## _wp1394

## Source details

**Canonical URL:** [_wp1394](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1394.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1394.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1394.pdf.json)

---

### I. Introduction and research questions
- Bank bailouts during and after the 2007–2009 financial crisis renewed debate on how government support affects banks’ management incentives and competition.
- Two theoretical channels:
  - market discipline hypothesis: government support decreases the incentive of outside investors (depositors, creditors, and shareholders) to monitor or influence bank risk taking (Merton, 1977; Flannery and Sorescu, 1996).
  - charter value hypothesis: government support decreases banks’ funding costs, increases interest margins and charter values, and leads banks to take fewer risks to protect future rents (Keeley, 1990).
- Primary empirical questions:
  - Do banks with more explicit or implicit government support take on more risk?
  - Does bank regulation limit the effect of government support on bank risk taking?

### II. Key definitions, sample, and measurement
- Bank risk:
  - z-score = (return on assets + capital to asset ratio) / standard deviation of return on assets.
  - Higher z-score indicates less risk; natural logarithm of z-score used due to skewness.
  - σ(ROA) computed using 5 year rolling windows.
- Government support:
  - Primary measure: difference (in rating notches) between a bank’s BFSR and its long-term foreign currency deposit rating (Moody’s-based).
  - Alternative: probability-of-bailout p = 1 − td/d using BFSR-based default probability (d) and deposit-rating-based total default probability (td); Fitch-based probabilities used as robustness.
- Sample and periods:
  - Data across 54 countries.
  - 286 banks for the period 2003–2004; 321 banks for the period 2009–2010.
  - Banks required to be rated by Moody’s or Fitch; subsidiaries excluded.
  - Results robust to excluding countries with less than 2 banks.
- Predictive evidence:
  - Median probability of support estimated by Moody’s increases from 0% to 40% (2002 to 2008 comparison noted).
  - Probit prediction of actual interventions (2008–2010): banks that enjoyed support in 2007 were more likely to be rescued in 2008–2010 by 30 percentage points; Moody’s measure drives baseline specifications.

### III. Hypothesis and empirical strategy
- Baseline specification:
  - ln(z)_b,c,t = β0 + β1 × GS_b,c,t−1 + β2 × X_b,c,t−1 + β3 × W_c,t−1 + ε_b,c,t
    - GS is government support lagged by at least one period; standard errors clustered at country level.
- Endogeneity/addressing identification:
  - Saturated regressions with bank and country measures.
  - Instrumental variables: instrument GS_b,c,t−1 with average GS of other n−1 banks in same country.
- Interaction hypothesis:
  - Regulation/supervision may alter GS impact:
    - ln(z)_b,c,t = β0 + β1 × GS_b,c,t−1 + β2 × R_c,t−1 + β3 × GS_b,c,t−1 × R_c,t−1 + controls.

### IV. Main empirical findings
- Benchmark and timing:
  - Larger government support associated with greater risk taking (negative GS coefficients for ln(z-Score)) in almost all specifications.
  - Relationship stronger and more statistically significant in 2009–2010 relative to 2003–2004.
  - Results robust to controls, country fixed effects, IV, and alternative GS measures.
- Magnitudes and statistical results (selected exact findings):
  - 2003–2004: a 4.5 percent decrease on the average z-score is reported for one specification but not statistically significant.
  - 2009–2010: a one standard deviation increase in government support associated with a 6.9 percent increase in bank risk taking (relative to the average z-Score); GS coefficient negative and statistically significant.
  - Table 6 selected coefficients (Government support in notches — dependent variable ln(z-Score)):
    - 2003–2004, simple bank controls (col 1): Government support = -0.048 [0.052] (not significant).
    - 2003–2004, country fixed effects (col 4): Government support = -0.133*** [0.034].
    - 2009–2010, simple bank controls (col 8): Government support = -0.081*** [0.030].
    - 2009–2010, country fixed effects (col 11): Government support = -0.134*** [0.037].
  - Table 7 (z-Score components):
    - 2003–2004, ROA: Government support = -0.087*** [0.027].
    - 2003–2004, Std. ROA: Government support = 0.050** [0.020].
    - 2009–2010, Equity/Assets: Government support = -0.495*** [0.152].
  - Probability-based GS (Table 8, Moody’s probability):
    - 2003–2004, bank controls (col 1): Government support = -0.468* [0.245].
    - 2003–2004, country fixed effects (col 3): Government support = -0.522** [0.222].
    - 2009–2010, bank controls (col 5): Government support = -0.494*** [0.182].
    - 2009–2010, country fixed effects (col 7): Government support = -0.474** [0.192].
  - Fitch probabilities (Table 9): GS generally not significant pre-crisis; GS negative and significant for 2009–2010 in country fixed effects (col 7): -0.458*** [0.159].
- Controls and mechanisms:
  - Bank-level controls:
    - Revenue growth captures charter value effect; size almost never significant on its own.
    - Banks with higher liquidity take significantly more risks.
  - Country-level:
    - Per capita income and inflation volatility significant across periods: inflation volatility increases always cause riskier banks.
  - IV and ownership:
    - Instrumental variables regressions confirm GS positive association with risk in the crisis period; magnitude similar.
    - Controlling for ownership (cash-flow rights, large shareholders, owner type) does not change main GS–risk result.
- Robustness:
  - Longer lags (2001/2007 supports) yield similar results.
  - Alternative risk measures: GS affects ROA pre-crisis (more volatile returns) and Equity/Assets during crisis (reduced capital buffers).
  - Probability-of-bailout measure: during crisis, one standard deviation increase in probability leads to an 8 percent increase in risk (relative to mean), significant at 1 percent.
  - Joint determination with Tobin’s Q (two-stage GMM): GS remains significant (albeit at ten percent) for the second period; bank value not significant.

### V. Regulation, supervision, and interaction results
- Regulation data: Barth, Caprio, and Levine indices for 2003 and 2008 (capital stringency, official supervisory power, activity restrictions).
- Cross-section interaction results (Table 11, selected 2009–2010 coefficients):
  - Government support coefficients across columns (examples): -0.017 [0.103]; -0.373* [0.186]; -0.432*** [0.082]; -0.481** [0.183].
  - Support × Activity restrictions (col 7): 0.035*** [0.007]; (col 8): 0.033*** [0.009].
- Interpretation:
  - Activity restrictions reduce the magnitude of moral hazard associated with GS during the crisis, but do not fully offset it.
  - Capital requirements and official supervisory power prior to the crisis were not sufficient to prevent additional risk taking by banks with more government support during the crisis.

### VI. Policy implications and recommendations
- Empirical support points to market-discipline channel as dominant in sample: GS intensity positively related to bank risk taking, stronger in 2009–2010.
- Recommended measures aimed at re-establishing private monitoring and limiting moral hazard:
  - Increase transparency and force more disclosure by bank managers.
  - Mandate periodic issuance of subordinated debt to create monitoring incentives.
  - Use market information to improve the quality of supervision (Rochet, 2005).
  - Consider restricting banks’ range of activities (security markets, insurance, real estate, ownership of non-financial firms) to limit risk taking by government-supported banks; simple rules like Glass-Steagall could be reconsidered (Haldane, 2012).
- Alternative policy if charter value channel dominated (contrasting view):
  - Capital surcharge on protected banks to decrease rents and competitors’ incentives to take on more risk (Gropp, Hakenes, and Schnabel, 2011).
  - Increased capital requirements to reduce gambling incentives by placing more equity at risk (but pre-crisis capital requirements did not achieve this in practice).
- Additional considerations:
  - Increased bank complexity over the past decade may have reduced effectiveness of investor monitoring, making disclosure and simplicity-focused regulation more valuable.
  - Limitations on governments’ ability to pledge support could be considered to enhance private monitoring.
- Governance and future research:
  - Investigate governance variables beyond ownership (CEO incentives, board size).
  - Determine optimal bank governance structure given a desired level of systemic risk remains open.

### VII. Selected sample composition and descriptive statistics (exact figures)
- Total banks in sample:
  - 2003–2004: 330
  - 2009–2010: 347
- Summary statistics (selected exact entries):
  - z-Score (2003–2004): N = 286, Mean = 3.30, Median = 3.41, Std. dev. = 1.14.
  - z-Score (2009–2010): N = 321, Mean = 3.02, Median = 3.17, Std. dev. = 0.94.
  - Return on assets (percent) (2003–2004): N = 313, Mean = 0.99, Median = 0.81, Std. dev. = 1.03.
  - Return on assets (percent) (2009–2010): N = 332, Mean = 0.66, Median = 0.56, Std. dev. = 2.44.
  - Size (in $ billions) (2003–2004): N = 313, Mean = 128.61, Median = 32.49, Std. dev. = 240.61.
  - Size (in $ billions) (2009–2010): N = 332, Mean = 225.02, Median = 49.29, Std. dev. = 456.03.
  - Moody’s support (in ratings notches) (2003–2004): N = 331, Mean = 1.03, Median = 0.00, Std. dev. = 2.60.
  - Moody’s support (in ratings notches) (2009–2010): N = 348, Mean = 1.66, Median = 2.00, Std. dev. = 2.64.
  - Moody’s support (in probability) (2003–2004): N = 331, Mean = 0.26, Median = 0.00, Std. dev. = 0.36.
  - Moody’s support (in probability) (2009–2010): N = 348, Mean = 0.42, Median = 0.40, Std. dev. = 0.36.
  - Fitch support (in probability) (2003–2004): N = 137, Mean = 0.54, Median = 0.62, Std. dev. = 0.39.
  - Fitch support (in probability) (2009–2010): N = 205, Mean = 0.55, Median = 0.61, Std. dev. = 0.37.
- Correlations (selected exact entries):
  - 2003–2004: Size–Moody’s support (notches) = 0.153***.
  - 2009–2010: Size–Moody’s support (notches) = 0.306***.
- Government interventions (Europe, 2008–2010) selected counts:
  - Germany: 193 / 322
  - Spain: 92 / 211
  - United Kingdom: 113 / 314
  - Total (listed): 934 intervened, 4,137 total
- Probit marginal effects (intervention regressions, Table 5):
  - Size coefficient: 0.327*** [0.112]
  - Government support in 2007: 22.694* [13.423]
  - Loan loss provisions: 3,865.50 [2,351.445]
  - Support in 2007 × Loan loss provisions: -3,798.37 [2,370.847]
  - Observations: 123; Countries: 17
  - Marginal effect of government support: 0.273*** [0.0735]

*Source: _wp1394 (PDF chapter/section content provided).*

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

### References

### I. Introduction and research questions
- Bank bailouts during and after the 2007–2009 financial crisis have reignited the debate on the effect of government support on banks’ management incentives and on competition in the banking sector.
- Two channels through which explicit and implicit government support can influence banks’ willingness to take on risk:
  - market discipline hypothesis: government support decreases the incentive of outside investors (depositors, creditors, and shareholders) to monitor or influence bank risk taking; risk-shifting may occur if deposit insurance is not fairly priced (Merton, 1977) or if governments provide guarantees to holders of bank debt (Flannery and Sorescu, 1996).
  - charter value hypothesis: government support decreases banks’ funding costs as both depositors and creditors demand lower rates; the decline in funding costs increases interest margins and raises banks’ charter values, leading banks to take fewer risks to protect future rents (Keeley, 1990).
- Primary empirical questions:
  - Do banks with more explicit or implicit government support take on more risk?
  - Does bank regulation limit the effect of government support on bank risk taking?

### II. Key definitions, sample, and measurement
- Bank risk is defined as the z-score: (return on assets plus capital to asset ratio) divided by the standard deviation of return on assets; a measure of distance to default.
- Government support is measured as the difference between each bank’s deposit rating and bank financial strength rating assigned by Moody’s Investors Service; the study notes this measure is able to predict actual bank bailouts.
- Sample periods: 2003-2004 (before the crisis) and 2009-2010 (during the crisis).
- Sample scope: two cross-country samples of banks covering several developed and emerging countries; sample excludes bank subsidiaries and includes the post-financial crisis period.

### III. Main empirical findings
- After controlling for bank size, liquidity, level of bank regulations, banks’ ownership structure, degree of market concentration in the banking sector, and country-specific macroeconomic conditions:
  - The intensity of government support is positively related to measures of bank risk taking (z-score).
  - The relationship is stronger for the 2009–2010 period relative to 2003–2004.
  - The result is robust to several checks, including addressing possible endogeneity of government support.
  - Conclusion: market discipline is the dominant factor shaping the relationship between support and risk in the banking industry in this sample.
- Regulatory scope findings:
  - Adoption of regulatory impediments restricting bank activities involving security markets, insurance, real estate, and ownership of non-financial firms reduces the magnitude of the moral hazard problem associated with government support.
  - Pre-crisis capital supervision and regulation were not enough to fully prevent additional risk taking by banks with more government support during the crisis.
  - Pre-crisis capital requirements did not weaken the link between government support and risk taking by banks.

### IV. Comparison with prior literature
- Prior studies focused on explicit support (deposit insurance, state ownership) or indirect measures (too-big-to-fail) with mixed results.
- Differences highlighted relative to other studies:
  - Forssbaeck (2011): focuses on 1995–2005 and finds no support for market discipline becoming more important during crises.
  - Dam and Koetter (2012): support for market discipline for 1995–2006 but restricted to German banks and use of actual bailouts to derive support measures.
  - Gropp, Hakenes, and Schnabel (2011): find expected government support induces more risk taking by competitors but not consistently by protected banks; their findings suggest charter value channel dominant for protected banks.
- This study’s distinctions:
  - Uses the z-score, a broader measure of risk encompassing credit risk and market risk.
  - Includes post-financial crisis period and excludes bank subsidiaries.
  - Focuses on expected, not actually received, government support.

### V. Policy implications and recommendations
- Given empirical support for the market-discipline channel and the ineffectiveness of pre-crisis capital requirements in reducing moral hazard:
  - Prefer measures that increase incentives for depositors and subordinated creditors to monitor or influence banks’ risk attitudes.
    - Examples include imposing more transparency and forcing more disclosure by bank managers, mandating periodic issuance of subordinated debt, or using market information to improve the quality of supervision (Rochet, 2005).
  - Consider restricting banks’ range of activities to limit risk taking even for government-supported banks; simple rules like those included in the Glass-Steagall Act could potentially be reconsidered (Haldane, 2012).
- Contrast with policy prescriptions if charter value channel dominated:
  - A capital surcharge on protected banks to decrease rents and competitors’ incentives to take on more risk (Gropp, Hakenes, and Schnabel, 2011).
  - Increased capital requirements could also reduce gambling incentives by putting more bank equity at risk, though the study finds pre-crisis capital requirements did not achieve this in practice.
- Note on complexity: increased bank complexity over the past decade may have decreased effectiveness of investor monitoring, making disclosure and simplicity-focused regulation relatively more valuable.

*Source: _wp1394 - References . . . (PDF chapter/section content provided).*

### Section V, we discuss how regulation may affect this result. Section VI concludes.

### _wp1394 - Section V, we discuss how regulation may affect this result. Section VI concludes.

### II. DATA — Bank risk taking, support, controls, and summary statistics
- Bank risk measure:
  - z-score = (ROA + CAR) / σ(ROA), where ROA is return on assets, CAR is the capital asset ratio, and σ(ROA) is the standard deviation of ROA.
  - A higher z-score indicates that the bank is less risky.
  - The natural logarithm of the z-score is used because the z-score is highly skewed.
  - The standard deviation of ROA is computed using 5 year rolling windows.
- Sample and periods:
  - Data across 54 countries.
  - 286 banks for the period 2003–2004.
  - 321 banks for the period 2009–2010.
  - Banks are required to be rated by Moody’s or Fitch.
  - Number of banks per country varies from 1 to 30.
  - Results are robust to excluding countries with less than 2 banks.
  - Focus on cross-sectional analysis due to a change in accounting standards (GAAP → IFRS) in the mid-2000s.
- Bank support measures:
  - Primary measure: difference (in rating notches) between a bank’s BFSR (Bank Financial Strength Rating) and its long-term foreign currency deposit rating (Moody’s-based).
  - Alternative/support robustness: probability of a government bailout using BFSR-based default probability (d) and deposit-rating-based total default probability (td) with bailout probability p = 1 − td/d; Fitch provides a similar measure used as an additional robustness check.
  - Subsidiaries excluded to eliminate support from parents and other external support.
- Predictive evidence (descriptive preview):
  - Figure 1 (referenced) shows average and median government support evolution since 1996; support increases during periods of economic distress.
  - Median probability of support estimated by Moody’s increases from 0% to 40% from the first sample period to the second one (2002 to 2008 comparison noted).
- Control variables:
  - Bank-level: revenue growth (growth in total revenues relative to previous period), size (log of total assets), liquidity (liquid assets to liquid liabilities), cash flow rights of large shareholders, ownership dummies (government, institutional, individual, other). Data from Bankscope, Capital IQ, SNL Financial, banks’ websites.
  - Industry/country-level: per capita income, inflation, inflation variability (World Bank Development Indicators), investor protection quality and contract enforcement (Doing Business Report 2003 and 2009).
  - Bank concentration: Hirsch-Herfindahl index (Bankscope).
  - Regulatory/banking institutional variables: existence of a deposit insurance scheme; level of capital requirements (minimum capital-asset ratio requirement). Deposit insurance data from Demirguc-Kunt, Kane, and Laeven (2008), Institute International Bankers (Global Surveys 2009 and 2010), IADI, Central Bank of Egypt, SDIC.
  - Regulatory intensity and breadth: Barth, Caprio, and Levine (2006) indices — capital stringency, official supervisory power, and activity restrictions. For 2003–2004 cross-section use 2003 regulatory database; for 2009–2010 cross-section use 2008 version.
- Summary statistics and changes across periods:
  - Statistics based on averages for periods 2003–2004 and 2009–2010 using annual data for z-score; other variables use annual data for 2002 and 2008.
  - Measured risk-taking increases by 0.3 standard deviations of the z-score when comparing 2003–2004 to 2009–2010.
  - Average size of banks increases by 0.5 standard deviations when comparing 2003–2004 to 2009–2010.
  - Noted sizable increase in average level of support from 2002 to 2008; Moody’s median support increase highlighted above.
- Predictive power of government support measures:
  - Moody’s and Fitch’s probability-based measures were mildly correlated before the crisis and more correlated with the crisis.
  - Before the crisis, Moody’s measure was uncorrelated with size while Fitch’s was significantly correlated with total assets.
  - Actual bailouts dataset:
    - Data on capital injections in Europe from Brei, Gambacorta, and von Peter (2011) complemented with Laeven and Valencia (2012) and FT.com.
    - Dataset includes 238 banks with capital injection information; ratings information for only 137 banks; roughly one third of these were intervened.
  - Probit model to predict actual government interventions (2008–2010):
    - Binary outcome y_it = 1 if bank i received a capital injection or was partially/totally nationalized between 2008 and 2010.
    - Key regressors: binary support dummy (Moody’s-based support measure > 0), loan loss provisions as % of average assets (LLP), interaction of support with LLP, GDP growth 2007–2009, and other 2007 controls.
    - Result: banks that enjoyed support in 2007 were more likely to be rescued in 2008–2010 by 30 percentage points; difference is highly significant.
    - Equivalent estimation using Fitch measure shows insignificant marginal effects; Moody’s measure is used in baseline specifications.

### III. HYPOTHESIS AND EMPIRICAL STRATEGY
- Primary hypothesis:
  - Bank risk taking is related to government support to banks.
- Baseline empirical specification:
  - Z_b,c,t = β0 + β1 × GS_b,c,t−1 + β2 × X_b,c,t−1 + β3 × W_c,t−1 + ε_b,c,t
    - Z_b,c,t is natural logarithm of the z-score of bank b in country c for period t.
    - GS_b,c,t−1 is government support for bank b from country c.
    - X_b,c,t−1 is matrix of bank-level controls.
    - W_c,t−1 are country-level controls.
    - Standard errors adjusted for clustering at the country level.
  - Causality claim based on using government support lagged by at least one period.
- Endogeneity and identification strategies:
  - Saturate regressions with many bank and country specific measures to capture omitted variables.
  - Instrumental variables approach:
    - Instrument each bank’s government support with the average GS of the other n−1 banks in the same country.
    - Instrument validity relies on assumption that an innovation in one bank’s risk taking does not affect government support to other banks.
- Second hypothesis — interaction effects:
  - Banking supervision and regulation affect the impact of government support on banks’ risk taking.
  - Specification with interaction:
    - Z_b,c,t = β0 + β1 × GS_b,c,t−1 + β2 × R_c,t−1 + β3 × GS_b,c,t−1 × R_c,t−1 + β4 × X_b,c,t−1 + β5 × W_c,t−1 + ε_b,c,t
    - R_c,t−1 are country-specific regulatory standards (e.g., capital stringency, official supervisory power, activity restrictions).
    - β3 captures the interaction effect between bank-specific government support and national regulations.

### IV. RESULTS — Benchmark regression and main findings
- Main empirical findings (benchmark regression, Table 6 referenced):
  - Larger government support is associated with greater risk taking by banks, reflected in the negative coefficient for government support (GS) in almost all specifications.
  - The relationship between government support and bank risk taking is present for both 2003–2004 and 2009–2010 periods.
  - Coefficients are generally more statistically significant during the 2009–2010 period.
  - Regressions 1 and 8 (which control for recent bank performance — revenue growth) show that a one standard deviation increase in government support is associated with a [text truncated in source at this point].

*Italic: Source: _wp1394 - Section V, we discuss how regulation may affect this result. Section VI concludes.*

### 4.5 percent decrease on the average z-score for the 2003–2004 period, but the relationship is

### _wp1394 - 4.5 percent decrease on the average z-score for the 2003–2004 period, but the relationship is

### Main empirical findings
- For 2003–2004, a 4.5 percent decrease on the average z-score is reported, but the relationship is not statistically significant.
- For 2009–2010, the government support coefficient is negative and statistically significant; a one standard deviation increase in government support is associated to a 6.9 percent increase in bank risk taking, relative to the average z-Score.
- The positive association between government support (GS) and risk holds when controlling for bank characteristics and country-level features, and after including country fixed effects.

### Bank-level controls and interpretation
- Regressions controlling for bank-specific characteristics (revenue growth, size, liquidity ratio) show:
  - The positive association between GS and banks’ risk-taking remains significant for 2009–2010 and insignificant for 2003–2004.
  - Revenue growth seems to capture the charter value effect; size on its own is almost never significant.
  - Banks with higher liquidity take (significantly) more risks — interpreted as liquidity capturing a bank-specific appetite for risk.
- Possible explanations noted:
  - The too-big-to-fail effect and the charter value hypothesis may cancel each other out for size.
  - Larger banks may be better at risk diversification but harder to monitor due to complexity.

### Country-level controls, endogeneity, and ownership
- Regressions with country fixed effects and with country-specific characteristics (per capita income, capital requirements, investment protection, deposit insurance presence, rule of law enforcement, Herfindahl concentration index):
  - The result that government support leads to riskier banks is robust to conditioning on either country controls or fixed effects.
  - Of country controls, only per capita income and inflation volatility are significant for both time periods.
    - Inflation volatility increase always causes riskier banks.
    - Change in sign for per capita income reflects that advanced economies were most affected by the 2007–2009 crisis.
- Instrumental variables results (regressions 3 and 10) confirm GS is positively and significantly associated with bank risk taking for the crisis period; the magnitude of the GS coefficient does not change.
- Controlling for bank ownership structure (cash-flow rights; large shareholders; owner types: government, institutions, individuals, others) — regressions 6 and 13 — the positive and significant association between bank risk taking and government support is robust.

### Timing and lag structure
- Benchmark specifications lag GS by one period (z-Score averages for 2003–2004 and 2009–2010 regressed on 2002 and 2008 supports, respectively).
- Longer lag regression (z-Score averages on 2001 and 2007 support) yields basically the same results (columns 7 and 14).

### Robustness exercises
- Alternative risk measures (individual z-score components: ROA, Capital to Assets, standard deviation of ROA):
  - Strong and statistically significant effect of GS on ROA regardless of period.
  - Pre-crisis (2003–2004): GS positively and significantly related to volatility of ROA.
  - Crisis sample (2009–2010): GS negatively and significantly related to capital to assets ratio.
  - Interpretation: Pre-crisis support encouraged riskier bets translating into more volatile returns; post-crisis, either increased leverage or losses reduced capital buffers.
  - Size matters for individual components (particularly in the second period): larger banks more leveraged (too-big-to-fail) but have less volatile ROA (diversification).
- Notches-based GS replaced with probability-of-bailout measure (Gropp, Hakenes, and Schnabel (2011) style):
  - During the crisis, a one standard deviation increase in the probability of a bailout led to an 8 percent increase in risk (relative to the mean); effect significant at the 1 percent significance level.
- Using Fitch Ratings based probabilities:
  - GS not significant for the pre-crisis period (2003–2004).
  - For the crisis period (2009–2010), strong evidence of moral hazard consistent with Moody’s-based results.
- Charter value channel and market concentration:
  - Market concentration (Hirsch-Herfindahl index) never significant in columns noted.
  - Additional tests with barriers to entry and number/change in number of banks normalized by GDP did not change results.
- Joint determination of bank risk and valuation:
  - Expanded specification includes Tobin’s Q as endogenous; Tobin’s Q = (total assets + market value of equity - book value of equity) / total assets.
  - Estimated by two-stage GMM with two excluded instruments: dummy for bank’s stock being widely held (Widely) and number of banks normalized by country GDP.
  - For the second period only (data availability): when explaining bank risk, GS remains significant (albeit at the ten percent level) but bank value is not.

### Regulation and government support
- Uses regulation data for 2003 and 2008 from Barth, Caprio, and Levine (2008): index of regulatory oversight of bank capital, capital stringency, official supervisory power, activity restrictions.
- Cross-section regressions for 2003–2004 and 2009–2010 including all bank and country controls show:
  - 2003–2004: GS not a significant factor for bank risk taking; regulation did not play a significant role.
  - 2009–2010 (crisis period): GS is associated with more risk taking by banks.
  - Interaction coefficient for activity restrictions and GS is positive and significant during the crisis, indicating that limiting permissible activities reduced bank risk taking, but did not fully offset the moral hazard from GS.
- Example cited: May 2012 J.P. Morgan trading losses underscore relevance of regulatory findings and debates (Volcker rule, proprietary trading limitations).

### Conclusion and policy implications
- Theoretical ambiguity: GS can increase risk via moral hazard (market discipline hypothesis) or reduce risk by increasing charter value (charter value hypothesis).
- Empirical conclusion:
  - Intensity of GS is positively related to measures of bank risk taking.
  - Relationship is stronger for 2009–2010 relative to 2003–2004.
  - Results robust to endogeneity and alternative risk measures.
  - Capital requirements regulation and enhanced supervisory powers failed to curb risk taking due to GS.
- Policy recommendations and potential measures:
  - Increase incentives for depositors, small shareholders, and subordinated creditors to monitor or influence banks to decrease moral hazard associated with GS.
  - Create regulatory environments that do not hinder private monitoring and consider limits on the amount of support governments can pledge.
  - Alternatively, restrict banks’ ability to engage in activities involving security markets, insurance, real estate, and ownership of non-financial firms to weaken link between GS and risk taking.
  - Further research needed on how such activity restrictions ameliorate the problem (reducing risky activities versus reducing complexity to facilitate monitoring).
- Governance and future research directions:
  - Investigate bank governance variables beyond ownership (e.g., CEO incentives, board size).
  - The optimal bank governance structure given a desired level of systemic risk remains an open question for future research.

*Source: IMF working paper content provided in the input.*

### REFERENCES

### _wp1394 - REFERENCES

### Key references cited
- Admati, Anat R., and Paul Pfleiderer, 2010, “Increased-Liability Equity: A Proposal to Improve Capital Regulation of Large Financial Institutions,” Stanford University Graduate School of Business Research Paper No. 2043.
- Altunbas, Yener, Leonardo Gambacorta, and David Marques-Ibanez, 2010, “Bank Risk and Monetary Policy,” Journal of Financial Stability, Vol. 6, No. 3, pp. 121 – 129.
- Barth, James R., Gerard Caprio, and Ross Levine, 2006, Rethinking Bank Regulation (Cambridge University Press).
- Basel Committee on Banking Supervision, 2009, Strengthening the Resilience of the Banking Sector, Consultative Document (Basel: Bank of International Settlements).
- Bitler, M.P., T.J. Moskowitz, and A. Vissing-Jørgensen, 2005, “Testing Agency Theory with Entrepreneur Effort and Wealth,” The Journal of Finance, Vol. 60, No. 2, pp. 539–576.
- Brei, Michael, Leonardo Gambacorta, and Goetz von Peter, 2011, “Rescue Packages and Bank Lending,” BIS Working Papers 357, Bank for International Settlements.
- Laeven, Luc, and Fabian Valencia, 2012, “Systemic Banking Crises Database: An Update,” IMF Working Papers 12/163, International Monetary Fund.
- Gropp, Reint, Hendrik Hakenes, and Isabel Schnabel, 2011, “Competition, Risk-shifting, and Public Bail-out Policies,” Review of Financial Studies, Vol. 24, No. 6, pp. 2084-2120.
- Haldane, Andrew G., 2012, “The Dog and the Frisbee,” Speech given at the Federal Reserve Bank of Kansas City’s 36th economic Policy symposium, Jackson Hole, Wyoming (August).
- Reinhart, Carmen M., and Kenneth S. Rogoff, 2009, “The Aftermath of Financial Crises,” The American Economic Review, Vol. 99, No. 2, pp. 466–472.
- Schich, Sebastian, and Sofia Lindh, 2012, “Implicit Guarantees for Bank Debt: Where Do We Stand?” OECD Journal: Financial Market Trends, Vol. 2012, No. 1, pp. 1–22.
- (Complete bibliography appears in the source PDF; above are select entries preserved exactly as in the source.)

### Sample composition and descriptive statistics
- Sample by country and period (Table 1): Total banks in sample:
  - 2003–2004: 330
  - 2009–2010: 347
- Summary statistics (Table 2) — selected exact figures:
  - z-Score (2003–2004): N = 286, Mean = 3.30, Median = 3.41, Std. dev. = 1.14.
  - z-Score (2009–2010): N = 321, Mean = 3.02, Median = 3.17, Std. dev. = 0.94.
  - Return on assets (percent) (2003–2004): N = 313, Mean = 0.99, Median = 0.81, Std. dev. = 1.03.
  - Return on assets (percent) (2009–2010): N = 332, Mean = 0.66, Median = 0.56, Std. dev. = 2.44.
  - Size (in $ billions) (2003–2004): N = 313, Mean = 128.61, Median = 32.49, Std. dev. = 240.61.
  - Size (in $ billions) (2009–2010): N = 332, Mean = 225.02, Median = 49.29, Std. dev. = 456.03.
  - Moody’s support (in ratings notches) (2003–2004): N = 331, Mean = 1.03, Median = 0.00, Std. dev. = 2.60.
  - Moody’s support (in ratings notches) (2009–2010): N = 348, Mean = 1.66, Median = 2.00, Std. dev. = 2.64.
  - Moody’s support (in probability) (2003–2004): N = 331, Mean = 0.26, Median = 0.00, Std. dev. = 0.36.
  - Moody’s support (in probability) (2009–2010): N = 348, Mean = 0.42, Median = 0.40, Std. dev. = 0.36.
  - Fitch support (in probability) (2003–2004): N = 137, Mean = 0.54, Median = 0.62, Std. dev. = 0.39.
  - Fitch support (in probability) (2009–2010): N = 205, Mean = 0.55, Median = 0.61, Std. dev. = 0.37.

### Correlations among main variables (Table 3) — selected exact entries
- Panel A: 2003–2004
  - Correlation Size–Moody’s support (notches): 0.153***.
  - Correlation Moody’s support (notches)–Moody’s support (probability): 0.843***.
  - Correlation Moody’s support (notches)–Fitch support (probability): 0.439***.
- Panel B: 2009–2010
  - Correlation Size–Moody’s support (notches): 0.306***.
  - Correlation Moody’s support (notches)–Moody’s support (probability): 0.714***.
  - Correlation Moody’s support (notches)–Fitch support (probability): 0.521***.
- Robust standard errors in brackets: * p<0.10, ** p<0.05, *** p<0.01.

### Government interventions and support: descriptive counts and probit marginal effects
- Government interventions in banks in Europe, 2008–2010 (Table 4) — selected country counts (Intervened / Total):
  - Germany: 193 / 322
  - Spain: 92 / 211
  - United Kingdom: 113 / 314
  - Total (listed): 934 intervened, 4,137 total
- Probit regression for intervention (Table 5):
  - Size coefficient: 0.327*** [0.112]
  - Government support in 2007: 22.694* [13.423]
  - Loan loss provisions: 3,865.50 [2,351.445]
  - Support in 2007 X Loan loss provisions: -3,798.37 [2,370.847]
  - Observations: 123
  - Countries: 17
  - Marginal effect of government support: 0.273*** [0.0735]
  - Robust standard errors in brackets: *** p<0.01, ** p<0.05, * p<0.1

### Main regression results on government support and bank risk-taking (selected exact coefficients and significance)
- Table 6: Government support (notches) — dependent variable: ln(z-Score)
  - 2003–2004, simple bank controls (column 1): Government support coefficient = -0.048 [0.052] (not significant).
  - 2003–2004, country fixed effects (column 4): Government support coefficient = -0.133*** [0.034].
  - 2009–2010, simple bank controls (column 8): Government support coefficient = -0.081*** [0.030].
  - 2009–2010, country fixed effects (column 11): Government support coefficient = -0.134*** [0.037].
  - Observations for 2009–2010 regressions range up to 321; Countries up to 54; R-squared values range (examples): 0.06, 0.07, 0.08, 0.40.
  - Robust standard errors in brackets: * p<0.10, ** p<0.05, *** p<0.01.
- Table 7: z-Score components (ROA, Std. ROA, Equity/Assets) — selected coefficients:
  - 2003–2004, ROA: Government support = -0.087*** [0.027].
  - 2003–2004, Std. ROA: Government support = 0.050** [0.020].
  - 2009–2010, Equity/Assets: Government support = -0.495*** [0.152].
  - Observations vary by regression (e.g., 198, 183, 312); R-squared values reported (examples): 0.43, 0.55, 0.65, 0.32, 0.41, 0.61.

### Results using probabilities of government support (Moody’s and Fitch)
- Table 8: Government support measured by Moody’s (probability) — dependent variable: ln(z-Score)
  - 2003–2004, bank controls (column 1): Government support = -0.468* [0.245].
  - 2003–2004, country fixed effects (column 3): Government support = -0.522** [0.222].
  - 2009–2010, bank controls (column 5): Government support = -0.494*** [0.182].
  - 2009–2010, country fixed effects (column 7): Government support = -0.474** [0.192].
  - Observations examples: 286, 250, 321, 317.
- Table 9: Government support measured by Fitch (probability) — dependent variable: ln(z-Score)
  - 2003–2004, country fixed effects (column 3): Government support = -0.260 [0.172].
  - 2009–2010, country fixed effects (column 7): Government support = -0.458*** [0.159].
  - 2009–2010, bank controls (column 5): Government support = -0.274 [0.199].
  - Observations examples: 175, 127, 268, 261.
  - Robust standard errors in brackets: * p<0.10, ** p<0.05, *** p<0.01.

### Valuation and instrumented relations (Table 10)
- Two-stage results (selected exact entries):
  - First stage: Government support coefficient on Tobin’s Q = -0.062* [0.036] (Tobin’s Q first-stage coefficient shown as -0.006* [0.003] in first-stage column).
  - Second stage: Tobin’s Q coefficient in z-Score regression = -1.078 [3.258].
  - Observations: 244 (both stages).
  - Hansen’s J statistic for over-identification: 1.364.
  - Angrist-Pischke multivariate F test of excluded instruments: 3.84**.
  - Robust standard errors in brackets: * p<0.10, ** p<0.05, *** p<0.01.

### Interaction with supervision and regulation (Table 11)
- Cross-section regressions including country-level regulation/supervision indices (Barth, Caprio, and Levine, 2006, 2008):
  - 2003–2004 (columns 1–4) and 2009–2010 (columns 5–8) present coefficients for Government support and interactions with:
    - Capital stringency
    - Official supervisory powers
    - Activity restrictions
  - Selected 2009–2010 coefficients (columns 5–8):
    - Government support = -0.017 [0.103] (col 5)
    - Government support = -0.373* [0.186] (col 6)
    - Government support = -0.432*** [0.082] (col 7)
    - Government support = -0.481** [0.183] (col 8)
    - Support x Activity restrictions (col 7): 0.035*** [0.007]; (col 8): 0.033*** [0.009].
  - Observations:
    - 2003–2004: 148 (columns 1–4)
    - 2009–2010: 246 (columns 5–8)
  - R-squared examples: 0.43, 0.45 (2003–2004); 0.15, 0.26 (2009–2010).
  - Bank controls: Yes. Country controls: Yes.
  - Countries covered: 41 (2003–2004) and 48 (2009–2010).
  - Robust standard errors in brackets: * p<0.10, ** p<0.05, *** p<0.01.

### Figure note
- Figure 1 (Government Support, 1995–2011):
  - Blue line: median government support (by year) measured by difference between a bank’s BFSR and its long-term foreign currency deposit rating, as measured by Moody’s (left-hand scale, “Government support in notches, Moody’s”).
  - Red line (right scale): median of the same measure converted to probabilities of default as in Gropp, Hakenes, and Schnabel (2011) (“Probability of government support, Moody’s”).
  - Green line (right scale): equivalent probability of government support using data from Fitch Ratings (“Probability of government support, Fitch”).
  - Year axis shown from 1994 to 2012 with corresponding plotted scales (0 to 3 left scale; 0 to 1 right scales).

*Content and numeric values preserved exactly as in the source PDF.*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1394.pdf_