## wp18162 — Bank Network Analysis in the ECCU

## Source details

**Canonical URL:** [wp18162 — Bank Network Analysis in the ECCU](https://www.imf.org/-/media/files/publications/wp/2018/wp18162.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2018/wp18162.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2018/wp18162.pdf.json)

---

### Abstract and research purpose
- Paper applies network analysis to assess systemic vulnerabilities in the ECCU banking system.
- Two illustrative stress-test sets:
  - Solvency and liquidity shocks to each individual bank and contagion via bilateral net asset exposures.
  - Country- and region-wide tail shocks to GDP affecting capital and liquidity of all banks in shocked jurisdictions, followed by network spillovers.
- Objective: identify systemic institutions that merit heightened regulatory attention based on connectivity and vulnerability to other banks’ failures.

### Context and key facts about the ECCU banking system
- ECCU membership: Antigua and Barbuda; The Commonwealth of Dominica; Grenada; St. Kitts and Nevis; Saint Lucia; St. Vincent and the Grenadines; Anguilla; Montserrat.
- Shared central bank: Eastern Caribbean Central Bank (ECCB).
- Currency peg: currently at XCD$2.70=US$1; historically EC$4.80=GBP1 under earlier arrangements.
- ECCB established on October 1, 1983.
- Banking sector size and composition:
  - Bank assets range from 100 to near 300 percent of GDP, depending on jurisdiction.
  - Overall ECCU banking sector size is about 150 percent of GDP.
  - In St. Kitts and Nevis, banking sector assets amounted to nearly 300 percent of GDP at end-2016 and constituted 1/4 of the ECCU banking sector.
  - In Antigua and Barbuda and Saint Lucia, commercial banks’ assets were 150 and 130 percent of GDP, respectively; each contributed about 20 percent to total ECCU banking sector.
  - Mid-2017: ECCU banking sector comprised 35 institutions, 17 locally incorporated banks (constituting about 53 percent of total assets) and 18 foreign incorporated banks.
- Nonperforming loans (NPLs) and provisioning:
  - At end-2017Q3, banks’ NPLs stood at 11 percent of total loans.
  - Region’s tolerance limit: 5 percent.
  - NPLs peaked at 19 percent in mid-2014 and have declined since.
  - Provisioning has remained insufficient; net NPLs in percent of capital remain elevated.
- Profitability, capital, liquidity, and expenses:
  - Capital accumulation increased in recent years; capital adequacy referenced via CAR and CAR ratio improvements.
  - Profitability remains low (ROA shown historically).
  - Operating expenses remain elevated (non-interest expense as percent of gross income is high).
  - Liquidity at all-time highs (liquid assets as percent of short-term liabilities increased).
- Exposures and linkages:
  - Commercial banks hold sizable shares of ECCU sovereign debt; St. Kitts and Nevis and Montserrat hold 16 and 12 percent of total assets in sovereign debt, respectively (end-2016).
  - Tight linkages exist with non-bank financial institutions (NBFIs) — credit unions, insurance companies, pension funds.
  - Credit unions are largest NBFI deposit holders; in Dominica credit unions maintain deposits in banks of about 5 percent of GDP.
  - In St. Vincent and the Grenadines, about 10 percent of commercial bank deposits belong to NBFIs.
  - Wholesale institutional deposits imply susceptibility to abrupt liquidity withdrawals.
- Scope limitations for analysis:
  - Study focuses on locally incorporated banks (about half the regional sector) because foreign branches do not hold independent capital.
  - Bank–credit union exposures, bank exposures to other NBFIs and sovereigns excluded due to data limitations.

### Key risks and motivation for network analysis
- Main risk analyzed: failure of an individual bank triggering cascades of bank failures via interbank bilateral exposures.
- Relevance: high output volatility, small size, undiversified production base of ECCU economies make macro and idiosyncratic tail shocks important.
- ECCB context: limited lender-of-last-resort (LOLR) facilities and no deposit insurance make liquidity shocks potentially more severe.

### Methodology overview
- Data basis: balance sheets of all locally incorporated bank institutions in the ECCU; interbank exposures captured by bilateral net asset positions; shock absorption captured by capital and liquid assets.
- Analytical framework: balance-sheet network analysis methodology in Espinosa-Vega and Sole (2010).
- Two shock types simulated:
  - Credit Shock (CS): loss on asset side; defaulted claims reduce asset values of creditor banks; assumes no funding shock replacement needed.
  - Funding Shock (FS): sudden withdrawal of funding from each and all banks (liability shock) that may force asset fire sales at depressed prices, reducing asset and equity values.
- Simulation procedure:
  - Individual-bank idiosyncratic shocks: shock to a bank’s asset value and tracking of propagation via bilateral net exposures.
  - Macroeconomic tail shocks: decline in output and increase in interest rates affecting asset values and capital of all banks in affected country, then spread through network.
  - Region-wide tail shock also tested for resilience to extreme external shocks.

### Individual-bank shock simulation setup and parameters
- Simultaneous credit and funding shocks applied to each bank individually, then contagion traced through the interbank network.
- Liquidity shock assumptions:
  - 50 percent haircut in fire sale of assets.
  - 65 percent roll-over ratio of interbank debt.
  - Corresponding to parameter values of δ=1 and ρ=0.35.
- Data based on end-June 2017, including 17 locally incorporated banks operating in 8 ECCU jurisdictions.
- Two sets of indicators used:
  - Systemic importance (impact of a bank’s failure on others): number of bank failures; failed capital (average percent of asset loss in other banks); number of contagion rounds; index of contagion (percent of capital loss in the system).
  - Vulnerability of each bank (exposure to failure of others): vulnerability level (number of other failures triggered); index of vulnerability (average percent of capital loss of a bank due to individual failure of all other banks).

### Individual-bank shock results (selected findings)
- Bank 12 (jurisdiction 6):
  - Its “failure” leads to “failure” of eight other banks within two contagion rounds.
  - Induces the highest capital loss in the system of nearly 50 percent.
  - Shows the highest index of contagion.
- Bank 14 (jurisdiction 6):
  - Its “failure” leads to “failure” of four other banks.
  - Produces losses of 17 percent to the system.
  - Leads to 3 contagion rounds (the highest number in the exercise).
- Bank 4 (jurisdiction 2):
  - Its “failure” in one contagion round would lead to “failure” of two other banks and produce losses of 18 percent of the banking system.
- Banks 2 and 14:
  - Appear systemic with systemic capital loss (failed capital) near 30 percent.
- Vulnerability findings:
  - Bank 3 (jurisdiction 2): vulnerability level = “failure” of six other institutions would lead to its “failure”.
  - Bank 2 (jurisdiction 1): vulnerability level = “failure” of four other institutions would lead to its “failure”.
  - Banks 6 and 8: appear vulnerable in terms of capital losses when other banks failed, indicating substantial net-exposures within the network.
- More than half of the banks induce at least one contagion round.

### Macroeconomic stress-test scenarios: design and channels
- Two macro scenarios analyzed:
  - Country-specific negative GDP shock applied individually to each jurisdiction.
  - Tail-risk negative shock applied to all ECCU members simultaneously.
- Transmission channels:
  - Credit risk: increase in NPLs and provisioning; initial collateral valuation adjustment assumes actual collateral worth half of reported value.
  - Interest rate risk: negative GDP shock assumed to lead to an interest rate spike informed by observed RGSM behavior; direct interest rate risk modeled via net interest income flow and bond repricing (average duration ≈ 5 years).
- Satellite panel VAR implication:
  - A 1 percentage point decline in real GDP growth is associated with about 3 percent increase in NPLs.
- Credit risk implementation:
  - Increase in NPLs for each bank = elasticity (3 percent per 1 percentage point) × country-specific GDP growth shock; increase proportional to a bank’s initial NPL level.
  - Provisioning assumption: 50% provisioning requirement on new NPLs is assumed.
  - Impact ratio assumption: 100% impact ratio is assumed.
- Interest rate risk implementation:
  - Increase in interest rates calculated as difference between maximum bid rate and actual T-bill yield, capped by a maximum bid rate typically set at 6 percent.
  - Net interest income impact uses asset-liability “gap” with only assets/liabilities repricing <12 months included; qualifying liabilities measured as 85 percent of total deposits.
  - Asset repricing impact proxied using duration = 5 years.

### Country-specific negative GDP shocks used (exact values)
- Antigua and Barbuda: -10.2
- Anguilla: -14.3
- Dominica: -4.1
- Grenada: -7.0
- Montserrat: -3.3
- St. Kitts and Nevis: -6.0
- St. Lucia: -3.8
- St. Vincent and the Grenadines: -3.5

### Customized scenario parameters (selected exact figures)
- Assumed collateral, % loan: 124%
- Haircut on collateral: 52%
- Assumed increase in NPLs (country-specific):
  - Antigua and Barbuda31%
  - Anguilla43%
  - Dominica12%
  - Grenada21%
  - Montserrat10%
  - St. Kitts and Nevis18%
  - St. Lucia11%
  - St. Vincent and the Grenadines11%
- Assumed provisioning of additional NPLs: 50%
- Assumed change in interest rate due to GDP shock (pptss):
  - Antigua and Barbuda3.4
  - Anguilla3.4
  - Dominica4.1
  - Grenada2.5
  - Montserrat3.4
  - St. Kitts and Nevis3.4
  - St. Lucia3.0
  - St. Vincent and the Grenadines4.0
- Assumed average duration of bonds held: 5.0

### Country-specific shock simulation outcomes (jurisdiction-level highlights)
- Jurisdiction 1
  - Two locally incorporated banks domiciled.
  - Negative GDP shock would result in a “failure” of bank 2, with spillovers contained.
  - Vulnerability: “failure” of 4 other regional banks could spill over to bank 2.
- Jurisdiction 2
  - Three locally incorporated banks domiciled.
  - Negative GDP shock would lead to “failure” of bank 3 with no further contagion.
  - Bank 3 vulnerability: “failure” of 6 other banks could spill over.
- Jurisdiction 3
  - One locally incorporated bank; it would “fail” after the shock.
  - Limited interconnectedness constrains spillovers.
  - Bank 6 vulnerability: “failure” of 7 other regional locally incorporate banks could lead to its insolvency.
- Jurisdiction 4
  - Three locally incorporated banks; only bank 8 would fail because of the negative GDP shock.
  - No further spillovers observed.
  - Bank 8 vulnerability: “failure” of 5 other institutions would lead to its insolvency.
- Jurisdiction 5
  - One locally incorporated bank (bank 10).
  - Bank 10’s buffers sufficient to withstand the simulated GDP shock.
- Jurisdiction 6
  - Three locally incorporated commercial banks domiciled.
  - Negative GDP shock would result in “failure” of two systemically important banks: bank 11 and bank 12; bank 13 withstands the shock.
  - Joint “failure” of banks 11 and 12 (only in two contagion rounds) would result in insolvency of 7 other banks from 6 other ECCU jurisdictions.
  - Vulnerability: exogenous “failure” of 9 other regional banks would result in the “failure” of banks 11 and 12.
- Jurisdiction 7
  - Two locally incorporated commercial banks (14 and 15).
  - Negative GDP shock would lead to “failure” of both banks 14 and 15.
  - Their “failure” spills over to 2 other institutions in jurisdictions 2 and 3.
  - Vulnerability: “failure” of 9 other institutions would lead to their demise.
- Jurisdiction 8
  - Two locally incorporated banks (16 and 17).
  - A shock to GDP would result in “failure” of both banks 16 and 17.
  - Only bank 16 would have a knock-on effect on another institution (bank 2 in jurisdiction 1).
  - Vulnerability: “failure” of 8 other banks could fail bank 16.

### ECCU-wide (regional) shock results
- Shock specification: all countries experience 2-standard deviations real GDP growth shocks below their corresponding historical averages.
- Result: region-wide shock leads to “failure” of 10 of the 17 locally incorporated banks in the ECCU; the remaining 7 banks continue operations with positive capital.
- Interpretation: under this tail risk scenario, over half of the locally incorporated banks become undercapitalized, reflecting vulnerabilities to extreme external shocks.

### Time-dimension findings (end-2014 to mid-2017)
- Time horizon covered: end-2014, end-2015, end-2016, and mid-2017; sample restricted to banks operational throughout 2014-2017 (15 banks in Appendix analysis).
- Overall trend:
  - Simulations highlight significant reduction in banks’ vulnerability over time.
  - Contagion across ECCU locally incorporated banks declined over time.
- Drivers of improvement:
  - Capital accumulation: number of undercapitalized banks declined significantly after 2015.
  - Regulatory intervention: ECCB interventions between 2015 and 2017 resulted in mergers/acquisitions of insolvent banks in Antigua and Barbuda and Anguilla.
  - Regulatory changes: response to higher capital requirements of the new Banking Act (2015).
  - Asset quality and provisioning: lower NPLs and higher provisioning rates helped lower banks’ vulnerability.
  - Macroeconomic environment: better economic outlook post-GFC contributed to improved bank resilience.
- Persistent systemic actor:
  - Bank 12 remained highly contagious across the period; as of mid-2017 failure of bank 12 would lead to undercapitalization of 7 other regional banks (nearly half of the sector in the simulation).
- Index measures highlight systemic roles (index of contagion) and vulnerability (indexes showing banks 3, 6, and 8 as vulnerable).

### Capital adjustment and collateral valuation methodology (selected exact steps and figures)
- Market-friction assumption: actual collateral worth about half of reported value.
- Adjustment steps (preserved order and parameters):
  1. Actual collateral value of NPLs = NPLs * required collateral-to-total loan ratio (124 percent).
  2. Collateral after the haircut = actual collateral value * (1- haircut); haircut assumed about 52 percent.
  3. Provisioning needed computed by loan category (substandard, doubtful, loss) using required provisioning rates.
  4. Underprovisioning = actual provisioning made – provisioning needed for collateral with haircut.
  5. Bank capital after adjustment for underprovisioning = initial capital – underprovisioning * impact ratio (impact ratio assumed 100%).
  6. RWA after adjustment for underprovisioning = initial RWA – initial RWA * impact ratio (100%).
- Provisioning guidance (ECCB Prudential Credit Guidelines, 1997):
  - Pass and special mention loans: 0%
  - Substandard (other than government or fully government-secured or by cash): 10%
  - Doubtful: 50%
  - Loss: 100%
- Haircut and NPV example:
  - Haircut of 52 percent; NPV of bank collection on NPL collateral of 48 percent.
  - Example parameters: loan = 100, loan repayment before default = 50, collateral value = 124, discount rate = 6.5%, time till collateral sale = 7 years; NPV calculation yields 0.48.

### Policy implications and recommendations
- Maintain and strengthen capital buffers:
  - Capital accumulation materially reduced contagion and vulnerability; capital buffers remain critical to absorb shocks.
- Continue regulatory vigilance and timely updating of network analysis:
  - Changes in balance sheet strength and bilateral exposures indicate network analysis should be updated regularly to reassess vulnerabilities and monitor progress.
- Address structural sectoral weaknesses:
  - Continue efforts to reduce vulnerabilities associated with property valuation difficulties, debtor-friendly foreclosure laws, and slow foreclosure processes.
- Sustain supervisory interventions and resolution frameworks:
  - ECCB regulatory interventions between 2015 and 2017 contributed to stability; continued supervisory capacity and effective resolution tools remain important.
- Research extensions:
  - Expand network to include NBFIs (notably credit unions) given their size and deposit links to banks; address weaker regulation and limited supervisory capacity for such institutions.

*Source: IMF Working Paper WP/18/162 — Bank Network Analysis in the ECCU (wp18162).*

### Section 1

### WP/18/162 — Bank Network Analysis in the ECCU (Section 1)

### Abstract and Research Purpose
- Paper applies network analysis to assess systemic vulnerabilities in the ECCU banking system.
- Two illustrative stress-test sets:
  - Solvency and liquidity shocks to each individual bank and contagion via bilateral net asset exposures.
  - Country- and region-wide tail shocks to GDP affecting capital and liquidity of all banks in shocked jurisdictions, followed by network spillovers.
- Objective: identify systemic institutions that merit heightened regulatory attention based on connectivity and vulnerability to other banks’ failures.

### Context and Key Facts about the ECCU Banking System
- ECCU membership: Antigua and Barbuda; The Commonwealth of Dominica; Grenada; St. Kitts and Nevis; Saint Lucia; St. Vincent and the Grenadines; Anguilla; Montserrat.
- Shared central bank: Eastern Caribbean Central Bank (ECCB).
- Currency peg: currently at XCD$2.70=US$1; historically EC$4.80=GBP1 under earlier arrangements.
- ECCB established on October 1, 1983.
- Banking sector size and composition:
  - Bank assets range from 100 to near 300 percent of GDP, depending on jurisdiction.
  - Overall ECCU banking sector size is about 150 percent of GDP.
  - In St. Kitts and Nevis, banking sector assets amounted to nearly 300 percent of GDP at end-2016 and constituted 1/4 of the ECCU banking sector.
  - In Antigua and Barbuda and Saint Lucia, commercial banks’ assets were 150 and 130 percent of GDP, respectively; each contributed about 20 percent to total ECCU banking sector.
  - Mid-2017: ECCU banking sector comprised 35 institutions, 17 locally incorporated banks (constituting about 53 percent of total assets) and 18 foreign incorporated banks (mainly headquartered in Canada).
- Nonperforming loans (NPLs) and provisioning:
  - At end-2017Q3, banks’ NPLs stood at 11 percent of total loans.
  - Region’s tolerance limit: 5 percent.
  - NPLs peaked at 19 percent in mid-2014 and have declined since.
  - Provisioning has remained insufficient; net NPLs in percent of capital remain elevated.
- Profitability, capital, liquidity, and expenses:
  - Capital accumulation increased in recent years; capital adequacy referenced via CAR and CAR ratio improvements.
  - Profitability remains low (ROA shown historically).
  - Operating expenses remain elevated (non-interest expense as percent of gross income is high).
  - Liquidity at all-time highs (liquid assets as percent of short-term liabilities increased).
- Exposures and linkages:
  - Commercial banks hold sizable shares of ECCU sovereign debt; St. Kitts and Nevis and Montserrat hold 16 and 12 percent of total assets in sovereign debt, respectively (end-2016).
  - Tight linkages exist with non-bank financial institutions (NBFIs) — credit unions, insurance companies, pension funds.
  - Credit unions are largest NBFI deposit holders; in Dominica credit unions maintain deposits in banks of about 5 percent of GDP.
  - In St. Vincent and the Grenadines, about 10 percent of commercial bank deposits belong to NBFIs.
  - Wholesale institutional deposits imply susceptibility to abrupt liquidity withdrawals.
- Scope limitations for analysis:
  - Study focuses on locally incorporated banks (about half the regional sector) because foreign branches do not hold independent capital.
  - Bank–credit union exposures, bank exposures to other NBFIs and sovereigns excluded due to data limitations.

### Key Risks and Motivations for Network Analysis
- Main risk analyzed: failure of an individual bank triggering cascades of bank failures via interbank bilateral exposures.
- Relevance: high output volatility, small size, undiversified production base of ECCU economies make macro and idiosyncratic tail shocks important.
- ECCB context: limited lender-of-last-resort (LOLR) facilities and no deposit insurance make liquidity shocks potentially more severe.

### Methodology Overview
- Data basis: balance sheets of all locally incorporated bank institutions in the ECCU; interbank exposures captured by bilateral net asset positions; shock absorption captured by capital and liquid assets.
- Analytical framework: balance-sheet network analysis methodology in Espinosa-Vega and Sole (2010).
- Two shock types simulated:
  - Credit Shock (CS):
    - Loss on asset side; defaulted claims reduce asset values of creditor banks.
    - If losses exceed capital, creditor defaults; contagion process iterates until no further defaults.
    - Assumption: other banks can fully replace funding lost from bankruptcies (no funding shock in CS).
  - Funding Shock (FS):
    - Sudden withdrawal of funding from each and all banks (liability shock).
    - Shortfall can force asset sales; under market stress asset fire sales may be at depressed prices, reducing asset and equity values.
- Simulation procedure:
  - Individual-bank idiosyncratic shocks: shock to a bank’s asset value and tracking of propagation via bilateral net exposures.
  - Macroeconomic tail shocks: decline in output and increase in interest rates affecting asset values and capital of all banks in affected country, then spread through network.
  - Region-wide tail shock also tested for resilience to extreme external shocks.

### Illustrative Findings and Diagnostic Focus (from Section 1)
- Interbank exposures relative to capital have diminished over time for locally incorporated banks, attributed to:
  - Increased capital accumulation.
  - Higher excess liquidity.
  - Lower demand for interbank loans.
- High NPLs underscore importance of ongoing monitoring of capitalization and provisioning.
- Large holdings of sovereign debt reinforce sovereign–bank nexus risk.
- Significant deposits from NBFIs (notably credit unions) increase potential for liquidity spillovers between NBFIs and commercial banks.
- Systemic importance identification:
  - Systemic institutions are those with high connectivity to the rest of the system and pronounced vulnerability to the failure of other banks.

*Source: IMF Working Paper WP/18/162 — Bank Network Analysis in the ECCU (Section 1), July 2018.*

### Section 2

### wp18162 - Section 2

### Funding shocks, fire sales, and bank default mechanics
- A bank forced to liquidate assets under fire sale conditions can suffer a reduced price and thus a capital loss, particularly when markets are under a liquidity shortage or a bank holds large positions and cannot find buyers.
- Losses in some banks could turn capital negative and cause a default; initial funding shocks (FS) can lead to further subsequent FS and contagion shocks (CS) as other banks default.
- The analysis assumes a bank “defaults” when the adjusted capital turns negative (adjusted for minimum requirement and NPLs net of provisioning, and adjusted for under-provisioning due to collateral undervaluation).

### Pre-stress-test capital adjustments and definitions
- Capital of each bank is adjusted by:
  - Deducting the minimum requirement: 8 percent risk-weighted assets.
  - Deducting non-performing loans (NPLs) net of provisioning.
  - Adjusting for under-provisioning due to collateral undervaluation (actual value of collateral assumed half of reported value).
- Bank “failure” definition: adjusted capital turns negative (total qualifying capital adjusted for under-provisioning due to collateral overvaluation, excluding net NPLs above the regulatory requirement minimum of 8 percent of risk-weighted assets).

### Individual bank shocks: simulation setup
- Simultaneous credit and funding shocks applied to each bank individually, then contagion traced through the interbank network.
- Liquidity shock assumptions:
  - 50 percent haircut in fire sale of assets.
  - 65 percent roll-over ratio of interbank debt.
  - Corresponding to parameter values of δ=1 and ρ=0.35.
- Data based on end-June 2017, including 17 locally incorporated banks operating in 8 ECCU jurisdictions.
- Two sets of indicators used:
  - Systemic importance (impact of a bank’s failure on others): number of bank failures; failed capital (average percent of asset loss in other banks); number of contagion rounds; index of contagion (percent of capital loss in the system).
  - Vulnerability of each bank (exposure to failure of others): vulnerability level (number of other failures triggered); index of vulnerability (average percent of capital loss of a bank due to individual failure of all other banks).

### Individual bank shocks: key simulation results
- Bank 12 (jurisdiction 6):
  - Its “failure” leads to “failure” of eight other banks within two contagion rounds (nearly half of the 17 locally incorporated banking system).
  - Induces the highest capital loss in the system of nearly 50 percent.
  - Shows the highest index of contagion.
- Bank 14 (jurisdiction 6):
  - Its “failure” leads to “failure” of four other banks.
  - Produces losses of 17 percent to the system.
  - Leads to 3 contagion rounds (the highest number in the exercise).
- Bank 4 (jurisdiction 2):
  - Its “failure” in one contagion round would lead to “failure” of two other banks and produce losses of 18 percent of the banking system.
- Banks 2 and 14:
  - Appear systemic with systemic capital loss (failed capital) near 30 percent.
- Vulnerability findings:
  - Bank 3 (jurisdiction 2): vulnerability level = “failure” of six other institutions would lead to its “failure”.
  - Bank 2 (jurisdiction 1): vulnerability level = “failure” of four other institutions would lead to its “failure”.
  - Banks 6 and 8: appear vulnerable in terms of capital losses when other banks failed, indicating substantial net-exposures within the network.
- More than half of the banks induce at least one contagion round.

### Macroeconomic stress-test scenario: motivation and design
- ECCU economies face significant exposure to large external shocks and natural disasters, producing high output volatility; banks are vulnerable to economy-wide shocks as well as idiosyncratic shocks.
- Two macro scenarios analyzed:
  - Country-specific negative GDP shock applied individually to each jurisdiction (examples: natural disaster; sudden decline in citizenship-by-investment flows; decrease in foreign debt access).
  - Tail-risk negative shock applied to all ECCU members simultaneously (example: regional shock affecting tourism; sharp increase in global interest rates triggering portfolio shift and credit crunch).
- In each scenario, the shock affects the economy and propagates in the banks’ network through:
  - Credit risk (impact on NPLs and provisioning).
  - Interest rate risk (impact on net interest revenues and repricing of bond holdings).
- Foreign exchange risk excluded due to fixed exchange rate arrangement (Eastern Caribbean dollar quasi-currency board vis-à-vis U.S. dollar).

### Customized scenario assumptions: implementation steps
- Two-step exercise:
  1. Quantify impact of GDP shock on each bank’s assets and capital and identify which banks “fail”.
  2. Trace spillovers from “failed” banks to others via interbank bilateral lending matrix; include default of interbank obligations and liquidity shock on banks funded by the “failed” bank, triggering asset fire sales.
- Credit risk channel:
  - Initial collateral valuation adjustment: actual value of collateral assumed half of reported value.
  - Adjustment captures debtor-friendly foreclosure laws, lack of specialized courts, and likely decline in collateral values in a negative tail macro environment.
  - Higher provisioning requirements result, lowering risk-weighted assets and capital.
  - Shock size: two standard deviations below the 2000-16 average of real GDP growth for each country (country-specific; depends on historical volatility).
  - Satellite panel VAR model (six independent ECCU economies) implies: a 1 percentage point decline in real GDP growth is associated with about 3 percent increase in NPLs.
  - Increase in NPLs for each bank = elasticity (3 percent per 1 percentage point) × country-specific GDP growth shock; increase proportional to a bank’s initial NPL level.
  - Provisioning assumption: 50% provisioning requirement on new NPLs is assumed.
  - Impact ratio assumption: 100% impact ratio is assumed.
- Interest rate risk channel:
  - Negative GDP shock assumed to lead to an interest rate spike informed by observed increases in the Regional Government Securities Market (RGSM).
  - Increase in interest rates calculated as difference between maximum bid rate and actual T-bill yield (coupon rate on 91-day T-bills on RGSM); capped by a maximum bid rate typically set at 6 percent.
  - Historical examples:
    - Tropical Storm Erika (Dominica, August 2015): 3-month T-bill yields rose from below 1 percent in June to maximum rate of 6 percent in September 2015.
    - Hurricane Maria (Dominica, September 2017): 91-day T-bill yield increased from 2 percent pre- to 6 percent post-storm.
  - Direct interest rate risk considered only (indirect interest rate risk excluded).
  - Direct interest rate risk components:
    - Net interest income impact (flow): calculated as product of asset-liability “gap” (only assets/liabilities with time to repricing below 12 months included, which includes only T-bills and liquid assets due to banks in other ECCB territories) and assumed change in interest rates; qualifying liabilities conservatively measured as 85 percent of total deposits (share of demand, savings, and foreign currency deposits in total constitutes 85 percent).
    - Asset repricing impact (stock): bonds marked-to-market; impact proxied using duration of bonds held by banks. Average ECCU-wide estimated duration ≈ 5 years applied to all banks. Change in bond portfolio value = bank bond holding × bond duration × assumed increase in interest rates.
  - Result: decline in net interest income and negative repricing impact reduce banks’ capital.

### Country-specific negative GDP shocks used in the customized scenario
- Assumed GDP shock defined as 2000-2016 average of real GDP growth minus two standard deviations (2000-2016 period).
- Country-specific GDP shock (Percent):
  - Antigua and Barbuda: -10.2
  - Anguilla: -14.3
  - Dominica: -4.1
  - Grenada: -7.0
  - Montserrat: -3.3
  - St. Kitts and Nevis: -6.0
  - St. Lucia: -3.8
  - St. Vincent and the Grenadines: -3.5

### Post-shock network contagion mechanics
- After initial impact on capital via credit and interest rate risks, simulations examine cascading effects:
  - Default of interbank obligations of undercapitalized banks.
  - Liquidity shock on banks funded by the “failed” bank triggers asset fire sales and further capital losses (50 percent fire-sale haircut applied in liquidity shock earlier).
- Bank “failure” in this stage: negative post-shock capital (qualifying capital adjusted for under-provisioning, net NPLs, and above the regulatory minimum of 8% of RWA).

*Source: IMF staff estimates and calculations, wp18162 - Section 2*

### Section 3

### wp18162 - Section 3

### Summary of simulation framework and interpretation
- Stress tests applied: combined asset and liability shocks to individual banks (individual-bank shock), tail GDP growth shocks to each country (country shock), and tail GDP growth shocks to the entire ECCU region (regional shock).
- Capital definition used in main exercises: total qualifying capital above 8 percent of risk-weighted assets regulatory minimum and excluding net NPLs; capital also adjusted for market frictions and underprovisioning in some analyses.
- Joint credit and liquidity shock assumption for fire-sale and interbank rollover: 50 percent haircut in the fire sale of assets and a 65 percent roll-over ratio of interbank debt.
- Purpose: warning benchmark for regulators focusing on capital shortfalls below the 8 percent minimum capital requirement.

### Jurisdiction-level findings (individual country shock outcomes)
- Jurisdiction 1
  - Two locally incorporated banks domiciled.
  - A negative GDP shock would result in a “failure” of bank 2, with spillovers contained.
  - Vulnerability indicators: “failure” of 4 other regional banks could spill over to bank 2.
- Jurisdiction 2
  - Three locally incorporated banks domiciled.
  - Negative GDP shock would lead to “failure” of bank 3 with no further contagion.
  - Bank 3 vulnerability: “failure” of 6 other banks could spill over.
- Jurisdiction 3
  - One locally incorporated bank; it would “fail” after the shock.
  - Limited interconnectedness constrains spillovers.
  - Bank 6 vulnerability: “failure” of 7 other regional locally incorporate banks could lead to its insolvency.
- Jurisdiction 4
  - Three locally incorporated banks; only bank 8 would fail because of the negative GDP shock.
  - No further spillovers observed.
  - Bank 8 vulnerability: “failure” of 5 other institutions would lead to its insolvency.
- Jurisdiction 5
  - One locally incorporated bank (bank 10).
  - Bank 10’s buffers sufficient to withstand the simulated GDP shock.
- Jurisdiction 6
  - Three locally incorporated commercial banks domiciled.
  - Negative GDP shock would result in “failure” of two systemically important banks: bank 11 and bank 12; bank 13 withstands the shock.
  - Joint “failure” of banks 11 and 12 (only in two contagion rounds) would result in insolvency of 7 other banks from 6 other ECCU jurisdictions.
  - Vulnerability: exogenous “failure” of 9 other regional banks would result in the “failure” of banks 11 and 12.
- Jurisdiction 7
  - Two locally incorporated commercial banks (14 and 15).
  - Negative GDP shock would lead to “failure” of both banks 14 and 15.
  - Their “failure” spills over to 2 other institutions in jurisdictions 2 and 3.
  - Vulnerability: “failure” of 9 other institutions would lead to their demise.
- Jurisdiction 8
  - Two locally incorporated banks (16 and 17).
  - A shock to GDP would result in “failure” of both banks 16 and 17.
  - Only bank 16 would have a knock-on effect on another institution (bank 2 in jurisdiction 1).
  - Vulnerability: “failure” of 8 other banks could fail bank 16.

### ECCU-wide (regional) shock results
- Shock specification: all countries experience 2-standard deviations real GDP growth shocks below their corresponding historical averages.
- Result: region-wide shock leads to “failure” of 10 of the 17 locally incorporated banks in the ECCU; the remaining 7 banks continue operations with positive capital.
- Interpretation: under this tail risk scenario, over half of the locally incorporated banks become undercapitalized, reflecting vulnerabilities to extreme external shocks (examples: tourism shock, tail shocks to global financial stability, sudden stop in financing flows).

### Key quantitative outcomes and indicators (from customized scenario summary)
- Induced systemic outcomes (selected highlights from simulations):
  - Individual-bank shock identified two potentially systemic banks that could induce “failure” of 8 and 4 other banks with capital losses in the system of nearly 50 and 30 percent, respectively.
  - Two most vulnerable banks identified: “failure” of six and four other banks would result in undercapitalization of these institutions.
- Regional shock headline:
  - 10 of the 17 locally incorporated banks fail under the ECCU-wide 2-standard deviations negative GDP shocks.
- Hazard and contagion indexes (reported in Figure 14 table, selected entries preserved verbatim as presented):
  - Examples from table: Bank 11 and Bank 12 entries show patterns such as ".....8..8.....3..2" under Induced Failures/Contagion Rounds and ".....9..12.....46..29" under HazardIndex of Contagion (table entries reproduced exactly as in source).

### Capital adjustment methodology for overvalued collateral (Appendix summary)
- Market-friction assumption: actual collateral worth about half of reported value to reflect shallow property markets, debtor-friendly foreclosure laws, lack of specialized courts, and absence of prior (lien) registries.
- Adjustment steps:
  1. Actual collateral value of NPLs = NPLs * required collateral-to-total loan ratio (124 percent).
  2. Collateral after the haircut = actual collateral value * (1- haircut); haircut assumed about 52 percent.
  3. Provisioning needed computed by loan category (substandard, doubtful, loss) using required provisioning rates.
  4. Underprovisioning = actual provisioning made – provisioning needed for collateral with haircut.
  5. Bank capital after adjustment for underprovisioning = initial capital – underprovisioning * impact ratio (impact ratio assumed 100%).
  6. RWA after adjustment for underprovisioning = initial RWA – initial RWA * impact ratio (100%).
- Provisioning guidance (ECCB Prudential Credit Guidelines, 1997):
  - Pass and special mention loans: 0%
  - Substandard (other than government or fully government-secured or by cash): 10%
  - Doubtful: 50%
  - Loss: 100%
- Haircut and NPV example used in calculations:
  - Haircut of 52 percent; NPV of bank collection on NPL collateral of 48 percent.
  - Example: assuming loan = 100, loan repayment before default = 50, collateral value = 124, discount rate = 6.5%, and time till collateral sale = 7 years, NPV of bank collection on collateral = ((124-(100-50))/(1.065)^7) = 0.48.

### Customized scenario assumptions (selected exact figures from Appendix Table A1.1)
- Country-specific GDP shocks (country-specific declines used in country shock simulations):
  - Antigua and Barbuda-10.2
  - Anguilla-14.3
  - Dominica-4.1
  - Grenada-7.0
  - Montserrat-3.3
  - St. Kitts and Nevis-6.0
  - St. Lucia-3.8
  - St. Vincent and the Grenadines-3.5
- Credit risk parameters:
  - Assumed collateral, % loan: 124%
  - Haircut on collateral: 52%
  - Assumed increase in NPLs (country-specific):
    - Antigua and Barbuda31%
    - Anguilla43%
    - Dominica12%
    - Grenada21%
    - Montserrat10%
    - St. Kitts and Nevis18%
    - St. Lucia11%
    - St. Vincent and the Grenadines11%
  - Assumed provisioning of additional NPLs: 50%
- Interest rate risk parameters:
  - Assets with time to repricing <12 months: T-bill holding
  - Liabilities with time to repricing <12: share of deposits 85%
  - Assumed change in interest rate due to GDP shock (pptss):
    - Antigua and Barbuda3.4
    - Anguilla3.4
    - Dominica4.1
    - Grenada2.5
    - Montserrat3.4
    - St. Kitts and Nevis3.4
    - St. Lucia3.0
    - St. Vincent and the Grenadines4.0
  - Assumed average duration of bonds held: 5.0

### Conclusions and implications for policy and further research
- Simulations identify systemic vulnerabilities in the ECCU locally incorporated banking network, with two jurisdictions (6 and 7) housing systemically important banks under country shocks and a region-wide shock causing undercapitalization of over half of locally incorporated banks.
- Observed drivers of improved resilience in recent years: increased capital accumulation, higher provisioning for non-performing loans, and lower NPLs, which reduced interbank exposures relative to capital.
- Recommended extensions for further research: expand the network to include non-bank financial institutions (notably credit union sectors) which may be of systemic importance given their size and net asset connections with banks; note concerns about weaker regulation and limited supervisory capacity for such institutions.

*Source: Authors' calculations (wp18162 - Section 3).*

### Section 4

### B. Appendix 2. Bank Network Analysis: Time Dimension

### Data, definitions, and assumptions
- Time horizon covered: end-2014, end-2015, end-2016, and mid-2017 (simulations repeated for each period).
- Sample restriction: banks operational during the entire period of 2014-2017; banks acquired, merged, or created during 2014-2017 were removed from the sample, yielding a total of 15 banks presented in Appendix Figure A2.
- Capital definition (primary): total qualifying capital above the regulatory minimum of 8 percent of risk-weighted assets, excluding NPLs net of provisioning.
- Alternate capital definition (not used for all periods due to data limitations): total qualifying capital adjusted for overvaluation of collateral, excluding net NPLs, and above 8 percent of risk-weighted assets regulatory minimum (footnoted in figure).
- Additional contextual assumptions and notes (from footnotes in source):
  - While the maximum rate varies by country, 6% rate is assumed for all jurisdictions for simplicity.
  - At end-2017Q2, 85% deposits constituted saving, demand, and foreign currency deposits in total deposits of ECCU commercial banks (excludes time deposits and EC$ cheques and drafts).
  - Assumed GDP shock is calculated as 2000-2016 average of real GDP growth minus two standard deviations (taken over the same period).
  - Required collateral in percent of loan as reported in World Bank Enterprise Survey; average taken for Antigua, Barbados, The Bahamas, Dominica, St. Kitts and Nevis, St. Lucia, and Trinidad and Tobago.
  - Large haircut reflects the difficulty to foreclose collateral (Cihak, 2007).
  - Source for prudential guidelines: ECCB Prudential Credit Guidelines, 1997.

### Sample, methodology, and caveats
- Interbank exposures of ECCU locally incorporated banks were analyzed to assess stability of prior results and the evolution of interconnectedness.
- Simulations: individual bank failure shock; outcomes measured include number of induced ‘failures’, failed capital (percent), number of contagion rounds (number), vulnerability level (number), index of contagion (number), and index of vulnerability (number).
- Capital in these simulations is the adjusted definition described above when feasible; however, results presented in this appendix are based on the unadjusted definition because data limitations impeded adjustment of capital for every period.
- Banks excluded if not continuously operational during 2014-2017, explaining omitted banks and 15-bank figure in Appendix Figure A2.

### Key simulation findings and dynamics over 2014–mid-2017
- Overall trend:
  - Simulations highlight significant reduction in banks’ vulnerability over time.
  - Contagion across ECCU locally incorporated banks declined over time.
- Drivers of improvement:
  - Capital accumulation: number of undercapitalized banks declined significantly after 2015.
  - Regulatory intervention: ECCB interventions between 2015 and 2017 resulted in mergers/acquisitions of insolvent banks in Antigua and Barbuda and Anguilla.
  - Regulatory changes: response to higher capital requirements of the new Banking Act (2015).
  - Asset quality and provisioning: lower NPLs and higher provisioning rates helped lower banks’ vulnerability.
  - Macroeconomic environment: better economic outlook post-GFC contributed to improved bank resilience.
- Quantitative/systemic observations (as presented in Appendix Figure A2):
  - Sample size depicted in figures includes Bank 1 through Bank 17 labels, with results reported for end-2014, end-2015, end-2016, and mid-2017; the plotted results show declines across:
    - Number of Induced Failures (Number)
    - Failed Capital (Percent)
    - Contagion Rounds (Number)
    - Vulnerability Level (Number)
    - Index of Contagion (Number)
    - Index of Vulnerability (Number)
  - The number of contagion rounds and the number of induced ‘failures’ declined over time for most banks.

### Systemic importance of Bank 12
- Persistent systemic role:
  - Simulation results emphasize the systemic importance of bank 12 across the period.
  - Number of induced ‘failures’ and number of contagion rounds for bank 12 declined over time, but as of mid-2017 bank 12 remained highly contagious.
- Mid-2017 specific outcome:
  - Failure of bank 12 would lead to undercapitalization of 7 other regional banks, amounting to nearly half of the sector in this simulation.
- Index measures:
  - Index of contagion highlights the systemic importance of bank 12.
  - Index of vulnerability highlights banks 3, 6, and 8.

### Policy implications and recommendations
- Maintain and strengthen capital buffers:
  - The results highlight the importance of capital buffers to absorb shocks; capital accumulation materially reduced contagion and vulnerability.
- Continue regulatory vigilance and timely updating of network analysis:
  - Changes in balance sheet strength and bilateral exposures indicate that network analysis should be updated regularly to reassess vulnerabilities and monitor progress.
- Address structural sectoral weaknesses:
  - Given the region’s susceptibility to external shocks and structural deficiencies (including difficulties associated with property valuation and costly and time-consuming foreclosure procedures), continue efforts to reduce vulnerabilities in the ECCU banking sector.
- Sustain supervisory interventions and resolution frameworks:
  - The ECCB’s regulatory interventions between 2015 and 2017 contributed to stability; continued supervisory capacity and effective resolution tools remain important.

*Source: IMF staff estimates and calculations (Appendix 2, "Bank Network Analysis: Time Dimension", wp18162 - Section 4).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2018/wp18162.pdf_