## _wp14185

## Source details

**Canonical URL:** [_wp14185](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp14185.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp14185.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp14185.pdf.json)

---

### I. Introduction — scope and purpose
- Timeframe of analysis: 1995–2012.
- Key focus: composition of cross-border bank lending with emphasis on syndicated loans and drivers of syndication versus non-syndicated lending.
- Main empirical approach: bilateral (country-pair) panel of 26 lender countries and 76 borrower countries; gravity-type empirical model.
- Data sources: Dealogic Loan Analytics for syndicated loans and BIS international banking statistics (IBS) for total cross-border bank claims.

### II. Stylized facts and key statistics
- Total cross-border loan claims almost tripled to reach 20 trillion U.S. dollars between 1995 and 2012.
- Syndicated loan exposures (SLEs):
  - Represented between 20 percent of total loan claims early in the sample to over 30 percent in later years.
  - Average share of SLEs in total loan claims: 30 percent for advanced economy (AE) borrowers and 18 percent for emerging market economy (EME) borrowers.
- Global syndicated loan volume:
  - Increased 160 percent between 1995 and 2012 to reach 3.5 trillion U.S. dollars.
  - Peak total deal volume in 2007: 4.5 trillion U.S. dollars; total deal volume fell in 2009 by more than 50 percent from the 2007 peak.
- Market concentration and composition:
  - Close to 90 percent of total deal volume accrues to AE borrowers.
  - Borrower types: roughly 75 percent to non-financial firms, 15 percent to financial firms, and 10 percent to sovereigns and public sector entities.
  - Market share of the top 100 borrowers declined from about 45 percent in the mid-1990s to about 25 percent in 2012.
- Syndicate structure and deal characteristics:
  - Average syndicate size during 1995–2012: 6.2 participants, including 2.7 lead banks (bookrunners or mandated arrangers).
  - Geographical composition of syndicates (from borrower perspective): 46 percent foreign banks and 54 percent domestic banks on average.
  - Close to 60 percent of syndicated loan deals had at least one foreign participant.
  - Most loans denominated in U.S. dollars; pricing typically over LIBOR.
  - Average loan deal maturity during 1995–2012: 4.7 years.
  - Estimated average maturity for bilateral cross-border loans: 3.1 years (upper bound estimate).
  - Estimated average maturity for all loans on AE banks’ balance sheets: 3 years (upper bound estimate).
  - Average loan extended at the 2007 peak amounted to almost half a billion U.S. dollars; loan size decreased during the global financial crisis, especially to AE borrowers.
  - Loan spreads before the global financial crisis hovered around 150-200 basis points over LIBOR; spreads doubled at the height of the crisis.

### III. Effects of the global financial crisis (2008–2012)
- Paradoxical finding: SLEs outstanding (stocks) increased during the crisis despite a collapse in syndicated loan origination (new deals).
- Mechanism: increase in drawdowns on existing syndicated loan commitments (credit lines).
- Estimated credit line usage rate:
  - Increased from approximately 25 percent before the global financial crisis to 52 percent by 2012.
- Crisis impact on deal volumes and pricing:
  - Total deal volume collapsed (see 2009 fall of more than 50 percent from 2007 peak).
  - Spreads doubled at the height of the crisis.
- During the global financial crisis, both SLEs and non-SLEs were higher for country pairs with lower information asymmetries.

### IV. Drivers of syndication vs. non-syndicated lending — empirical findings
- Information asymmetries and geography:
  - Greater informational asymmetries (less economic integration and greater geographical distance) are associated with lower total cross-border loan activity.
- Lender balance sheet characteristics:
  - Banks with lower levels of capital in lender countries favor syndicated loans over other kinds of cross-border loans.
  - A 1 percentage point increase in the capital-to-assets ratio reduces SLEs by 4.9 percent (column 1 baseline).
- Borrower country characteristics:
  - Level of development, economic size, and capital account openness play a lesser role for SLEs compared to non-SLEs, consistent with a diversification motive for syndications.
- Gravity-type correlates (baseline regression magnitudes preserved):
  - A 10 percent increase in bilateral trade is associated with a 2.4-2.6 percent increase in cross-border loan exposures.
  - A 10 percent decrease in geographical distance brings about an increase in loan exposures by between 4 and 8 percent.
- Model fit:
  - Baseline gravity-type models yield R2 in excess of 70 percent.

### V. Market structure, incentives, and secondary market
- Lead banks and incentives:
  - Lead banks (bookrunners) incur reputational costs from loan defaults, providing incentives to hold greater portions of loans as a signal of screening and monitoring.
- Secondary market:
  - Syndicated loans can be sold in an active secondary market, allowing originate-to-distribute behavior in principle.
  - Empirical evidence suggests securitization or trading of syndicated loans does not necessarily lead to worse loan performance for syndicated loans, possibly due to the syndication structure and reputational incentives.

### VI. Data construction and methodology highlights
- Main datasets:
  - Dealogic Loan Analytics: more than 150,000 syndicated loan deals downloaded between 1990 and 2012 to estimate country-pair SLEs during 1995–2012.
  - BIS IBS: locational and consolidated banking statistics capturing cross-border assets and liabilities of creditor banking systems vis-à-vis borrower countries.
- Key data challenges and adjustments:
  - BIS data report stocks at a point in time, while syndicated loan data record origination flows and include both disbursed loans and loan commitments (credit lines).
  - Authors estimate the stock of outstanding cross-border syndicated loans at the lender-borrower country-pair level using deal volume and maturity, applying the same aggregation criteria as BIS IBS.
  - Aggregation performed on both consolidated and locational bases; primary analysis focuses on locational aggregation due to longer time series.
  - Imputation: loan volumes are split equally across syndicate participants to obtain lender-specific loan amounts and exposures when lender-specific shares are missing; validation exercise shows pro-rata imputation produces ratios close to 1 across top country-pair aggregates (see Table A1 values).
  - Use of consolidated data to estimate on-balance sheet share of syndicated credit lines; adjusted SLEs constructed by applying estimated credit line usage rates.

### VII. Appendix findings — adjustments, validation, and decomposition
- Imputation validation (2006–2010 sample; Table A1 exact ratios preserved):
  - top 100 country pairs — cumulative share: 41.6% — (i) Pro-rata: 0.98 — (ii) Regression: 0.99 — (iii) Average-to-lead: 0.98
  - top 200 country pairs — cumulative share: 54.3% — (i) Pro-rata: 0.99 — (ii) Regression: 0.99 — (iii) Average-to-lead: 0.99
  - top 300 country pairs — cumulative share: 62.9% — (i) Pro-rata: 1.00 — (ii) Regression: 1.01 — (iii) Average-to-lead: 0.99
  - top 400 country pairs — cumulative share: 69.0% — (i) Pro-rata: 1.00 — (ii) Regression: 1.00 — (iii) Average-to-lead: 0.98
  - top 500 country pairs — cumulative share: 73.8% — (i) Pro-rata: 1.00 — (ii) Regression: 1.00 — (iii) Average-to-lead: 0.98
  - top 1000 country pairs — cumulative share: 94.2% — (i) Pro-rata: 1.06 — (ii) Regression: 1.04 — (iii) Average-to-lead: 1.03
  - Median (across country-pair cutoffs): (i) Pro-rata: 1.01 — (ii) Regression: 1.01 — (iii) Average-to-lead: 1.00
- Adjustments to make Loan Analytics SLEs comparable to BIS loan claims:
  - Restrict SLEs to syndicated loans reported as credit lines or “Term Loan A”-type term loans (extended almost exclusively by banks).
  - Adjust credit line amounts to reflect drawn (on-balance sheet) amounts using credit line utilization information from BIS consolidated statistics.
- Estimated credit line usage rates (Table A2 annual values preserved):
  - 2005: 25.65, 28.69, 27.96, 36.44, 32.08, 24.65
  - 2006: 28.12, 32.95, 27.73, 34.97, 33.40, 27.92
  - 2007: 25.14, 24.22, 23.21, 43.65, 36.70, 24.24
  - 2008: 39.62, 43.65, 37.30, 51.96, 43.31, 38.25
  - 2009: 40.20, 36.78, 47.35, 62.93, 54.25, 39.26
  - 2010: 50.46, 49.51, 50.00, 62.49, 48.03, 49.85
  - 2011: 57.40, 56.90, 50.61, 68.74, 44.29, 56.83
  - 2012: 52.16, 52.74, 52.40, 66.16, 44.52, 51.17
- Decomposition of total cross-border loan exposures (1995–2012):
  - Using adjusted SLEs and BIS locational statistics, syndicated loan exposures, bilateral loan exposures, and intragroup loan exposures each accounted for roughly one third of total cross-border bank loan claims.
  - Intragroup loans account for 28.8 percent of total claims using BIS locational statistics.

### VIII. Empirical results — key coefficients, interactions, and robustness
- Baseline regression highlights (Table 4 magnitudes preserved where reported):
  - Log-real trade coefficients: 0.241*** (log-SLE, full sample), 0.256*** (log-non-SLE, full sample).
  - Log-geographical distance coefficients: -0.405*** (log-SLE, full sample), -0.777*** (log-non-SLE, full sample).
  - Common language: 0.305*** (log-SLE, full sample); mixed/non-significant for log-non-SLE in some specs.
  - Foreign affiliate: 0.525*** (log-SLE, full), 0.753*** (log-non-SLE, full).
  - Bank capital-to-assets ratio (lender): -0.049*** (log-SLE, full sample); no association with non-SLEs in baseline.
  - Log-total other lending: 0.617*** (log-SLE, full sample).
  - Borrower bank assets (% GDP): 0.002** (log-SLE, full), 0.006*** (log-non-SLE, full).
  - Capital account openness: 0.009*** (log-non-SLE, full).
- Crisis interactions (Table 5 preserved magnitudes):
  - Log-real trade*GFC: 0.081*** (log-SLE, full sample); 0.062*** (log-non-SLE, full sample).
  - Log-geographical distance*GFC: -0.084*** (log-SLE, full).
  - Capital account openness*GFC: 0.005*** (log-SLE, full).
  - Institutional quality*GFC: 0.014 (not always significant).
  - When all interactions included, only bilateral trade effect remains statistically significant: during the crisis, country pairs with total trade flows higher by 10 percent experienced cross-border loan exposures higher by 2.3-2.6 percent.
- Robustness:
  - Country-pair fixed effects largely leave baseline results intact though foreign affiliate dummy is largely absorbed and borrower institutional quality can lose significance.
  - Dynamics: autoregressive coefficients around 0.5 (moderate persistence); bias-corrected LSDV estimator confirms main patterns.
  - Additional lender variables: net charge-offs positively associated with SLEs (e.g., coefficient 0.0603** in one spec).
  - Additional borrower variables: stock market return volatility negatively associated with both SLEs and non-SLEs (e.g., -0.0002*** in some specs); % internationally rated banks linked positively to non-SLEs in some specs (e.g., 0.0033***).
  - Pre-crisis (1995-2007) results similar to full sample for key findings.

### IX. Policy-relevant interpretations and implications
- Information and monitoring costs matter:
  - Lower information asymmetries (proxied by bilateral trade, geographic proximity, common language) raise cross-border syndicated and non-syndicated loan exposures.
- Lender capitalization and lending structure:
  - Lower capital levels in lender banks are associated with greater reliance on syndication (syndicated loan exposures increase as capital falls), suggesting capital constraints shape banks’ incentives to syndicate large loans.
- Borrower risk and lending composition:
  - Syndication appears tied to diversification motives: borrower country development, economic size, and capital account openness matter less for SLEs than for non-SLEs.
  - Bilateral lenders may be less willing to serve riskier borrowers because they bear full credit risk on their balance sheets.
- Crisis-period behavior:
  - During crises, economic proximity (especially bilateral trade) becomes more important for maintaining cross-border lending, indicating that stronger trade ties reduce perceived information asymmetries when monitoring needs rise.
- Data and measurement caution:
  - The share of cross-border syndicated lending ranges from 20 percent to more than 30 percent of total cross-border loan claims in the full sample, and syndicated loan exposures increased during the global financial crisis due to large credit line drawdowns.
  - Significant heterogeneity in the share of syndicated exposures across countries and over time implies caution in using syndicated loan data as representative of total cross-border bank lending.

*Content derived from _wp14185 - References and Appendix (IMF Working Paper).*

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

### References

### I. Introduction — scope and purpose
- Timeframe of analysis: 1995–2012.
- Key focus: composition of cross-border bank lending with emphasis on syndicated loans and drivers of syndication versus non-syndicated lending.
- Main empirical approach: bilateral (country-pair) panel of 26 lender countries and 76 borrower countries; gravity-type empirical model.
- Data sources: Dealogic Loan Analytics for syndicated loans and BIS international banking statistics (IBS) for total cross-border bank claims.

### II. Stylized facts and key statistics
- Total cross-border loan claims almost tripled to reach 20 trillion U.S. dollars between 1995 and 2012.
- Syndicated loan exposures (SLEs):
  - Represented between 20 percent of total loan claims early in the sample to over 30 percent in later years.
  - Average share of SLEs in total loan claims: 30 percent for advanced economy (AE) borrowers and 18 percent for emerging market economy (EME) borrowers.
- Global syndicated loan volume:
  - Increased 160 percent between 1995 and 2012 to reach 3.5 trillion U.S. dollars.
  - Peak total deal volume in 2007: 4.5 trillion U.S. dollars; total deal volume fell in 2009 by more than 50 percent from the 2007 peak.
- Market concentration and composition:
  - Close to 90 percent of total deal volume accrues to AE borrowers.
  - Borrower types: roughly 75 percent to non-financial firms, 15 percent to financial firms, and 10 percent to sovereigns and public sector entities.
  - Market share of the top 100 borrowers declined from about 45 percent in the mid-1990s to about 25 percent in 2012.
- Syndicate structure and deal characteristics:
  - Average syndicate size during 1995–2012: 6.2 participants, including 2.7 lead banks (bookrunners or mandated arrangers).
  - Geographical composition of syndicates (from borrower perspective): 46 percent foreign banks and 54 percent domestic banks on average.
  - Close to 60 percent of syndicated loan deals had at least one foreign participant.
  - Most loans denominated in U.S. dollars; pricing typically over LIBOR.
  - Average loan deal maturity during 1995–2012: 4.7 years.
  - Estimated average maturity for bilateral cross-border loans: 3.1 years (upper bound estimate).
  - Estimated average maturity for all loans on AE banks’ balance sheets: 3 years (upper bound estimate).
  - Average loan extended at the 2007 peak amounted to almost half a billion U.S. dollars; loan size decreased during the global financial crisis, especially to AE borrowers.
  - Loan spreads before the global financial crisis hovered around 150-200 basis points over LIBOR; spreads doubled at the height of the crisis.

### III. Effects of the global financial crisis (2008–2012)
- Paradoxical finding: SLEs outstanding (stocks) increased during the crisis despite a collapse in syndicated loan origination (new deals).
- Mechanism: increase in drawdowns on existing syndicated loan commitments (credit lines).
- Estimated credit line usage rate:
  - Increased from approximately 25 percent before the global financial crisis to 52 percent by 2012.
- Crisis impact on deal volumes and pricing:
  - Total deal volume collapsed (see 2009 fall of more than 50 percent from 2007 peak).
  - Spreads doubled at the height of the crisis.
- During the global financial crisis, both SLEs and non-SLEs were higher for country pairs with lower information asymmetries.

### IV. Drivers of syndication vs. non-syndicated lending — empirical findings
- Information asymmetries and geography:
  - Greater informational asymmetries (less economic integration and greater geographical distance) are associated with lower total cross-border loan activity.
- Lender balance sheet characteristics:
  - Banks with lower levels of capital in lender countries favor syndicated loans over other kinds of cross-border loans.
- Borrower country characteristics:
  - Level of development, economic size, and capital account openness play a lesser role for SLEs compared to non-SLEs, consistent with a diversification motive for syndications.
- Interpretation and link to literature:
  - Findings support motives emphasized in prior literature: lender balance sheet strength as determinant of syndication (Simons, 1993); diversification motive and relevance of borrower institutions/financial development (Godlewski and Weill, 2007).

### V. Market structure, incentives, and secondary market
- Lead banks and incentives:
  - Lead banks (bookrunners) incur reputational costs from loan defaults, providing incentives to hold greater portions of loans as a signal of screening and monitoring.
- Secondary market:
  - Syndicated loans can be sold in an active secondary market, allowing originate-to-distribute behavior in principle.
  - Empirical evidence suggests securitization or trading of syndicated loans does not necessarily lead to worse loan performance for syndicated loans, possibly due to the syndication structure and reputational incentives.

### VI. Data construction and methodology highlights
- Main datasets:
  - Dealogic Loan Analytics: more than 150,000 syndicated loan deals downloaded between 1990 and 2012 to estimate country-pair SLEs during 1995–2012.
  - BIS IBS: locational and consolidated banking statistics capturing cross-border assets and liabilities of creditor banking systems vis-à-vis borrower countries.
- Key data challenges and adjustments:
  - BIS data report stocks (banking-sector exposures) at a point in time, while syndicated loan data record origination flows and include both disbursed loans and loan commitments (credit lines).
  - The authors estimate the stock of outstanding cross-border syndicated loans at the lender-borrower country-pair level using deal volume and maturity, applying the same aggregation criteria as BIS IBS.
  - Aggregation performed on both consolidated and locational bases using lender/borrower location and lender parent nationality; primary analysis focuses on locational aggregation due to longer time series.
  - Imputation: loan volumes are split equally across syndicate participants to obtain lender-specific loan amounts and exposures when lender-specific shares are missing.
  - Use of consolidated data to estimate on-balance sheet share of syndicated credit lines, a crucial step to make SLEs comparable to BIS loan claims.

### VII. Additional methodological notes and scope limitations
- Period limitation: analysis limited to 1995–2012 because Loan Analytics data pre-1990 are of lower quality and to ensure maximum availability of country characteristic data.
- Data limitations: unable to further break down non-SLEs into single-lender and intragroup components due to data constraints.
- Terminology: the paper uses “claims” and “exposures” interchangeably.

*Content derived from _wp14185 - References (IMF Working Paper).*

### Appendix we show that the approach of splitting loan deal amounts equally across lenders produces estimates of

### _wp14185 - Appendix we show that the approach of splitting loan deal amounts equally across lenders produces estimates of

### Constructing syndicated loan exposures
- Comparison made between total cross-border loan claims on a locational basis from the BIS and estimated syndicated loan exposures (SLEs) over 1995–2012, both expressed in trillions of constant U.S. dollars (2005 prices).
- Between 1995 and 2007, total cross-border loan claims rose three-fold; estimated SLEs increased by a comparable amount.
- During the global financial crisis there was a significant decrease in total bank loan claims (by about 5 trillion U.S. dollars from their 2007 peak to 2012) while SLEs did not experience the same decrease.
- Estimated SLEs exceed BIS total cross-border loan claims for about half of the time. Two explanations provided:
  - Some syndication participants are non-bank institutional investors while BIS IBS only capture banks’ positions.
  - Syndicated loan deals often involve credit lines that are not fully drawn over the life of the loan.
- Two adjustments made to make SLEs comparable with BIS loan claims:
  - Restrict SLEs to syndicated loans reported as either credit lines or “Term Loan A”-type term loans (extended almost exclusively by banks).
  - Adjust credit line amounts to reflect drawn (on-balance sheet) amounts using credit line utilization information obtained by comparing syndicated credit line exposures with undrawn credit lines from the BIS consolidated banking statistics (available for a limited number of lender countries from 2005 onwards).
- Estimated credit line usage rates:
  - Approximately 25 percent before the global financial crisis.
  - 40 percent by 2009.
  - 57 percent by 2011 at the height of the European sovereign debt crisis.
- These usage rate trends are consistent with literature documenting cyclical variation in credit line utilization (cited studies: Mian and Santos (2012); Ivashina and Scharfstein (2010); Berrospide and Meisenzahl (2013); Correa, Sapriza, and Zlate (2013)).
- The credit line utilization rates are used to adjust the credit line component of SLEs downward to obtain “adjusted SLEs” comparable to BIS loan claims.

### The composition of cross-border bank loan exposures
- BIS total cross-border loan claims are decomposed into adjusted SLEs and non-SLEs (residual = BIS loan claims − adjusted SLEs); panels consider all borrowers, AE borrowers, and EME borrowers.
- From 1995 to 2012, the share of syndicated lending in total loan claims:
  - Fluctuated between 17 and 41 percent in the full sample.
  - Between 19 and 47 percent for AE borrowers.
  - Between 10 and 27 percent for EME borrowers.
- Relative importance of the syndicated loan market has grown over time.
- Co-movement evidence:
  - SLEs and BIS total loan claims co-move significantly.
  - Regressing BIS total loan claims on (unadjusted) SLEs: about 50 percent of the variation in total exposures is explained by variation in SLEs; country-pair and year fixed effects explain an additional 25 percent.
- Despite lower loan volumes to EMEs than AEs, syndicated loans remain significant for both AE and EME borrowers.
- During the crisis:
  - BIS total loan claims show a steep increase in the run-up to the crisis and a reduction in exposures to AE borrowers during the crisis consistent with deleveraging.
  - SLEs increased during the crisis due to increased drawdowns on existing syndicated credit lines and lengthening maturities (average maturity rose from 3.8 years in 2002 to 5.3 years in 2007).
  - These factors induced stickiness in SLE dynamics.
- Non-SLEs refer to bilateral and intragroup loans. Using BIS locational statistics:
  - Intragroup loans account for 28.8 percent of total claims.
- Using adjusted SLEs, the estimated contribution of intragroup lending, and the residual (bilateral loans), a rough estimate for 1995–2012: these components each contribute about one third of total cross-border loan exposures.
- Heterogeneity in SLE shares across jurisdictions:
  - Median share during 1995–2012 varies from almost zero for the Cayman Islands, Cyprus, and Panama to over 20 percent for Australia, Japan, and South Africa.
  - SLEs are zero for offshore financial centers where loan origination is mainly by non-bank institutional lenders.
- Additional note: In 2012, 86 percent of loan deals were extended by syndicates with at least one domestic bank.

### V. Drivers of cross-border bank lending activity — Hypotheses and variables
- Analysis framework: gravity-type empirical model comparing determinants of SLEs and non-SLEs, drawing on literature on cross-border capital flows and banks’ syndication decisions.
- Baseline covariates include country-pair, lender, and borrower characteristics; focus on push-pull and gravity-type drivers.
- Observation 1 (information asymmetries):
  - Greater information asymmetries reduce cross-border bank lending.
  - Proxies: bilateral trade, geographical distance, common language indicator.
  - Syndication may reduce information asymmetries because most syndicates include domestic banks.
  - Expect information asymmetries to matter more during financial stress.
- Observation 2 (balance sheet constraints and lender size):
  - Banking system capacity and balance sheet strength influence ability to intermediate cross-border credit.
  - Syndication helps reduce concentration risk; when capital and liquidity constraints bind, banks may prefer syndication.
  - Larger and more profitable banks may prefer bilateral lending.
  - Expect capital constraints to be more binding during the global financial crisis.
  - Lender characteristics included: per capita income, size of banking system (total banking system assets), bank total regulatory capital-to-asset ratio (main proxy for lender capital), net interest margins, returns on assets, non-performing assets. Banking regulation controlled via lender country fixed effects.
- Observation 3 (borrower country risk):
  - Borrower risk characteristics affect lenders’ willingness to extend credit and borrowers’ access.
  - Syndication allows diversification and may make borrower risk less relevant for SLEs versus non-SLEs.
  - Borrower variables included: per capita income, size of domestic banking system, institutional quality, capital account openness.

### V. Drivers of cross-border bank lending activity — Regression specification
- Regression form estimated for 1995–2012 bilateral dataset:
  - log(SLE)ijt = αi + δj + λt + Xit β + Xjt γ + Zijt β + εijt
  - Analogous specification for non-SLEs.
- Sample: 26 lender countries and 76 borrower countries; analysis performed on country-pairs with non-zero syndicated loan activity (unbalanced panel).
- All coefficients estimated using Ordinary Least Squares (OLS); standard errors clustered on country pair.
- Control variables included in all regressions:
  - Indicator for presence of foreign affiliates of lender country i in borrower country j (captures intragroup lending mechanical effect and potential co-syndication).
  - Total other lending (log) for lender i and total other borrowing (log) for borrower j (sums of exposures to countries other than the pair partner) to control for heterogeneity in dynamics.
- Additional model features:
  - Interaction terms between a global financial crisis dummy (2008–2012) and selected covariates in subsequent specifications.
  - Country fixed effects included in all baseline specifications; country-pair fixed effects used in robustness checks.
  - Year fixed effects included to capture global variables such as uncertainty and risk aversion.
  - Robustness section and Appendix present specifications controlling for a wider range of characteristics and methodological choices.

### V. Drivers of cross-border bank lending activity — Empirical results (Baseline)
- Baseline results reported in Table 4 for dependent variables SLEs (columns 1–3) and non-SLEs (columns 4–6).
- Purpose of baseline specifications: explore correlates of different types of cross-border loan exposures rather than establish causality; results should not be interpreted causally.

*Source: _wp14185 - Appendix we show that the approach of splitting loan deal amounts equally across lenders produces estimates of*

### Appendix Table B2 reports unconditional correlations between the dependent variables and selected regressors.

### _wp14185 - Appendix Table B2 reports unconditional correlations between the dependent variables and selected regressors.

### Main empirical findings on drivers of SLEs and non-SLEs
- Higher volume of trade between lender and borrower countries, lower geographical distance between the capitals, and sharing a common language—indicators of lower information asymmetries—are associated with higher SLEs. Results extend to non-SLEs, although the coefficient on “common language” remains statistically significant only for exposures to EME borrowers.
- A 10 percent increase in bilateral trade is associated with a 2.4-2.6 percent increase in cross-border loan exposures (columns 1, 4).
- A 10 percent decrease in geographical distance brings about an increase in loan exposures by between 4 and 8 percent (columns 1, 4).
- Lender country characteristics:
  - Income level and size of the lender’s banking system do not influence cross-border loan exposures of either kind.
  - Higher levels of capital (measured with the regulatory capital-to-assets ratio) are associated with lower SLEs: a 1 percentage point increase in the capital-to-assets ratio—about half a standard deviation in the sample—reduces SLEs by 4.9 percent (column 1).
  - There is no association between the degree of capitalization and non-SLEs (columns 4-6).
- Borrower country characteristics:
  - Institutional quality appears to systematically favor syndicated lending (columns 1-3), though this result is not very robust to subsequent specifications.
  - Non-SLEs are higher for borrowers from higher income countries, countries with larger banking systems, and countries with more open capital accounts (columns 4-6).
  - Non-SLEs are larger vis-à-vis higher-income AE borrowers (column 5) and larger vis-à-vis EME borrowers with higher capital account openness (columns 4, 6).
- Control variables:
  - Presence of foreign affiliates of the lender country in the borrower country raises loan exposures of both kinds.
  - Variables measuring total exposure of each lender and borrower generally yield positive and statistically significant coefficients.
- Overall model fit:
  - Baseline results provide support for gravity-type models of cross-border bank assets, with R2 in excess of 70 percent.

### Interactions with the global financial crisis (2008-2012)
- Models include interaction terms between an indicator for the 2008-2012 period and, respectively, bilateral trade, geographical distance, lender bank capital ratio, borrower capital account openness, and borrower institutional quality.
- Country pairs with higher bilateral trade and lower geographical distance had higher cross-border loan exposures during the crisis.
- Higher borrower capital account openness and better institutions are associated with higher loan exposures during the crisis (institutional quality statistically significant only for non-SLEs).
- When all interaction terms are included simultaneously (columns 6, 12), only the effect of bilateral trade remains statistically significant:
  - During the crisis, country pairs with total trade flows higher by 10 percent experienced cross-border loan exposures that were higher by 2.3-2.6 percent.
- Interpretation: stronger economic ties reduced information asymmetries during the crisis; economic proximity became more relevant when loan screening and monitoring needs rose.
- Subsample note: these crisis-interaction results are driven by the subsample of exposures vis-à-vis AE borrowers (Appendix Tables B3 and B4).
- The size of the borrower country banking system becomes positively correlated with both types of cross-border loan exposures when interaction terms are included (coefficient magnitudes lower for SLEs).

### Robustness analysis and alternative specifications
- Country-pair fixed effects:
  - Including country-pair fixed effects largely leaves baseline results intact.
  - Exceptions: the foreign affiliate dummy is largely absorbed by pair effects; borrower institutional quality coefficients remain positive but lose statistical significance.
- Dynamics and lagged dependent variable:
  - Autoregressive coefficient around 0.5, indicating moderate persistence.
  - Results robust across: (i) OLS with country fixed effects, (ii) OLS with country-pair fixed effects, and (iii) OLS with bias correction for short panels (bias-corrected least squares dummy variable dynamic panel data estimator with initial Anderson-Hsiao coefficient values and bootstrapped standard errors).
  - Higher lender capital reduces SLEs; higher borrower income, larger banking system, and higher capital account openness increase non-SLEs.
  - Borrower institutional quality loses statistical significance when adding country-pair fixed effects or using the bias-corrected estimator.
  - System GMM produced unreliable results due to a large set of instruments and unstable coefficients with large standard errors.
- Additional lender and borrower characteristics:
  - Lender balance sheet variables examined: net interest margins, return on assets, liquid assets-to-deposits ratio, loan loss reserves, net charge-offs.
  - The only consistently relevant lender balance-sheet variable is loan portfolio quality (net charge offs in percent of average gross loans), which is positively associated with SLEs and unrelated to non-SLEs (columns 6, 12).
  - Inclusion of loan portfolio quality removes statistical significance of the capital coefficient (columns 5-6), though the capital coefficient remains negative.
  - Additional borrower variables: exchange rate volatility, stock market capitalization, stock market return volatility, % of internationally rated banks, and S&P sovereign credit rating.
  - Stock market return volatility is the only factor that matters for both SLEs and non-SLEs: higher borrower stock market volatility is associated with lower cross-border loan exposures.
  - When all variables included simultaneously (columns 6, 12), the share of internationally rated banks in the borrower’s banking system is positively linked to non-SLEs.
- Robustness to alternative time period:
  - Restricting the sample to 1995-2007 yields similar main findings:
    - High levels of bank capital in lender countries are negatively associated with SLEs (columns 1-3).
    - Borrower country income, banking system size, and capital account openness are positively related to non-SLEs (columns 4-6).
    - Better institutional quality in the borrower country increases syndicated loans to EME borrowers only (column 3).

### Policy-relevant interpretations and implications
- Information and monitoring costs matter:
  - Lower information asymmetries (proxied by bilateral trade, geographic proximity, common language) raise cross-border syndicated and non-syndicated loan exposures.
- Lender capitalization and lending structure:
  - Lower capital levels in lender banks are associated with greater reliance on syndication (syndicated loan exposures increase as capital falls), suggesting capital constraints shape banks’ incentives to syndicate large loans.
- Borrower risk and lending composition:
  - Syndication appears tied to diversification motives: borrower country development, economic size, and capital account openness matter less for SLEs than for non-SLEs.
  - Bilateral lenders may be less willing to serve riskier borrowers because they bear full credit risk on their balance sheets.
- Crisis-period behavior:
  - During crises, economic proximity (especially bilateral trade) becomes more important for maintaining cross-border lending, indicating that stronger trade ties reduce perceived information asymmetries when monitoring needs rise.
- Data and measurement caution:
  - The share of cross-border syndicated lending ranges from 20 percent to more than 30 percent of total cross-border loan claims in the full sample, and syndicated loan exposures increased during the global financial crisis due to large credit line drawdowns.
  - Significant heterogeneity in the share of syndicated exposures across countries and over time implies caution in using syndicated loan data as representative of total cross-border bank lending.

*Source: _wp14185 - Appendix Table B2 reports unconditional correlations between the dependent variables and selected regressors.*

### REFERENCES

### _wp14185 - REFERENCES

### Bibliographic scope
- References cite empirical and theoretical literature on syndicated loans, cross-border banking, financial globalization, crisis transmission, and related econometric methods. Examples include works by Adrian and Shin (2011); Altman, Gande, and Saunders (2010); Benmelech, Dlugosz and Ivashina (2012); Brunnermeier (2009); Cerutti, Claessens, McGuire (2012); Giannetti and Laeven (2012); Hale (2007, 2012); Laeven and Valencia (2013); Minoiu and Reyes (2013); and many others listed in the reference section.

### Data sources and variable definitions (Table 2)
- Syndicated loan exposures (SLE): Computed from loan-level data on syndicated loan deals using information on deal amount and maturity. Expressed in constant terms using US CPI. Sources: Authors' calculations using Dealogic Loan Analytics and IMF's INS database for US CPI.
- BIS international banking statistics: Total claims of banking systems of BIS reporting countries vis-à-vis residents in other countries, aggregated on a locational or consolidated basis. Expressed in constant terms using US CPI. Source: BIS locational and consolidated banking statistics and IMF's INS database for US CPI.
- Non-syndicated loan exposures (non-SLE): Computed as the difference between total loan claims of banking systems of BIS reporting countries (locational basis) and syndicated loan exposures. Source: Authors' calculations.
- Within-pair and country characteristic variables: Log-real (bilateral) trade (UN-COMTRADE); common language and geographical distance (Mayer and Zignago (2011)); foreign affiliate indicator (BIS consolidated statistics); bank assets (% GDP), bank capital-to-assets ratio, % internationally-rated banks, stock market capitalization (% GDP), and other financial variables (Global Financial Development Database (World Bank, 2013), Bankscope, Penn World Tables 8.0, IFS, Polity IV, Standard & Poor's, Datastream).

### Key descriptive statistics (Table 3)
- Sample observations: 9,213 for many cross-border exposure and within-pair variables.
- BIS claims (total): Obs. 9,213; Mean 4331.40; St. Dev. 4331.40; Min 0.92; P25 68.50; P50 762.88; P75 3081.97; Max 149440.35.
- BIS claims (loans): Obs. 9,213; Mean 17088.12; St. Dev. 17088.12; Min 0.90; P25 346.30; P50 1913.31; P75 96909.00; Max 1035620.00.
- SLE (total, unadjusted): Obs. 9,213; Mean 12678.98; St. Dev. 12678.98; Min 0.90; P25 250.88; P50 1275.82; P75 6333.66; Max 1032253.90.
- SLE (total, adjusted): Obs. 9,213; Mean 1134.77; St. Dev. 1134.77; Min 0.03; P25 17.12; P50 102.97; P75 628.61; Max 60363.54.
- Log-SLE: Obs. 9,213; Mean 3.44; St. Dev. 1.93; Min 0.08; P25 1.95; P50 3.14; P75 4.76; Max 9.58.
- Log-non-SLE: Obs. 9,213; Mean 5.86; St. Dev. 2.41; Min 0.00; P25 4.04; P50 5.92; P75 7.55; Max 12.41.
- Borrower institutional quality: Obs. 9,213; Mean 5.92; St. Dev. 6.11; Min -10.00; P25 6.00; P50 9.00; P75 10.00; Max 10.00.
- Capital account openness (borrower): Obs. 9,213; Mean 77.96; St. Dev. 25.48; Min 25.00; P25 50.00; P50 87.50; P75 100.00; Max 100.00.

### Main regression findings — baseline (Table 4)
- Dependent variables: log-SLE (columns 1–3) and log-non-SLE (columns 4–6). Sample period: 1995-2012. All regressions include country and year fixed effects. Standard errors clustered on country pair.
- Within-pair variables:
  - Log-real trade: Positive and significant for both SLE and non-SLE. Coefficients: 0.241*** (log-SLE, full sample), 0.256*** (log-non-SLE, full sample).
  - 1: Common language: Positive for log-SLE (0.305*** full sample) and mixed for log-non-SLE (0.161 full sample, not significant).
  - Log-geographical distance: Negative and significant for both SLE and non-SLE. Coefficients: -0.405*** (log-SLE, full sample), -0.777*** (log-non-SLE, full sample).
  - 1: Foreign affiliate: Positive and significant for both SLE and non-SLE. Coefficients: 0.525*** (log-SLE, full), 0.753*** (log-non-SLE, full).
- Lender characteristics:
  - Bank capital-to-assets ratio: Negative and significant for log-SLE (coefficient -0.049*** full sample).
  - Log-total other lending: Strong positive for log-SLE (0.617*** full sample) and small or mixed for log-non-SLE (0.034 full sample).
- Borrower characteristics:
  - Bank assets (% GDP): Positive and significant across SLE and non-SLE. Coefficients: 0.002** (log-SLE, full), 0.006*** (log-non-SLE, full).
  - Capital account openness: Positive and significant for log-non-SLE (0.009*** full) and mixed for log-SLE (0.002 full).

### Interactions with the global financial crisis (GFC) (Table 5)
- Sample period: 1995-2012; interactions reported for GFC effects (GFC dummy effect absorbed in year fixed effects).
- Key interaction effects (log-SLE, full sample):
  - Log-real trade*GFC: 0.081*** (positive, significant).
  - Log-geographical distance*GFC: -0.084*** (negative, significant).
  - Capital account openness*GFC: 0.005*** (positive, significant).
  - Institutional quality*GFC: 0.014 (not always significant across specifications).
- For log-non-SLE (full sample), interaction effects include:
  - Log-real trade*GFC: 0.062*** (positive, significant).
  - Capital account openness*GFC: 0.003 (positive, significance varies).
  - Institutional quality*GFC: 0.016* in some specifications.

### Robustness and alternative specifications (Tables 6–8)
- Country-pair fixed effects (Table 6): Strong R-squared values (e.g., 0.938 for log-SLE full sample), while many substantive signs (e.g., negative bank capital-to-assets ratio coefficients) persist.
- Additional lender characteristics (Table 7): Inclusion of net interest margin, liquid assets, return on assets, loan loss reserves, and net charge-offs; net charge-offs show a positive coefficient 0.0603** in one specification for log-SLE.
- Additional borrower characteristics (Table 8): Inclusion of exchange rate volatility, stock market measures, S&P sovereign rating, and % internationally rated banks; stock market return volatility shows negative and significant coefficients (e.g., -0.0002*** in some log-SLE specifications). % internationally rated banks is positive and significant in some specifications (e.g., 0.0033*** for log-non-SLE in column subsets).

### Figures and tables (captions and notes)
- Figure 1: Syndicated vs. total loan exposures (trillions of U.S. dollars). Notes: Unadjusted SLEs vs. BIS loan claims during 1995–2012; figures in trillions of constant U.S. dollars (2005 prices). Sources: Dealogic Loan Analytics, BIS locational banking statistics (Table 7A).
- Figure 2 (panels A–C): Syndicated vs. non-syndicated loan exposures (trillions of U.S. dollars) for all borrowers, advanced economy borrowers, and emerging market borrowers. Notes: Adjusted SLEs vs. non-SLEs during 1995–2012; figures in trillions of constant U.S. dollars (2005 prices). Sources: Dealogic Loan Analytics, BIS locational banking statistics, and authors’ calculations.
- Table 1: Lists 26 lender countries and 76 borrower countries. Countries classified per IMF World Economic Outlook, September 2013.
- Tables 3–8: Provide descriptive statistics, baseline regressions, crisis interactions, country-pair effects, and alternative specifications. Sample period consistently 1995-2012 across regressions; standard errors clustered on country pair; most regressions include year and country fixed effects.

### Appendix (I. DATA SOURCES AND DESCRIPTION)
- Syndicated loan data: Dealogic’s Loan Analytics; downloaded data for 153,255 syndicated loan deals signed during the 1990-2012 period to compute cross-border exposures for 1995- (text truncated in source).

*Source: _wp14185 - REFERENCES (IMF PDF)._

### 2012. For each loan, the database offers detailed information on contractual characteristics including

### _wp14185 - 2012. For each loan, the database offers detailed information on contractual characteristics including

### Data coverage and loan-level variables
- Database fields include lender and borrower identity, loan type (credit line, term loan), size, maturity, interest rate, and currency.
- Tranche-level variable “tranche instrument type” indicates whether syndicated-loan tranches are credit lines (CL) or upfront term loans (TL).
- When multiple tranche classifications exist, the first classification is prioritized in labeling a tranche as CL or TL.
- Tranches labeled “multiple facility” are classified as CLs because they include revolvers and may not be fully drawn.
- Other facility labels classified as CLs: Bridge Facility, Credit Facility, L/C Facility, Reducing Revolving Credit, and Revolving Credit.
- Main facility types classified as TL: “Term Loans” (A, B, C, D, etc.).
- When in doubt, classification was conservative and leaned toward classifying more tranches as CLs.
- For further processing details, refer to the appendix of Hale, Kapan and Minoiu (2014).

### Imputation of missing lender shares and validation exercise
- Challenge: splitting loan deal amounts across lenders by pro-rating should not create systematic biases in estimated country-pair loan volumes.
- Validation exercise sample: loans issued during 2006-2010 for which individual loan shares are observed.
- Three imputation methods compared:
  - (i) Pro-rata: splitting loan amounts equally across syndicate members (“pro-rata”).
  - (ii) Regression: predicting shares using a regression of log-shares on loan characteristics (“regression”), following de Haas and van Horen (2013) and Kapan and Minoiu (2013).
  - (iii) Average-to-lead: assigning larger shares to lead banks (30 percent to mandated arrangers and 40 percent to bookrunners) (“average-to-lead”), approximating sample averages for these roles.
- Procedure: pool 2006-2010 data for all lenders, sum bank-specific loan volumes at the country-pair level, and compare imputed volumes to reported shares.

### Imputation performance (Table A1 summary)
- Table A1 reports cumulative share of lending by top country-pairs and the ratio of imputed to actual loan origination volume for each method.
- Key values exactly as reported:
  - top 100 country pairs — cumulative share: 41.6% — (i) Pro-rata: 0.98 — (ii) Regression: 0.99 — (iii) Average-to-lead: 0.98
  - top 200 country pairs — cumulative share: 54.3% — (i) Pro-rata: 0.99 — (ii) Regression: 0.99 — (iii) Average-to-lead: 0.99
  - top 300 country pairs — cumulative share: 62.9% — (i) Pro-rata: 1.00 — (ii) Regression: 1.01 — (iii) Average-to-lead: 0.99
  - top 400 country pairs — cumulative share: 69.0% — (i) Pro-rata: 1.00 — (ii) Regression: 1.00 — (iii) Average-to-lead: 0.98
  - top 500 country pairs — cumulative share: 73.8% — (i) Pro-rata: 1.00 — (ii) Regression: 1.00 — (iii) Average-to-lead: 0.98
  - top 1000 country pairs — cumulative share: 94.2% — (i) Pro-rata: 1.06 — (ii) Regression: 1.04 — (iii) Average-to-lead: 1.03
  - Median (across country-pair cutoffs): (i) Pro-rata: 1.01 — (ii) Regression: 1.01 — (iii) Average-to-lead: 1.00
- Interpretation:
  - For the top 100 country pairs (41.6% of global issuance) the pro-rata imputation produces an average ratio of 0.98 relative to actual origination volume.
  - Ratios are close to 1 for country pairs accounting up to 73.8% of global origination, and rise to 1.06 for country pairs that account for 94.2%.
  - Biases become larger for country-pairs contributing small amounts to loan origination; this positive-bias pattern is common across imputation methods.
  - Conclusion: errors from pro-rata imputation at the country-pair aggregation level are not systematically worse than alternatives and are within acceptable bounds for the analysis.

### Structure and evolution of the syndicated loan market (1995–2012)
- The appendix provides charts (Figure A1) on:
  - borrower composition,
  - currency composition,
  - loan terms for AE vs. EME borrowers,
  - cross-border vs. domestic loans.
- Cross-border loans definition: at least one syndicate member is a foreign bank (nationality different from the borrower’s nationality) (Gadanecz and von Kleist, 2002).
- Notes about panels (exact text preserved):
  - Panels E and H: deal prices are average spreads (in basis points) over LIBOR for 5-year loans.
  - Panel F, G, and H: cross-border loans are loans for which at least one lender does not have the same nationality as the borrower.
  - Panel I: Gini coefficient for distribution of total deal amounts by individual borrowers.
  - Panel J: market share of largest 100 borrowers; both concentration measures computed at the borrower parent level.
  - Panel K: “lead banks” refer to bookrunners and mandated arrangers.
  - Panel L: “foreign borrowers” are borrowers whose nationality is different from their parent’s nationality.
- Source: Dealogic Loan Analytics.

### Adjustments to make Loan Analytics SLEs comparable to BIS loan claims
- Four adjustments applied to syndicated loan data:
  - (i) Aggregate syndicated loan data following same criteria used for CBS and LBS.
  - (ii) Compare the same type of exposures—loans with loans.
  - (iii) Ensure the same reporting group—only banks as lenders.
  - (iv) Adjust for the fact that a large part of loan commitments (credit lines) may not be drawn and therefore represent off-balance sheet exposures.

- A. Aggregation on consolidated and locational basis:
  - Construct SLEs at individual lender-borrower level using individual lender and borrower nationalities, lender parent nationality, loan amounts and maturities.
  - Aggregate into two bilateral SLE series: consolidated basis and locational basis.
  - Locational series used to examine relative importance of loan syndication market in total cross-border loan exposures.
  - Figure A2 compares BIS locational claims and estimated (unadjusted) SLEs aggregated at the lender-borrower country-pair level on a locational basis.
  - Figures expressed in trillions of constant U.S. dollars (2005 prices). Sources: Dealogic’s Loan Analytics, BIS locational banking statistics (Table 7A).

- B. Comparing like with like: loans with loans
  - BIS total claims include securities and other bank assets; to improve comparability focus on BIS component referring to bank loans only (BIS Table 6a).
  - This comparison appears in Figure 1 of the paper.

- C. Ensuring same reporting group: banks vs. non-banks
  - BIS IBS reports cross-border exposures of banks only; non-bank institutional lenders are excluded.
  - Non-bank participation in syndicated loans increased during the 2000s and inflates raw SLEs.
  - Adjustment approach: retain only syndicated loans reported as credit lines or “Term Loan A”-type loans, as these are extended almost exclusively by banks.
  - “Term Loan A” represents, on average, 8 percent of total term loan volume over 1995-2012.

- D. Estimating credit line usage rates (to remove off-balance sheet component)
  - Many syndicated credit lines are not fully drawn; undrawn amounts are off-balance-sheet commitments and not included in LBS.
  - Method: compare BIS undrawn portion of credit lines (from CBS for 18 reporting countries over 2005–2012) with syndicated credit lines on a consolidated basis.
  - Under the assumption most undrawn cross-border credit lines are syndicated, the ratio BIS undrawn credit lines / syndicated credit lines provides an estimate of (1 - utilization rate).
  - Calculated utilization rates on syndicated credit lines increased from 25.6 percent in 2005 to 52.1 percent in 2012 (Table A2).
  - For 1995–2004 and each borrower country, assume credit line usage rate equal to the 2005–2007 average for that country.
  - The uncovered trend is consistent with existing literature on credit line drawdowns and business-cycle sensitivity.

### Credit line usage rates (Table A2 — average credit line usage rates, 2005–2012)
- Notes: Average credit line usage rates for columns 1–4 obtained by comparing BIS off-balance sheet credit commitments and syndicated credit lines on a consolidated basis over 2005–2012. BIS off-balance sheet commitments available for 18 reporting countries. Column 5 uses Shared National Credits Program 2013 Review for US syndicated loans. Sources: Dealogic Loan Analytics, BIS consolidated banking statistics, and Shared National Credits Program.
- Reported annual values (rows exactly as in the source):
  - 2005: 25.65, 28.69, 27.96, 36.44, 32.08, 24.65
  - 2006: 28.12, 32.95, 27.73, 34.97, 33.40, 27.92
  - 2007: 25.14, 24.22, 23.21, 43.65, 36.70, 24.24
  - 2008: 39.62, 43.65, 37.30, 51.96, 43.31, 38.25
  - 2009: 40.20, 36.78, 47.35, 62.93, 54.25, 39.26
  - 2010: 50.46, 49.51, 50.00, 62.49, 48.03, 49.85
  - 2011: 57.40, 56.90, 50.61, 68.74, 44.29, 56.83
  - 2012: 52.16, 52.74, 52.40, 66.16, 44.52, 51.17
- Interpretation: utilization rates on syndicated credit lines increased over 2005–2012, consistent with broader evidence of drawdowns during tight credit conditions.

### Breaking down total cross-border loan exposures
- Total cross-border loan exposures comprise syndicated, bilateral, and intragroup loan exposures.
- Analysis of BIS locational statistics decomposes banks’ international positions into:
  - assets vis-à-vis related offices (indicative of intragroup lending),
  - assets vis-à-vis unrelated banks,
  - assets vis-à-vis non-banks,
  - assets vis-à-vis official monetary authorities.
- Finding: assets vis-à-vis related offices represented almost 30 percent of total loan exposures during 1995–2012.
- Under the assumption intragroup loans are rarely syndicated, bilateral cross-border loan exposures = BIS total loan claims − syndicated loan exposures − intragroup loan exposures.
- Figure A4 (1995–2012) shows syndicated loan exposures, bilateral loan exposures, and intragroup loan exposures each accounted for roughly one third of total cross-border bank loan claims (figures in trillions of constant U.S. dollars, 2005 prices).

### Additional empirical results and robustness
- Table B1: co-movement between (unadjusted) SLEs and BIS total loan claims (sample period 1995–2012). Key coefficient magnitudes (preserving reported values and significance):
  - Log-SLE-Total: columns report coefficients 0.897***, 0.901***, 0.653***, 0.652*** with standard errors (0.042), (0.042), (0.050), (0.055) respectively.
  - Log-SLE-Credit lines: 0.458***, 0.458***, 0.273***, 0.263*** with standard errors (0.031), (0.031), (0.022), (0.025).
  - Log-SLE-Term loans: 0.359***, 0.362***, 0.345***, 0.353*** with standard errors (0.045), (0.045), (0.033), (0.035).
  - Sample sizes and R-squared values are reported across specifications (e.g., Observations: 28,251; 19,323; R-squared values 0.494, 0.741, etc.) and split by AE and EME borrowers in panel rows.
- Figure B1: distribution of the share of adjusted SLEs in BIS cross-border bank loan claims (panels by lender, AE borrower, EME borrower, and year). Notes:
  - Figures show inter-quantile ranges with medians; ratio top-winsorized at 90th percentile.
  - Sample includes all country pairs with non-zero cross-border syndicated activity.
- Table B2: unconditional correlation matrices for selected regression variables (preserves reported correlation magnitudes and significance indicators in the source). Variables include Log-SLE, Log-real trade, common language indicator, log-geographical distance, foreign affiliate indicator, lender and borrower economic and banking characteristics, borrower institutional quality, and borrower capital account openness. (Correlation values and significance markers preserved as in the source tables.)

*Source: Dealogic Loan Analytics; BIS locational and consolidated banking statistics; Shared National Credits Program; appendix material from the provided content unit.*

### 1. Foreign

### _wp14185 - 1. Foreign

### Drivers of syndicated and non-syndicated loan exposures — AE borrowers (Table B3)
- Dependent variables: log-SLE (columns 1-6) and log-non-SLE (columns 7-12). Sample period: 1995-2012. Year and country fixed effects; standard errors clustered on country pair.
- Consistent positive and significant association of trade with syndicated exposure:
  - Log-real trade: 0.194*** (0.033) … 0.203*** (0.034) across columns for log-SLE; 0.150*** (0.036) … 0.159*** (0.038) across columns for log-non-SLE.
- Foreign affiliate effect:
  - 1: Foreign affiliate (log-SLE): 0.410*** (0.078) … 0.411*** (0.077).
  - 1: Foreign affiliate (log-non-SLE): 0.749*** (0.098) … 0.751*** (0.097).
- Distance effect:
  - Log-geographical distance (log-SLE): -0.525*** (0.078) … -0.553*** (0.079).
  - Log-geographical distance (log-non-SLE): -0.845*** (0.084) … -0.887*** (0.085).
- Lender bank capital-to-assets ratio:
  - Negative and significant for log-SLE: -0.050** (0.020) … -0.064*** (0.020).
  - Insignificant or small positive for log-non-SLE: 0.021 (0.023) … 0.017 (0.023).
- Log-total other lending:
  - Strong positive for log-SLE: 0.660*** (0.058) … 0.647*** (0.059).
  - Negative and significant for log-non-SLE: -0.124** (0.058) … -0.141** (0.062).
- Borrower characteristics:
  - Borrower bank assets (% GDP) positive for log-non-SLE: 0.004*** (0.001) … 0.004*** (0.001).
  - Institutional quality positive for log-non-SLE in some specifications: 0.161* (0.088) … 0.147* (0.087).
- GFC interaction highlights (selected):
  - Log-real trade*GFC: 0.119*** (0.026) … 0.158*** (0.033).
  - Geographical distance*GFC: -0.057 (0.038) … 0.182*** (0.055).
  - Bank capital-to-assets ratio*GFC: 0.028* (0.017) … 0.033* (0.020).

- Observations: 4,601 across columns; R-squared: 0.855 (log-SLE) and 0.764 (log-non-SLE).

### Drivers of syndicated and non-syndicated loan exposures — EME borrowers (Table B4)
- Same setup as Table B3 for EME borrowers.
- Log-real trade:
  - log-SLE: 0.223*** (0.071) … 0.231*** (0.070).
  - log-non-SLE: 0.436*** (0.071) … 0.436*** (0.069).
- Common language large positive effect for EME pairs:
  - 1: Common language (log-SLE): 0.542*** (0.186) … 0.542*** (0.187).
  - (log-non-SLE): 0.611*** (0.155) … 0.599*** (0.154).
- Foreign affiliate:
  - log-SLE: 0.466*** (0.097) … 0.458*** (0.096).
  - log-non-SLE: 0.581*** (0.089) … 0.553*** (0.088).
- Distance effect:
  - log-SLE: -0.510*** (0.136) … -0.503*** (0.141).
  - log-non-SLE: -0.853*** (0.121) … -0.762*** (0.122).
- Bank capital-to-assets ratio:
  - Negative and statistically significant across most columns for both log-SLE and log-non-SLE: e.g., -0.048** (0.022) … -0.037* (0.022); -0.050** (0.025) … -0.047* (0.026).
- Log-total other lending:
  - Positive for log-SLE: 0.594*** (0.038) … 0.572*** (0.037).
  - Positive but smaller for log-non-SLE: 0.180*** (0.046) … 0.107** (0.050).
- Borrower bank assets (% GDP) positive and significant:
  - log-SLE: 0.003 (0.002) … 0.004** (0.002).
  - log-non-SLE: 0.008*** (0.002) … 0.008*** (0.002).
- GFC interaction highlights (selected):
  - Geographical distance*GFC: -0.077 (0.062) … -0.285*** (0.060).
  - Capital account openness*GFC: 0.006** (0.002) … 0.007*** (0.002).
  - Institutional quality*GFC: 0.016 (0.012) … 0.030*** (0.011).

- Observations: 4,200 across columns; R-squared: ~0.748 (log-SLE) and ~0.717 (log-non-SLE).

### Robustness: Adding lagged dependent variable (Table B5)
- Dependent variables: log-SLE (columns 1-3) and log-non-SLE (columns 4-6). Sample period: 1995-2012. Estimators: OLS with country fixed effects, OLS with country-pair fixed effects, bias-corrected LSDV dynamic panel (Bruno, 2005).
- Lagged dependent variable:
  - Log-SLE: 0.780*** (0.012) [OLS within-country], 0.474*** (0.019) [within-pair], 0.545*** (0.007) [LSDV bias-correct].
  - Log-non-SLE: 0.808*** (0.011), 0.493*** (0.020), 0.584*** (0.016).
- 1: Foreign affiliate remains positive and significant in several specifications:
  - log-SLE: 0.141*** (0.021) … 0.0170 (0.014).
  - log-non-SLE: 0.156*** (0.024) … 0.070*** (0.015).
- Log-real trade remains positive and significant in OLS specifications:
  - log-SLE: 0.085*** (0.009) … 0.0540 (0.057).
  - log-non-SLE: 0.075*** (0.011) … 0.027 (0.066).
- Log-total other lending positive and significant for log-SLE: 0.302*** (0.021) … 0.476*** (0.016).

- Observations: 7,895 (or 8,072 in some columns); reported R-squared: 0.942, 0.960, etc. (see table).

### Robustness: Alternative sample period 1995-2007 (pre-global financial crisis) (Table B6)
- Excluded global financial crisis years (2008-2012). Year and country fixed effects; standard errors clustered on country pair.
- Log-real trade:
  - log-SLE full: 0.213*** (0.029); AE borrowers: 0.164*** (0.031); EME borrowers: 0.205*** (0.078).
  - log-non-SLE: 0.210*** (0.034); AE borrowers: 0.120*** (0.033); EME borrowers: 0.344*** (0.071).
- 1: Foreign affiliate:
  - log-SLE full: 0.472*** (0.067); AE borrowers: 0.393*** (0.084); EME borrowers: 0.359*** (0.112).
  - log-non-SLE full: 0.691*** (0.073); AE borrowers: 0.644*** (0.106); EME borrowers: 0.532*** (0.098).
- Bank capital-to-assets ratio negative for log-SLE: -0.061*** (0.017) … -0.045** (0.022).
- Log-total other lending positive for log-SLE: 0.570*** (0.036) … 0.607*** (0.058); but often small/negative for log-non-SLE in pre-GFC sample.

- Observations: 6,772 (full), 3,320 (AE borrowers), 3,040 (EME borrowers). R-squared ~0.812 (full log-SLE).

### Threshold effects of lender regulatory capital (Table B7)
- Explores lender bank regulatory capital (total regulatory capital to RWA) and threshold indicators for percentiles.
- Using bank capital-to-RWA (columns 1-3):
  - Bank capital-to-RWA: 0.039*** (0.012) in full sample; -0.041*** (0.013) for AE borrowers; -0.025 (0.018) for EME borrowers.
- Using percentile indicators (columns 4-6), omitted category: above 75th percentile. Indicators:
  - 1: Bank capital-to-RWA < P25: 0.199*** (0.059) full; 0.197*** (0.076) AE; 0.143 (0.090) EME.
  - P25 < Bank capital-to-RWA < P50: 0.116** (0.054) full; 0.118* (0.066) AE; 0.010 (0.080) EME.
  - P50 < Bank capital-to-RWA < P75: 0.055 (0.037) full; 0.056 (0.046) AE; 0.003 (0.054) EME.
- Remaining controls maintain expected signs: Log-real trade 0.241*** (0.032); 1: Foreign affiliate 0.524*** (0.062); Log-total other lending 0.618*** (0.031).
- Observations: 9,108 (full); 4,558 (AE); 4,150 (EME). R-squared ~0.812 (full).

### Probability of no within-pair syndication activity (Table B8)
- Dependent variable: probability that SLE=0 within a country pair. Sample period: 1995-2012. Columns 1-3: linear probability model with year and country fixed effects. Columns 4-6: probit with year dummies and random country effects. Standard errors clustered on country pair.
- Key results (linear probability model):
  - Log-real trade: -0.011** (0.005) full; -0.006 (0.004) AE; -0.017* (0.009) EME.
  - 1: Common language: -0.035** (0.016) full; -0.003 (0.008) AE; -0.041* (0.022) EME.
  - 1: Foreign affiliate: -0.016 (0.011) full; -0.020 (0.014) AE; 0.003 (0.014) EME.
  - Lender log-per capita GDP: -0.317*** (0.052) full; -0.334*** (0.061) AE; -0.361*** (0.070) EME.
  - Bank capital-to-assets ratio: 0.012*** (0.004) full; 0.018*** (0.005) AE; 0.015*** (0.005) EME.
  - Borrower bank assets (% GDP): -0.000*** (0.000) full; -0.000 (0.000) AE; -0.000 (0.000) EME.
- Probit specifications (magnitude interpretation differs; coefficients reported):
  - Log-real trade: -0.162** (0.077) full; -0.024 (0.116) AE; -0.367** (0.150) EME.
  - 1: Common language: -4.313 (7.401) full; -1.336*** (4.475) AE; -6.424*** (1.421) EME.
  - Bank capital-to-assets ratio: 0.674*** (0.148) full; 0.688*** (0.176) AE; 0.600*** (0.115) EME.
- Observations: 9,639 (full linear probability), 4,704 (AE), 4,466 (EME); R-squared (linear probability) 0.270 (full), 0.196 (AE), 0.461 (EME).

*Content derived from tables B3–B8 in the provided PDF chapter.*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2014/_wp14185.pdf_