## _wp1691

## Source details

**Canonical URL:** [_wp1691](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp1691.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp1691.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp1691.pdf.json)

---

### Purpose and research question
- Examine how systemic banking crises are transmitted through the “global banking network” constructed from interbank long-term lending.
- Analyze effects of direct (first-order) and indirect (second-order) exposures to banks in countries that experience systemic banking crises on bank profitability and lending decisions.

### Data sources and network construction
- Transaction-level Dealogic Loan Analytics data on 170,274 syndicated loan deals signed between 1990 and 2012; 16,526 loans are loans issued by banks to banks for a total of 6,083 distinct banks.
- Final unbalanced panel dataset: about 2,000 banks merged to Bankscope; regression sample contains 1,869 banks.
- Global banking network constructed annually for 6,083 banks; empirical analysis focuses on 1997-2012.
- Loan amounts expressed in U.S. dollars at 2005 prices using the U.S. consumer price index.
- Foreign interbank exposures measured primarily as counts (number of exposures) rather than dollar values.
- Definitions:
  - Direct exposures: number of banks to which bank i has direct exposures at time t (out-degree).
  - Indirect/second-degree exposures: number of banks to which bank i’s direct counterparties have direct exposures (two-step away).
  - Crisis exposures: exposures to banks in countries experiencing systemic banking crises in year t; non-crisis exposures: exposures to banks in countries not experiencing a crisis in year t.
- Network properties:
  - Network density ranges between 0.3 percent (1998) and 0.48 percent (2007).

### Market context and size
- Interbank loans account for about 10 percent of global syndicated loan volume, which peaked at 4.3 trillion U.S. dollars in 2007.
- The average loan in this market amounts to 500 million U.S. dollars and matures in 5 years.
- Based on BIS bilateral positions, interbank long-term lending estimated to account for 12.5 percent of total interbank loan claims.
- In the authors’ matched subsample:
  - Foreign interbank exposures represent 3.2 percent of total gross loans during 1997-2012.
  - Foreign interbank loans represent 5 percent of total liabilities and 8 percent of total liabilities less deposits.
  - Interbank funding share examples: 12.5 percent of non-deposit liabilities for banks in Turkey, 20 percent for banks in Iceland, and 40 percent for banks in Latvia.

### Empirical approach and identification
- Profitability analysis:
  - Outcomes: return on assets (ROA), return on equity (ROE), net interest margins (NIM).
  - Estimate bank-year panel over 1997-2012 with bank country*year fixed effects to account for time-varying country-level unobservables.
  - Estimation: OLS with bank country*year fixed effects and standard errors clustered at the bank level.
- Lending analysis:
  - Aggregate individual loan amounts at the bank-borrower-year level and regress on interbank crisis and non-crisis exposures.
  - Borrower*year fixed effects included to isolate loan supply effects from loan demand.
  - Bank country*year fixed effects included as in profitability regressions.
- Modeling choices:
  - Baseline structural framework expands Yi = Xi β + λ Ci + γ ∑j1 Eij1 Yj1 to an infinite series; empirical implementation includes only first- and second-degree exposures (γ assumed to decay exponentially).
  - Parsimonious specifications exclude counterparty bank characteristics (jointly statistically insignificant under most specifications); results robust to their inclusion.
- Endogeneity strategies:
  - Decompose exposures into “stock” (end of t−1) and “flow” (originated during t); statistically significant results mainly for stock exposures.
  - Remove largest banks (top 5 percent) and G-SIBs as robustness checks; results broadly unchanged.

### Main empirical findings — profitability
- Direct crisis exposures:
  - An increase of one direct crisis exposure (holding total number of exposures constant) reduces ROA by 0.03 percentage points (Table 4, Column 1).
  - For a bank leveraged 30 times, an ROA reduction of 0.03 percentage points becomes an ROE reduction of 0.9 percentage points, corresponding to almost 11 percent of a mean ROE of 8.4 percent.
  - For a bank with total assets of one trillion U.S. dollars, an additional crisis connection translates into a reduction in annual returns of 300 million U.S. dollars.
- Indirect (second-degree) crisis exposures:
  - Baseline direct crisis effect on ROA in conditional specification: -0.022.
  - Additional indirect exposure through a crisis-country bank to a crisis-country bank (C-C) further reduces ROA by 0.007 (32 percent of the base effect).
  - Additional indirect exposure through a crisis-country bank to a non-crisis-country bank (C-NC) mitigates the negative direct crisis effect by 0.005 (dampening equal to 23 percent of the base effect).
- Non-crisis exposures:
  - Adding a non-crisis exposure does not affect ROA in the baseline comparison.

### NIM channel and contribution to ROA declines
- Turning a non-crisis exposure into a crisis exposure reduces NIMs by about 20 basis points (Table 7, columns 1-4).
- Coefficient estimates for ROA effects lie between 0.02 and 0.03 (used in back-of-the-envelope calculations).
- The impact of direct crisis exposures on NIMs accounts for 41-56 percent of their impact on ROA (calculation uses a NIM coefficient of 0.02, ROA between 0.02 and 0.03, and an estimated ratio of interest bearing assets to total assets of 62 percent).

### Lending effects (loan supply)
- Ten additional direct crisis exposures reduce a bank’s supply of loans by 2.4 percent (coefficient -0.0024, column 1).
- Indirect crisis exposures further reduce the ability to extend new loans; indirect non-crisis exposures dampen lending less or mitigate effects.
- A larger total number of non-crisis exposures is associated with an increase in loan supply (positive and significant coefficient on total number of exposures).
- Domestic vs cross-border:
  - Direct exposures are harmful for both domestic and cross-border loans.
  - Indirect exposures have statistically insignificant effects on domestic loans (p-values for F-tests that coefficients on indirect exposures C-C and C-NC are jointly insignificant in column 4 are 0.3095 and 0.3671).
  - Cross-border lending is more sensitive to higher-degree exposures, consistent with a “flight-home” tendency.

### Interpretation of exposures and transmission channels
- Direct interbank exposures capture both “idiosyncratic risk” (credit risk to foreign bank counterparties) and “country risk” (risks associated with exposure to foreign markets more generally).
- Given the crisis exposure definition (country-wide systemic banking crisis) and the rarity of bank defaults, the documented negative effect is largely driven by the “country risk” component.
- Plausible mechanisms for profit declines:
  - Loan restructuring in severely impaired syndicated loans reduces lenders’ present-value cash flows, lowering NIMs.
  - Information contagion and reputational effects reduce future business and increase funding costs.
  - Marked-to-market losses on syndicated loans placed in the securities book are a potential channel (data limitations prevent direct testing).
- Syndicated loan market statistics (Standard & Poor’s, 2011):
  - During 2011-2012, loan default rates were 2 percent.
  - Five-year default rates: AAA-rated firms 0.38 percent; B-rated firms 21.76 percent during 1981-2010.
  - Loan recovery rates have been 71 percent compared to 43.5 percent for unsecured lending during 1989-2009.

### Heterogeneity and robustness
- Large banks:
  - According to 2014 Bankscope data, there are 30 banks with total assets of at least 1 trillion U.S. dollars.
  - No differential effect of direct crisis exposures for large banks; indirect crisis exposures have a lower effect on larger balance sheets.
- Robustness checks:
  - “Leave-one-year-out” regressions for bank ROA show confidence bounds for key coefficients never cross zero; results robust to removing clustered-crisis years (e.g., 2001-2002, 2008-2009).
  - Profitability results not driven by countries with few banks or banks with few observations.
  - Lending regressions for the subsample of non-financial borrowers yield the same broad findings.

### Correlation with total foreign exposures (U.S. banks)
- FFIEC009a used for total foreign exposures for 1997-2012; sample includes 214 individual U.S. banks vis-à-vis 183 destination countries; 114 banks matched with interbank exposures.
- Regression findings:
  - Estimated elasticity of total cross-border exposures with respect to foreign interbank exposures measured in dollars is 0.09 percent (Table 3, column 1).
  - Semi-elasticity of total exposures with respect to the number of bank borrowers is 32 percent (Table 3, column 2).
- Interpretation: long-term interbank exposures are a useful proxy for total exposures to foreign markets and more strongly correlated with total exposures than exposures to non-banks.

### Key sample counts and descriptive statistics (selected)
- Dealogic Loan Analytics deals signed between January 1990 and December 2012: 170,274; retained for foreign interbank network: 16,526.
- Final regression sample: 1,869 banks.
- Table 3: Observations 8,397; R-squared 0.662 (column 1) and 0.657 (column 2).
- Table 4 (ROA): Observations 14,448; R-squared 0.440–0.442 across columns; number of banks 1,869.
- Table 5 (ROE): Observations 14,445; R-squared 0.345–0.347; number of banks 1,869.
- Table 7 (NIM): Observations 14,135; 12,415; 10,771; 10,637 across columns 1–4; R-squared 0.659; 0.664; 0.664; 0.698; number of banks 1,849; 1,740; 1,527; 1,519 respectively.
- Table 8 (Loan supply): Observations columns 1–4: 167,958; columns 5–8: 295,204; R-squared 0.889–0.906 across specifications.
- Descriptive statistics (Table 2, selected):
  - ROA: N = 14,448; Mean = 0.809; St. Dev. = 1.560; Min = -6.850; Max = 8.850.
  - ROE: N = 14,445; Mean = 8.398; St. Dev. = 16.44; Min = -78.09; Max = 53.17.
  - NIM: N = 14,315; Mean = 2.759; St. Dev. = 2.238; Min = -0.910; Max = 15.87.
  - Equity/Assets: N = 14,448; Mean = 8.753; St. Dev. = 9.333; Min = 0.320; Max = 81.51.
  - Assets (USD bn): N = 14,448; Mean = 72.342; St. Dev. = 236.40; Min = 0.45; Max = 3,808.
  - Banks: # indirect exposures: N = 14,448; Mean = 15.626; St. Dev. = 8.190; Min = 1; Max = 1,981.
  - Systemic banking crisis indicator: N = 14,448; Mean = 0.209; St. Dev. = 0.407; Min = 0; Max = 1.
- Estimation conventions:
  - All bank balance sheet variables winsorized at the 1st and 99th percentiles.
  - Standard errors clustered at the bank level.
  - Significance indicators: *** 1% level, ** 5% level, * 10% level.
  - Fixed effects typically include bank nationality*year and borrower*year or bank*destination country FE where indicated.

### Summary conclusions
- A larger number of direct loan exposures to bank borrowers in countries experiencing systemic banking crises reduces bank returns and the granting of large corporate loans, controlling for time-varying borrower demand.
- Indirect, second-degree exposures to borrowers in crisis countries have additional negative impacts, while indirect exposures to borrowers in non-crisis countries can mitigate negative effects.
- Loan restructuring and information contagion plausibly squeeze net interest margins and prompt banks to reduce loan supply, especially cross-border lending.
- Long-term interbank market interactions enable international transmission of financial-sector shocks and interbank exposures provide useful signals for monitoring vulnerabilities to foreign-market stress.

*Source: _wp1691 (IMF working paper, Introduction, Shock Transmission Mechanism, Results, and References excerpts).*

### 1. Introduction  3

### _wp1691 - 1. Introduction  3

### Purpose and research question
- Examine how systemic banking crises are transmitted through the “global banking network” constructed from interbank long-term lending.
- Analyze effects of direct (first-order) and indirect (second-order) exposures to banks in countries that experience systemic banking crises on bank profitability and lending decisions.

### Data sources and construction
- Exploit transaction-level data from the interbank long-term lending market to compute time-varying exposures for more than 6,000 banks.
- Draw on information for more than 170,000 loans extended during 1990-2012 to borrowers in more than 200 countries (Dealogic Loan Analytics).
- Of these, 16,526 are loans issued by banks to banks for a total of 6,083 distinct banks.
- Two thirds of interbank loans and three quarters of all loans are syndicated; the rest are single-lender loans.
- Use loan origination, signing dates, and maturity dates to compute bilateral bank-borrower exposures and construct measures of direct (one step away) and indirect (two steps away) exposures to banks in countries experiencing systemic banking crises (“crisis exposures”); exposures to banks in countries not experiencing a crisis are “non-crisis exposures.”

### Market context and size
- Interbank loans account for about 10 percent of global syndicated loan volume, which peaked at 4.3 trillion U.S. dollars in 2007.
- The average loan extended in this market amounts to 500 million U.S. dollars and matures in 5 years.
- Based on BIS bilateral positions, interbank long-term lending is estimated to account for 12.5 percent of total interbank loan claims.
- In the authors’ matched subsample:
  - Foreign interbank exposures represent 3.2 percent of total gross loans during 1997-2012 (with substantial cross-country variation).
  - Foreign interbank loans represent 5 percent of total liabilities and 8 percent of total liabilities less deposits.
  - Interbank funding is relatively more important for some emerging market banks: 12.5 percent of non-deposit liabilities for banks in Turkey, 20 percent for banks in Iceland, and 40 percent for banks in Latvia.

### Empirical approach
- Profitability analysis:
  - Use return on assets (ROA), return on equity (ROE), and net interest margins (NIM).
  - Estimate bank-year panel over 1997-2012 with bank country*year fixed effects to account for time-varying country-level unobservables.
- Lending analysis:
  - Aggregate individual loan amounts at the bank-borrower-year level and regress on interbank crisis and non-crisis exposures.
  - Isolate loan supply effects from loan demand by exploiting multiple bank relationships and adding borrower*year fixed effects.
  - Control for bank country*year fixed effects as in profitability regressions.

### Main empirical findings
- Crisis exposures are associated with lower bank profitability and lower supply of new loans.
- Profitability effects:
  - Holding the total number of (direct and indirect) exposures constant, an additional direct exposure to a bank in a crisis country reduces bank ROA by 0.03 percentage points in the same year.
  - This effect is 32 percent larger for an additional indirect crisis exposure through banks in a crisis country.
  - An additional indirect exposure to banks in non-crisis countries through a crisis country mitigates the negative effect of the direct crisis exposure by 23 percentage points.
- NIM channel:
  - NIMs are lower by 23 basis points for each additional direct crisis exposure.
  - A back-of-the-envelope calculation indicates the impact of direct crisis exposures on NIMs accounts for about half of that impact on ROA.
- Lending effects:
  - Holding the total number of exposures constant, turning ten non-crisis direct exposures into crisis exposures reduces a bank’s supply of loans by 2.4 percent.
  - The reduction in loan supply is larger for banks with second-degree crisis exposures on top of first-degree crisis exposures, and smaller for banks with second-degree non-crisis exposures.
  - The loan supply impact of second-degree crisis exposures is stronger for cross-border loans than for domestic loans (effects on domestic loans are statistically indistinguishable from zero).

### Interpretation of exposures and channels
- Direct interbank exposures capture both:
  - “Idiosyncratic risk” (credit risk to foreign bank counterparties), and
  - “Country risk” (risks associated with exposure to foreign markets more generally).
- Given the crisis exposure definition (country-wide systemic banking crisis) and the rarity of bank defaults, the documented negative effect is largely driven by the “country risk” component.
- One mechanism for the negative performance impact is lower profit margins linked to loan restructuring in severely impaired syndicated loans; this is consistent with the observed NIM compression.

### Contribution to literature
- Provide empirical evidence on cross-border propagation of financial sector shocks using transaction-level syndicated interbank loan data rather than simulations.
- Expand the analysis beyond first-order (direct) effects to include second-order (indirect) effects on bank performance and credit supply.
- Document that interbank long-term loan exposures are strongly correlated, for U.S. banks, with total bank-level exposures aggregated at the destination country level; in the dataset interbank exposures are more strongly correlated with total exposures than are exposures to firms and sovereigns, suggesting usefulness for real-time monitoring of vulnerabilities to foreign-market stress.

*Source: _wp1691 - 1. Introduction  3 (IMF working paper, Introduction section).*

### 3.1    Shock Transmission Mechanism

### _wp1691 - 3.1    Shock Transmission Mechanism

### Shock transmission framework
- Bank performance is measured by Y (bank i: Yi).
- Exposure indicator: Eij1 denotes presence of an exposure from bank i to bank j (E is an indicator).
- Crisis indicator: Ci denotes an indicator for a financial crisis in the country of bank i.
- Bank characteristics: Xi is the (1×K) matrix of bank i’s K characteristics.
- Baseline hypothesis for bank returns (omitting time subscript):
  - Yi = Xi β + λ Ci + γ ∑j1 Eij1 Yj1. (Equation (1))
- Expansion of (1) yields an infinite series showing direct and indirect (higher-degree) transmission of shocks:
  - Yi = Xi β + λ Ci + γ ∑j1 Eij1 Xj1 β + γ ∑j1 Eij1 λ Cj1 + γ^2 ∑j2 Eij1 Ej1j2 Xj2 β + γ^2 ∑j2 Eij1 Ej1j2 λ Cj2 + ... + γ^n ∑jn Eij1 Ej1j2 ... Ejn−1jn Xjn β + γ^n ∑jn Eij1 Ej1j2 ... Ejn−1jn λ Cjn. (Equation (2))
- Interpretations:
  - j1: first-degree (direct) connections of bank i.
  - j2: second-degree (indirect) connections of bank i, etc.
  - The union of first-degree connections of all j1 banks corresponds to bank i’s second-degree connections.
  - Equation (2) describes how bank i’s performance depends on its direct and indirect exposures to borrowers in countries experiencing banking crises.

### Empirical specification and estimation strategy
- Complete empirical specification links bank performance to:
  - bank-specific controls (Xiht),
  - indicator for bank located in a country experiencing a banking crisis (Ciht),
  - first-, second-, and higher-degree exposures and characteristics of counterparty banks.
- Practical modeling choices:
  - γ decays exponentially, so higher-degree connections have drastically reduced impact; empirical implementation includes only first and second-degree exposures.
  - Bank country*year fixed effects (αht) are included to absorb home-country time-varying shocks.
- Full panel specification with time and bank-country subscripts (parsimonious forms emphasized):
  - Profitability specification (parsimonious, estimated by OLS with bank country*year fixed effects and standard errors clustered at the bank level):
    - Yiht = αht + X_iht β0 + λ0 C_iht + λ1 ∑j1 Eij1t Cj1t + λ2 ∑j2 Eij1t Ej1j2t Cj2t + εiht. (Equation (4))
  - Lending specification for loan volumes (log-transformed) with borrower*year fixed effects to isolate credit-supply effects:
    - Lihjt = αht + Xiht β0 + λ0 Ciht + λ1 ∑j1 Eij1t Cj1t + λ2 ∑j2 Eij1t Ej1j2t Cj2t + γjt + εihjt. (Equation (5))
- Counterparty bank characteristics:
  - When estimating Equation (3) initially, counterparty bank characteristics were jointly statistically insignificant under most specifications.
  - Parsimonious specifications therefore exclude these characteristics; regression results are robust to their inclusion.
- Controls:
  - Exposures to non-bank borrowers (non-financial firms and sovereigns) and bank size are included as components of Xiht.
- Estimation method:
  - OLS with bank country*year fixed effects and standard errors clustered at the bank level for profitability regressions.
  - Lending regressions also estimated by OLS with standard errors clustered at the bank level.

### Hypotheses on effects of crisis exposures
- Coefficients of interest:
  - λ1 and λ2 represent impacts of first- and second-degree crisis connections on bank profitability and loan supply (from Equations (4) and (5)).
- Expected signs and channels:
  - Negative shocks via foreign interbank exposures are expected to reduce bank earnings (lower net income and returns).
    - Direct channels: valuation effects and write-downs on non-performing exposures.
    - Indirect channels: loss of other business.
  - Bank lending is expected to be negatively affected as shocks erode capital via write-offs and lower earnings, or raise cost of funds.
  - In a networked financial system, shocks can cascade through higher-order exposures; indirect spillovers are possible even when direct exposures do not experience crises.

### Data and construction of foreign interbank exposures
- Data sources and sample construction:
  - Individual loan deals: Dealogic’s Loan Analytics, syndicated and single-lender loans since the early 1980s.
  - For 1997-2012 period, information on 170,274 syndicated loan deals signed between 1990 and 2012 used to construct interbank exposures.
  - Loan amounts expressed in U.S. dollars at 2005 prices using the U.S. consumer price index.
  - Global banking network constructed annually for 6,083 banks (details in Data Appendix).
- Measurement caveats and choices:
  - Only loans at origination observed; no data on drawdowns, liquidation, prepayments, side-arrangements, or loan sales.
  - To limit measurement error, empirical analysis uses the number of exposures (counts) rather than dollar values.
- Definitions:
  - Direct exposures: number of banks to which bank i has direct exposures at time t (out-degree).
  - Indirect/second-degree exposures: number of banks to which bank i’s direct counterparties have direct exposures (two-step away).
  - Crisis and non-crisis exposures: sums of exposures to banks in countries experiencing crises in year t versus other countries.
  - Exposures to non-banks constructed from Dealogic loans to non-financial borrowers; only direct exposures for non-banks computed and used as control variables.
- Network properties:
  - Network density (observed connections divided by possible connections) ranges between 0.3 percent (1998) and 0.48 percent (2007).
  - The global network is quite sparse; density comparable to domestic interbank markets.
- Supplementary data:
  - Bank balance sheet information from Bankscope; name/nationality matching between Dealogic and Bankscope with manual inspection and adjustment.
  - Final unbalanced panel dataset comprises about 2,000 banks; regression sample contains 1,869 banks (missing balance sheet data for some banks).
  - Main outcome variable: bank ROA; alternatives considered: ROE and NIMs.
  - Control variables: bank capital (equity/assets), size (log-total assets), exposures to non-banks, indicators for bank type, and indicators for bank business model.
  - Data on systemic banking crises from Laeven & Valencia (2013); systemic banking crises defined by significant domestic banking stress plus at least three of six listed public interventions.
  - Loan origination data used to construct bank-borrower-year loan volumes from Dealogic’s Loan Analytics.
- Correlation with total foreign exposures (U.S. banks exercise):
  - FFIEC009a used for total foreign exposures (loans, securities, derivatives, other claims) for 1997-2012.
  - Sample: 214 individual U.S. banks vis-à-vis 183 destination countries; matched 114 banks with interbank exposures.
  - Regressions of total cross-border exposures on syndicated loan exposures (dollars and counts), controlling for bank*destination country and year fixed effects.
  - Findings (reported in Table 3):
    - Estimated elasticity of total cross-border exposures with respect to foreign interbank exposures measured in dollars is 0.09 percent (column 1).
    - Semi-elasticity of total exposures with respect to the number of bank borrowers is 32 percent (column 2).
  - Interpretation:
    - Long-term interbank exposures are a good proxy for total exposures to foreign markets, even after controlling for non-bank exposures.
    - Exposures to banks are more strongly correlated with total exposures than exposures to non-banks, suggesting greater informativeness about total foreign activity.

### Key empirical findings on profitability (summary of Results 5.1.1)
- Baseline profitability regressions use crisis exposures measured as counts; home country effects absorbed by bank country*year fixed effects.
- Main reported effect (Table 4, Column 1):
  - An increase of one direct crisis exposure (keeping total number of connections constant) reduces ROA by 0.03 percentage points.
  - Economic magnitudes:
    - For a bank leveraged 30 times, an ROA reduction of 0.03 percentage points becomes an ROE reduction of 0.9 percentage points, corresponding to almost 11 percent of a mean ROE of 8.4 percent.
    - For a bank with total assets of one trillion U.S. dollars, an additional crisis connection translates into a reduction in annual returns of 300 million U.S. dollars.
- Indirect (second-degree) exposures (Columns 2–3, Table 4):
  - Column 2: coefficients on second-degree exposure variables (counts of crisis and non-crisis exposures of first-degree counterparties) are not statistically significant.
  - Column 3 (conditional paths through a first-degree crisis exposure):
    - Baseline direct crisis effect on ROA: -0.022 (in this specification).
    - Additional indirect exposure through a crisis-country bank to a crisis-country bank (C-C) further reduces ROA by 0.007 (32 percent of the base effect).
    - Additional indirect exposure through a crisis-country bank to a non-crisis-country bank (C-NC) reduces the negative effect on ROA by 0.005 (dampening equal to 23 percent of the base effect).
  - Preferred baseline specification used in subsequent analysis is the model in Column 3 (conditions on first-degree crisis exposure and distinguishes C-C and C-NC second-degree paths).

*Source: _wp1691 - 3.1    Shock Transmission Mechanism*

### 0.03 percentage points.  By contrast, adding a non-crisis exposure does not affect ROA. A possible explanation for

### _wp1691 - 0.03 percentage points.  By contrast, adding a non-crisis exposure does not affect ROA. A possible explanation for

### Key findings on bank profitability
- Turning a non-crisis exposure into a crisis exposure reduces ROE by 0.3 percentage points (Table 5, column 1).
- Turning a non-crisis exposure into a crisis exposure reduces NIMs by about 20 basis points (Table 7, columns 1-4).
- Coefficient estimates for ROA effects lie between 0.02 and 0.03 (used in a back-of-the-envelope calculation).
- The impact of direct crisis exposures on NIMs accounts for 41-56 percent of their impact on ROA (calculation using NIM coefficient 0.02, ROA between 0.02 and 0.03, and an estimated ratio of interest bearing assets to total assets of 62 percent).
- Indirect, second-degree crisis exposures have an additional negative impact on bank returns; adding a non-crisis exposure does not affect ROA.
- The magnitudes are economically meaningful given large and highly leveraged bank balance sheets and the potential for correlated exposures across banks to amplify shocks.

### Heterogeneity and bank size
- According to 2014 Bankscope data, there are 30 banks with total assets of at least 1 trillion U.S. dollars.
- Interactions of first- and second-degree exposures with log(total assets) reveal:
  - No differential effect of direct crisis exposures for large banks.
  - Indirect crisis exposures have a lower effect on larger balance sheets.
- In unreported specifications, no differential effects were found by bank business model, entity type, or bank capital.

### Tackling endogeneity
- Endogeneity sources considered: banks reducing exposures after past or anticipated shocks; endogenous network formation to mitigate risk; use of credit portfolio management tools (hedging, loan sales, securitization).
- Two empirical strategies:
  - Decompose interbank exposure at time t into a “stock” exposure (in place as of the end of t−1) and a “flow” exposure (loans originated during t). Stock exposures are less likely to be endogenously adjusted.
    - Splitting exposures yields statistically significant results only for the stock exposure (Table 6, column 1).
  - Exploit differences in bank size and business model by removing the largest banks (top 5 percent of size distribution) and removing G-SIBs (columns 2-3 of Table 6).
    - Removing these banks leaves results broadly unchanged.

### Potential mechanisms for profit declines
- Direct balance sheet losses from individual borrower defaults in syndicated loan markets are less frequent but loan restructuring is common and reduces lenders’ present-value cash flows, lowering NIMs.
- Syndicated loan market default/recovery statistics (Standard & Poor’s, 2011):
  - During 2011-2012, loan default rates were 2 percent.
  - Over five years, default rate for firms rated AAA was 0.38 percent while that for firms rated B was 21.76 percent during 1981-2010.
  - Loan recovery rates have been 71 percent compared to 43.5 percent for unsecured lending during 1989-2009.
- Information contagion and reputational effects can reduce future business and increase funding costs (evidence from Saunders et al., 2003; Gopalan et al., 2011).
- Another potential channel is marked-to-market losses on syndicated loans placed in the securities book, but lack of accounting designation data prevents direct testing.

### Effects on lending (loan supply)
- Data: loans extended by banks during 1997-2012; aggregated at the bank-borrower-year level with borrower*year fixed effects.
- Ten additional direct crisis exposures reduce the supply of loans by 2.4 percent (coefficient -0.0024, column 1).
- Indirect crisis exposures further reduce ability to extend new loans; indirect non-crisis exposures also dampen lending (column 2).
- A larger number of non-crisis exposures is associated with an increase in loan supply (positive and significant coefficient on total number of exposures).
- Differential effects for domestic versus cross-border loans:
  - Direct exposures are harmful for both domestic and cross-border loans.
  - Indirect exposures have statistically insignificant effects on domestic loans (p-values for F-tests that coefficients on indirect exposures C-C and C-NC are jointly insignificant in column 4 are 0.3095 and 0.3671).
  - Cross-border lending is more sensitive to higher-degree exposures, consistent with a “flight-home” tendency.

### Robustness
- “Leave-one-year-out” regressions for bank ROA show confidence bounds for key coefficients never cross zero (Figure A2), robust to removing years with clustered crises (e.g., 2001-2002, 2008-2009).
- Profitability results are not driven by countries with few banks or by banks with few observations (Table A3).
- Lending regressions for the subsample of non-financial borrowers yield the same broad findings as the full borrower sample (Table A4).

### Summary conclusions
- Using detailed long-term interbank loan data, a global banking network for 1997-2012 was constructed comprising more than 6,000 banks; financial statement data are available for a sample of 1,869 banks.
- A larger number of direct loan exposures to bank borrowers in countries experiencing systemic banking crises reduces bank returns and the granting of large corporate loans, controlling for time-varying borrower demand.
- Indirect, second-degree exposures to borrowers in crisis countries have additional negative impacts, while exposures to borrowers in non-crisis countries have a dampening effect.
- Loan restructuring and information contagion are plausible channels squeezing net interest margins and prompting banks to reduce loan supply, especially cross-border lending.
- Findings indicate banks are unable to fully shield balance sheets from foreign risk, illustrating how long-term interbank market interactions enable international shock transmission.

*Source: IMF working paper content (excerpt)._

### References

### _wp1691 - References

### References (bibliographic list)
- Full bibliographic reference list provided (authors, titles, journals, volumes, pages) including, among others:
  - Acemoglu, D., Ozdaglar, A., & Tahbaz-Salehi, A. (2015). Systemic risk and stability in financial networks. American Economic Review, 105(2), 564–608.
  - Allen, F. & Gale, D. (2000). Financial contagion. Journal of Political Economy, 108(1), 1–33.
  - Bernanke, B. S. (2013). Monitoring the financial system, Speech at the 49th Annual Conference on Bank Structure and Competition, Chicago, IL, May 10, 2013.
  - Cetorelli, N. & Goldberg, L. (2011). Global banks and international shock transmission: Evidence from the crisis. IMF Economic Review, 59, 41–76.
  - Laeven, L. & Valencia, F. (2013). Systemic banking crises database. IMF Economic Review, 61, 225–270.
  - Standard & Poor’s (2011). A guide to the loan market, Standard and Poors Financial Services LLC. Available on https://www.lcdcomps.com/d/pdf/LoanMarketguide.pdf (accessed February 20, 2014).
- (The complete reference list as presented in the source PDF is included in the content unit.)

### Data Appendix — dataset construction and key sample counts
- Source of loan deals: Dealogic’s Loan Analytics data on 170,274 loan deals signed between January 1990 and December 2012.
- Retained for foreign interbank network: 16,526 loans extended from banks to banks.
- After dropping deals with lender recorded as “unknown”, “undisclosed syndicate”, or “undisclosed investor (unknown)”, deals involving multiple borrowers (representing less than 1 percent of the sample), and deals with missing maturity information, the sample is reduced to 148,378 deals.
- Territories removed for lacking an IFS code: Guernesey, Isle of Man, Jersey, and occupied Palestinian Territory.
- Bank identification:
  - Lender country variable: “Lender nationality” from Loan Analytics.
  - Borrower country variable: “Deal nationality” (validated against banks appearing as both borrowers and lenders).
  - Bank borrowers identified via general industry group “Finance” and sub-classifications: commercial and savings banks, provincial banks, municipal banks, savings and loans, and investment banks.
  - Financial firms classified as investment managers, special purpose vehicles, development banks, multilateral agencies, and miscellaneous are excluded from foreign interbank exposures.
- Bank name cleaning rules:
  - If a bank changed name during 1990-2012, retain its Bankscope name as of end-2012 throughout the entire sample period.
  - Merged banks: constituent banks kept distinct until year of merger; post-merger bank kept thereafter.
  - Acquired banks appear distinct until year of acquisition.
  - Lending from multiple branches of the same bank in a foreign country is aggregated.
  - Lending from off-shore branches of a bank is aggregated.
- Matching to balance sheets:
  - Banks matched by name and country with Bankscope balance sheet data.
  - Non-automatically matched banks inspected and matched manually using bank websites, the Federal Reserve Board National Information Center, and Bloomberg Businessweek.
  - Subsidiaries, branches, and other banking group entities with Bankscope balance sheet information are treated as distinct entities (not linked to parent financials).
  - Federal Reserve NIC lookup URL provided: http://www.ffiec.gov/nicpubweb/nicweb/SearchForm.aspx
  - Bloomberg Businessweek URL provided: http://investing.businessweek.com/research/company/overview/overview.asp
- Global banking network construction and sample sizes:
  - Full set of banks in Dealogic Loan Analytics during 1990-2012: 6,083 banks appearing as lenders or borrowers.
  - Sample of banks merged to Bankscope: about 2,200 distinct banks.
  - Final regression sample: 1,869 banks (due to missing balance sheet variables).
- Treatment of loans and exposures:
  - Loans treated as (non-amortizing) bullet loans.
  - Foreign interbank exposures constructed using lender and borrower identity, loan amount, and loan maturity.
  - For bank-borrower pairs where borrowers are non-financial firms or sovereigns, same approach used to construct foreign exposures.
  - Empirical analysis predominantly uses the number of (crisis and non-crisis) exposures rather than their dollar value because individual loan amounts contributed by each lender in the syndicate are observed for 40 percent of loan deals.
- Estimation of missing individual loan amounts:
  - For analyses requiring dollar exposures (Tables 3 and 8), individual loan amounts for remaining loans are estimated by regression.
  - Regression on loans with reported shares over 1990-2012; predictors include log-loan amount, original loan currency indicators, number of syndicate participants, borrower country and industry indicators, lender role in syndicate (bookrunner, mandated arranger, arranger, participant), lender country, prior lending/borrowing relationship indicator, same-country lender-borrower indicator, and year*quarter dummies.
  - Regression R-squared: 74 percent.
- U.S. bank foreign exposure correlation:
  - Data source: Federal Reserve RSSD lookup form on the Federal Reserve Board National Information Center website.
  - For foreign exposures use FFIEC 009a Column 4 representing “Total Amount of Cross-Border Claims and Foreign Office Claims on Local Residents,” defined precisely in the source.

### Key empirical sample counts and model fit statistics (selected from tables)
- Figure and table sample periods and data sources: generally 1997-2012; data sources include Dealogic Loan Analytics, BIS locational banking statistics, Bankscope, Federal Reserve (FFIEC009a), and Laeven and Valencia (2013).
- Table 3 (Correlation of interbank exposures with total foreign exposures)
  - Observations: 8,397; R-squared: 0.662 (column 1) and 0.657 (column 2).
- Table 4 (Effect of crisis exposures on bank performance - Baseline (ROA))
  - Observations: 14,448 across columns; R-squared: 0.440, 0.440, 0.441, 0.442.
  - Number of banks: 1,869.
- Table 5 (Effect of crisis exposures on bank performance - Baseline (ROE))
  - Observations: 14,445 across columns; R-squared: 0.345, 0.345, 0.346, 0.347.
  - Number of banks: 1,869.
- Table 6 (Addressing endogeneity; ROA and ROE robustness)
  - Observations vary by specification: 14,450; 13,754; 13,026; 14,447; 13,751; 13,023.
  - R-squared: 0.441; 0.442; 0.451; 0.346; 0.350; 0.363.
  - Number of banks varies: 1,869; 1,822; 1,703.
- Table 7 (NIM channel)
  - Observations: 14,135; 12,415; 10,771; 10,637 across columns 1–4.
  - R-squared: 0.659; 0.664; 0.664; 0.698.
  - Number of banks: 1,849; 1,740; 1,527; 1,519.
- Table 8 (Effect on banks’ supply of corporate loans)
  - Observations: columns 1–4: 167,958; columns 5–8: 295,204.
  - R-squared: 0.889; 0.889; 0.889; 0.890; 0.906; 0.906; 0.906; 0.906.
  - Number of banks (lead banks / all banks): examples listed include 1,211; 1,213; 1,333; 1,333.
  - Number of borrowers: examples listed include 2,252; 2,727; 3,030; 3,013 (as reported).
- Descriptive statistics (Table 2, selected)
  - Return on assets (ROA): N = 14,448; Mean = 0.809; St. Dev. = 1.560; Min = -6.850; Max = 8.850.
  - Return on equity (ROE): N = 14,445; Mean = 8.398; St. Dev. = 16.44; Min = -78.09; Max = 53.17.
  - Net interest margins (NIM): N = 14,315; Mean = 2.759; St. Dev. = 2.238; Min = -0.910; Max = 15.87.
  - Equity/Assets: N = 14,448; Mean = 8.753; St. Dev. = 9.333; Min = 0.320; Max = 81.51.
  - Assets (USD bn): N = 14,448; Mean = 72.342; St. Dev. = 236.40; Min = 0.45; Max = 3,808.
  - Banks: # indirect exposures: N = 14,448; Mean = 15.626; St. Dev. = 8.190; Min = 1; Max = 1,981.
  - Banks: Syndicated loan exposures (bank-destination country exposures, USD mn): N = 8,397; Mean = 842; Max = 1,773 (note: table formatting in source).
  - Systemic banking crisis indicator: N = 14,448; Mean = 0.209; St. Dev. = 0.407; Min = 0; Max = 1.
- Table notes and estimation conventions:
  - All bank balance sheet variables are winsorized at the 1st and 99th percentiles.
  - Standard errors clustered at the bank level (unless otherwise noted).
  - Significance indicators: *** 1% level, ** 5% level, * 10% level (as noted in table captions).
  - Fixed effects typically include bank nationality*year and borrower*year or bank*destination country FE where indicated.

*Source: _wp1691 - References (PDF)._

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp1691.pdf_