## _wp13102

## Source details

**Canonical URL:** [_wp13102](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp13102.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp13102.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp13102.pdf.json)

---

### I. Purpose and research questions
- Examine whether bank balance sheet strength mattered for the extent banks reduced lending during the 2007–08 financial crisis.
- Investigate how banks’ exposure to market-based funding shocks during 2007–08 influenced subsequent lending and how this interacted with balance sheet features such as capital.
- Quantify the link between bank health and credit supply during the crisis to inform Basel III-related regulation.

### Data, sample, and identification
- Loan-level data: close to 39,000 syndicated loan deals signed between January 2006 and March 2010 from Dealogic’s Loan Analytics; final merged sample: 24,082 deals collapsed at bank-industry level; final sample 803 banks lending to firms from 448 country-specific industries in 48 countries.
- Sample period partition:
  - Before: January 2006–June 2007 (six quarters).
  - Shock window: July 2007–September 2008 (dropped from estimation).
  - After: October 2008–March 2010 (six quarters).
- Identification strategy:
  - Use Libor-OIS spread as indicator of tightened liquidity conditions.
  - Regress change in total loan volume between the “before” and “after” periods on pre-shock balance sheet indicators and interactions.
  - To separate supply from demand effects, retain borrowers who borrowed from the same lender both before and after the shock and exploit within-borrower variation across banks.
  - Baseline specification: ΔCijk = α + βLij + ηk + δj + εijk, with standard errors clustered at the bank level.

### Balance sheet indicators and hypotheses
- Funding-structure measures:
  - Non-deposit liabilities: (Total liabilities - total customer deposits) / Total liabilities.
  - Non-deposit funding: (Total funding - total customer deposits) / Total funding.
- Structural liquidity: NSFR (Net Stable Funding Ratio) approximated using stylized balance sheet weights (Vazquez and Federico (2012) weights).
  - NSFR = (Σi wi Li) / (Σj wj Aj), where wi and wj are weights between 0 and 1.
- Capital measures:
  - Tier 1 ratio (Tier 1 / RWA), total regulatory capital ratio ((Tier 1 + Tier 2) / RWA), and tangible common equity ratio.
- Hypotheses:
  - Greater reliance on market funding before the crisis → greater reduction in credit supply during the crisis.
  - Higher NSFR → less constrained during crisis and better able to maintain lending.
  - Better-capitalized banks → able to sustain syndicated lending despite funding-market stress.
  - Interaction: complementarities between capital and structural liquidity—stable funding effective primarily for well-capitalized banks.

### Main empirical findings
- Wholesale funding exposure and lending supply:
  - When controlling for borrower fixed effects, banks that relied more on market-based funding before the crisis reduced syndicated loan supply more than other banks.
  - Estimated magnitude: one percentage point increase in the share of non-deposit funding led to a decrease in syndicated credit supply by between 0.7 and 0.9 percent (across samples or sets of controls).
  - Robustness-representative magnitude: one percentage point increase in pre-crisis reliance on wholesale funding led to a decline in syndicated loan supply of 0.6–0.7 percent (statistically significant at the 1 percent level) in robustness checks.
- Capital as a mitigating factor:
  - Interaction effects:
    - Using Tier 1 or total regulatory capital ratios: interaction coefficients not statistically different from zero.
    - Using tangible common equity: exposed banks with more tangible common equity reduced syndicated loan supply less.
  - Evaluated at the sample mean of the tangible common equity ratio (6.90 percent), coefficients imply that with every percentage point increase in reliance on wholesale funding, the supply of credit fell by 0.65 to 0.82 percent (columns 9–12).
  - Example calculation (Column 10): –1.392 + 0.083 * 6.90 = –0.82 percent.
  - Capital elasticity example: an increase in bank capital by 1 standard deviation (4.29) from the sample mean (6.90 percent) reduces the negative effect of the liquidity shock from 0.82 to 0.46 percent; a reduction in bank capital by 1 standard deviation amplifies the effect to 1.18 percent.
  - Interpretation: tangible common equity (closer to Basel III concept) provides buffer; Basel II Tier 1 and total capital ratios are less informative for sustaining lending under liquidity shocks.
- Structural liquidity (NSFR) results:
  - Full-sample: for each percent increase in the NSFR, banks increased lending post-crisis by 0.43 to 0.46 percent (coefficients statistically significant at conventional levels).
  - Interaction with capital:
    - Positive NSFR effect larger among well-capitalized banks when capital measured by Tier 1 or tangible common equity.
    - For poorly-capitalized banks, higher NSFR did not translate into better ability to sustain lending.
  - Implication: complementarities between capital and structural liquidity—stable funding structure is effective only with a strong capital base.
  - Caveat: NSFR constructed from approximate weights; some NSFR coefficients only statistically significant at the 14.5 and 12.1 percent levels in certain specifications.

### Robustness checks
- Addressed concerns that results are driven by:
  - Crisis-specificity to the syndicated loan market: controlled for change in total loans between pre- and post-shock periods.
  - Strong banks expanding by acquisitions: controlled for change in total assets between pre- and post-shock periods.
- Robustness results summary:
  - Funding-structure specifications: one percentage point increase in pre-crisis reliance on wholesale funding → decline in syndicated loan supply of 0.6–0.7 percent (statistically significant at the 1 percent level) (columns 1–2, 8–9).
  - Negative impact cushioned by higher tangible common equity (columns 3–4, 10–11).
  - Results robust when using approximate NSFR as the vulnerability indicator (columns 5–7, 12–14), noting some NSFR coefficients have weaker statistical significance.

### Key statistics and sample metrics
- Syndicated loan market peak new issuances: USD 4.5 trillion in 2007.
- Loan Analytics sample: initial universe 39,626 deals; retained 38,538 loan deals with non-missing loan amount; final merged sample 24,082 deals; final sample 803 banks.
- Table 1 selected summary statistics:
  - Dependent variable Log-change in lending (x100): Obs = 7,156; Mean = -21.411; St. Dev. = 42.71; P5 = -254.68; P50 = -21.79; P95 = 214.72.
  - Non-deposit liabilities: Obs = 7,103; Mean = 2.33; St. Dev. = 26.81; P5 = 1.49; P50 = 23.91; P95 = 86.91.
  - Non-deposit funding: Obs = 7,105; Mean = 29.53; St. Dev. = 26.85; P5 = 0.18; P50 = 20.48; P95 = 86.52.
  - Log-NSFR: Obs = 4,444; Mean = 4.44; St. Dev. = 0.22; P5 = 3.90; P50 = 4.46; P95 = 4.77.
- Model performance highlights:
  - Share-imputation regression for individual syndicate shares: adjusted R-squared = 74 percent (in-sample).
  - Baseline regression observation and fit ranges:
    - Table 3 observations range 4,788–6,694 with R-squared up to 0.309.
    - Table 4 observations 5,756–6,051 with R-squared ≈ 0.277–0.280.
    - Table 5 observations 852–4,249 with R-squared up to 0.513.
    - Table 6 observations 3,019–5,706 with R-squared up to 0.517.

### Policy-relevant implications
- Empirical evidence supports raising both the quantity and quality of regulatory capital (emphasis on tangible common equity) consistent with Basel III to improve banks’ ability to sustain lending during funding-market stress.
- Structural liquidity (NSFR) enhances resilience only when paired with strong capital—policy frameworks should consider complementarities between capital and liquidity requirements.
- Market-based wholesale funding can amplify shocks; regulatory attention to funding structure, rollover risk, and liquidity buffers is warranted to limit procyclicality in intermediation.

*Source: _wp13102 - References and Sections I–VI (PDF chapter content as provided).*

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

### I. INTRODUCTION

### Purpose and research questions
- Examine the proposition that bank balance sheet strength matters for the extent to which banks reduce lending during a financial crisis.
- Investigate how banks’ exposure to market-based funding shocks during 2007–08 influenced their subsequent lending decisions and how this interacts with key balance sheet features such as capital.
- Quantify the link between bank health and the supply of credit during the crisis to inform the discussion on bank regulation under the Basel III framework.

### Main findings
- Banks that relied more on market-based funding, and thus were more vulnerable to liquidity shocks during the crisis, reduced their supply of credit more than other banks.
- Banks with a higher net stable funding ratio (NSFR) were better able to weather the crisis.
- Bank capitalization mattered: banks that were exposed to shocks but held more tangible common equity reduced lending less than other similarly exposed banks.
- Complementarities exist between capital and structural liquidity: higher structural liquidity helps sustain lending only for well-capitalized banks.

### Data and sample
- Loan-level data from the syndicated loan market.
- Analysis covers more than 800 financial institutions lending to borrowers from 55 advanced economies and emerging markets.
- Sample period: 2006–2010.
- Syndicated loan market peak new issuances: USD 4.5 trillion in 2007.

### Empirical strategy and identification
- Partition sample period into “before,” “shock,” and “after” windows; the “shock” window runs from the early signs of financial distress in the summer of 2007 to the fall of Lehman Brothers.
- Use Libor-OIS spread as an indicator that liquidity conditions tightened substantially during the crisis.
- Regress change in total loan volume between the “before” and “after” periods on indicators of balance sheet strength—both in levels and interactions—measured before the shock period.
- To separate supply from demand effects, retain only borrowers who borrowed from the same lender both in the “before” and “after” periods for at least two lenders; exploit within-borrower variation across differently exposed banks.
- Identification relies on the assumption that disruptions in funding markets caused by the US subprime crisis were unanticipated by banks and that each borrower reduced demand proportionately across its lenders.

### Conceptual and regulatory context
- Study evaluates balance sheet variables proposed under Basel III and their role in sustaining post-crisis lending.
- Addresses debate on whether tighter capital and liquidity rules lower banks’ profitability and restrict credit in the short run versus making banks safer and reducing funding costs in the medium run.
- Emphasizes both quantity and quality of capital (tangible common equity) and a complex measure of structural liquidity drawn from regulatory proposals.

### Contributions to the literature
- Documents the role of balance sheet variables proposed under Basel III in sustaining lending during crisis periods.
- Explicitly examines interactions between capital and structural liquidity.
- Expands research on:
  - Bank lending channel and international transmission of funding-market shocks using micro-data on lending to firms across many countries.
  - The literature tracing liquidity shocks in bank funding markets during the global financial crisis by incorporating a comprehensive view of balance sheet strength and testing the bank balance sheet channel in a large international syndicated loan sample.

*Source: _wp13102 - References*

### Section IV discusses the baseline results. Robustness checks are presented in Section V. Section

### _wp13102 - Section IV discusses the baseline results. Robustness checks are presented in Section V. Section

### Balance sheet strength: indicators and hypotheses
- Bank health indicators considered:
  - Measures of funding structure: reliance on wholesale funding (non-deposit liabilities as a share of total liabilities; non-deposit funding as a share of total funding).
  - Structural liquidity measure: NSFR (Net Stable Funding Ratio) as defined under Basel III.
- Interpretation and hypotheses:
  - Indicators measured before the crisis are treated as proxies for banks’ exposure to the heightened liquidity risk of 2007–08 when funding markets nearly shut down.
  - Expectation: reliance on market funding is negatively associated with the supply of bank credit (supported by literature: Aiyar, 2012; Cornett et al., 2011; Ivashina and Scharfstein, 2010; Raddatz, 2010).
  - Expectation: banks with higher structural liquidity (NSFR) were less constrained during the crisis and hence better able to maintain lending.
- Bank capital as a mitigating factor:
  - Hypothesis: better-capitalized banks are better able to sustain the supply of syndicated lending despite funding-market stress.
  - Theoretical channels:
    - Bank lending channel: well-capitalized banks can raise debt under more favorable terms due to lower agency costs (Kishan and Opiela, 2000; Stein, 1998; Holmstrom and Tirole, 1997; Bernanke and Blinder, 1988).
    - Bank capital channel: increases in cost of funds erode profits and over time capital; banks may forego lending to avoid future undercapitalization (Van den Heuvel, 2012, 2006, 2002; Bolton and Freixas, 2006).
  - Empirical expectation: interaction between capital and liquidity-exposure measures—well-capitalized banks reduce lending less than other banks with similar exposure.
- Capital measurement and Basel III context:
  - Alternative capital measures used: traditional Tier 1, total regulatory capital, and tangible common equity (closer to Basel III spirit).
  - Basel III proposals: higher regulatory capital; focus on instruments with high loss-absorbing ability such as common equity; removal of components with limited loss-absorbing ability under Basel II.

### Data and empirical strategy
- Lending data:
  - Loan-level data: close to 39,000 syndicated loan deals signed between January 2006 and March 2010 from Dealogic’s Loan Analytics.
  - Sample periods:
    - Before: January 2006–June 2007 (six quarters).
    - Shock window: July 2007–September 2008 (dropped from estimation).
    - After: October 2008–March 2010 (six quarters).
  - Rationale: use a shock window rather than a single event to avoid assuming no portfolio adjustments during the shock.
  - Bank-level loan volumes: obtained by splitting syndicated loan amounts across syndicate participants and aggregating at the bank level (for each borrower).
  - Share prediction: for ~30 percent of loans Loan Analytics reports individual loan shares; for remainder, log-shares are regressed on a comprehensive set of variables (loan amount, syndicate size, dummies for year-quarters, loan currency, lender role, lender and borrower nationality, borrower industry). The model performs well in-sample with an adjusted R-squared of 74 percent.
  - Aggregation of borrowers: borrowers defined as country-specific industries (maximum 25 industries per country); analysis retains industries that borrowed from at least two banks both before and after the shock to allow borrower fixed effects.
- Bank balance sheet data:
  - Matched lenders in Loan Analytics to Bankscope using lender names and nationalities, adjusting for name and ownership changes.
  - Final dataset: syndicated loans extended by 803 unique banks from 55 countries to firms from 448 country-specific industries in 48 countries.
- Variables capturing exposure to the shock:
  - Wholesale funding measures: non-deposit liabilities (share of total liabilities) and non-deposit funding (share of total funding).
  - NSFR: computed as an approximation based on a stylized balance sheet consistent with Basel III formulas due to lack of fine-grained balance sheet items in Bankscope.
  - Empirical observations:
    - Figure 3 (referenced) shows a significant increase in average reliance on wholesale funding prior to 2007–08 followed by a downward adjustment during the crisis.
    - Table 2 (referenced) shows NSFR is negatively correlated with measures of reliance on wholesale funding.
    - Figure 4 (referenced) shows average NSFR improved after the 2000–01 crisis, deteriorated before the recent crisis, and started improving in 2008.
- Empirical model:
  - Baseline specification: ΔCijk = α + βLij + ηk + δj + εijk, where ΔCijk is the log-change in syndicated bank credit (bank i in country j to borrower k), and Lij is the proxy for bank-specific liquidity shock exposure.
  - Identification assumptions:
    - Banks did not anticipate the US subprime crisis and did not adjust lending in anticipation (Corr(Lij, εijk) = 0).
    - Borrower fixed effects ηk control for demand-side shifts; bank nationality effects δj control for exchange rate movements and home-country macro policies.
  - Additional controls in variants: return on assets, non-performing loans, liquid-to-total assets ratio, bank size (total assets), risk profile (RWA/total assets), capital, specialization dummies, legal entity type dummies.
  - Standard errors clustered at the bank level.

### Results
- Baseline findings on bank funding structure and lending:
  - OLS estimates reported in Table 3:
    - Columns 1–4: OLS without borrower fixed effects (do not control for demand shifts).
    - Columns 5–10: OLS with borrower fixed effects (control for demand shifts).
  - Evidence of bias from demand correlation:
    - Coefficient magnitudes on wholesale funding are larger in columns 1–4 than in columns 5–10, indicating potential bias when not controlling for demand.
  - Main result:
    - When controlling for demand (borrower fixed effects), banks that relied more heavily on market-based funding before the crisis reduced the supply of syndicated loans more than other banks.
    - According to the estimated magnitudes, one percentage point increase in the share of non-deposit funding led to a decrease in the supply of syndicated credit by between

*Source: _wp13102 - Section IV discusses the baseline results. Robustness checks are presented in Section V. Section*

### 0.7 and 0.9 percent (across samples or sets of controls).

### _wp13102 - 0.7 and 0.9 percent (across samples or sets of controls).

### Wholesale funding effect and transmission through the bank lending channel
- Results corroborate previous studies on transmission of shocks through the bank lending channel and provide evidence for the so-called “dark side” of wholesale funding (Acharya, Gale and Yorulmazer, 2011; Huang and Ratnovski, 2011).
- The identified effect persists in a significantly larger cross-section of market participants than considered before.
- Potential confounding concern: non-US banks reliant on market-based funding also may have faced U.S. dollar shortages; results for non-US banks (columns 9–10) could reflect both reliance on non-deposit funding and dollar shortages (Kapan and Minoiu, 2012).
- Counterarguments alleviating the concern:
  - Cornett, McNutt, Strahan and Tehranian (2011) and Ivashina and Scharfstein (2010) show US banks (liabilities denominated in dollars) that were more dependent on wholesale funding were more likely to reduce lending during the crisis—evidence of a “wholesale funding effect” for US banks.
  - Federal Reserve Bank dollar liquidity swap lines (with ECB, Bank of England, Bank of Japan, Swiss National Bank) in early crisis stages likely reduced US dollar shortfalls for non-US banks; Goldberg, Kennedy and Miu (2010) argue these policies supported attribution of the baseline effect to wholesale funding rather than confounding factors.
- Representative magnitudes reported in robustness checks: a one percentage point increase in pre-crisis reliance on wholesale funding led to a decline in the supply of syndicated loans of 0.6–0.7 percent (statistically significant at the 1 percent level).

### B. The Mitigating Effect of Bank Capital
- Capital measures considered:
  - Regulatory Tier 1 ratio (Tier 1 / RWA) based on Basel II.
  - Total regulatory capital ratio (Tier 1 + Tier 2) / RWA based on Basel II.
  - Tangible common equity ratio (definition given in text), closer to Basel III concept.
- Method: interact capital measures with exposure to liquidity shocks; control for banks’ risk profile.
- Findings:
  - Using Tier 1 or total regulatory capital ratios (columns 1–8): interaction coefficients not statistically different from zero—extent of credit reduction unrelated to degree of capitalization by these measures.
  - Using tangible common equity (columns 9–12): exposed banks with more tangible common equity reduced syndicated loan supply less than other banks.
  - Evaluated at the sample mean of the tangible common equity ratio (6.90 percent), coefficients imply that with every percentage point increase in reliance on wholesale funding, the supply of credit fell by 0.65 to 0.82 percent (columns 9–12).
  - Column 10 calculation: –1.392+0.083*6.90 = –0.82 percent.
  - Example elasticity: an increase in bank capital by 1 standard deviation (4.29) from the sample mean (6.90 percent) reduces the negative effect of the liquidity shock from 0.82 to 0.46 percent.
  - Conversely, a reduction in bank capital by 1 standard deviation from the mean would amplify the effect of the liquidity shock to 1.18 percent.
- Interpretation:
  - Basel II regulatory Tier 1 and total regulatory capital ratios are not helpful for gauging banks’ ability to sustain lending under liquidity shocks.
  - Findings support Basel III proposals to raise quantity and quality of regulatory capital—especially high loss-absorbing capital like common equity.
  - Robustness checks with alternative capital ratios generally did not show statistically significant interaction effects, except for statistically weak results for Tier 1 as a share of total assets.

### C. The Role of Structural Liquidity: NSFR
- NSFR definition and interpretation:
  - NSFR represents the share of long-term illiquid assets funded with liabilities that are long-term or stable.
  - Calculated as NSFR = (Σi wi Li) / (Σj wj Aj) where i denotes liabilities, j denotes assets, and wi and wj are weights between 0 and 1 assigned to different liabilities and assets.
  - Higher NSFR indicates more stable funding relative to asset liquidity profile and lower liquidity risk.
- Measurement approach:
  - Accurate NSFR requires detailed balance sheet data unavailable in Bankscope; an approximate NSFR constructed using weights from Vazquez and Federico (2012) and a stylized balance sheet based on Bankscope variables.
  - Asset-side assumptions: entire loan portfolio assigned weight 1; fixed assets and non-interest earning assets treated similarly; “cash and due from banks” receives zero weight; other earning assets receive weight 0.35.
  - Liability-side assumptions: long-term funding, reserves, and equity receive maximum weight; short-term funding items (deposits from banks, other deposits and short-term borrowings, derivatives and trading liabilities) receive zero weight; customer deposits receive higher weights than short-term funding.
- Empirical findings:
  - Full-sample results (Table 5, columns 1–2): for each percent increase in the NSFR, banks increased lending post-crisis by 0.43 to 0.46 percent (coefficients statistically significant at conventional levels).
  - Interaction of NSFR with capital (columns 3–11):
    - Positive NSFR effect larger among well-capitalized banks when capital measured by Tier 1 or tangible common equity (columns 4, 10).
    - Banks with higher pre-crisis NSFR and strong capital reduced credit supply less than other banks.
    - For poorly-capitalized banks, higher NSFR did not translate into better ability to sustain lending.
  - Implication: complementarities between capital and structural liquidity—stable funding structure is effective only with a strong capital base.
- Caveats:
  - Approximate NSFR may be imperfect due to unavailable fine-grained balance sheet data; authors refrain from comparing estimated NSFR levels to regulatory 100 percent requirement.
  - Estimated NSFR coefficients in some specifications are only statistically significant at the 14.5 and 12.1 percent levels (columns 5, 12).

### V. Robustness checks
- Two robustness concerns addressed:
  - Whether crisis-induced credit supply shock was specific to syndicated loan market; addressed by controlling for change in total loans between pre- and post-shock periods (with caveat on demand shifts).
  - Whether strong banks sustained credit supply by acquiring other banks and expanding balance sheets; addressed by controlling for change in total assets between pre- and post-shock periods.
- Results:
  - Specifications with funding structure as proxy for exposure indicate a one percentage point increase in pre-crisis reliance on wholesale funding led to a decline in syndicated loan supply of 0.6–0.7 percent (statistically significant at the 1 percent level) (columns 1–2, 8–9).
  - The negative impact was cushioned by higher tangible common equity (columns 3–4, 10–11).
  - Results robust to using structural liquidity as the indicator of vulnerability to shocks (columns 5–7, 12–14), noting some NSFR coefficients only statistically significant at 14.5 and 12.1 percent levels (columns 5, 12).

### VI. Conclusions
- Main findings:
  - Banks that relied more heavily on wholesale funding and had lower structural liquidity reduced syndicated lending more than other banks following liquidity shocks.
  - Bank capital cushioned the negative impact: better-capitalized banks exposed to shocks decreased credit supply less than other banks.
  - Evidence that bank equity acts as a buffer is stronger when capitalization is measured with tangible common equity (close to Basel III concept); Tier 1 and total regulatory capital ratios are less useful for distinguishing banks’ abilities to sustain lending.
  - Banks with stronger balance sheets from the Basel III perspective were better able to perform intermediation during the crisis.
- Theoretical and policy implications:
  - Findings lend support to Disyatat’s (2011) re-interpretation of the bank lending channel in modern markets where market-based funding is prevalent—banks can act as amplifiers or absorbers of shocks depending on financial strength.
  - Results underscore importance of bank capital for post-crisis credit recovery and support proposals to raise regulatory standards on both quantity and quality of capital.

*Source: _wp13102 - 0.7 and 0.9 percent (across samples or sets of controls).*

### REFERENCES

### _wp13102 - REFERENCES

### Core literature and topics covered
- Rollover risk, market freezes, bank capital, liquidity, and transmission of financial shocks (selected authors: Acharya; Bernanke; Basel Committee on Banking Supervision; Cetorelli and Goldberg; Ivashina and Scharfstein; Huang and Ratnovski; Vazquez and Federico).
- Empirical evidence on bank lending and liquidity during the 2007–09 crisis across syndicated loan markets, deposit and non-deposit funding, dollar funding, sovereign exposures, and international shock transmission (selected authors: Chui et al.; De Haas and Van Horen; Giannetti and Laeven; Schnabl; Raddatz; Brei et al.).
- Regulatory and policy analyses of capital and liquidity frameworks including Basel III and NSFR (Basel Committee on Banking Supervision 2010a, 2010b, 2012; Vazquez and Federico 2012; King 2010, 2012).

### Datasets, sample sizes, and data sources
- Syndicated loan data: Loan Analytics (Dealogic). Initial universe: 39,626 syndicated loan deals signed between January 1, 2006 and March 31, 2010. After dropping lenders recorded as ―unknown,‖ ―undisclosed syndicate,‖ or ―undisclosed investor,‖ retained 38,538 loan deals with non-missing loan amount.
- Final merged sample after cleaning and matching with Bankscope: 24,082 deals collapsed at bank-industry level; matched to 853 lenders; after removing multilateral government banks, specialized government credit institutions, two Islamic banks, and banks with missing balance sheet variables, final sample is 803 banks lending to firms from 448 country-specific industries in 48 countries.
- Coverage note: Over 90 percent of the world’s largest 1,000 banks report to Bankscope.

### Key variable definitions and computation
- Non-deposit liabilities: (Total liabilities - total customer deposits) / Total liabilities.
- Non-deposit funding: (Total funding - total customer deposits) / Total funding.
- Tier 1 ratio: as reported in Bankscope.
- Total regulatory capital ratio: as reported in Bankscope.
- Tangible common equity ratio: as reported in Bankscope.
- Return on assets: Return on average assets (ROAA) as reported in Bankscope.
- Non-performing loans: Impaired loans / Gross loans as reported in Bankscope.
- Liquid asset ratio: Liquid assets / Total assets.
- Risk-weighted assets (RWA): Computed from Tier 1 regulatory ratio and Tier 1 regulatory capital variables (RWA not directly available in Bankscope).
- Total loans: Gross loans (as reported in Bankscope).
- NSFR (approximate): Computed in line with the Basel III proposal using weights from Vazquez and Federico (2012). Stylized balance-sheet weights reproduced (Assets and Liabilities weights), including 100% weights for Total earning assets, Total long-term funding, Preferred shares and hybrid capital, Equity, and fixed 0% weights for some deposit and bank-related items as shown in the NSFR schematic.

### Empirical model and imputation of syndicate shares
- For roughly 70 percent of loans LA does not report individual syndicate shares; shares imputed using a regression model estimated on the available 30 percent of loan observations (N=100,938 individual shares).
- Regression specification (log-shares): ln(sijkt) = β1 X1i + β2 X2j + β3 X3k + β4 X4t + ε
  - X1: log-loan amount, dummies for loan currency, number of lenders.
  - X2: borrower nationality and borrower industry dummies.
  - X3: lender role (mandated arranger/arranger/bookrunner/participant) and lender nationality dummies.
  - X4: year-quarter dummies.
- Model performance: adjusted R-squared = 74 percent.
- Out-of-sample prediction: predicted shares re-scaled to sum to 100 percent per loan; final dataset uses actual shares for 30 percent of loans and predicted shares for remaining loans.

### Data cleaning and name uniformization procedures
- Bank name uniformization to merge LA and Bankscope: retained Bankscope name if a bank changed name during 2006–2010Q1; keep merging banks distinct up to year of merger; acquired banks kept distinct until year of acquisition.
- Lender nationality: use LA variable ―Lender nationality‖; when missing, impute via cross-checking other deals, Bankscope, and online sources; offshore branches imputed to nationality of onshore parent if available.
- Matching procedure: exact match between LA and Bankscope for almost 40 percent of 853 lenders; remaining matched manually. Subsidiaries and branches kept if they report balance sheet information in Bankscope; subsidiaries not linked to parent balance sheets.

### Lists of countries and industries in the sample
- Advanced Economies (IMF WEO 2010): Australia, Austria, Belgium, Canada, Croatia, Czech Republic, Denmark, Finland, France, Germany, Greece, Hong Kong SAR, Ireland, Italy, Japan, Luxembourg, Malta*, The Netherlands, New Zealand, Norway, Portugal, Singapore, Slovenia, South Korea, Spain, Sweden, Switzerland, Taiwan Province of China, United Kingdom, United States.
- Emerging Market Economies (FTSE list): Argentina*, Brazil, Chile*, China, Colombia*, Egypt, Hungary, India, Indonesia, Malaysia, Mexico, Morocco*, Pakistan*, Peru, Philippines, Poland, Russian Federation, South Africa, Thailand*, Turkey, United Arab Emirates.
- Borrowing country-specific industries and 2007 syndicated loan volume shares (selected): aerospace (0.71); agribusiness (0.27); auto/truck (3.34); chemicals (3.26); computers and electronics (5.32); construction and building (11.46); finance (13.27); mining (10.04); telecommunications (5.44); utility and energy (7.80); others as listed.
- Full list of lender countries includes: Australia, Austria, Bahrain, Bangladesh, Belgium, Brazil, Canada, China, Croatia, Czech Republic, Denmark, Egypt, Finland, France, Germany, Greece, Hong Kong SAR, Hungary, India, Indonesia, Ireland, Israel, Italy, Japan, Jersey, Jordan, Luxembourg, Macao SAR, Malaysia, Malta, Mexico, The Netherlands, Norway, Pakistan, Peru, Philippines, Poland, Portugal, Qatar, Russian Federation, Saudi Arabia, Singapore, South Africa, South Korea, Spain, Sweden, Switzerland, Taiwan Province of China, Thailand, Turkey, United Arab Emirates, United Kingdom, United States.

### Figures, tables, and sample statistics (high-level)
- Figure 1: Syndicated loan market total global issuance (1992–2010), Loan Analytics vs BIS series; note difference due to BIS exclusion of loans with maturity greater than 3 months.
- Figure 2: Three-month dollar Libor-OIS spread (2004–2011); ―before‖ period Jan 2006-Jun 2007; ―shock‖ window Jul 2007-Sep 2008; ―after‖ Oct 2008-Mar 2010.
- Figures 3–5: Trends in bank reliance on wholesale funding (1999–2010), approximate NSFR trends (1999–2010), and NSFR stylized calculation (weights by balance-sheet category).
- Table 1 (Summary statistics): Dependent variable Log-change in lending (x100), Obs = 7,156; Mean = -21.411; St. Dev. = 42.71; P5 = -254.68; P50 = -21.79; P95 = 214.72. Non-deposit liabilities Obs = 7,103; Mean = 2.33; St. Dev. = 26.81; P5 = 1.49; P50 = 23.91; P95 = 86.91. Non-deposit funding Obs = 7,105; Mean = 29.53; St. Dev. = 26.85; P5 = 0.18; P50 = 20.48; P95 = 86.52. Log-NSFR Obs = 4,444; Mean = 4.44; St. Dev. = 0.22; P5 = 3.90; P50 = 4.46; P95 = 4.77. (Additional bank characteristics and percentiles reported in table.)
- Tables 3–6: Baseline regressions, interactions with capital, NSFR role, robustness checks. Selected empirical notes:
  - Baseline dependent variable: change in log-lending between Jan 2006-Jun 2007 and Oct 2008-Mar 2010.
  - Standard errors clustered at bank level. Significance notation: * 10%, ** 5%, *** 1% (Table 6: # indicates 15%).
  - Observations and R-squared vary across specifications; examples: Table 3 observations range 4,788–6,694 with R-squared up to 0.309; Table 4 observations 5,756–6,051 with R-squared ≈ 0.277–0.280; Table 5 observations 852–4,249 with R-squared up to 0.513; Table 6 observations 3,019–5,706 with R-squared up to 0.517.

### Data appendix: merging, winsorization, and measurement timing
- Winsorization: All variables (apart from log-change in lending, liquid asset ratio, risk profile, and total assets) are winsorized at the 1st and 99th percentiles.
- Covariate timing: All covariates are measured as of 2007Q2 or 2007Q1 if Q2 data missing; if quarterly data missing, variables measured as midpoint of 2006 and 2007 data, or as of end-2006.
- Matching note: LA and Bankscope have no common unique identifier; matching based on lender name and nationality with manual matching for non-exact cases.
- Final analytic restrictions: borrowers must borrow in syndicated market before and after shock window (2007Q3–2008Q3) from at least two distinct financial institutions.

*Italic: Source: _wp13102 - REFERENCES (PDF chapters and data appendix as provided).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp13102.pdf_