## _wp0909

## Source details

**Canonical URL:** [_wp0909](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2009/_wp0909.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2009/_wp0909.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2009/_wp0909.pdf.json)

---

### I. Introduction — purpose, novelty, and policy relevance
- Purpose: create a database of observed situations of distress in European Union (EU) banks and build an early warning system for bank distress.
- Novelty: comprehensive data on bank distress not previously made publicly available on an EU-wide basis; no literature on predicting bank failures in the EU as a whole prior to this work.
- Approach:
  - Cover banks in all EU countries to create a wider sample than single-country studies.
  - Control for cross-country differences so distress situations in some parts of the EU can serve as benchmarks for others.
  - Include recent high-profile cases of bank distress in the database.
- Uses and policy relevance:
  - Useful for depositors, creditors, rating agencies, and prudential supervisors.
  - If a bank is flagged as “weak,” supervisors should focus on that institution and, if necessary, enforce corrective action (which in extreme cases may include taking over the bank).
  - Argues for EU-wide benchmark criteria for banking sector soundness and more rules-based, uniform supervision across the EU to reduce country-level supervisory discretion and forbearance.
- Preview of main findings:
  - Empirical support for establishing EU-wide benchmark criteria.
  - Plausible thresholds (“trigger points”) for increased supervisory attention can be established.
  - Capital adequacy alone is insufficient; combinations of several relevant variables are needed to capture individual institution riskiness.

### II. Methodology and data
- Estimation methodology:
  - Model: logistic probability (logit) model estimating probability of bank distress (PD).
  - Dependent variable: Yijt = 1 when bank i headquartered in country j experiences financial distress in period t; 0 otherwise.
  - Explanatory variables: lagged values Xijt-1 (baseline one-year lag; two-year and three-year lags used in robustness checks with weaker significance).
  - Estimation details:
    - Log-likelihood: log L(β) = Σi Σt [Yijt log F(β'Xijt-1) + (1 − Yijt) log(1 − F(β'Xijt-1))].
    - Log odds: log (Pijt / (1 − Pijt)) = β0 + Σk βk Xk,ijt-1 where Pijt = Prob(Yijt = 1 | Xijt-1).
    - Use heteroscedasticity robust variance-covariance matrix allowing for correlated errors within banks.
    - Random effects logit estimated with random intercept variation across banks i or countries j; u ~ N(0, σu^2).
- Data:
  - Financial data source: Bureau Van Dijk’s BankScope database.
    - Extracted financial data on 5,708 banks in the EU-25 countries in 1996–2007.
  - Distress data source: NewsPlus database (powered by Factiva).
    - For each of the 5,708 banks and for each year, searches used bank name plus keywords: “rescue,” “bailout,” “financial support,” “liquidity support,” “government guarantee,” and “distressed merger.”
    - Hits manually examined; supervisory authority website searches conducted; compiled into unique distress dataset.
- Distress measure and sample:
  - Bank distress dummy (Yijt): =1 if at least one reference to distress for that bank in that year; 0 otherwise.
  - Identified distress events: 79 distress events for 54 EU banks during 1997–2008.
  - Database coverage: 5,708 banks, with 29,862 bank-year observations in total.
  - NewsPlus/Factiva coverage: collection of 14,000 sources.
  - Average distress frequency: about 0.3 percent per year.

### III. Determinants analyzed (CAMEL and additional variables)
- CAMEL covariates used (definitions as in source):
  - Capitalization: (Total equity)/(Total assets).
  - Asset quality: (Loan loss provisions)/(Total loans).
  - Managerial quality: (Total costs)/(Total income) — cost-to-income ratio.
  - Earnings: (Profit before taxes)/(Total equity) — ROE; ROA used in robustness checks.
  - Liquidity: (Liquid assets)/(Deposits and short-term funding).
- Additional covariates:
  - Market discipline: (Interest expenses)/Deposits (average deposit rate proxy).
  - Contagion dummy: =1 if a failure occurred in a similar bank in the same country (similar size defined as total assets within EUR ±200 million; robustness with ±100 million and loan-share ±5 percent).
  - Macro variables: country-level variables from IMF’s International Financial Statistics.
  - Market concentration: Herfindahl index (based on bank total assets).
  - Stock market indicators: ratios of bank stock indices relative to FTSE-100 for listed banks (222 EU banks with stock info).
  - Wholesale financing: share of wholesale financing in bank total liabilities.
- Data processing note: independent variables winsorized at the 1 percent level.

### IV. Descriptive differences between distressed and nondistressed banks (selected statistics)
- (Total equity)/(Total assets):
  - Nondistressed mean 0.0778, median 0.0583.
  - Distressed mean 0.0445, median 0.0350.
  - Median difference p-value 0.0000.
- (Loan loss provisions)/(Total loans):
  - Nondistressed mean 0.0076, median 0.0058.
  - Distressed mean 0.0293, median 0.0047.
  - Mean difference p-value 0.0045.
- (Total costs)/(Total income):
  - Nondistressed mean 0.7953, median 0.8068.
  - Distressed mean 1.1334, median 0.8919.
  - Median difference p-value 0.0006.
- (Profit before taxes)/(Total equity):
  - Nondistressed mean 0.1120, median 0.1034.
  - Distressed mean -0.2566, median 0.0245.
  - Median difference p-value 0.0000.
- (Liquid assets)/(Total assets):
  - Nondistressed mean 0.2700, median 0.2288.
  - Distressed mean 0.3205, median 0.1908.
  - Median difference p-value 0.6079.
- (Interest expences)/Deposits:
  - Nondistressed mean 0.0440, median 0.0330.
  - Distressed mean 0.1240, median 0.0730.
  - Mean difference p-value 0.0000; median difference p-value 0.0000.
- Dataset composition highlights:
  - Distress events concentrated unevenly across countries and years; most distress episodes occurred in Germany.
  - Most distressed banks are commercial; some are specialized banks and credit institutions.
  - Table 1 counts as presented in source: Total bank-year observations 29,862; Total banks 5,708; Distressed bank-year observations 79; Distressed banks 54; Non-hitters 2,639.

### V. Baseline estimation results (pooled logit specification I) — selected coefficients and fit
- Estimation approach: pooled logit, errors clustered at bank level; heteroscedasticity robust.
- Key coefficient signs and significance (Specification I, Baseline):
  - Capitalization: -26.578**
  - Asset quality: 20.443**
  - Managerial quality: -0.109 (not significant)
  - Earnings: -1.911***
  - Liquidity: -0.405 (not significant)
  - Market discipline: 4.957***
  - Contagion dummy: 6.072***
  - Intercept: -5.494***
- Model fit and sample:
  - Number of observations: 29,862
  - Log likelihood: -284.6
  - Baseline pseudo R-squared = 0.48
- Qualitative interpretation:
  - PD negatively associated with capitalization and earnings.
  - PD positively associated with loan loss provisions (deteriorating asset quality).
  - Higher deposit rates (market discipline) associated with higher PD.
  - Contagion dummy positive and highly significant.
  - Managerial quality and basic liquidity not significant in baseline; wholesale financing share significant in robustness checks.

### VI. Robustness checks and alternative specifications — major results
- Excluding “non-hitters” (specification II, N = 18,164): results similar to baseline; pseudo R-squared = 0.469.
- Adding macro variables (III, N = 29,155): macroeconomic variables not significant in presence of contagion dummy; pseudo R-squared = 0.613. (When contagion excluded, inflation and GDP become significant — results not shown.)
- Time dummies (IV): most time dummies not significant; main findings unchanged; pseudo R-squared = 0.551.
- Excluding repeated distress observations (V): main results corroborated; pseudo R-squared = 0.462.
- Market concentration (VI): concentration (Herfindahl) coefficient 5.136** (positive and significant) but becomes insignificant when macro variables included; pseudo R-squared = 0.490.
- Z-score (VII): coefficient insignificant (-203.95 reported); Z-score does not add information beyond CAMEL; pseudo R-squared = 0.497.
- Stock market information (VIII, N = 29,862): market information coefficient 4.965*** (positive and significant); pseudo R-squared = 0.485.
- Wholesale financing share (IX, N = 27,800): wholesale liabilities (share) coefficient 0.163*** — banks relying more on wholesale financing more likely to experience distress; pseudo R-squared = 0.506.
- Random effects at bank level (X, N = 29,862): random error (log of st. dev.) 2.077***; results qualitatively similar to pooled model; pseudo R-squared = 0.487.
- Random effects at country level (XI): random intercept insignificant at country level; pseudo R-squared = 0.520.
- Excluding Germany (XII, N = 13,924): main results hold except asset quality impact; pseudo R-squared = 0.516.
- Only commercial banks (XIII, N = 7,556): qualitative findings largely unchanged except capitalization impact; pseudo R-squared = 0.443.
- Only 2008 distress (XIV, N = 4,358): cross-section predicting 2008 failures performs well; pseudo R-squared = 0.337; 15 failures out of 19 identified at cutoff PD = 1 percent. Differences: capitalization significant; managerial quality may play a role; asset quality not significant.

### VII. Prediction performance and cutoff trade-offs (selected cutoff outcomes)
- Cutoff PD = 10 percent (baseline):
  - Correctly classified distress: 44 out of 79 (55.7 percent).
  - Correctly classified nondistress: 29,706 out of 29,783 (99.7 percent).
  - Type I errors (missed distress): 35 out of 79.
  - Type II errors (false positives): 77 out of 29,783.
- Cutoff PD = 1 percent:
  - Correctly classified distress: 50 out of 79.
  - Correctly classified nondistress: 29,525 out of 29,783.
  - Classified distress total: 308; Type II errors increased to 258.
- Cutoff PD = 0.5 percent:
  - Correctly classified distress: 54 out of 79.
  - Correctly classified nondistress: 29,366 out of 29,783.
  - Classified distress total: 471; Type II errors increased to 417.
- Authors’ note: in absence of substantial reduction in Type I errors, a case could be made for adopting the 10 percent cutoff point.

### VIII. Distribution of predicted PDs and assets at risk
- Majority of banks belong to lowest risk category (PD < 0.1 percent).
- Share of total bank assets at PD > 1 percent grew steadily from 1997 to 2004, then declined by end of sample.
- Assets-weighted risk distribution shows economic impact can be high even when number of high-PD banks is small.

### IX. Marginal effects and trigger points
- Marginal impacts computed at sample mean for significant CAMEL covariates: capitalization, asset quality, earnings.
- Example trigger:
  - For PD cutoff at 10 percent, asset quality corresponds to a trigger point of 14.3 percent of loan loss provisions relative to bank loans (holding other covariates at sample means).
- Trade-off visualizations (as described):
  - Capitalization vs. Loan loss provision vs. Bank PD.
  - Loan loss provision vs. Return on equity vs. Bank PD.
  - Capitalization vs. Return on equity vs. Bank PD.
- Concept: increases in capitalization can offset increases in loan loss provisioning to keep PD unchanged.

### X. Conclusions and policy recommendations
- Feasibility and signals:
  - An EU-wide early warning system using CAMEL covariates (capitalization, asset quality, profitability) can establish plausible thresholds for identifying weak banks.
  - Cost-to-income ratios and basic liquidity indicators show limited predictive power; however, share of wholesale financing is informative about PDs.
  - Depositor discipline is an important signal: higher deposit rates associate with higher PDs.
  - Contagion effects important: recent distress in similarly sized banks in same country significantly increases PD.
  - Stock prices (when available) contain useful predictive information; Z-score adds no incremental information beyond CAMEL indicators.
  - More concentrated banking markets associated with relatively higher likelihood of bank distress.
- Policy implications:
  - Supports potential for common benchmark criteria and possibly common EU supervisory framework given limited heterogeneity in baseline hazard after accounting for financial indicators.
  - Recommended use of early warning system:
    - Provide criteria for depositors and creditors.
    - Help policymakers limit supervisory forbearance and set more rules-based supervisory triggers: banks exceeding trigger points would receive closer scrutiny and potential intervention.
  - Caution: avoid purely mechanical application; system must be reestimated and reassessed regularly.
- Further work recommended:
  - Creation of a unified supervisory database across EU with more detailed indicators (exposures by sector, maturity, currency, performance) to improve the early warning system.

### XI. Appendices — supervisory systems, EU banking structure, and resolution principles (high-level summaries)
- Appendix I (survey of early warning systems):
  - Supervisory models include financial ratio/peer group analysis, statistical models (ratings, failure/survival, expected loss).
  - Examples: SEER (Federal Reserve), Bundesbank hazard models, Bank of Italy duration models, French SAABA expected loss model.
  - Limitations: qualitative factors often not captured; models are typically unconditional; back-testing and supervisory judgment are integral.
- Appendix II (European banking system):
  - EU has approximately 8,000 banks.
  - Some 46 LCFIs hold about 68 percent of EU banking assets; 16 key cross-border players account for about one-third of EU banking assets and hold an average of 38 percent of their EU banking assets outside home countries.
  - Policy gap: legal, regulatory, and supervisory frameworks lag centralization of treasury and risk management; cross-border crisis management principles and MoU exist (ECOFIN October 2007) but challenges remain.
- Appendix III (European Structured Early Intervention and Resolution — SEIR):
  - Objectives: prevent failures and restore failed banks to health; aim for efficiency and cost minimization with mandatory action at trigger points.
  - Principles: immediate supervisory action, capital restoration, mandatory resolution triggers, custody/receivership focused on continuity of core services, bridge bank use, recapitalization and sale within short time frames.

*Source: _wp0909 - References*

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

### _wp0909 - References

### I. Introduction
- Purpose: create a database of observed situations of distress in European Union (EU) banks and build an early warning system for bank distress.
- Novelty: comprehensive data on bank distress not previously made publicly available on an EU-wide basis; no literature on predicting bank failures in the EU as a whole prior to this work.
- Approach:
  - Cover banks in all EU countries to create a wider sample than single-country studies.
  - Control for cross-country differences to use distress situations in some parts of the EU as benchmarks for others.
  - Include recent high-profile cases of bank distress in the database.
- Uses and policy relevance:
  - Useful for depositors, creditors, rating agencies, and prudential supervisors.
  - If a bank is flagged as “weak,” supervisors should focus on that institution and, if necessary, enforce corrective action (which in extreme cases may include taking over the bank).
  - Argues for EU-wide benchmark criteria for banking sector soundness and more rules-based, uniform supervision across the EU to reduce country-level supervisory discretion and forbearance.
- Preview of main findings:
  - Empirical support for establishing EU-wide benchmark criteria.
  - Possible to establish plausible thresholds (“trigger points”) for increased supervisory attention.
  - Capital adequacy alone is insufficient; combinations of several relevant variables are needed to capture individual institution riskiness.

### II. Literature overview and motivation
- A. Early Warning Systems for Banking Soundness
  - Leading indicators grouped into three categories:
    - CAMEL variables: “capital, asset quality, management, earnings, and liquidity” — standard balance sheet and income statement financial ratios; asset quality important for medium- to long-term horizons; profitability, liquidity, and solvency useful for short-run.
    - Market prices: bank stocks and subordinated debt can contain predictive information (evidence stronger for U.S. studies; mixed for non-U.S. banks).
    - Other measures: deposit rates or indicators of the economic environment; rating agency assessments combine financial ratios and market indicators.
  - Observations:
    - CAMEL indicators are useful but there is no consensus on how to combine them into a bottom-line assessment; these measures are rarely back tested on actual distress situations.
    - Traditional CAMEL grades have limits and may need complements (e.g., Rojas-Suarez, 2001).
- B. Examples of Uses of the Early Warning Systems
  - Two main potential uses:
    - Strengthening rules in banking supervision and decreasing discretion (link system outputs to corrective actions as trigger points are reached).
    - Enhancing market discipline by publishing banks’ performance relative to the system.
  - Practical experience:
    - FDIC in the United States is a notable exception with stronger links between assessments and prompt corrective action (FDIC, 2003; Jones and King, 1995).
    - Most supervisory systems subject identified risky institutions to greater surveillance and on-site examination before formal enforcement.
    - Surveys indicate substantial scope for deviations between rules “on the book” and implementation.
  - EU-specific considerations:
    - Supervisors in individual EU member countries have different approaches to dealing with weak banks (Čihák and Decressin, 2007).
    - Cross-border large financial institutions (LCFIs) dominate the EU landscape, raising coordination issues.
    - EU banking structure statistics: the EU has some 8,000 banks; 46 LCFIs hold about 68 percent of EU banking assets; of these, 16 key cross-border players account for about one-third of EU banking assets, hold an average of 38 percent of their EU banking assets outside their home countries, and operate in just under half of the other EU countries (Appendix II).
  - Proposal parallels:
    - Analogous to Maastricht criteria for macro performance, proponents call for simple, transparent benchmarks for banks (liquidity, regulatory capital requirement, asset diversification, corporate governance) with minimum and target rates to promote market discipline and peer pressure.

### III. Methodology and data
- A. Estimation methodology
  - Model: logistic probability (logit) model estimating probability of bank distress (PD).
  - Notation and setup:
    - Yijt is a dummy = 1 when bank i headquartered in country j experiences financial distress in period t; 0 otherwise.
    - PD estimated as a function of lagged explanatory variables Xijt-1.
    - Log-likelihood function given as log L(β) = Σi Σt [Yijt log F(β'Xijt-1) + (1 − Yijt) log(1 − F(β'Xijt-1))] (equation (1) in text).
    - Log odds specification: log (Pijt / (1 − Pijt)) = β0 + Σk βk Xk,ijt-1 (equation (2) in text), where Pijt = Prob(Yijt = 1 | Xijt-1).
  - Interpretation:
    - Signs of β coefficients indicate direction of marginal impact on PD.
    - Magnitude of PD impact depends on initial values of explanatory variables and coefficients; economic magnitude assessed via marginal effects evaluated at the sample mean.
  - Estimation choices to address panel data issues:
    - Use heteroscedasticity robust variance-covariance matrix allowing for correlated errors within banks to correct downward-biased standard errors if independence assumption is violated.
    - Also estimate a random effects logit model with random intercept variation either across individual banks i (β0 + ui) or countries j (β0 + uj); u ~ N(0, σu^2).
    - Significance of σu^2 can confirm heterogeneity of baseline hazard across banks or countries.
  - Lag structure:
    - Baseline uses one-period (one year) lags of explanatory variables.
    - Robustness checks with two-year and three-year lags produced very similar results but weaker statistical significance (especially three-year lags), suggesting predictive power declines for longer horizons.
- B. Data
  - Financial data source: Bureau Van Dijk’s BankScope database.
    - Extracted financial data on 5,708 banks in the EU-25 countries in 1996–2007.
  - Distress data source: NewsPlus database (powered by Factiva, a Dow Jones company).
    - For each of the 5,708 banks and for each year, searches used bank name plus keywords: “rescue,” “bailout,” “financial support,” “liquidity support,” “government guarantee,” and “distressed merger.”
    - Hits were manually examined to confirm relevance; searches of relevant supervisory authority websites were also conducted (search procedure described; results compiled into unique distress dataset).

*Source: _wp0909 - References*

### references to banks that failed. Based on all these searches, we created a bank distress

### _wp0909 - references to banks that failed. Based on all these searches, we created a bank distress

### Data and definition of distress
- Bank distress dummy variable (Yijt): equal to 1 if there is (at least one) reference to distress in the particular bank in that particular year, and 0 otherwise.
- Method: NewsPlus/Factiva searches for negative media references; robustness check excludes banks for which searches returned no hits.
- Identified distress events: 79 distress events for 54 EU banks during 1997–2008.
- Database coverage: 5,708 banks from the EU-25 countries, with 29,862 bank-year observations in total.
- NewsPlus/Factiva: collection of 14,000 sources (as described in the source).
- Average distress frequency: about 0.3 percent per year.

### Dataset composition and distribution
- Distress events concentrated unevenly across countries and years; most distress episodes occurred in Germany.
- Most distressed banks are commercial; some are specialized banks and credit institutions.
- Table 1 (excerpted counts as presented in source): Total bank-year observations 29,862; Total banks 5,708; Distressed bank-year observations 79; Distressed banks 54; Non-hitters 2,639.

### Determinants analyzed (CAMEL and additional variables)
- CAMEL covariates used:
  - Capitalization: (Total equity)/(Total assets) — simple unweighted leverage ratio.
  - Asset quality: (Loan loss provisions)/(Total loans) as proxy (stock of nonperforming loans and reserves largely unavailable).
  - Managerial quality: (Total costs)/(Total income) — cost-to-income ratio.
  - Earnings: (Profit before taxes)/(Total equity) — (after-tax) return on average equity (ROE); ROA used in robustness checks.
  - Liquidity: (Liquid assets)/(Deposits and short-term funding).
- Additional covariates:
  - Market discipline: (Interest expenses)/Deposits — average deposit rate proxy.
  - Contagion dummy: =1 if a failure occurred in a similar bank in the same country (similar size defined as total assets within EUR ±200 million; robustness with ±100 million and loan-share ±5 percent).
  - Macro variables: country-level variables from IMF’s International Financial Statistics.
  - Market concentration: Herfindahl index (based on bank total assets).
  - Stock market indicators: ratios of bank stock indices relative to FTSE-100 for listed banks (222 EU banks with stock info).
  - Wholesale financing: share of wholesale financing in bank total liabilities.

### Descriptive differences between distressed and nondistressed (Table 2 findings)
- (Total equity)/(Total assets): Nondistressed mean 0.0778, median 0.0583; Distressed mean 0.0445, median 0.0350; median difference p-value 0.0000.
- (Loan loss provisions)/(Total loans): Nondistressed mean 0.0076, median 0.0058; Distressed mean 0.0293, median 0.0047; mean difference p-value 0.0045.
- (Total costs)/(Total income): Nondistressed mean 0.7953, median 0.8068; Distressed mean 1.1334, median 0.8919; median difference p-value 0.0006.
- (Profit before taxes)/(Total equity): Nondistressed mean 0.1120, median 0.1034; Distressed mean -0.2566, median 0.0245; median difference p-value 0.0000.
- (Liquid assets)/(Total assets): Nondistressed mean 0.2700, median 0.2288; Distressed mean 0.3205, median 0.1908; median difference p-value 0.6079.
- (Interest expences)/Deposits: Nondistressed mean 0.0440, median 0.0330; Distressed mean 0.1240, median 0.0730; mean difference p-value 0.0000; median difference p-value 0.0000.
- Note: independent variables winsorized at the 1 percent level.

### Baseline estimation results (logit, pooled; specification I)
- Model robust to heteroscedasticity; errors clustered at bank level.
- Key qualitative findings:
  - PD negatively associated with capitalization and earnings.
  - PD positively associated with loan loss provisions (deteriorating asset quality).
  - Market discipline (higher deposit rates) positively associated with PD.
  - Contagion dummy positive and highly significant.
  - Managerial quality (cost-to-income) not significant in baseline.
  - Liquidity not significant in baseline; wholesale financing share found significant in robustness checks.
- Baseline fit: pseudo R-squared = 0.48.
- Table 3 selected coefficient signs and significance indicators (I, Baseline):
  - Capitalization: -26.578**
  - Asset quality: 20.443**
  - Managerial quality: -0.109 (not significant)
  - Earnings: -1.911***
  - Liquidity: -0.405 (not significant)
  - Market discipline: 4.957***
  - Contagion dummy: 6.072***
  - Intercept: -5.494***
  - Number of observations: 29,862
  - Log likelihood: -284.6

### Robustness checks and alternative specifications (summary of major findings)
- Excluding “non-hitters” (specification II, N = 18,164): results similar to baseline; pseudo R-squared = 0.469.
- Adding macro variables (III, N = 29,155): macroeconomic variables not significant in presence of contagion dummy; pseudo R-squared = 0.613.
  - When contagion dummy excluded, inflation and GDP coefficients become significant (results not shown).
- Time dummies (IV): most time dummies not significant; main findings unchanged; pseudo R-squared = 0.551.
- Excluding repeated distress observations (V): main results corroborated; pseudo R-squared = 0.462.
- Market concentration (VI): concentration (Herfindahl) positive and significant (5.136**), but becomes insignificant when macro variables included; pseudo R-squared = 0.490.
- Z-score (VII): coefficient insignificant (-203.95 reported); Z-score does not add information beyond CAMEL; pseudo R-squared = 0.497.
- Stock market information (VIII, N = 29,862): market information positive and significant (4.965***); pseudo R-squared = 0.485.
- Wholesale financing share (IX, N = 27,800): wholesale liabilities (share) 0.163*** — banks relying more on wholesale financing more likely to experience distress; pseudo R-squared = 0.506.
- Random effects at bank level (X, N = 29,862): random error (log of st. dev.) 2.077***; results qualitatively similar to pooled model; pseudo R-squared = 0.487.
- Random effects at country level (XI): random intercept insignificant at country level; pseudo R-squared = 0.520.
- Excluding Germany (XII, N = 13,924): main results hold except asset quality impact; pseudo R-squared = 0.516.
- Only commercial banks (XIII, N = 7,556): qualitative findings largely unchanged except capitalization impact; pseudo R-squared = 0.443.
- Only 2008 distress (XIV, N = 4,358): cross-section predicting 2008 failures performs well; pseudo R-squared = 0.337; 15 failures out of 19 identified at cutoff PD=1 percent. Differences: capitalization significant; managerial quality may play a role; asset quality not significant.

### Prediction performance and cutoff trade-offs (Table 4)
- Cutoff PD = 10 percent (baseline table):
  - Correctly classified distress: 44 out of 79 (55.7 percent).
  - Correctly classified nondistress: 29,706 out of 29,783 (99.7 percent).
  - Type I errors (missed distress): 35 out of 79.
  - Type II errors (false positives): 77 out of 29,783.
- Cutoff PD = 1 percent:
  - Correctly classified distress: 50 out of 79.
  - Correctly classified nondistress: 29,525 out of 29,783.
  - Classified distress total: 308; Type II errors increased to 258.
- Cutoff PD = 0.5 percent:
  - Correctly classified distress: 54 out of 79.
  - Correctly classified nondistress: 29,366 out of 29,783.
  - Classified distress total: 471; Type II errors further increased to 417.
- Authors’ note: in absence of substantial reduction in Type I errors, a case could be made for adopting the 10 percent cutoff point.

### Distribution of predicted PDs and assets at risk
- Majority of banks belong to lowest risk category (PD < 0.1 percent).
- Share of total bank assets at PD > 1 percent grew steadily from 1997 to 2004, then declined by end of sample.
- Assets-weighted risk distribution shows economic impact can be high even when number of high-PD banks is small.

### Marginal effects and trigger points
- Marginal impacts computed at sample mean for significant CAMEL covariates: capitalization, asset quality, earnings.
- Example trigger: for PD cutoff at 10 percent, asset quality corresponds to a trigger point of 14.3 percent of loan loss provisions relative to bank loans (holding other covariates at sample means).
- Trade-offs illustrated in three-dimensional plots:
  - Capitalization vs. Loan loss provision vs. Bank PD.
  - Loan loss provision vs. Return on equity vs. Bank PD.
  - Capitalization vs. Return on equity vs. Bank PD.
- Concept: increases in capitalization can offset increases in loan loss provisioning to keep PD unchanged.

### Conclusions and policy recommendations
- An EU-wide early warning system using CAMEL covariates (capitalization, asset quality, profitability) can establish plausible thresholds for identifying weak banks.
- Cost-to-income ratios and basic liquidity indicators show limited predictive power; however, share of wholesale financing is informative about PDs.
- Depositor discipline is an important signal: higher deposit rates associate with higher PDs.
- Contagion effects important: recent distress in similarly sized banks in same country significantly increases PD.
- Stock prices (when available) contain useful predictive information; Z-score adds no incremental information beyond CAMEL indicators.
- More concentrated banking markets associated with relatively higher likelihood of bank distress.
- Macroeconomic differences among EU countries play some, but relatively small, role in predicting individual bank PDs; contagion dummy may capture much of macro effect.
- EU countries relatively homogeneous in bank baseline hazard after accounting for financial indicators; limited heterogeneity in slope coefficients — supports potential for common benchmark criteria and possibly common EU supervisory framework.
- Recommended use of the early warning system:
  - Provide criteria for depositors and creditors.
  - Help policymakers limit supervisory forbearance and set more rules-based supervisory triggers: banks exceeding trigger points would receive closer scrutiny and potential intervention.
  - Caution: avoid purely mechanical application; system must be reestimated and reassessed regularly to respond to new developments.
- Further work: creation of a unified supervisory database across EU with more detailed indicators (exposures by sector, maturity, currency, performance) to improve the early warning system.

*Source: authors, based on BankScope and NewsPlus/Factiva, as presented in the supplied content.*

### APPENDIX I. EARLY WARNING SYSTEMS FOR BANKING SUPERVISION: SURVEY

### APPENDIX I. EARLY WARNING SYSTEMS FOR BANKING SUPERVISION: SURVEY

### Financial ratio and peer group analysis systems
- Core financial variables used: measures of capital adequacy, asset quality, profitability and liquidity.
- Financial ratio analysis generates warnings if a ratio:
  - exceeds a predetermined critical level;
  - lies within a set interval; or
  - is an outlier relative to the bank’s past performance.
- Peer group analysis:
  - peer groups based on asset size (e.g., small versus large banks) or specialization (e.g., domestic commercial banks, foreign banks, cooperative banks, or savings banks);
  - each bank’s ratios compared with its peer group averages;
  - approaches include identifying worst performers relative to peer average or sorting ratios and calculating percentile rankings;
  - individual banks whose ratios have deteriorated relative to peer-group averages can be identified.

### Statistical models — overview and classification
- Broad categories of supervisory statistical models:
  - (i) models estimating ratings or rating downgrades;
  - (ii) failure or survival prediction models;
  - (iii) expected loss models.
- Agencies using formally estimated models include: the U.S. Federal Reserve, the FDIC, the OCC, the Financial Services Authority (UK), the French Banking Commission, the Deutsche Bundesbank, and the Bank of Italy.

### Models estimating ratings or rating downgrades (example: SEER)
- SEER (System for Estimating Examination Ratings) used by the U.S. Federal Reserve (since 1993):
  - employs a multinomial logistic regression to estimate the probability that the bank’s next composite CAMEL rating will be each of the five possible ratings (1–5);
  - the SEER rating equals the sum of the five rating levels multiplied by their respective probabilities.
- Methodology:
  - determines historical relationship using call report data from two previous quarters and corresponding latest examination data;
  - estimates statistical relationship between the latest composite CAMEL on-site rating and about 45 financial and nonfinancial variables;
  - variables not statistically significant for the current quarter are eliminated;
  - example variables: past-due loans, nonaccrual loans, foreclosed real estate loans, tangible capital, net income, investment securities, an asset growth variable, prior management rating, and the previous composite CAMEL rating.
- Use of outputs:
  - combines variable weights with current call report values to estimate probable composite CAMEL ratings;
  - banks with SEER estimates significantly different from the most recent on-site examination rating are singled out for further review.

### Models predicting failure or survival rates
- Estimation and deployment:
  - estimated on samples of failed/troubled banks, tested on hold-out samples, then used out of sample to identify banks resembling failed banks in the model.
- Example: Federal Reserve failure probability model
  - estimates probability a bank will become “critically undercapitalized” during the subsequent two years using most recent call report data;
  - employs a bivariate probit regression;
  - originally used characteristics of bank failures during the period 1985–91;
  - when failures declined in the 1990s, model was developed on pooled cross-section and time-series data for the period 1985–91;
  - uses 11 explanatory variables to calculate a risk rank;
  - banks with risk rank higher than a predetermined threshold are flagged for more intensive review.
- Model outputs and enhancements:
  - initially produced a listing of contributing variables; updated in 1997 to include a detailed “risk profile analysis” with “peer analysis” and “change analysis” for each bank;
  - distribution and average of risk ranks across banks provide measures of current industry risk.

- Deutsche Bundesbank approach:
  - uses a hazard rate model for German savings and cooperative banks to estimate probability that an institution’s existence is endangered within one year without affiliated network support;
  - determinants based on CAMEL ratings and supplemented by regional and macroeconomic factors (Deutsche Bundesbank, 2004).

- Duration models (Bank of Italy and Bundesbank):
  - estimate probability of failure and probable time to failure;
  - dependent variable is time to failure; model computes probability a bank will survive longer than a specified time or fail at a specified future time.

### Expected loss models
- Motivation:
  - used where few historical bank failures exist, making failure/survival prediction models hard to estimate.
- French Banking Commission’s SAABA model (in use since 1997):
  - premise: credit risk is the major risk faced by banks;
  - complements quantitative diagnosis with qualitative assessments of ownership/shareholder quality, management and internal controls;
  - input data from Banking Commission databases, Bank of France database, and external sources;
  - methodology:
    - adjusts all outstanding individual and corporate loans by a potential future loss amount based on default probability for each credit;
    - sums individual potential losses to arrive at a total for the entire credit portfolio over a three-year period;
    - total potential loss figure is adjusted against current level of reserves;
    - the unadjusted balance represents potential future loss, which is deducted from current level of the bank’s own funds;
    - if bank’s own funds fall below the 8 percent requirement after quantitative analysis, the bank’s future solvency is considered questionable;
  - SAABA also assesses shareholders’ ability to support the institution, management, internal controls, and liquidity.

### Issues and model limitations
- Qualitative factors:
  - statistical models typically do not represent qualitative factors such as management quality, internal control, fraud, or financial misconduct, although these can be significant causes of bank failure;
  - few models attempt to quantify management quality or include realistic surrogates for management performance.
- Use and validation:
  - statistical models’ outputs are generally supplemented with other systems and supervisory judgment;
  - models are back-tested and validated periodically by several authorities:
    - Federal Reserve undertakes annual validation for SEER rating and risk rank models, comparing predictions with actual outcomes and computing Type I and Type II error rates;
    - French Banking Commission carries out periodic back testing to verify correct identification of banks likely to face serious problems;
    - FDIC compares composite scores with future bank failure rates and finds banks in the lowest composite score decile usually fail at the highest rate during the two years immediately after scores were measured, while those in the highest decile fail at the highest rate between three and five years after the scores are assigned.
- Conditioning and forecasting:
  - most models are unconditional, predicting failure given current independent variable values and not conditioning forecasts on assumed future paths of variables;
  - some authorities are attempting to develop models based on forecasts of individual bank variables and resultant failure/survival probabilities.
- Practical performance:
  - while some early warning models have achieved satisfactory results, successes have been in limited contexts; accurate prediction of rating downgrades, failure/survival probabilities, expected losses, or insolvency across a wide range of institutions and time periods remains difficult.

*Source: APPENDIX I. EARLY WARNING SYSTEMS FOR BANKING SUPERVISION: SURVEY (content supplied).*

### APPENDIX II. EUROPEAN BANKING SYSTEM

### Structure and concentration
- The European Union has approximately 8,000 banks.
- Large cross-border banks (LCFIs) have emerged with substantial market share.
- Mapping of EU cross-border banking groups (Banking Supervision Committee of the European System of Central Banks) findings:
  - some 46 LCFIs hold about 68 percent of EU banking assets;
  - of these, 16 key cross-border players account for about one-third of EU banking assets;
  - these 16 hold an average of 38 percent of their EU banking assets outside their home countries;
  - they operate in just under half of the other EU countries.

### Integration, risks, and policy gaps
- Cross-border banking business concentrated in wholesale markets (interbank and corporate bond markets are relatively well integrated; equity, securitization markets, and arms’ length financing have scope for further integration).
- Legal, regulatory, and supervisory frameworks have lagged behind the centralization of treasury and risk management functions of LCFIs.
- IMF position:
  - IMF has argued that the EU needs a more integrated approach to financial stability (e.g., IMF, 2007);
  - completing single market objectives and managing associated risks require an integrated approach to financial stability.
- Constraints on progress:
  - political preferences, legal and institutional considerations limit progress on cross-border financial stability arrangements;
  - national supervisors’ fiduciary responsibilities toward national governments and parliaments limit incentives to work toward common EU objectives.

### Cross-border crisis management principles and MoU
- ECOFIN (October 2007) adopted cross-border crisis management principles committing member states to act in crises to minimize “potential harmful economic impacts at the lowest overall collective costs.”
- If public resources are needed, direct budgetary net costs are to be “shared among Member States on the basis of equitable and balanced criteria.”
- The accompanying Memorandum of Understanding (MoU):
  - commits member states to national and cross-border arrangements to manage financial stability problems;
  - sets common guidelines for crisis management and a common assessment framework to determine systemic nature of a crisis.
- Ongoing legal framework work includes:
  - improvements to deposit guarantee schemes;
  - framework for early intervention and reorganization measures;
  - assessment of obstacles to cross-border asset transferability.

### Lamfalussy framework and reform roadmap
- Lamfalussy framework aims at regulatory and supervisory convergence through EU-level rule making and consistent national application.
- Level 3 Committees bring together national supervisors to achieve convergence.
- December 2007 ECOFIN reforms:
  - reinforce Level 3 Committees by giving them more resources, introducing scope for qualified majority voting, and strengthening national application of committee guidelines while keeping the guidelines nonbinding.

### Remaining challenges
- Despite principles and MoU recognizing collective responsibility and cost sharing, in a severe crisis national interests may still prevail.
- The MoU may add complexity to cross-border financial stability arrangements.
- Key challenge: align legal underpinnings of nationally anchored financial stability frameworks and incentives of relevant agents with commonly agreed principles.

*Source: APPENDIX II. EUROPEAN BANKING SYSTEM (content supplied).*

### APPENDIX III. EUROPEAN STRUCTURED EARLY INTERVENTION AND RESOLUTION

### Objectives of SEIR / prompt corrective action
- Aim: achieve efficiency and cost minimization in bank resolution by prescribing mandatory action at certain trigger points.
- SEIR frameworks for cross-border systemic banks in the EU should aim at:
  - preventing failures; and
  - restoring failed banks to health.

### Principles for efficient resolution procedures
- Immediate supervisory action:
  - prudential authorities need to act as soon as a solvency shortfall or other warning signals are detected;
  - if solvency shortfall is not large, bank should be given a grace period to restore solvency to regulatory minimum under intensified supervision and restrictions on actions.
- Capital restoration:
  - if no improvement after the grace period, a capital injection should be imposed;
  - in absence of controlling shareholders or if they are unable to mobilize capital, shareholders would accept dilution of ownership.
- Mandatory resolution triggers:
  - if no private sector solution and solvency drops below a certain level or another trigger point is met:
    - mandatory and prompt suspension of shareholder rights;
    - a bank resolution agency should take custody or receivership of the bank;
    - new management should be installed.
- Custody/receivership objectives:
  - bank resolution agency must make a quick early assessment to allow continuity in the bank’s core operations and minimal or no disruption in availability of most deposits;
  - if estimated solvency is negative, some liabilities could be blocked or separated from the (bridge) bank pending a more complete audit and final determination of losses.
- Continuity of core services:
  - systemic and core operations, including basic retail services, should continue uninterrupted or after a minimal interruption not exceeding one or two days;
  - continuation could be assured by a new entity (a bridge bank).
- Reopening, recapitalization and sale:
  - the reopened bank should be recapitalized, restructured, and prepared for sale, as a whole or in parts, to private acquirers within a relatively short time period;
  - proceeds from sale, net of recapitalization and management costs, should be used to pay off liabilities not assumed by the reopened bank according to legal priority;
  - any remaining funds at end of resolution process should be disbursed to the bank’s original shareholders.

*Source: APPENDIX III. EUROPEAN STRUCTURED EARLY INTERVENTION AND RESOLUTION (content supplied).*

### REFERENCES

### _wp0909 - REFERENCES

### Competition and financial stability
- Allen, F., and D. Gale, 2004, “Competition and Financial Stability,” Journal of Money, Credit, and Banking, Vol. 36 (3), pp. 453–80. 
- Beck, T., Demirguc-Kunt, A. and Levine, R., 2006, “Bank Concentration, Competition, and Crises: First Results,” Journal of Banking and Finance, Vol. 30, pp. 1581–1603. 
- Boyd, J. H., and G. de Nicoló, 2005, “The Theory of Bank Risk-taking and Competition Revisited,” Journal of Finance, Vol. 60, pp. 1329–42. 
- Gropp, Reint, and Florian Heider, 2008, “The Determinants of Capital Structure: Some Evidence from Banks,” Center for European Economic Research, Vol. 8, No. 15. 
- Koetter, M., Bos, J. W. B., Heid, F., Kolari, J. W., Kool, C. J. M., and D. Porath, 2007, “Accounting for Distress in Bank Mergers,” Journal of Banking and Finance, Vol. 31, No. 10, pp. 3200–17. 

### Early warning systems, supervisory indicators, and market information
- Berg, Andrew Eduardo Borensztein, and Catherine Pattillo, 2004, “Assessing Early Warning Systems: How Have They Worked in Practice?” IMF Working Paper 04/52 (Washington: International Monetary Fund). 
- Bongini, P., L. Laeven, and G. Majnoni, 2002. “How Good is the Market at Assessing Bank Fragility? A Horse Race Between Different Indicators,” Journal of Banks and Finance, Vol. 26, pp. 1011–31. 
- Čihák, Martin, 2007, “Systemic Loss: A Measure of Financial Stability,” Czech Journal of Economics and Finance, Vol. 57, No.1–2, pp. 5–26. 
- Čihák, Martin, and Jörg Decressin, 2007, “The Case for a European Banking Charter”, IMF Working Paper No. 07/173 (Washington: International Monetary Fund).  
- Čihák, Martin, and Klaus Schaeck, 2007, “How Well Do Aggregate Bank Ratios Identify Banking Problems?” IMF Working Paper No. 07/275 (Washington: International Monetary Fund).  
- Čihák, Martin, and Alexander Tieman, 2006, “How Good is Financial Sector Supervision and Regulation in Europe?” Euro Area Selected Issues Paper, IMF Country Report No. 06/288 (Washington: International Monetary Fund).  
- Čihák, Martin, and Alexander Tieman, 2008, “Quality of Financial Sector Regulation and Supervision Around the World,” IMF Working Paper No. 08/190 (Washington: International Monetary Fund).  
- Curry, Timothy, Peter Elmer, and Gary Fissel, 2003, “Using Market Information to Help Identify Distressed Institutions: A Regulatory Perspective.” FDIC Banking Review, Vol. 15, no.3, pp. 1–16. 
- Flannery, Mark J., 1998, “Using Market Information in Prudential Bank Supervision: A Review of the US Empirical Evidence,” Journal of Money, Credit and Banking, August, pp. 273–302. 
- Gilbert, R.A., 2002, “Could a CAMELS downgrade model improve off-site surveillance?” Rev./Federal Res. Bank St. Louis, Vol. 84, No. 1, pp. 47–63. 
- Persons, Obeua, 1999, “Using Financial Information to Differentiate Failed vs. Surviving Finance Companies in Thailand: An Implication for Emerging Economies, Multinational Finance Journal, 1999, Vol. 3, No. 2, pp. 127–145 
- Rojas-Suarez, L., 2001, “Rating Banks in Emerging Markets: What Credit Agencies Should Learn from Financial Indicators,” Institute for International Economics Working Paper, 01-6, May. 
- Sahajwala, Ranjana, and Paul Van den Bergh, 2000, “Supervisory Risk Assessment and Early Warning Systems,” Basel Committee on Banking Supervision, Working Paper No. 4. 
- Martin, Daniele, 1977, “Early Warning of Bank Failure: A Logit Regression Approach,” Journal of Banking and Finance, Vol. 1, No. 3, December, pp. 249–276. 
- Hardy, Daniel, 1998, “Are Banking Crises Predictable?” Finance & Development (Washington: International Monetary Fund).  

### Banking distress, crises, and resolution
- Calomiris, C., and J. Mason, 2000, “Causes of U.S. Bank Distress During the Depression,” NBER Working Paper 7919 (Cambridge, MA: NBER). 
- Gonzalez-Hermosillo, B., Pazarbasioglu, C., and R. Billings, 1997, “Determinants of Banking System Fragility: A Case Study of Mexico,” International Monetary Fund Staff Papers. 44 (September): 295–314. 
- Eisenbeis, Robert A., and George G. Kaufman, 2005, “Bank crisis resolution and foreign-owned banks,” Economic Review, Federal Reserve Bank of Atlanta, Q4, pp. 1–18. 
- King, Thomas B., Daniel A. Nuxoll, and Timothy J. Yeager, 2006, “Are the Causes of Bank Distress Changing? Can Researchers Keep Up?” Federal Reserve Bank of St. Louis Review, January/February, pp. 57–80. 
- Koetter, M., Bos, J. W. B., Heid, F., Kolari, J. W., Kool, C. J. M., and D. Porath, 2007, “Accounting for Distress in Bank Mergers,” Journal of Banking and Finance, Vol. 31, No. 10, pp. 3200–17. 
- Kraft, E. and Galac, T., 2007, “Deposit Interest Rates, Asset Risk and Bank Failure in Croatia,” Journal of Financial Stability, 2, pp. 312–336. 
- Kick, T. and M. Koetter, 2007, “Slippery Slopes of Stress: Ordered Failure Events in German Banking,” Journal of Financial Stability, 3(2), pp. 132–48. 
- Calomiris, C., and J. Mason, 2000, “Causes of U.S. Bank Distress During the Depression,” NBER Working Paper 7919 (Cambridge, MA: NBER). 

### European banking structures, regulation, and policy
- De Haan, Jakob, Sander Oosterloo, and Dirk Schoenmaker, 2009, European Financial Markets and Institutions (Cambridge: Cambridge University Press). 
- Decressin, Jörg, Hamid Faruqee, and Wim Fonteyne (eds), 2007, Integrating Europe’s Financial Markets (Washington: International Monetary Fund). 
- Deutsche Bundesbank, 2004, Report on the Stability of the German Financial System, Monthly Report, October. 
- European Central Bank, 2005, EU Banking Structures, October (Frankfurt: European Central Bank) http://www.ecb.int/pub/pdf/other/eubankingstructure102005en.pdf.  
- European Central Bank, 2006, EU Banking Structures, October (Frankfurt: European Central Bank). http://www.ecb.int/pub/pdf/other/eubankingstructures2006en.pdf.  
- Nieto, Maria, David Mayes, and Larry Wall, 2008, “Multiple Safety Net Regulators and Agency Problems in the EU: Is Prompt Corrective Action Partly the Solution?” Bank of Spain Working Paper No. 08/19. 
- Schoenmaker, Dirk, and Sander Oosterloo, 2007, “Cross-Border Issues in European Financial Supervision,” in David Mayes and Geoffrey Wood (eds.), The Structure of Financial Regulation (London: Routledge). 
- Lannoo, Karel, 2008, “The Maastricht Criteria for Banking,” CEPS Commentary, February 4 (www.ceps.eu). 
- International Monetary Fund, 2007, Euro Area Policies: 2007 Article IV Consultation—Staff Report, IMF Country Report No. 07/260 (Washington: International Monetary Fund). 
- Čihák, Martin, and Jörg Decressin, 2007, “The Case for a European Banking Charter”, IMF Working Paper No. 07/173 (Washington: International Monetary Fund).  

### Regulatory handbooks and institutional reports
- Federal Deposit Insurance Corporation, 2003, “Resolutions Handbook.” Available at: http://www.fdic.gov/bank/historical/reshandbook/index.html. 
- Sahajwala, Ranjana, and Paul Van den Bergh, 2000, “Supervisory Risk Assessment and Early Warning Systems,” Basel Committee on Banking Supervision, Working Paper No. 4. 
- Curry, Timothy, Peter Elmer, and Gary Fissel, 2003, “Using Market Information to Help Identify Distressed Institutions: A Regulatory Perspective.” FDIC Banking Review, Vol. 15, no.3, pp. 1–16. 

*Source: _wp0909 - REFERENCES*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2009/_wp0909.pdf_