005: 90.0, 2006: 70.0, 2007: 40.0, 2008: 40.6
  - VSTOXX by year: 2004: 18.9, 2005: 14.1, 2006: 16.5, 2007: 19.7, 2008: 33.7
  - Bloomberg Euro 500 Stocks Index by year: 2004: 175, 2005: 202, 2006: 240, 2007: 270, 2008: 202
- Observed pattern:
  - Market indicators detected problems earlier than financial soundness indicators; CDS spread tripled in 2007, while financial soundness indicators did not show crisis signs until 2008.

### Key explanatory variables and expected signs (CAMEL and market risk indicators)
- Capital adequacy measures (expected negative correlation with credit risk): Tier I ratio; Tier II ratio; Leverage ratio (expected positive correlation); Z-score (higher → lower risk).
- Asset quality measures (expected positive correlation): Loan-loss Provisions to Total Loans; Non-performing Loans to Total Loans. Loan-loss Reserves to Non-performing Loans expected negative.
- Management and earnings indicators:
  - Efficiency Ratio (Operating Expenses / Total Revenues): higher implies higher default risk.
  - Share of Trading Income to Total Revenues: higher implies higher default risk.
  - Net Interest Income ratio: higher may signal higher risk-taking and higher default risk.
  - ROA / ROE: higher implies lower default risk.
- Liquidity/funding indicators:
  - Loan to Deposit ratio: higher implies greater reliance on nondeposit funding → higher risk.
  - Share of Wholesale Funds to Total Liabilities: measure of wholesale funding reliance.
  - Short-term Borrowing to Total Liabilities: higher share implies higher vulnerability and higher CDS spread.
  - Liquid Assets to Total Assets: higher share implies greater resilience.
- Market and economic uncertainty indicators:
  - Bank-level: Variance of Equity Price; Options Implied Volatility; Cost of Funds; Moody’s KMV EDF (5 years); Effective GDP growth and GDP volatility weighted by bank exposures.
  - Country-level/common: Slope of the Yield Curve; Actual and expected GDP growth and volatility; Bloomberg Euro 500 stocks index; VSTOXX.
- Expected correlations summarized:
  - Equity variance: positive with credit default risk.
  - Options Implied Volatility: positive with credit default risk.
  - KMV EDF: higher EDF implies higher CDS spread.
  - Cost of Funds: positive with default risk.
  - Slope of the Yield Curve: steepening (higher slope) correlates negatively with credit default risk; flattening increases CDS spread.
  - GDP growth (country and weighted): negative with default risk; GDP volatility: positive with default risk.
  - Bloomberg Euro 500: negative with credit default premium; VSTOXX: positive.

### Baseline estimation results (dynamic unrestricted structural model)
- Model composition:
  - Core structural variables: leverage ratio and market risk indicators (Options Implied Volatility, Slope of the Yield Curve, Bloomberg 500 Stocks Index).
  - Additional regressors: Loan-loss Provisions Ratio, Share of Non-performing Loans, Efficiency Ratio, Trading Income Ratio, Net Interest Income Ratio, ROA, Loans to Deposits Ratio.
- Diagnostics:
  - Hansen test of over-identifying restrictions: does not reject validity of instruments at the 26 percent significance level (example).
  - First-order serial correlation present; second-order autocorrelation not present (probabilities 0.18 and 0.33 for One- and Two-step estimators, respectively).
- Statistically significant determinants of CDS spread (Two-step inference emphasized; coefficient estimates preserved as presented):
  - Lagged dependent variable (persistence close to unity): examples 1.13*(0.16;0.6) and 1.07***(0.1;0.3).
  - Efficiency Ratio: 0.3***(0.05;0.1) — a percentage reduction in the ratio of cost to revenues lowers the CDS spread by 0.3 bps.
  - Trading Income Ratio: 0.33***(0.05;0.1) — a percentage increase in the share of income from trading increases the CDS spread by 0.33 bps.
  - Net Interest Income Ratio: 23.2**(4.6;11) — a percentage increase in net interest margin increases the CDS spread by 23.2 bps.
  - ROA: -21.7*(4;11) — an increase in ROA by 1 percent lowers the spread by 21.7 bps.
  - Equity Options Implied Volatility: 1.86***(0.1;0.3) — a percentage increase increases CDS spread by 1.86 bps.
  - Slope of the Yield Curve: -12**(2.6;5.2) — a percentage decrease (flattening) increases the CDS spread by 12 bps (steepening correlates negatively).
- Magnitude interpretations (Two-step):
  - A percentage increase in the share of income from trading increases the CDS spread by 0.33 bps.
  - A percentage increase in net interest margin increases the CDS spread by 23.2 bps.
  - A percentage reduction in the ratio of cost to revenues lowers the CDS spread by 0.3 bps.
  - An increase in ROA by 1 percent lowers the spread by 21.7 bps.
  - A percentage increase in Options Implied Volatility increases the CDS spread by 1.86 bps.
  - A percentage decrease in the Slope of the Yield Curve increases the CDS spread by 12 bps.
- Model fit:
  - R2 statistic computed on first differences over 2006-2008 reaches about 89 percent.
  - Contribution decomposition: lagged dependent variable explains 37 percent of time-dimension variation; market risk indicators contribute 37 percent; management and earnings potential jointly account for 16 percent.

### Robustness tests and alternative indicators
- Parsimonious baseline:
  - Derived by dropping insignificant variables from the unrestricted model.
  - F-test probability of not rejecting joint restriction: 0.99 (One-step) and 0.66 (Two-step).
  - Parsimonious baseline results robust and used for subsequent robustness analyses.
- Alternative CAMEL indicators (nine variants tested) — summary:
  - Capital adequacy alternatives (Tier I, Tier II, Z-score): statistically insignificant.
  - Asset quality alternatives (Loan-loss Reserves Ratio): insignificant.
  - Management quality (Long-term Issuer Default Rating): insignificant.
  - Earnings potential (ROE): insignificant; ROA has superior information content.
  - Liquidity alternatives:
    - Short-term Borrowing to Total Liabilities: associated with higher CDS spread and marginally statistically significant but adds negligible explanatory power to baseline.
    - Wholesale Funds to Total Liabilities: insignificant.
    - Liquid Assets to Total Assets: insignificant.
  - Dummy for year 2008: tested; baseline specification robust to this dummy.
  - Diagnostic outcomes: Hansen tests do not reject instrument validity; first-order autocorrelation significant and second-order insignificant as theory requires.
- Alternative market risk indicators (nine variants tested) — summary:
  - Historical Variance of Bank’s Equity Price: insignificant.
  - Moody’s KMV EDF5: positive correlation with CDS spread (a percentage increase in EDF5 leads to a rise in CDS spread by 5.2 percent in regression A19) but negligible contribution to R2.
  - Cost of Funds: not significant.
  - GDP growth measures (actual, expected, weighted): largely insignificant; weighted GDP growth has negative sign but insignificant.
  - Volatility of GDP growth: marginally significant and positive — a percentage increase implies an increase in CDS spread by 4.8 percent (regression A16) with negligible added explanatory power.
  - VSTOXX and other common measures: often uninformative; VSTOXX likely dominated by bank-specific implied volatility from options.
  - Diagnostics: Hansen tests support instrument validity across regressions; second order autocorrelation rejected in all nine regressions.

### Findings on financial soundness indicators and limits
- Capital adequacy, asset quality, and liquidity largely insignificant in estimations for LCFIs.
- Suggested reasons:
  - Very little variation in leverage and loan-to-deposit ratio within LCFIs during sample.
  - Sample covers outbreak of crisis and may not capture adverse effects on loan portfolios yet.
- Cross-sectional drivers:
  - Variation in CDS spreads largely explained by differences across LCFIs rather than time variation.
  - Market indicators detected problems earlier than financial soundness indicators.
- Caveats and transferability:
  - Analysis limited to European LCFIs to preserve sample homogeneity.
  - Extrapolation to other banks or regions may require adjustments (e.g., differences in cost of funds).
  - Application to banks without CDS spreads may need adaptation.

### Policy implications, monitoring, and early warning applications
- Key implication: LCFIs’ business models, earnings potential, and economic uncertainty are among the most significant determinants of credit risk.
- Empirical associations for higher vulnerability (higher CDS spreads):
  - Higher net interest rate margin.
  - Higher share of trading income.
  - Higher cost efficiency ratio (operating expenses-to-total revenue).
  - Higher share of short-term borrowing in total liabilities.
  - Lower ROA.
- Market measures (implied volatility, slope of the yield curve) embed expectations about future bank value and economic outlook and help explain CDS spreads.
- Potential applications:
  - Use CDS pricing mechanism for vulnerability assessment and as an early warning tool.
  - Out-of-sample forecasts based on statistically significant determinants could provide forward-looking assessments of credit default risk across LCFIs.
  - Approach could possibly be applied to banks without CDS spreads with appropriate adjustments.

### Appendix highlights
- Appendix I — Interventions in European LCFI in the sample (2008/9–2009/5): selected intervention entries with Percent of total assets and CDS in bps (avg. of 30 days before intervention) include:
  - Dexia SA; €6.4bn capital (Sep 30, 2008); 1.062
  - UBS AG; $54bn liquidity (Oct 16, 2008); 2.7236
  - Royal Bank of Scotland; ≈£33bn capital (Oct 19, 2008); 1.4206
  - Lloyds; £4bn capital (Oct 19, 2008); 0.9138
  - Commerzbank; €8.2bn capital & €15bn guarantees (Nov 4, 2008); 1.382
  - Anglo-Irish Bank; €1.5bn capital (Dec 21, 2008); 1.5187
  - Raiffeisen Zentralbank; €1.75bn capital (Jan 30, 2009); 2.0240
  - Royal Bank of Scotland; £325bn liquidity; £19bn capital (Feb 26, 2009); 0.8155
  - Unicredit SPA; €4bn capital (March18, 2009); 0.4228
  - BNP Paribas; €5.1bn capital (March 31, 2009); 0.2113
  - Natixis; €5bn capital (May 13, 2009); 0.9288
- Appendix II — Rationale for baseline regression specification:
  - Banking framed as leveraged portfolio management; CDS premiums hinge on portfolio risk and the balance between risk and return.
  - Regression specification derived from the bank’s risk objective and the break-even condition with Lagrange multiplier ߣ, preserving parameter notation as in the source.

*Source: _wp10153 (PDF chapter content provided).*

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

### _wp10153 - References

### Introduction and purpose
- Objective: identify factors affecting credit default risk of selected European large complex financial institutions (LCFIs) to provide a fundamentals-based surveillance tool and potential early warning device.
- Rationale:
  - EWS models aim to detect underlying economic weaknesses and vulnerabilities and to trigger preemptive interventions.
  - LCFIs played a central role in the ongoing global financial crisis, motivating focused analysis of their business, funding models, risks, and spillovers.
- Approach: combine a standard structural credit risk model with financial soundness indicators to estimate determinants of credit default risk for European LCFIs.

### Choice of credit risk measure and empirical motivation
- Dependent variable: Credit Default Swap (CDS) spread (5-year maturity, yearly average of daily last price quotes from Bloomberg).
- Empirical support:
  - Figure 1 shows a positive correlation between pre-intervention CDS spreads and the extent of needed recapitalization (percent of total assets).
  - Regression reported: Recapitalization = 0.62(s.e. 0.09) CDS; R2 (adj.) = 0.73.
- Advantages of CDS spreads over corporate bond credit spreads (as stated):
  - CDS spread is a more direct measure of credit risk.
  - CDS spread does not seem to contain liquidity components or tax distortions that affect corporate bond spreads.
  - CDS spread leads in the credit risk price-discovery process and represents the upper bound of credit risk.

### Theoretical foundation and additional explanatory factors
- Baseline theoretical framework: structural approach based on Merton (1974), with applicability to CDS markets following Benkert (2004).
- Extensions noted in literature: inclusion of stock market returns and slope of the yield curve to address the credit spread puzzle (Collin-Dufresne et al., 2001; Blanco et al., 2005; Ericsson et al., 2009).
- Additional factors commonly considered for banks’ credit risk and EWS:
  - Credit rating agency assessments (e.g., Fitch Ratings).
  - CAMELS components: Capital Adequacy, Asset Quality, Management Quality, Earnings Potential, Liquidity, Sensitivity to Market Risk.
- Appendix II: provides formal motivation for choice of baseline variables (referenced in text).

### Methodology
- Dynamic panel specification (equations reproduced in source):
  - CDS_{i,t} = ρ CDS_{i,t-1} + X_{i,t}'β + Z_t'γ + u_{i,t}
  - First-differenced form: ∆CDS_{i,t} = ρ ∆CDS_{i,t-1} + ∆X_{i,t}'β + ∆Z_t'γ + ∆ε_{i,t}
- Estimator: Arellano-Bond GMM dynamic panel data estimator (Arellano and Bond, 1991).
- Reasons for choice of Arellano-Bond:
  - Endogeneity of some explanatory variables (e.g., ratings, variance of equity).
  - Need to remove bank-specific time-invariant fixed effects via first-differencing.
  - Lagged dependent variable instrumented to address autocorrelation in residuals.
  - Financial market variables exhibiting random walk behavior require instrumentation by lagged levels.
  - Appropriate for panels with T/N → 0 (sample has T=3, N=29).
- Estimation procedure:
  - One-step estimator assumes i.i.d. ε_{i,t} and uses weighting matrix W1.
  - Two-step estimator uses optimal weighting matrix W2 based on one-step residuals.
  - Two-step standard errors are downward biased in small samples; inference uses Windmeijer (2005) finite sample correction.

### Sample characteristics and variance decomposition
- Sample: 29 European LCFIs over 2006–2008.
- Observed changes in CDS spreads between 2006 and 2008:
  - Average CDS spread increased by a factor of 11.
  - Variability (standard deviation) increased 18 times.
- Variance decomposition in dynamic specification:
  - Lagged dependent variable explains 37 percent of time-dimension variation.
  - Bank-specific and market fundamentals explain 52 percent of the remaining variability (as reported).

### Key explanatory variables and expected signs (CAMEL and market risk indicators)
- Capital adequacy (four alternative measures): Tier I ratio; Tier II ratio; Leverage ratio; Z-score.
  - Tier I and Tier II: higher values should correlate negatively with credit risk.
  - Leverage (Average Total Assets / Average Total Common Equity): higher leverage correlates positively with default risk.
  - Z-score (ROA + Equity/Total Assets scaled by sd(ROA)): higher Z-score implies lower credit risk.
- Asset quality (three ratios): Loan-loss Provisions to Total Loans; Non-performing Loans to Total Loans; Loan-loss Reserves to Non-performing Loans.
  - Loan-loss Provisions and Non-performing Loans ratios: expected positive correlation with credit risk.
  - Loan-loss Reserves to Non-performing Loans: expected negative correlation with CDS spread.
- Management quality and business model indicators:
  - Efficiency Ratio (Operating Expenses / Total Revenues): higher Efficiency Ratio → higher default risk.
  - Long-term Issuer Default Rating (Fitch Ratings): higher numerical rating value means higher credit default risk (rating not available for DnB NOR ASA; no change in rating assumed in estimation).
  - Share of Trading Income to Total Revenues: higher share (investment banking orientation) expected to increase default risk due to greater revenue volatility.
- Market and economic uncertainty indicators referenced as significant:
  - Implied Volatility from Options.
  - KMV EDF.
  - Slope of the Yield Curve.
  - Volatility of GDP growth relevant to banks’ exposures.

### Main empirical findings (summary statements contained in source)
- Variation in CDS spreads largely explained by differences across LCFIs rather than time variation.
- Most important bank-specific fundamentals identified: Efficiency Ratio; Share of Trading Income; Net Interest Income Ratio; return on assets (ROA); share of Short-term Borrowing in Total Liabilities (representing Management, Earnings potential, and Liquidity elements of CAMEL).
- Significant market fundamentals: Implied Volatility from Options; KMV EDF; Slope of the Yield Curve; Volatility of GDP growth.

*Source: _wp10153 - References (content unit extracted from the IMF chapter PDF)_*

### 21.      Earnings potential is represented by three profitability measures: Net Interest

### _wp10153 - 21.      Earnings potential is represented by three profitability measures: Net Interest

### Earnings potential measures
- Earnings potential is represented by three profitability measures: Net Interest Income ratio, Return on Equity (ROE), and Return on Assets (ROA).
- Definitions:
  - Net Interest Income ratio: percentage interest yield of interest bearing assets.
  - ROE (ROA): Net Income divided by Total Common Equity (Total Assets).
- Interpretations:
  - Net Interest Income ratio:
    - Rough measure for the lending margin charged by a bank—a risk taking measure.
    - Higher Net Interest Income ratio may signal higher risk-taking; assuming negligible differences in funding costs across LCFIs, a higher Net Interest Income ratio would imply higher default risk.
  - ROE (ROA):
    - Measures profit generation given shareholders’ capital (total assets); should correlate negatively with default risk.
    - A higher ROE or ROA indicates better profit prospects for growth and resilience to shocks, and thus should be associated with lower credit risk.

### Liquidity and funding indicators
- Four liquidity/funding ratios:
  - Loan to Deposit ratio
  - Share of Wholesale Funds to Total Liabilities
  - Short-term Borrowing to Total Liabilities
  - Liquid Assets to Total Assets
- Interpretations:
  - Higher Loan to Deposit ratio implies higher dependence on nondeposit funding; greater reliance on wholesale funding may signal higher funding risk and higher bank riskiness.
  - Share of Wholesale Funds to Total Liabilities is an alternative measure of reliance on wholesale funding.
  - Short-term Borrowing to Total Liabilities:
    - Higher share implies greater vulnerability to a bank run and hence a higher CDS spread.
  - Liquid Assets to Total Assets:
    - Higher share implies greater resilience to liquidity pressures.

### Market risk factors (bank-specific and common)
- Bank-level measures:
  - Variance of a bank’s Equity Price (historical variance of bank’s equity price in a particular year).
  - Options Implied Volatility (extracted from equity options with one year maturity using Bloomberg).
  - Cost of Funds (ratio of Interest Expense to Total Liabilities).
  - Moody’s KMV Expected Default Frequency (EDF) (5 years).
  - Effective GDP growth and GDP volatility affecting a bank (bank-specific weighted measures).
- Country-level / common measures:
  - Slope of the Yield Curve (return on 10-year government bonds minus overnight interbank market rate approximated by the overnight rate).
  - Actual and expected GDP growth and GDP volatility (by country of incorporation).
  - Bloomberg Euro 500 stocks index (common market factor).
  - European stock market volatility (VSTOXX).
- Interpretations and expected correlations with CDS spread:
  - Variance of Equity Price:
    - Approximates variance of firm’s value in structural pricing theory; firm-specific equity volatility is an important determinant of bond credit spreads.
    - Expected to correlate positively with credit default risk.
  - Options Implied Volatility:
    - Contains additional information beyond historical equity volatility; measures uncertainty of equity returns; expected to correlate positively with credit default risk.
  - Moody’s KMV EDF:
    - Derived from market value and volatility of equity and book value of liabilities; a higher EDF generally implies a higher CDS spread.
  - Cost of Funds:
    - Increasing ratio signals willingness to pay premium on funds, often linked to liquidity management problems; expected to correlate positively with default risk.
  - Slope of the Yield Curve:
    - Reflects growth prospects and expected future short-term interest rates; a steepening signals positive economic prospects and should correlate negatively with credit default risk.
    - Derived from 10-year government bond return and overnight interbank market rate; overnight rate used to approximate risk-free short-term rate to filter out 3-month credit/liquidity risks during crisis years.
  - GDP growth and volatility:
    - GDP growth (country of incorporation) and weighted GDP growth (based on revenue exposure shares) expected to correlate negatively with default risk.
    - GDP volatility and weighted GDP volatility expected to correlate positively with default risk.
    - One-year ahead expected GDP growth (country and weighted exposures) extracted from December Consensus Forecasts used as economic outlook indicators.
  - Stock market index and volatility:
    - Bloomberg Euro 500 stocks index expected to correlate negatively with credit default premium.
    - VSTOXX expected to correlate positively with credit default premium.

### Data sample
- Definition and selection of LCFIs:
  - No commonly agreed definition of a Large Complex Financial Institution (LCFI).
  - Criteria used: size, interconnectedness, substitutability; banks large in domestic markets (assets as percent of domestic claims exceeding 10 percent) and/or large cross-border exposures (share of foreign revenues exceeds 30 percent).
- Sample composition:
  - 29 LCFIs in 12 European countries, balanced panel over 2004–2008.
  - Banks listed in sample (as provided): Erste Group Bank; Raiffeisen International; Dexia SA; KBC Groep; HSBC Plc; Barclays Plc; Royal Bank of Scotland; Lloyds; Standard Charter; Danske Bank; BNP Paribas; Societe Generale; Credit Agricole; Natixis; Deutsche Bank AG; Commerzbank; Bank of Ireland; Anglo-Irish Bank; Allied-Irish Bank; Unicredit SPA; Intesa Sanpaolo; Banca Monte dei Paschi di Siena; DnB NOR ASA; Banco Santander; BBVA; Nordea Bank AB; Svenska Handelsbanken; UBS AG; Credit Suisse.
- Data sources and coverage:
  - Data from Bloomberg, Economic Intelligence Unit, Consensus Forecasts, Datastream, Stoxx.com, and Credit Edge.
  - Bloomberg coverage for 2004-2008: about 70 percent of needed yearly data; missing data filled from banks’ publicly available statements.
- Descriptive statistics highlights (cross-section averages, as presented in Table 1):
  - Dependent variable CDS (basis points) by year: 2004: 13.8, 2005: 12.3, 2006: 9.6, 2007: 27.3, 2008: 108
  - Standard deviation of CDS by year: 2004: 3.9, 2005: 3.1, 2006: 2.7, 2007: 17.5, 2008: 47.4
  - Tier I ratio by year: 2004: 8.3, 2005: 8.1, 2006: 8.3, 2007: 7.9, 2008: 8.4
  - Leverage (multiple of equity) by year: 2004: 25.8, 2005: 27.1, 2006: 27.2, 2007: 27.8, 2008: 31.2
  - Z-score by year: 2004: 3.5, 2005: 3.5, 2006: 3.8, 2007: 3.7, 2008: 1.9
  - Ratio of Loan-loss Provisions to Total Loans by year: 2004: 0.34, 2005: 0.29, 2006: 0.33, 2007: 0.39, 2008: 0.85
  - Share of Non-performing Loans in Total Loans by year: 2004: 1.88, 2005: 1.8, 2006: 1.75, 2007: 1.68, 2008: 2.3
  - Efficiency Ratio (operating costs to revenues) by year: 2004: 61.6, 2005: 58.9, 2006: 56.2, 2007: 60.6, 2008: 104.4
  - Trading income as percent in revenues by year: 2004: 13.98, 2005: 18.6, 2006: 19.38, 2007: 12.98, 2008: -41.24
  - Net Interest Income (percent of average earning assets) by year: 2004: 1.78, 2005: 1.62, 2006: 1.52, 2007: 1.49, 2008: 1.65
  - ROA by year: 2004: 0.7, 2005: 0.7, 2006: 0.8, 2007: 0.7, 2008: 0.2
  - ROE by year: 2004: 16.32, 2005: 18.11, 2006: 19.86, 2007: 16.49, 2008: 0.19
  - Loans to Deposits Ratio by year: 2004: 128, 2005: 134, 2006: 138, 2007: 139, 2008: 150
  - Short-term Borrowing to Total Liab. by year: 2004: 22, 2005: 22, 2006: 21, 2007: 21, 2008: 20
  - Wholesale Funds to Total Liab. by year: 2004: 59, 2005: 62, 2006: 63, 2007: 63, 2008: 65
  - Liquid Assets to Total Assets by year: 2004: 13, 2005: 11.9, 2006: 11.3, 2007: 10.4, 2008: 9.8
  - Options Implied Variance of Equity by year: 2004: 16.8, 2005: 18.3, 2006: 21.6, 2007: 25.6, 2008: 52.5
  - Slope of the Yield Curve by year: 2004: 1.85, 2005: 1.09, 2006: 0.83, 2007: 0.29, 2008: 0.26
  - Variance of Bank's Equity Price by year: 2004: 99, 2005: 41, 2006: 02, 2007: 114, 2008: 973345 (as presented)
  - Cost of Funds by year: 2004: 2.4, 2005: 2.5, 2006: 2.9, 2007: 3.5, 2008: 3.3
  - GDP growth by year: 2004: 2.5, 2005: 2.8, 2006: 3.1, 2007: 3.1, 2008: -1.9
  - Volatility of GDP growth by year: 2004: 0.41, 2005: 0.55, 2006: 0.45, 2007: 0.5, 2008: 0.15
  - Expected GDP growth by year: 2004: 2.3, 2005: 2.4, 2006: 2.4, 2007: 2.5, 2008: 2.2
  - Moodys' KMV EDF by year: 2004: 50.0, 2005: 90.0, 2006: 70.0, 2007: 40.0, 2008: 40.6
  - VSTOXX by year: 2004: 18.9, 2005: 14.1, 2006: 16.5, 2007: 19.7, 2008: 33.7
  - Bloomberg Euro 500 Stocks Index by year: 2004: 175, 2005: 202, 2006: 240, 2007: 270, 2008: 202
- Observed patterns:
  - Market indicators detected problems earlier than financial soundness indicators; CDS spread tripled in 2007, while financial soundness indicators did not show crisis signs until 2008.
  - Average CDS spread increased 11 times during the crisis (from 9.6 in 2006 to 108 in 2008); standard deviation of CDS spread across LCFIs increased by a multiple of 18, indicating strong cross-sectional differences.

### Estimation results — Baseline model (dynamic unrestricted structural model)
- Model composition:
  - Core structural variables: leverage ratio and market risk indicators (Options Implied Volatility, Slope of the Yield Curve, Bloomberg 500 Stocks Index).
  - Additional variables: Loan-loss Provisions Ratio, Share of Non-performing Loans, Efficiency Ratio, Trading Income Ratio, Net Interest Income Ratio, ROA, Loans to Deposits Ratio.
- Estimation approach:
  - Two estimators: One-step (assumes i.i.d. errors) and Two-step (accounts for residual structure). Two-step preferred when differences exist; Two-step standard errors reported as asymptotic and Windmeijer (2005) finite sample corrected.
  - Instruments include GMM-style regressors (2nd and 3rd lag) of lagged dependent variable, ROA, GDP growth, expected GDP growth, and lagged differenced Leverage, Tier I, Tier II, Loan-loss Provisions Ratio, Share of Non-performing Loans, Efficiency Ratio, Share of Trading Income in Revenues, Slope of the Yield Curve, Net Interest Income Ratio, Loans to Deposits Ratio.
- Diagnostics:
  - Hansen test of over-identifying restrictions: does not reject validity of instruments at the 26 percent significance level.
  - First-order serial correlation present (not rejected); second-order autocorrelation not present (probabilities 0.18 and 0.33 for One- and Two-step estimators, respectively).
- Key statistically significant determinants of CDS spread (Two-step inference emphasized):
  - Lagged dependent variable (persistence close to unity): examples of estimates reported include 1.13*(0.16;0.6) and 1.07***(0.1;0.3).
  - Management quality:
    - Efficiency Ratio: 0.3***(0.05;0.1) — a percentage reduction in the ratio of cost to revenues lowers the CDS spread by 0.3 bps.
    - Trading Income Ratio: 0.33***(0.05;0.1) — a percentage increase in the share of income from trading increases the CDS spread by 0.33 bps.
  - Earnings potential:
    - Net Interest Income Ratio: 23.2**(4.6;11) — a percentage increase in net interest margin increases the CDS spread by 23.2 bps.
    - ROA: -21.7*(4;11) — an increase in ROA by 1 percent lowers the spread by 21.7 bps.
  - Market risk:
    - Equity Options Implied Volatility: 1.86***(0.1;0.3) — a percentage increase increases CDS spread by 1.86 bps.
    - Slope of the Yield Curve: -12**(2.6;5.2) — a percentage decrease (flattening) increases the CDS spread by 12 bps (equivalently, steepening correlates negatively with CDS spread).
- Magnitude interpretations (from Two-step estimates and summary discussion):
  - A percentage increase in the share of income from trading increases the CDS spread by 0.33 bps.
  - A percentage increase in net interest margin increases the CDS spread by 23.2 bps.
  - A percentage reduction in the ratio of cost to revenues lowers the CDS spread by 0.3 bps.
  - An increase in ROA by 1 percent lowers the spread by 21.7 bps.
  - A percentage increase in Options Implied Volatility increases the CDS spread by 1.86 bps.
  - A percentage decrease in the Slope of the Yield Curve increases the CDS spread by 12 bps.
- Model fit and tests (selected):
  - Number of observations / Banks reported as 87 / 29 in model tables.
  - R2 reported around 0.9 in table entries.
  - Test statistics examples: 1st order autocorrelation -2.6***(0.01) in One-step, -1.64*(0.1) in Two-step; 2nd order autocorrelation statistics not significant.

*Italic: Source: _wp10153 (PDF chapter content provided).*

### 33.      On the other hand, other key financial soundness indicators, such as capital

### _wp10153 - 33.      On the other hand, other key financial soundness indicators, such as capital

### Findings on financial soundness indicators
- Capital adequacy, asset quality, and liquidity turn out to be insignificant in the estimations for LCFIs.
- Insignificance explanations offered:
  - Very little variation in leverage (the capital measure used) within the LCFIs group during the sample period.
  - Very little variation in loan-to-deposit ratio (the liquidity measure used) within the LCFIs group during the sample period.
  - Sample period covers only the outbreak of the crisis and does not yet include the adverse effect on the loan portfolio; this likely explains insignificance of asset quality indicators (Loan-loss Provisions and the Share of Nonperforming Loans).

### Model fit and explanatory contributions
- The model fits the data well: the R2 statistic computed on first differences over the three-year estimation period during 2006-2008 reaches about 89 percent.
- Contribution to explanatory power:
  - Lagged dependent variable contributes 37 percent.
  - Market risk indicators contribute 37 percent.
  - Management and earnings potential (financial soundness indicators) jointly account for 16 percent.

### Parsimonious baseline model and tests
- A baseline parsimonious model was derived by dropping insignificant variables from the unrestricted model.
- F-test results: probability of not rejecting the joint restriction is 0.99 for One-step estimator and 0.66 for the Two-step estimator.
- Results in the parsimonious baseline model remain robust to the restrictions and are used in subsequent robustness analyses.

### Robustness to alternative CAMEL indicators (summary)
- Nine alternative CAMEL variables tested across categories:
  - Capital adequacy: Tier I, Tier II, Z-score.
  - Asset quality: Loan-loss Reserves.
  - Management quality: Long-term Issuer Default Rating.
  - Earnings potential: ROE.
  - Liquidity: Short-term Borrowing to Total Liabilities, Wholesale Funds to Total Liabilities, Liquid Assets to Total Assets.
  - Additionally: a dummy for year 2008 (A10) tested.
- Baseline specification proved robust to all alternative CAMEL indicators and to the dummy variable for 2008.
- Diagnostic tests:
  - Hansen test of over-identifying restrictions never rejects the validity of instruments at conventional significance levels across regressions.
  - Tests for serial correlation show significant first order and insignificant second order autocorrelation (the required serial correlation structure by the theory).
- Specific results:
  - Alternative capital adequacy measures (Tier I, Tier II, Z-score) are statistically insignificant — pointing to homogeneity of LCFIs in capitalization.
  - Alternative measures of asset quality (Loan-loss Reserves Ratio), management quality (Long-term Issuer Default Rating), and earnings potential (ROE) do not add value; insignificance attributed to homogeneity and stability during the outbreak of the crisis, and in ROE’s case subordinate information content compared to ROA.
  - Alternative liquidity measures show mixed results:
    - Liquid Assets to Total Assets: insignificant.
    - Wholesale Funds to Total Liabilities: insignificant.
    - Short-term Borrowing to Total Liabilities: associated with a higher CDS spread and marginally statistically significant, but its contribution to overall explanatory power of the baseline is negligible.

### Robustness to alternative market risk indicators (summary)
- Nine market risk variables tested: Historical Variance of Bank’s Equity Price, Moody’s KMV EDF5, Cost of Funds, Actual and Expected GDP growth, Actual and Expected GDP growth weighted by bank’s revenues, Volatility of GDP growth, Volatility of Weighted GDP growth.
- Most alternative market risk indicators are statistically insignificant and those significant add negligible explanatory power.
- Diagnostics for the Two-step estimator (Table 4):
  - Hansen test does not reject instrument validity at conventional significance levels across nine regressions.
  - Second order autocorrelation is rejected in all nine regressions; absence of first order autocorrelation in regression A19 is marginal.
- Specific findings:
  - Historical Variance of Bank’s Equity Price is insignificant; implied volatility from options is statistically significant in the baseline.
  - Moody’s KMV EDF5 correlates positively with the CDS spread: a percentage increase in EDF5 leads to a rise in the CDS spread by 5.2 percent (regression A19), but its contribution to R2 is negligible.
  - Cost of Funds is not significant for LCFIs, suggesting homogeneous access to funds (regression A12).
  - GDP growth of parent bank’s country and its volatility are statistically insignificant (regressions A13 and A14).
  - Weighted GDP growth is insignificant but has a negative sign (regression A15).
  - Volatility of GDP growth is marginally significant and positive: a percentage increase in volatility implies an increase in CDS spread by 4.8 percent (regression A16), with negligible added explanatory power.
  - Expected GDP growth, Weighted Expected GDP growth, and European stock market volatility (VSTOXX) are uninformative (regressions A17, A18, and A20); VSTOXX likely dominated by bank-specific implied volatility from options.

### Concluding remarks and policy implications
- The paper focuses exclusively on LCFIs, a group often excluded in the bank-failure literature, adding specific evidence on determinants of credit risk for large, complex, foreign-incorporated institutions.
- Key finding: LCFIs’ business models, earnings potential, and economic uncertainty are among the most significant determinants of credit risk.
- Empirical associations for more vulnerable LCFIs (higher CDS spreads):
  - Higher net interest rate margin.
  - Higher share of trading income.
  - Higher cost efficiency ratio (operating expenses-to-total revenue).
  - Higher share of short-term borrowing in total liabilities.
  - Lower ROA.
- Economic uncertainty measures (implied volatility from options and market views on economic growth via slope of the yield curve) also explain CDS spreads.
- Results are driven by cross-sectional variability in CDS spreads during the crisis outbreak, allowing markets to discriminate across LCFIs hit to varying degrees.
- Relationship to CAMELS literature:
  - While capital adequacy and asset quality are not statistically significant in the estimations, forward-looking market variables (implied volatility, slope of the yield curve) embed expectations about future bank value (equity capital) and economic outlook (impacting asset quality), implying default risk of LCFIs relates to the entire CAMELS structure.
- Comparison with broader European bank distress literature shows consistent coverage of CAMELS factors as distress determinants.

*Source: _wp10153 - 33.      On the other hand, other key financial soundness indicators, such as capital*

### 47.      The paper’s findings also suggest that overall; most variables suggested by the

### _wp10153 - 47.      The paper’s findings also suggest that overall; most variables suggested by the

### Determinants of LCFIs’ CDS spreads (main findings)
- Most variables suggested by the structural approach to the pricing of default risk are significant determinants of the LCFIs’ CDS spreads.
- The slope of the yield curve is a significant determinant of LCFIs’ CDS spreads.
- Implied volatility from option prices is a significant determinant of LCFIs’ CDS spreads.
- The above findings are consistent with the literature cited (Ericsson et al., 2009; Abid and Naifar, 2006; Blanco et al., 2005; Aunon-Nerin et al., 2002).
- Leverage is not significant as a determinant of CDS spreads, due to its high persistence and little variation across the LCFIs during the sample period.

### Use for monitoring, vulnerability assessment, and early warning
- Establishing a link between financial and market fundamentals of LCFIs and their CDS spreads provides a potential tool for fundamentals-based monitoring of LCFIs.
- Rationale:
  - The need for such tools is amplified by the potential systemic risk LCFIs can generate domestically and across borders where they operate.
  - The CDS spread contains information about the probability of default given the recovery rate; therefore, determinants of CDS spreads could indicate the extent of LCFIs’ vulnerability.
- Potential applications:
  - The identified CDS pricing mechanism can be used for vulnerability assessment and as an early warning tool.
  - An out of sample forecast for statistically significant CDS spread determinants of LCFIs would produce forward looking credit default risk assessment of their financial vulnerability in cross-section and time dimension.
  - The approach could possibly be applied even to banks that do not have a CDS spread.

### Limitations, sample scope, and transferability
- Analysis limited to European LCFI to preserve a high degree of sample homogeneity.
- Focus rationale:
  - Most LCFIs that have CDS spreads covering 2004–008 are domiciled in Europe.
  - Results therefore pertain to European LCFIs.
- Transferability cautions:
  - Extrapolation of results to other banks and countries might require adjustments.
  - Banks in other regions might not be homogeneous in terms of cost of funds and other aspects to the same degree as the European sample of LCFIs.
  - Some adjustments may be needed when using the identified pricing mechanism to evaluate banks that do not have CDS spreads.
- Note: Variables driving CDS spreads (e.g., yield curve and implied volatility) are usually missing in CAMELS bank failure and distress models, and likely contain expectations about the impact of the crisis on asset quality and equity capital.

### Appendix I — Interventions in European LCFI in the sample (2008/9–2009/5)
- Source: IMF, Bloomberg, and own calculations.
- Table 5 — Interventions in European LCFI in the sample (2008/9–2009/5): selected entries (Financial institution; Measure; Percent of total assets; CDS in bps. (avg. of 30 days before intervention))
  - Dexia SA; €6.4bn capital (Sep 30, 2008); 1.062
  - UBS AG; $54bn liquidity (Oct 16, 2008); 2.7236
  - Royal Bank of Scotland; ≈£33bn capital (Oct 19, 2008); 1.4206
  - Lloyds; £4bn capital (Oct 19, 2008); 0.9138
  - Credit Agricole; €3bn capital (Oct 22, 2008); 0.291
  - BNP Paribas; €2.55bn capital (Oct 22, 2008); 0.169
  - Societe Generale; €1.7bn capital (Oct 22, 2008); 0.2100
  - Erste Group Bank; €2.7bn capital (Oct 30, 2008); 1.3186
  - Commerzbank; €8.2bn capital & €15bn guarantees (Nov 4, 2008); 1.382
  - Anglo-Irish Bank; €1.5bn capital (Dec 21, 2008); 1.5187
  - Commerzbank; €10bn capital (Jan 7, 2009); 1.685
  - Raiffeisen Zentralbank; €1.75bn capital (Jan 30, 2009); 2.0240
  - Royal Bank of Scotland; £325bn liquidity; £19bn capital (Feb 26, 2009); 0.8155
  - Unicredit SPA; €4bn capital (March18, 2009); 0.4228
  - Intesa Sanpaolo; €4bn capital (March 20, 2009); 0.6170
  - Monte dei Paschi di Siena; €1.9bn capital (March 27, 2009); 0.9142
  - BNP Paribas; €5.1bn capital (March 31, 2009); 0.2113
  - Natixis; €5bn capital (May 13, 2009); 0.9288

### Appendix II — Rationale for baseline regression specification (summary)
- Banking framed as leveraged portfolio management: banks leverage on depositors and invest in risky projects.
- CDS premiums hinge on the level of risk in the bank’s portfolio and on the effective balance between risk and return (interest rate margin on asset j, fees per dollar of assets ܨ, size of assets ܣ).
- Decisions constrained by a break-even condition and the pricing mechanism of risky projects linking ݅௝ to the risk premium ߪ௝, the share of impaired assets׎௝, and the expected default rate׎൫ܧ௝൯.
- Return from trading ݅ே is considered more volatile; the rate of return from trading relative to the rest of the portfolio ߱ proxies the trading risk surcharge.
- The regression specification for the baseline model of CDS pricing results from writing the bank’s risk objective function along with the break-even condition and Lagrange multiplier ߣ.
- Parameter definitions and symbols preserved exactly as presented in the source (e.g., c=ൌൣܣܱܴ, ߣ, ݍ, ߚ, ߙ, N, P, ܨ, ܣ).

*Source: _wp10153 - 47.      The paper’s findings also suggest that overall; most variables suggested by the*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2010/_wp10153.pdf_