## _wp1229

## Source details

**Canonical URL:** [_wp1229](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp1229.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp1229.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp1229.pdf.json)

---

### Regulatory context and motivation
- Global financial crisis prompted revisions including:
  - Introduction of liquidity standards for internationally-active banks.
  - Binding leverage ratios.
  - Revision of capital requirements under Basel III.
- Two prudential liquidity ratios discussed:
  - Liquidity Coverage Ratio (LCR): resilience over a 30-day period.
  - Net Stable Funding Ratio (NSFR): resilience over a one-year horizon.
- A leverage ratio computed as shareholders’ capital over total assets introduced to ensure a hard minimum capital level.

### Contributions and approach of the paper
- Two primary contributions:
  - Measure structural liquidity and leverage in bank balance sheets consistent with the NSFR and the leverage ratio (EQUITY) proposed in Basel III.
  - Explore systematic differences in the relationship between structural liquidity, leverage, and subsequent probability of failure across bank types: Global banks versus Domestic banks.
- Rationale for partition:
  - Global banks: systemically important, complex, cross-border, often enjoy more stable funding in distress.
  - Domestic banks: typically smaller, retail-focused, potentially less stable funding base.

### Data, sample, and target variables
- Bank-level dataset: about 11,000 U.S. and European banks during 2001−09.
- Series yearly, spanning 2001–09; captures run-up (2001–07) and crisis (2008–09).
- Classification:
  - Global banks: internationally-active institutions with consolidated assets surpassing US$10 billion at end-2009; sample covers 91 institutions.
  - Domestic banks: domestically-owned institutions; sample tracks 10,805 institutions during 2001−09.
- Balance sheet treatment and indicators:
  - NSFR = (sum_i w_i L_i) / (sum_j w_j A_j); weights w bounded between zero and one; proposed regulation requires NSFR > 1.
  - EQUITY: shareholders’ equity to assets (leverage ratio).
  - Approximation weights used due to limited granularity:
    - Loan portfolio weight = 1.00.
    - Other earning assets weight = 0.35.
    - Fixed assets and non-earning assets weight = 1.00 (except cash and due from banks).
  - Liabilities: customer deposits split by type; other liabilities split by maturity. Core retail deposits treated as stable; other short-term funding given weight zero. Long-term liabilities and equity considered stable at the one-year horizon.
- Alternatives and robustness measures:
  - STFR (Short-Term Funding Ratio) = liabilities maturing within one-year / total liabilities.
  - Basel CAR = regulatory capital / risk-weighted assets.
- Outlier treatment: observations with liquidity or leverage below the 0.5 percentile and above the 99.5 percentile filtered out.
- Currency: balance sheets and income statements in U.S. dollar terms using market rate at bank-specific accounting closing dates.
- Failure identification: combined criteria from BankScope status changes, Basel CAR < 8 percent during 2008−09, Moody’s downgrades to E+ or E, and Laeven and Valencia (2010) for failing Global banks.

### Stylized facts and descriptive statistics
- Sample breadth and size:
  - About 11,000 banks; yearly series 2001–09.
- Size differences at end-2009:
  - Average Domestic bank balance sheet: US$0.7 billion.
  - Average Global bank balance sheet: US$527.1 billion.
  - Institution at the 99 percentile: US$2.9 trillion.
- Pre-crisis differences:
  - Global banks: thinner capital cushions and weaker structural liquidity indicators; greater reliance on non-deposit and short-term funding.
  - Gap between EQUITY and Basel CAR:
    - Global banks, first percentile: EQUITY = 1.4 percent; Basel CAR = 9.2 percent.
    - Domestic banks, first percentile: EQUITY = 2.5 percent; Basel CAR = 10.1 percent.
- Time evolution:
  - Median NSFR before the crisis relatively stable and close to one, but wide dispersion; lower extreme displayed extremely weak structural liquidity.
  - EQUITY: averages comfortable, but lower tail extremely leveraged.
  - Failed banks exhibited lower NSFR and lower EQUITY (Global failed banks mean EQUITY close to 4 percent).
  - NSFR declined pre-crisis, reverted from 2007 for Domestic banks and from 2008 for Global banks; Global banks show sharp drop at crisis peak and sharp increase thereafter (liquidity hoarding).
  - Domestic banks: equity collapsed for failed group after crisis eruption; failed Global banks’ equity increased reflecting capital injections/public support.

### Empirical approach
- Probit specification for probability of failure during 2008−09:
  - Pr(F_i = 1 | x_i) = Φ(β x_i), where F_i = 1 if bank i failed during 2008−09.
  - X_i includes pre-crisis NSFR and EQUITY; bank-level controls: CREDIT GROWTH (yearly average), NON-INTEREST INCOME (ratio of non-interest income to total income, measured by absolute value), Z-SCORE (distance to default); country variables: GDP GROWTH (yearly average 2001−07) and MONEY MARKET RATES (Monetary Conditions 2001-07).
- Pre-crisis averages used; cross-sectional specification; no bank fixed effects.
- Interpretation framework tests whether pre-crisis buffers reflect optimal risk management or failure to internalize risk.

### Baseline empirical results (selected estimates and marginal effects)
- Main findings:
  - NSFR and EQUITY in the years before the crisis negatively and significantly associated with probability of failure (coefficients negative and significant at the 1 percent level).
  - Credit growth positively associated with probability of failure; Z-score negatively associated (higher Z-score reduces failure probability).
  - NON-INTEREST INCOME not statistically significant in baseline regressions.
  - Country macro variables significant: higher pre-crisis GDP growth associated with higher failure probability; easier monetary conditions associated with higher failure probability (monetary conditions coefficient negative, thus tighter conditions reduce failure probability).
  - Model fit: pseudo R-square low; model explains less than five percent of variation in failure probability in baseline.
- Marginal impacts (change in probability of failure associated with a 0.5 standard deviation change; Table reporting column [4] values):
  - NSFR: Regression Coefficient -0.0430; Mean 0.986; Change in Variable 0.104; Change in Pr. Failure -0.45 percentage points.
  - Equity Capital to Assets: Regression Coefficient -0.1990; Mean 0.107; Change in Variable 0.031; Change in Pr. Failure -0.63 percentage points.
  - Credit Growth: Regression Coefficient 0.0490; Mean 0.130; Change in Variable 0.081; Change in Pr. Failure 0.39 percentage points.
  - Z-Score: Regression Coefficient -0.0470; Mean 0.469; Change in Variable 0.230; Change in Pr. Failure -1.08 percentage points.
  - Non-Interest Income to Assets: Regression Coefficient -0.2530; Mean 0.010; Change in Variable 0.006; Change in Pr. Failure -0.16 percentage points.
  - GDP Growth 2001-07: Regression Coefficient 4.0480; Mean 0.026; Change in Variable 0.005; Change in Pr. Failure 2.16 percentage points.
  - Monetary Conditions 2001-07: Regression Coefficient -0.9730; Mean 0.031; Change in Variable 0.007; Change in Pr. Failure -0.67 percentage points.
- Example simulations (column [6] referenced):
  - A 10.4 percentage point increase in the NSFR, from 0.99 to 1.09, would cause a 0.46 percentage point drop in the probability of failure of the average bank, all else equal.
  - A 3.1 percentage point increase in EQUITY, from 10.7 percent to 13.8 percent, would cause a 0.64 percentage point drop in bank probability of failure.
- Quantitative importance: effects characterized as small and consistent with quantitative impact studies (BCBS, 2010); caveat of potential non-linear or threshold effects for banks at distribution extremes.

### Threshold effects and subsample analysis
- Sample splits and thresholds:
  - NSFR below one used as liquidity threshold.
  - EQUITY below seven percent used as capital threshold.
- Findings on threshold effects:
  - Complementarity of liquidity and capital for financial stability; benefits concentrated for banks at lower extremes.
  - In subsamples with weak structural liquidity:
    - Coefficients on EQUITY two to four times higher than baseline.
    - Relationship between structural liquidity and failure probability reverses sign and becomes statistically insignificant for banks with low liquidity and capital.
  - Partition by EQUITY:
    - Strongest marginal benefits of capital cushions for banks operating below the seven percent threshold.
    - Explanatory variables account for a larger share of failure probability in lower capital subsamples (higher pseudo R-squared).
    - For EQUITY above twelve percent, coefficients generally not statistically significant.
- Table 7 illustrative coefficients (selected):
  - NSFR coefficients across partitions: 0.0090 [0.036]; 0.0704 [0.046]; -0.1091* [0.057]; -0.0218 [0.018]; -0.0858*** [0.025]; -0.0018 [0.023].
  - Equity coefficients across partitions: -0.3490*** [0.112]; -1.1667*** [0.447]; -0.5937* [0.353]; -1.0045** [0.404]; -0.6343** [0.262]; -0.1475 [0.105].
  - Observations in partitions: 6,744 | 1,421 | 4,391 | 1,798 | 6,637 | 2,461.
  - Pseudo R2 in partitions: 0.0478 | 0.1460 | 0.0489 | 0.1280 | 0.0424 | 0.0492.

### Differences across bank types (Global vs Domestic)
- Separate regressions show heterogeneity:
  - Capital shortages more important in failures of Global banks; liquidity more important for Domestic banks.
  - For Global banks:
    - Equity coefficient ~ -5.9592*** [2.236] (Table 8, column [1]); coefficient almost 25 times larger than baseline.
    - Using this value, a one percent increase in Global bank capital in the pre-crisis period would cause a material 4.8 percent drop in their probability of failure (paper contextualizes large marginal effect for Global banks).
    - Credit growth coefficient much larger: 1.6543*** [0.447].
    - Country macro variables not systematically significant for Global banks.
  - For Domestic banks (Table 8, column [2]) key coefficients:
    - NSFR: -0.0370** [0.015].
    - Equity: -0.1024 [0.065].
    - Credit Growth: 0.0392*** [0.012].
    - Z-Score: -0.0545*** [0.010].
  - Domestic subgroups:
    - Commercial banks: more vulnerable to weak structural liquidity (NSFR -0.0619*** [0.020]).
    - Savings banks: capital shortages relatively more important (Equity -0.4274*** [0.142]).
    - Cooperatives: failure linked to non-traditional activities (NON-INTEREST INCOME positive and significant 1.8486*** [0.585]).
- Observations and fit by type:
  - Observations: Global 911; Domestic 10,805; Domestic Commercial 8,243; Domestic Savings 1,550; Domestic Cooperatives 1,012.
  - Pseudo R2: Global 0.200; Domestic 0.043; Commercial 0.033; Savings 0.099; Cooperatives 0.161.

### Robustness checks
- Alternative liquidity and capital measures:
  - NSFR and STFR for liquidity; EQUITY and CAR for capital.
  - Table 9 selected coefficients:
    - NSFR/Equity liquidity coefficient: -0.0431*** [0.015]; Equity -0.1993*** [0.073].
    - STFR/Equity liquidity coefficient: 0.0449* [0.025]; Equity -0.2487*** [0.070].
    - STFR/CAR liquidity coefficient: 0.1523*** [0.037]; CAR capital coefficient -0.1840*** [0.055].
  - Positive coefficients on STFR indicate reliance on short-term funding associated with increased fragility.
- Alternative failure definitions (Table 10):
  - Variants: Baseline; Baseline ex-CAR; Laeven-Valencia only; CAR only.
  - Results broadly consistent with baseline; some coefficients vary in magnitude and significance but qualitative conclusions unchanged.
  - Example: NSFR Baseline -0.0431*** [0.015]; Baseline ex-CAR -0.0186 [0.015]; Laeven-Valencia -0.0282*** [0.005]; CAR -0.0313*** [0.007].
- Overall robustness summary: main inferences about the protective role of higher NSFR and higher EQUITY hold under alternative measures and failure definitions.

### Policy implications and interpretation
- Findings support Basel III initiatives on structural liquidity (NSFR, LCR) and leverage, and their complementary nature.
- Effects concentrated among weaker institutions:
  - Stability benefits of tighter liquidity and capital moderate for the average bank but substantially more relevant for banks at the lower tails.
  - Capital regulations likely play a more critical role for systemically important (Global) banks; liquidity regulations more relevant for smaller domestic banks.
- Macroprudential considerations:
  - Pre-crisis GDP growth and monetary conditions significantly affected failure probabilities, supporting macro-prudential policies complementary to micro-prudential regulation.
  - Further work recommended to deepen understanding of the macroeconomic environment’s role for financial stability.

*Source: Excerpts from IMF working paper _wp1229 (Section 1 and concluding sections) provided in the input.*

### 1. Stylized Balance-Sheet and Weights to Compute the NSFR ...............................................23

### 1. Stylized Balance-Sheet and Weights to Compute the NSFR ...............................................23

### Regulatory context and motivation
- The global financial crisis prompted revisions to regulatory and supervisory frameworks governing bank liquidity risk and capital buffers, including:
  - Introduction of liquidity standards for internationally-active banks.
  - Binding leverage ratios.
  - Revision of capital requirements under Basel III.
- Two prudential liquidity ratios discussed in the source:
  - Liquidity Coverage Ratio (LCR): aimed at promoting banks’ resilience to liquidity risk over the short-term (a 30-day period).
  - Net Stable Funding Ratio (NSFR): aimed at promoting resilience over a one-year horizon.
- A leverage ratio computed as shareholders’ capital over total assets was introduced to ensure a hard minimum capital level, regardless of risk-weight structure.

### Contributions and approach of the paper
- Two primary contributions:
  - Measures structural liquidity and leverage in bank balance sheets consistent with the formulations of the NSFR and the leverage ratio (EQUITY) proposed in Basel III.
  - Explores systematic differences in the relationship between structural liquidity, leverage, and subsequent probability of failure across bank types, distinguishing:
    - Global banks (large, internationally-active).
    - Domestic banks (typically smaller, retail-focused).
- Rationale for the partition:
  - Global banks are systemically important, complex to resolve, operate across borders, and may exploit internal capital markets; they often enjoy a more stable funding base during market distress.
  - Domestic banks may differ in funding stability and exposure to liquidity transformation risks.

### Data and scope
- Bank-level dataset covering about 11,000 U.S. and European banks during 2001−09.
- Structural liquidity is proxied as the ratio of long-term stable funding sources to structural asset positions.

### Key empirical findings
- Pre-crisis levels and distribution:
  - The average structural liquidity in bank balance sheets in the run up to the global financial crisis (as measured by a proxy of the NSFR) was close to the target values proposed in Basel III recommendations.
  - There was a wide dispersion in structural liquidity across banks.
  - A mild (albeit sustained) increase in structural liquidity mismatches in the run up to the crisis was driven by banks located at the lower extreme of the distribution.
  - Pre-crisis leverage was widely uneven across banks; Global banks displayed thinner capital buffers and wider gaps between leverage ratios and Basel capital to risk-weighted assets.
- Relationship with bank failures:
  - Banks with weaker structural liquidity in the run up to the crisis were more vulnerable to failure, controlling for pre-crisis risk-taking.
  - Banks with higher leverage ratios in the run up to the crisis were more vulnerable to failure, controlling for pre-crisis risk-taking.
  - The average effects of stronger structural liquidity and capital buffers on the likelihood of bank failure are not large.
  - Evidence of substantial threshold effects: benefits of stronger buffers are substantial for banks located at the lower extremes of the distributions.
- Cross-sectional differences:
  - Global banks were more susceptible to failure from excessive leverage.
  - Domestic banks were more susceptible to failure from weak structural liquidity (i.e., excessive liquidity transformation) and overreliance on short-term wholesale funding.
- Role of controls:
  - Bank-level controls for pre-crisis risk-taking and country-specific macroeconomic conditions were included.
  - Banks engaging in more aggressive risk taking in the run-up to the crisis—as measured by the rate of growth of their credit portfolios and by their pre-crisis distance to default—were more likely to fail afterward.
  - Pre-crisis macroeconomic conditions affected bank probabilities of default, indicating banks may have failed to internalize risks from overheated economic activity and exuberant asset prices.

### Policy implications and interpretation
- Findings provide support for:
  - Proposed regulations on liquidity and capital (e.g., NSFR, LCR, leverage ratio).
  - Introduction of a macro-prudential approach to bank regulation.
- From a financial stability perspective:
  - Regulations on capital—particularly for the larger banking groups—are likely to be more relevant.

*Source: IMF working paper section "1. Stylized Balance-Sheet and Weights to Compute the NSFR ...............................................23"*

### Section VI concludes.

### _wp1229 - Section VI concludes.

### Related literature and empirical hypotheses
- Liquidity creation is an essential role of banks and is strongly connected to financial stability (Bryant, 1980; Diamond and Dybvig, 1983).
- Banks create liquidity by financing long-term projects with relatively liquid liabilities; this exposes them to liquidity risk, which can discipline behavior and support efficiency (Diamond and Rajan, 2000).
- Bank capital (i.e., lower leverage) reduces liquidity creation but provides a buffer against asset value changes, increasing survival probabilities under distress (Diamond and Rajan, 2001).
- Structural liquidity mismatches (long-term illiquid assets financed with short-term funding and non-core deposits) are linked to bank liquidity creation; higher leverage (higher EQUITY) is also associated with greater liquidity creation.
- Empirical findings cited:
  - Periods of abnormal liquidity creation preceded U.S. banking crises (Berger and Bouwman, 2008, 2009).
  - Reliance on wholesale funding had negative effects on stock performance after the crisis (Raddatz, 2010) and increased financial fragility measured by distance to default and stock return volatility (Demirgüç-Kunt and Huizinga, 2009).
  - Banks with more stable funding continued to lend and were less likely to fail during the crisis (Cornett et al., 2010; Bologna, 2011).
  - Larger capital cushions correlated with better stock returns during the crisis (Demirgüç-Kunt, Detragiache, and Merrouche, 2010); capital had heterogeneous survival effects by bank size (Berger and Bouwman, 2010).
  - Prior studies showed capital ratios have strong explanatory power for subsequent bank failure and possible non-linear effects (Estrella, Park, and Peristaki, 2000; Gomez-Gonzalez and Kiefer, 2007).
- Competing theoretical views exist on the effect of wholesale funding on vulnerability and market discipline (Calomiris, 1999; Huang and Ratnovski, 2010).
- Paper’s empirical questions:
  - (i) Are structural liquidity and leverage in the pre-crisis period connected to the probability of subsequent failure?
  - (ii) Are there systematic differences across bank types (large globally-active vs domestic retail-oriented)?
- Hypotheses:
  - If banks optimally manage risk, pre-crisis risk-taking should be positively correlated with liquidity and capital buffers, and those buffers should have weak connections to probability of failure.
  - Proper risk management would weaken links between macroeconomic pre-crisis conditions and subsequent bank failure.

### Data and target variables
- Source: Bank-level financial statements from the Bankscope database.
  - Coverage advantage: sampled banks account for about 90 percent of total assets in each country (per source).
  - Data standardized across accounting/reporting differences.
  - Limitation: lack of granularity in some balance-sheet accounts (loan breakdowns, securities by asset class/maturity).
  - Liabilities side: deposits classified by type; non-deposit funding classified as short-term (residual maturity shorter than one year) versus long-term (residual maturity longer than one year).
- Sample:
  - About 11,000 banks incorporated in the U.S. and Europe.
  - Series yearly, spanning 2001–09; able to capture run-up (2001–07) and crisis (2008–09).
  - Sample splits: (i) large internationally active (Global) vs domestically-oriented (Domestic); (ii) splits by target levels of structural liquidity and leverage to explore threshold effects.
- Currency and reporting:
  - Balance sheets and income statements in U.S. dollar terms, using market rate at bank-specific accounting closing dates.
  - Consolidated figures used when possible.
- Outlier treatment:
  - Observations with liquidity or leverage below the 0.5 percentile and above the 99.5 percentile are filtered out.

### Indicators of bank liquidity and leverage
- Primary measures:
  - Net Stable Funding Ratio, NSFR (ratio between weighted sum of liabilities L_i and assets A_j: NSFR = (sum_i w_i L_i) / (sum_j w_j A_j)). Weights w bounded between zero and one; higher NSFR implies lower liquidity risk. Proposed regulations require NSFR higher than one.
  - EQUITY: leverage ratio measured as equity capital to assets (shareholders’ equity to assets).
  - Interpretation: higher NSFR and higher EQUITY imply lower bank liquidity creation.
- Approximation approach and weight assumptions (due to limited granularity):
  - Loan portfolio treated conservatively with weight 1.00 (total loan portfolio requires stable funding).
  - Other earning assets assigned weight 0.35 (average within Basel III proposed range).
  - Fixed assets and non-earning assets (except cash and due from banks) receive weight 1.00.
  - Liabilities: customer deposits split by type; other liabilities split by maturity. Core retail deposits assumed more stable; other short-term funding given weight zero. Long-term liabilities and equity considered stable at the one-year horizon.
- Robustness checks:
  - Liquidity alternative: Short-Term Funding Ratio (STFR) = liabilities maturing within one-year / total liabilities.
  - Capitalization alternative: Basel CAR = regulatory capital / risk-weighted assets.

### Global banks versus domestic banks (sample composition and adjustments)
- Classification criteria:
  - Global banks: internationally-active institutions with consolidated assets surpassing US$10 billion at end-2009; parent banking groups identified by majority stakes in foreign subsidiaries and no financial institutions as ultimate owners.
  - Domestic banks: domestically-owned institutions with no majority stakes in subsidiaries abroad.
- Coverage:
  - Domestic banks: tracks 10,805 institutions during 2001−09; more than eight years of time coverage for about 57 percent of these banks.
  - Global banks: covers 91 institutions; more than six years of information for 60 percent of these banks.
- Data break and checks:
  - Apparent break in the European subsample in 2005 attributable to IFRS adoption; pre-crisis variables computed using three alternative criteria for robustness: (i) means over entire available data for each bank; (ii) means over 2004−07; (iii) values as of end-2007. Results broadly consistent.

### Bank failure: identification criteria
- Failure identification uses complementary sources:
  - BankScope ongoing status: banks changing from “active” to “under receivership”, “bankruptcy”, “dissolved”, “dissolved by merger”, or “in liquidation”.
  - Basel capital (CAR): banks with CAR dropping below the 8 percent threshold between 2008−09.
  - Moody’s bank financial strength ratings: banks downgraded to ratings E+ or E (in distress).
- Limitations and adjustments:
  - These criteria capture banks allowed to fail and subject to resolution (typically smaller non-systemic institutions).
  - Failing Global banks often assisted by governments and not fully captured; to address this, failing Global banks are identified using Laeven and Valencia (2010) information.

### Empirical approach
- Probit model estimating probability of bank failure during the crisis (2008−09):
  - Model: Pr(F_i = 1 | x_i) = Φ(β x_i), where F_i = 1 if bank i failed during 2008−09 and zero otherwise.
  - Vector X_i includes:
    - Target variables measured prior to crisis: NSFR and EQUITY.
    - Bank-level controls: CREDIT GROWTH (yearly average of credit growth), NON-INTEREST INCOME (ratio of non-interest income to total income), Z-SCORE (distance to default measured as number of standard deviations that bank ROA must drop to trigger insolvency).
    - Country-specific variables: GDP GROWTH (yearly average rate of GDP growth) and MONEY MARKET RATES.
- Specification notes:
  - Pre-crisis averages used to ameliorate endogeneity at the cost of neglecting time dynamics.
  - Purely cross-sectional specification; does not include bank-level fixed effects.
- Interpretative framework:
  - If banks optimally manage liquidity and capitalization, expect positive correlation between ex-ante risk-taking and the size of liquidity and capital buffers, and weak or no connection between these buffers and probability of failure.
  - Evidence to the contrary would suggest banks failed to account properly for risk-taking, supporting a stronger prudential framework for capital and liquidity buffers.
  - A systematic link between macroeconomic conditions and failure probabilities would support macro-prudential complementarity to microprudential regulation.

### Stylized facts (sample statistics and key numbers)
- Sample breadth and timing:
  - Sample covers about 11,000 banks; yearly series spanning 2001–09, capturing run-up (2001–07) and crisis (2008–09).
- Size differences:
  - Average balance sheet, Domestic banks at end-2009: US$0.7 billion.
  - Average balance sheet, Global banks at end-2009: US$527.1 billion.
  - Institution at the 99 percentile had a balance sheet of US$2.9 trillion at end-2009.
- Pre-crisis differences between Global and Domestic banks:
  - Global banks displayed thinner capital cushions and weaker indicators of structural liquidity.
  - Global banks’ liabilities relied more heavily on non-deposit funding and were more short-term.
  - Gap between EQUITY and Basel CAR:
    - Global banks, first percentile: EQUITY = 1.4 percent; Basel CAR = 9.2 percent (Basel CAR 6.6 times higher).
    - Domestic banks: EQUITY = 2.5 percent; Basel CAR = 10.1 percent.

*Source: _wp1229 - Section VI concludes.*

### 4.0 times higher. Other risk indicators, such as the Z-score and credit growth are broadly

### _wp1229 - 4.0 times higher. Other risk indicators, such as the Z-score and credit growth are broadly

### Correlations and pre-crisis cross-sectional patterns
- Pair-wise correlations (Table 4) are shown separately for Global banks (lower triangle) and Domestic banks (upper triangle).
- Liquidity measures are closely related for both bank types:
  - Stronger structural liquidity is associated with lower reliance on short-term funding (and with money market funding) and positively correlated with deposit funding.
- Two measures of bank capital convey similar information despite gaps stemming from risk weights.
- Correlation between bank capital and credit growth is positive and statistically significant in both subsamples.
  - Interpreted as consistent with bank governance and risk management mechanisms: a more aggressive credit expansion was associated with stronger capital cushions.
- Differences across bank types:
  - For Global banks, higher structural liquidity is associated with more moderate credit growth and a larger distance to default.
  - For Domestic banks, the relationship between these variables is not significant.
- Overall implication:
  - Expansion of bank balance sheets in the pre-crisis period was associated with riskier liquidity profiles, particularly for Global banks, but did not suggest an immediate connection with potential shortages in capital buffers.
- The analysis motivates a more rigorous exploration of the link between bank probability of failure and pre-crisis levels of liquidity and capital.

### Time evolution of structural liquidity and leverage
- Figure 1 plots medians and the 10th and 90th percentiles for NSFR and EQUITY across bank types.
  - The average NSFR before the crisis is relatively stable and close to one, but with wide dispersion; banks at the lower extreme display extremely weak structural liquidity.
  - A similar pattern holds for EQUITY: average banks had relatively comfortable equity to asset ratios, but banks at the low end were extremely leveraged.
- Figure 2 (by bank type and Failed vs Non-Failed):
  - Failed banks had lower structural liquidity and higher leverage than non-failed banks.
  - NSFR shows a declining trend in the pre-crisis period, reverting from 2007 for Domestic banks and from 2008 for Global banks.
  - In Global banks, there is a sudden drop at the peak of the crisis followed by an equally sharp increase reflecting hoarding of liquidity for precautionary purposes.
  - Regarding EQUITY:
    - Domestic banks display more comfortable cushions than Global banks and an upward trend in the pre-crisis period.
    - After the eruption of the crisis, equity collapses in the group of failed Domestic banks, but increases in the group of failed Global banks, reflecting capital injections and public support due to their systemic importance.

### Distributions of pre-crisis liquidity and leverage (Failed vs Non-Failed)
- For Figure 3 comparisons, banks with NSFR above 1.5 and banks with EQUITY above 20 percent are excluded.
- Distributional properties:
  - All distributions have positive skewness and excess kurtosis; normality tests reject the null in all cases.
- Key contrasts:
  - Failed Global banks show substantially lower EQUITY, with the mean close to 4 percent.
  - Distributions of NSFR for Failed banks are displaced to the left, but differences tend to be smaller than for EQUITY.
- Tests of differences of means (not shown) suggest:
  - Insufficient EQUITY was associated with failure in the case of Global banks.
  - Insufficient structural liquidity was a problem associated with the Domestic banks.
- The text proceeds to develop an empirical model to test these conjectures.

### Baseline probit regressions: determinants of probability of failure
- Main results (Table 5):
  - Banks with higher NSFR and EQUITY in the years before the crisis were less susceptible to fail during the turmoil.
    - Coefficients on NSFR and EQUITY are negative and statistically significant at the one percent level in all cases.
    - Results are robust to inclusion of control variables.
  - Banks with higher risk taking in the pre-crisis period were more likely to fail afterward:
    - Credit growth is positively associated with the probability of failure.
    - The Z-score (distance to default) operates in the opposite direction (reduces failure probability).
  - The ratio of non-interest income to total income is not statistically significant in these regressions.
    - This contrasts with Demirguc-Kunt and Huizinga (2010); difference likely due to variable construction.
      - Footnote on construction: NON-INTEREST INCOME is measured by taking the absolute value of non-interest income to total income, to account for trading income possibly taking negative values. A bank with either large non-interest gains or losses relative to total income is assumed to be riskier.
  - Macroeconomic variables (common to all banks in a country) are highly significant and operate in the expected direction:
    - Banks incorporated in countries with higher pre-crisis economic growth and with easier monetary conditions were more likely to fail during the crisis.
    - Interpreted as banks failing to fully internalize risks from their external environment and as a potential justification for macro-prudential regulations complementary to micro-prudential rules.
  - Model fit:
    - The pseudo R-square of the regression tends to be low, with the model explaining less than five percent of the variation in bank probability of failure.

### Economic significance (simulation based on column [6] coefficients)
- Using coefficients from column [6], estimated change in the probability of failure from a 0.5 standard deviation change in explanatory variables (Table 6):
  - A 10.4 percentage point increase in the NSFR, from 0.99 to 1.09, would cause a drop 0.46 percentage point drop in the probability of failure of the average bank, all else equal.
  - A 3.1 percentage point increase in EQUITY, from 10.7 percent to [text truncated in source].

*Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp1229.pdf*

### 13.8 percent would cause a drop of 0.64 percentage point drop in bank probability of failure.

### _wp1229 - 13.8 percent would cause a drop of 0.64 percentage point drop in bank probability of failure.

### Quantitative importance of liquidity and capital effects
- A 13.8 percent (presumably change in a target variable referenced by the source) would cause a drop of 0.64 percentage point drop in bank probability of failure.
- The quantitative importance of these effects is characterized as small and consistent with quantitative impact studies (BCBS, 2010).
- Caveat: potential presence of non-linear or threshold effects operating more severely for banks in the extremes of the distribution.

### Drivers of bank failure probability (pre-crisis)
- Bank risk profiles and operating environments more strongly influence failure probability:
  - Banks incorporated in countries with a pre-crisis GDP growth 0.5 percentage points higher than the average were 2.2 percentage points more likely to fail.
  - Tighter monetary conditions operated in the opposite direction (reducing failure probability).
- Interpretation: higher pre-crisis GDP growth may reflect unsustainable economic activity and/or potential asset bubbles.

### Threshold effects (NSFR and EQUITY)
- Method: sample split by pre-crisis NSFR and EQUITY using dummies; NSFR below one and EQUITY below seven used as threshold references.
- Findings:
  - Results support complementarity of liquidity and capital for financial stability and indicate threshold effects.
  - In subsamples of banks with weak structural liquidity:
    - Coefficients associated with EQUITY are two and four times higher than in matching baseline regressions.
    - Relationship between structural liquidity and probability of failure reverses sign and becomes statistically insignificant for banks with low liquidity and capital, suggesting capital shortages were critical for failing banks in this subsample.
  - Partition by EQUITY levels:
    - Strongest marginal benefits of capital cushions originate from banks operating below the seven percent threshold (column [4]).
    - Explanatory variables account for a more significant proportion of failure probability in lower capital subsamples (higher pseudo R-squared).
    - For intermediate EQUITY levels, structural liquidity and capital both contribute; target coefficients are two and three times larger than baseline.
    - Coefficients are not statistically significant for subsample with EQUITY above twelve percent (column [6]).
- Interpretation: stability benefits of tighter liquidity and capital regulations are moderate for the average bank but substantially more relevant for institutions at the lower extreme of the distribution; capital regulations likely play a more critical role than liquidity regulations for financial stability.

### Differences across bank types (Global vs Domestic)
- Separate regressions for Global and Domestic banks (Domestic further split into commercial banks, savings banks, cooperatives).
- Key findings:
  - Capital shortages were more important in failure of Global banks; liquidity was key for Domestic banks.
  - Magnitude for Global banks:
    - Coefficient associated with EQUITY for Global banks is almost 25 times larger than baseline.
    - Using this value, a one percent increase in Global bank capital in the pre-crisis period would cause a material 4.8 percent drop in their probability of failure.
  - Credit growth coefficient also substantially larger for Global banks — aggressive expansion linked to higher failure likelihood.
  - Country-specific macroeconomic conditions do not play a systematic role for Global banks (diversification from international operations).
- Domestic bank subgroups:
  - Capital shortages relatively more important for savings banks.
  - Commercial banks more vulnerable to weak structural liquidity.
  - Credit cooperatives more likely to fail if more heavily engaged in non-traditional activities (proxied by non-interest to total income ratio).

### Robustness checks
- Alternative liquidity and capital measures:
  - Short Term Funding Ratio (STFR): liabilities with less than one year residual maturity divided by total liabilities.
  - Basel Tier 1 capital ratio (CAR): Tier 1 regulatory capital to risk-weighted assets (larger than baseline equity measure due to risk weighting).
- Alternative failure definitions (three variations):
  1. Banks that ceased active status during the crisis or reclassified to risk categories E and E+ by Moody’s.
  2. Banks with regulatory CAR ratios dropping below 8 percent between 2008–09.
  3. Banks included in the failed list compiled by Laeven and Valencia (2009).
- Summary of robustness results:
  - Full set of results broadly consistent with baseline.
  - Coefficients associated with STFR are positive, indicating reliance on short-term funding before the crisis was associated with increased financial fragility.
  - Regressions using alternative failure definitions are broadly consistent with baseline results.

### Baseline regression highlights (selected values from Table 5)
- Sample size and model fit:
  - Observations: 10,896 (for columns reported).
  - Pseudo R2 values range up to 0.0399 in presented specifications.
- Marginal impacts (expressed as change in probability of failure associated with marginal change from pre-crisis means):
  - NSFR coefficients (columns shown): -0.0690***, -0.0585***, -0.0461***, -0.0421***, -0.0431***.
    - Robust standard errors: [0.015] (repeated across columns).
  - Equity coefficients (columns shown): -0.2341***, -0.1452**, -0.2512***, -0.2008***, -0.1993***.
    - Robust standard errors: [0.068], [0.069], [0.073], [0.072], [0.073].
  - Credit Growth: 0.0590***, 0.0491***, 0.0487***. [0.015], [0.014], [0.014]
  - Z-Score: -0.0471***, -0.0477***. [0.011], [0.011]
  - GDP Growth 2001-07: 4.1008***, 4.4759***, 4.3866***, 4.2473***, 3.9885***, 4.0479***. [robust SEs range ~0.336–0.366]
  - Monetary Conditions 2001-07: -0.9138***, -0.8543***, -0.8646***, -0.9469***, -1.0286***, -0.9738***. [robust SEs range ~0.264–0.341]

### Concluding policy-relevant implications
- Results broadly support Basel III initiatives on structural liquidity and leverage and highlight their complementary nature.
- Banks with weaker structural liquidity and higher leverage pre-crisis were more vulnerable to failure; effects concentrated among weaker institutions (threshold effects).
- Regulatory focus implications:
  - Ensure capital buffers of systemically important banks are commensurate with their risk-taking.
  - Regulatory and supervisory emphasis should account for heterogeneity across bank types: capital focus for large cross-border groups, liquidity focus for smaller domestic banks.
- Macroprudential considerations:
  - Evidence of bank risk-taking and country-specific macroeconomic conditions contributing to vulnerability supports introducing a macro-prudential framework complementary to micro-prudential regulation.
  - Further work recommended to deepen understanding of the macroeconomic environment’s role on financial stability.

*Italic: Source — Excerpts from the IMF working paper content provided in the input.*

### 0.5 standard deviation is displayed column [3]. The estimate defects, measured in percentage point changes

### _wp1229 - 0.5 standard deviation is displayed column [3]. The estimate defects, measured in percentage point changes

### Probit regressions: baseline marginal effects (0.5 standard deviation change)
- Table reports change in probability of bank failure associated with a 0.5 standard deviation change in each variable (column [4]).
- Variables and associated regression coefficients, means, change in variable, and change in probability of failure (percentage points) 1/:
  - NSFR: Regression Coefficients -0.0430; Mean of Variable 0.986; Change in Variable 0.104; Change in Pr. Failure (Percentage points) -0.45
  - Equity Capital to Assets: Regression Coefficients -0.1990; Mean of Variable 0.107; Change in Variable 0.031; Change in Pr. Failure (Percentage points) -0.63
  - Credit Growth: Regression Coefficients 0.0490; Mean of Variable 0.130; Change in Variable 0.081; Change in Pr. Failure (Percentage points) 0.39
  - Z-Score: Regression Coefficients -0.0470; Mean of Variable 0.469; Change in Variable 0.230; Change in Pr. Failure (Percentage points) -1.08
  - Non-Interest Income to Assets: Regression Coefficients -0.2530; Mean of Variable 0.010; Change in Variable 0.006; Change in Pr. Failure (Percentage points) -0.16
  - GDP Growth 2001-07: Regression Coefficients 4.0480; Mean of Variable 0.026; Change in Variable 0.005; Change in Pr. Failure (Percentage points) 2.16
  - Monetary Conditions 2001-07: Regression Coefficients -0.9730; Mean of Variable 0.031; Change in Variable 0.007; Change in Pr. Failure (Percentage points) -0.67
- 1/ Associated with a 0.5 standard deviation change in the corresponding variable.

### Table 7 — Probit regressions by subsamples of liquidity and leverage
- Sample partitions: leftmost three columns = subsample of banks with pre-crisis NSFR below one and EQUITY intervals; rightmost three columns = partitions by intervals of EQUITY.
- Columns and key coefficient estimates (robust standard errors in brackets). Starred coefficients denote statistical significance at 1% (***) , 5% (**) , and 10% (*).

- Column headings: NSFR < 1 | NSFR<1 and Equity below 7 Percent | NSFR<1 and Equity between 7 and 12 Percent | Average Equity below 7 Percent | Average Equity between 7 and 12 Percent | Average Equity above 12 Percent

- NSFR coefficients:
  - 0.0090 [0.036]
  - 0.0704 [0.046]
  - -0.1091* [0.057]
  - -0.0218 [0.018]
  - -0.0858*** [0.025]
  - -0.0018 [0.023]

- Equity coefficients:
  - -0.3490*** [0.112]
  - -1.1667*** [0.447]
  - -0.5937* [0.353]
  - -1.0045** [0.404]
  - -0.6343** [0.262]
  - -0.1475 [0.105]

- Credit Growth coefficients:
  - 0.1301*** [0.023]
  - 0.0885 [0.056]
  - 0.2897*** [0.037]
  - 0.0881* [0.048]
  - 0.1902*** [0.025]
  - 0.0180* [0.010]

- Z-Score coefficients:
  - -0.0543*** [0.017]
  - 0.0017 [0.012]
  - -0.0544*** [0.020]
  - 0.0029 [0.009]
  - -0.0542*** [0.016]
  - -0.0657*** [0.018]

- Non-Interest Income coefficients:
  - -0.2159 [0.412]
  - -0.4481 [0.838]
  - 0.1243 [0.516]
  - -0.3947 [0.598]
  - 0.094 [0.407]
  - -0.5872 [0.447]

- GDP Growth 2001-07 coefficients:
  - 4.0177*** [0.479]
  - 4.7385*** [0.679]
  - 2.9141*** [0.959]
  - 4.5008*** [0.585]
  - 3.1336*** [0.755]
  - 3.5580*** [0.715]

- Monetary Conditions 2001-07 coefficients:
  - -0.8515** [0.378]
  - 2.1617*** [0.800]
  - -3.0919*** [1.002]
  - 2.1699*** [0.809]
  - -3.2266*** [0.960]
  - -0.8456 [0.542]

- Observations:
  - 6,744 | 1,421 | 4,391 | 1,798 | 6,637 | 2,461

- Pseudo R2:
  - 0.0478 | 0.1460 | 0.0489 | 0.1280 | 0.0424 | 0.0492

### Table 8 — Probit regressions by bank types
- Subsamples: Global Banks (column [1]); Domestic Banks (column [2]); Domestic Commercial Banks (column [3]); Domestic Savings Banks (column [4]); Domestic Cooperatives (column [5]).
- Key coefficients (robust standard errors in brackets). Starred coefficients denote statistical significance at 1% (***) , 5% (**) , and 10% (*).

- NSFR:
  - Global Banks: -0.3565 [0.232]
  - Domestic Banks: -0.0370** [0.015]
  - Domestic Commercial Banks: -0.0619*** [0.020]
  - Domestic Savings Banks: 0.0520** [0.024]
  - Domestic Cooperatives: -0.0322* [0.017]

- Equity:
  - Global Banks: -5.9592*** [2.236]
  - Domestic Banks: -0.1024 [0.065]
  - Domestic Commercial Banks: -0.0714 [0.081]
  - Domestic Savings Banks: -0.4274*** [0.142]
  - Domestic Cooperatives: -0.1476 [0.135]

- Credit Growth:
  - Global Banks: 1.6543*** [0.447]
  - Domestic Banks: 0.0392*** [0.012]
  - Domestic Commercial Banks: 0.0372*** [0.013]
  - Domestic Savings Banks: 0.0707** [0.034]
  - Domestic Cooperatives: 0.0761** [0.034]

- Z-Score:
  - Global Banks: 0.0261 [0.063]
  - Domestic Banks: -0.0545*** [0.010]
  - Domestic Commercial Banks: -0.0777*** [0.013]
  - Domestic Savings Banks: -0.0454*** [0.016]
  - Domestic Cooperatives: 0.0064** [0.003]

- Non-Interest Income:
  - Global Banks: -2.8068 [3.066]
  - Domestic Banks: -0.3172 [0.309]
  - Domestic Commercial Banks: -0.403 [0.370]
  - Domestic Savings Banks: -0.1616 [0.455]
  - Domestic Cooperatives: 1.8486*** [0.585]

- GDP Growth 2001-07:
  - Global Banks: 0.2084 [6.940]
  - Domestic Banks: 0.1505*** [0.360]
  - Domestic Commercial Banks: 2.9101*** [0.590]
  - Domestic Savings Banks: 5.7008*** [0.778]
  - Domestic Cooperatives: 1.5062*** [0.507]

- Monetary Conditions 2001-07:
  - Global Banks: -1.1448 [1.512]
  - Domestic Banks: -1.1965*** [0.449]
  - Domestic Commercial Banks: -0.7675* [0.440]
  - Domestic Savings Banks: -0.7383* [0.383]
  - Domestic Cooperatives: 2.8008 [2.688]

- Observations:
  - 911 | 10,805 | 8,243 | 1,550 | 1,012

- Pseudo R2:
  - 0.200 | 0.043 | 0.033 | 0.099 | 0.161

### Table 9 — Robustness checks by alternative definitions of liquidity and capital
- Liquidity measures: NSFR and STFR. Capital measures: EQUITY and CAR. Coefficients transformed to convey change in probability of failure associated with marginal change from pre-crisis mean values. Starred coefficients denote statistical significance at 1% (***) , 5% (**) , and 10% (*).

- Column headings: NSFR / Equity | NSFR / CAR | STFR / Equity | STFR / CAR

- Liquidity Measure coefficients:
  - NSFR/Equity: -0.0431*** [0.015]
  - NSFR/CAR: -0.0409* [0.022]
  - STFR/Equity: 0.0449* [0.025]
  - STFR/CAR: 0.1523*** [0.037]

- Capital Measure coefficients:
  - NSFR/Equity: -0.1993*** [0.073]
  - NSFR/CAR: -0.1453** [0.061]
  - STFR/Equity: -0.2487*** [0.070]
  - STFR/CAR: -0.1840*** [0.055]

- Credit Growth coefficients:
  - 0.0487*** [0.014]
  - 0.0501*** [0.016]
  - 0.0495*** [0.014]
  - 0.0486*** [0.016]

- Z-Score coefficients:
  - -0.0477*** [0.011]
  - -0.0626*** [0.015]
  - -0.0495*** [0.011]
  - -0.0641*** [0.014]

- Non-Interest Income coefficients:
  - -0.253 [0.302]
  - -0.1859 [0.346]
  - -0.3128 [0.315]
  - -0.4828 [0.375]

- GDP Growth 2001-07 coefficients:
  - 4.0479*** [0.366]
  - 3.6582*** [0.567]
  - 4.1631*** [0.374]
  - 3.9135*** [0.582]

- Monetary Conditions 2001-07 coefficients:
  - -0.9738*** [0.334]
  - -2.0121*** [0.728]
  - -1.0647*** [0.366]
  - -2.8030*** [0.945]

- Observations:
  - 10,896 | 9,441 | 10,896 | 9,441

- Pseudo R2:
  - 0.0399 | 0.0352 | 0.0390 | 0.0379

### Table 10 — Robustness checks by sub-components of bank failure
- Definitions of failure:
  - Baseline: combines four criteria (Bankscope status change; drop in regulatory CAR below 8 percent; Moody’s bank financial strength rating downgrade to E+ or E; bank identified as failed in Laeven and Valencia (2010)).
  - Baseline ex-CAR: excludes criterion (ii) (drop in CAR below 8 percent).
  - Laeven-Valencia: criterion (iv) only.
  - CAR: criterion (ii) only.
- Coefficients transformed to convey change in probability of failure associated with marginal change from pre-crisis mean values. Starred coefficients denote statistical significance at 1% (***) , 5% (**) , and 10% (*).

- Columns: Baseline | Baseline ex-CAR | Laeven-Valencia | CAR

- NSFR:
  - Baseline: -0.0431*** [0.015]
  - Baseline ex-CAR: -0.0186 [0.015]
  - Laeven-Valencia: -0.0282*** [0.005]
  - CAR: -0.0313*** [0.007]

- Equity:
  - Baseline: -0.1993*** [0.073]
  - Baseline ex-CAR: -0.1720** [0.072]
  - Laeven-Valencia: -0.1465*** [0.035]
  - CAR: -0.0527* [0.029]

- Credit Growth:
  - Baseline: 0.0487*** [0.014]
  - Baseline ex-CAR: 0.0327*** [0.011]
  - Laeven-Valencia: 0.0182*** [0.004]
  - CAR: 0.0119*** [0.004]

- Z-Score:
  - Baseline: -0.0477*** [0.011]
  - Baseline ex-CAR: -0.0348*** [0.011]
  - Laeven-Valencia: -0.0016 [0.003]
  - CAR: -0.0238*** [0.004]

- Non-Interest Income:
  - Baseline: -0.253 [0.302]
  - Baseline ex-CAR: -0.3357 [0.293]
  - Laeven-Valencia: 0.1739** [0.075]
  - CAR: -0.0446 [0.119]

- GDP Growth 2001-07:
  - Baseline: 4.0479*** [0.366]
  - Baseline ex-CAR: 3.9072*** [0.360]
  - Laeven-Valencia: 0.2065* [0.107]
  - CAR: 0.7779*** [0.114]

- Monetary Conditions 2001-07:
  - Baseline: -0.9738*** [0.334]
  - Baseline ex-CAR: -0.7906*** [0.287]
  - Laeven-Valencia: 0.0261 [0.029]
  - CAR: -0.4991*** [0.114]

- Observations:
  - Baseline: 10,896
  - Baseline ex-CAR: 9,713
  - Laeven-Valencia: 10,896
  - CAR: 10,896

- Pseudo R2:
  - Baseline: 0.0399
  - Baseline ex-CAR: 0.0330
  - Laeven-Valencia: 0.0910
  - CAR: 0.0814

*Source: _wp1229 - 0.5 standard deviation is displayed column [3]. The estimate defects, measured in percentage point changes*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp1229.pdf_