## Non-Performing Loans in the ECCU: Determinants and Macroeconomic Impact — WP/16/229

## Source details

**Canonical URL:** [Non-Performing Loans in the ECCU: Determinants and Macroeconomic Impact — WP/16/229](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16229.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16229.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16229.pdf.json)

---

### I. Overview and context
- As of end-2015, banks’ NPLs stood at 17 percent of total loans; prudential guideline = 5 percent.
- Insolvency and resolution episode:
  - Three banks intervened/resolved (all intervened in 2011; formally resolved between November, 2015 and April, 2016): Antigua and Barbuda Investment Bank (ABIB) — ECCB assumed control in July, 2011; Caribbean Commercial Bank (CCB); National Bank of Anguilla (NBA).
- Financial structure and weaknesses:
  - Bank-dominated financial intermediation; underdeveloped financial markets.
  - Key infrastructure gaps: absence of regional credit bureau and credit-rating agencies; shallow property markets; long collateral resolution periods; outdated foreclosure and bankruptcy frameworks; lack of orderly prior pledge (lien) registries.
- ECCB-led regional initiatives:
  - Establish Eastern Caribbean Asset Management Corporation (ECAMC) to purchase NPLs and expedite collection of NPLs/collateral.
  - Develop regional foreclosure legislation and appraisal guidelines.
  - Establish a regional credit bureau.
  - Strengthen regulation and supervision, including risk-based supervision under new regional legislation.

### II. Data, sample, and methodology
- Bank-level quarterly panel: 1996Q1-2015Q4.
  - Observations = 2,359 (bank-level panel), 34 banks, six countries (Antigua and Barbuda, Dominica, Grenada, St. Kitts and Nevis, St. Lucia, St. Vincent and the Grenadines), 80 quarters.
  - Bank ownership composition: 12 domestic banks, 6 subsidiaries of foreign banks, 15 branches of foreign banks.
  - Foreign banks held 47 percent of total banking system assets (or 109 percent of GDP) and 40 percent of total deposits (or 79 percent of GDP) as of end-2015.
- Dependent variable: logit transformation of the NPL ratio to restrict predictions to [0,1].
- Estimation approaches:
  - Fixed effects, random effects, and GMM-IV (to control for endogeneity of lagged dependent variable and bank-specific endogenous variables).
  - Interaction terms with foreign bank dummy to assess ownership effects.
- Panel VAR (annual, country-aggregate): variables in levels growth rates: ∆npl, ∆credit, ∆FDI, ∆GDP, ∆CPI; exogenous: advanced economies’ real GDP growth and natural disaster dummy; sample 1997–2015, total observations = 114; orthogonalized impulse responses via Cholesky ordering (NPLs, credit, FDI, GDP, CPI); 300 Monte Carlo simulations for confidence intervals.

### III. Stylized facts and descriptive statistics
- NPL levels and composition:
  - NPL ratios elevated across all ECCU countries and above prudential guideline of 5 percent.
  - Sector shares at end-2015: tourism = 18 percent of total loans, construction = 18 percent, personal loans = 43 percent.
  - Sectoral concentration of NPL increases: construction, tourism, agricultural industries.
- Dispersion across banks (2015):
  - Median NPL ratio across individual banks = 12.8 percent; minimum = 4.5 percent; maximum = 24.8 percent.
  - Domestic banks generally have higher NPL ratios than foreign banks (exceptions: Grenada, St. Vincent and the Grenadines).
- Panel VAR descriptive statistics (annual, 1997-2015; panel overall unless country-specified):
  - NPLs to total loans: 10th = 4.3; 50th = 7.9; 90th = 15.3; Average = 8.9; Standard deviation = 4.5; Observations = 114.
  - Non-performing loan growth, percent: 10th = -23.4; 50th = 8.8; 90th = 57.1; Average = 14.4; Standard deviation = 36.3; Observations = 114.
  - Credit growth, percent: 10th = -2.8; 50th = 4.9; 90th = 13.8; Average = 5.6; Standard deviation = 6.8; Observations = 114.
  - Real GDP growth, percent (panel overall): 10th = -2.3; 50th = 2.3; 90th = 7.1; Average = 2.3; Standard deviation = 4.1; Observations = 114.
  - Inflation, percent (panel overall): 10th = 0.0; 50th = 2.0; 90th = 4.8; Average = 2.2; Standard deviation = 2.2; Observations = 114.

### IV. Key empirical findings — determinants (bank-level panel)
- Persistence and crisis-era shift:
  - NPLs display high autocorrelation; lagged NPLs coefficients reported across specifications: 0.828***, 0.788***, 0.858***, 0.835***, 0.917***, 0.901***, 0.803***, 0.771***, 0.834***, 0.816***, 0.900***, 0.894***.
  - Global financial crisis association: 2009-2015 dummy positive and significant with coefficients reported = 0.192***, 0.144***, 0.0951***, 0.186***, 0.139***, 0.105*** in various specifications.
- Macroeconomic drivers:
  - Stronger growth in advanced economies lowers ECCU NPLs (open-economy exposure).
  - Tourism growth (used as proxy for domestic activity) in lagged form positively associated with NPLs; tourism growth t-2 coefficients reported: 0.00224***, 0.00283***, 0.00217***, 0.00260***, 0.00205**, 0.00234**, 0.00238***, 0.00290***, 0.00239***, 0.00280***, 0.00219**, 0.00247**.
  - Credit to the private sector (lagged) tends to reduce NPLs: coefficients include -0.00172***, -0.00134**, -0.00146***, -0.00123**, and in some GMM-IV specs -0.00351*, -0.00333*.
- Bank-specific drivers:
  - Higher bank profitability (ROA) associated with lower NPLs in baseline, but insignificance emerges when ROA endogeneity is controlled in GMM-IV (examples: ROA coefficients -0.0369**, -0.0158, -0.0381***, -0.0257*; some GMM-IV estimates not significant e.g., 0.0405, 0.0851).
  - Loan portfolio composition matters:
    - Construction Loans/Total Loans t-1 coefficients reported include 0.0148, -0.000277, 0.0276***, 0.0246***, 0.0320**, 0.0318**.
    - Household Loans/Total Loans t-1 coefficients reported include 0.162***, 0.162***, 0.0491*, 0.0350, 0.0541**, 0.0439*.
  - Lagged credit growth found to result in lower NPLs in this sample (denominator effect).
  - Expense-to-income ratio, loans-to-assets, loans-to-deposits, and interest rates were not significant determinants in the ECCU sample.
- Ownership effects:
  - Foreign bank dummy negative and significant across baseline specifications: examples -0.0905***, -0.106***, -0.0502**, -0.0611***, -0.103***, -0.117***, -0.0666***, -0.0752***.
  - Interaction analysis: determinants broadly comparable across domestic and foreign banks; foreign banks’ asset quality more responsive to macro developments and profitability in some interaction terms (e.g., "Foreign banks -0.0502*" for ROA interaction), and foreign banks more sensitive to household lending concentration while domestic banks more sensitive to construction lending—though many differences are not statistically significant.

### V. Key empirical findings — macroeconomic dynamics (panel VAR and quarterly sector models)
- Aggregate annual panel VAR:
  - A deterioration in asset quality negatively affects real GDP growth, CPI inflation, and FDI growth; statistical significance strongest for FDI growth.
  - Deterioration in asset quality leads to a decline in credit (response not always statistically significant in aggregate).
  - Baseline elasticity: an increase in 1 percentage point in real GDP growth, holding other factors constant, leads to a decline in NPL growth by about 1.8 percent (driven largely by personal and tourism industries).
- Sectoral (quarterly) models (2005Q1-2015Q4; 263 quarterly observations across sectors):
  - Deterioration in asset quality → statistically significant decline in credit to tourism, agriculture, construction, and manufacturing.
  - Personal loans and trade sectors show weaker negative credit response to asset quality deterioration.
  - Improvements in asset quality in agricultural and construction sectors may have significant positive effects on real GDP growth.
- Variance decomposition (5-year horizon):
  - About 7 percent of NPLs explained by economic performance variables.
  - NPL growth explains about 1 percent of real GDP growth over the medium term (aggregate); sectoral quarterly models show under 4 percent of real GDP growth explained by NPL growth for agricultural and construction sectors.
- Persistence and feedback:
  - NPLs are persistent; shocks to NPLs can have prolonged effects on banking system and credit cycle.
  - Strong macro-financial feedback loops: elevated NPLs reduce credit supply, depressing economic activity, which can further weaken asset quality.

### VI. Policy implications and recommendations
- Financial infrastructure reforms (priority):
  - Establish credit-information-sharing institutions (regional credit bureau, credit-rating agencies).
  - Improve foreclosure and bankruptcy resolution frameworks; reduce collateral resolution periods.
  - Develop orderly prior pledge (lien) registries; deepen property markets and improve collateral valuation (appraisal guidelines).
- ECCB-led interventions:
  - Operationalize the ECAMC to purchase NPLs and expedite workouts and collateral recovery.
  - Enact regional foreclosure legislation to reduce time and cost of resolving problem loans.
  - Strengthen regulation and supervision under new regional legislative framework, including risk-based supervision and improved supervisory capacity.
- Bank-level measures:
  - Improve bank profitability and risk management to enable adequate provisioning.
  - Monitor and limit excessive exposures to volatile sectors (construction, tourism) and to unsecured household lending.
- Growth and structural reforms:
  - Pursue structural reforms to boost growth; stronger economic growth is imperative to improve asset quality and break negative macro-financial feedback loops.

*Source: WP/16/229 — Non-Performing Loans in the ECCU: Determinants and Macroeconomic Impact (Sections 1–5, IMF staff estimates and calculations).*

### Section 1

### Non-Performing Loans in the ECCU: Determinants and Macroeconomic Impact — Section 1

### I. Introduction
- As of end-2015, banks’ NPLs stood at 17 percent of total loans, well above the region’s prudential guideline of 5 percent.
- The deterioration in asset quality has contributed to a marked deterioration in banks’ profitability and overall financial soundness, culminating in the insolvency of three banks in the region and their intervention and resolution by the Eastern Caribbean Central Bank (ECCB). All were intervened in 2011 and formally resolved between November, 2015 and April, 2016:
  - Antigua and Barbuda Investment Bank (ABIB) — ECCB assumed control of ABIB in July, 2011.
  - Caribbean Commercial Bank (CCB).
  - National Bank of Anguilla (NBA).
- NPLs remain elevated across the regional banking system and have prompted bank deleveraging and a significant contraction in credit to the private sector that persists and continues to restrain the region’s economic growth.
- The region’s financial intermediation is bank-dominated; credit from banks is the largest source of financing for businesses and households, while other intermediaries play complementary roles and financial markets are relatively underdeveloped.
- Financial infrastructure weaknesses that have exacerbated and hindered resolution of NPLs include:
  - Lack of a credit bureau and credit-rating agencies, restricting availability of borrower financial history.
  - Shallow property markets limiting banks’ ability to value collateral at market value.
  - Long collateral resolution periods and outdated foreclosure and bankruptcy resolution frameworks.
  - Need for orderly prior pledge (lien) registries and improved screening, collateral use, and monitoring.
- ECCB-led regional initiatives underway:
  - Establishing the Eastern Caribbean Asset Management Corporation (ECAMC) to help improve banks’ asset quality and clean-up balance sheets through purchases of banks' NPLs; once operational, the ECAMC will have comprehensive powers to expedite the collection of NPLs or their collateral.
  - Developing regional foreclosure legislation to reduce time and cost of resolving problem loans.
  - Developing appraisal guidelines to help banks value collateral more effectively.
  - Establishing a regional credit bureau.
  - Strengthening regulation and supervision, including risk-based supervision under new strengthened regional legislation.
- Analytical aims of the paper:
  - Evaluate determinants of NPLs in the ECCU, using a unique bank-level panel dataset with universal coverage of all banks (domestic and foreign) operating in the ECCU.
  - Assess feedback effects between the banking sector (via NPLs) and the real economy using a panel vector autoregression (PVAR) approach.

### II. Stylized facts and recent dynamics
- NPL ratios are elevated across all ECCU member countries and above the prudential guideline of 5 percent.
- The high level of NPLs is in part a legacy of the global financial crisis, which burst the domestic credit cycle that had expanded rapidly prior to the crisis, mainly spurred by tourism activity and related construction.
- Sectoral composition and drivers:
  - The increase in NPLs since the global financial crisis has been driven largely by tourism, construction, and personal loans.
  - At the end of 2015, tourism, construction, and personal loans accounted for 18, 18 and 43 percent of total loans, respectively.
  - The increase in NPLs was concentrated in the construction, tourism and agricultural industries.
- Country-specific episode:
  - In St. Kitts and Nevis, the debt-for-land swap completed between the government and domestic banks contributed to the sharp rise in the NPL ratio over 2011-2015.
- Dispersion across banks:
  - In 2015, the median NPL ratio across individual banks was 12.8 percent; the lowest ratio was 4.5 percent and the highest was 24.8 percent.
  - While NPL ratios are high across both domestic (indigenous) and foreign banks, domestic banks tend to have higher NPL ratios in most ECCU countries (exceptions: Grenada and St. Vincent and the Grenadines where NPL ratio is higher for foreign-owned banks).
- Profitability and ownership patterns:
  - More profitable banks tend to have lower NPL ratios; stronger bank profitability, evidenced by return on assets, is correlated with lower NPL ratios.
  - Among commercial banks, foreign owned banks have generally exhibited stronger profitability and lower NPL ratios relative to domestic banks; stronger profitability of foreign banks is partly attributable to their relatively lower cost of funds compared to domestic banks.
- Provisioning and balance-sheet capacity:
  - Low profitability has restricted banks’ ability to raise provisioning, which remains inadequate throughout the region.
- Credit supply and macro-financial feedback:
  - Elevated NPLs have resulted in adverse macro-financial feedback loops by reducing credit supply: accumulating NPLs force banks to tighten underwriting standards and limit supply of credit to the private sector.
  - NPLs appear to be negatively correlated with credit growth in the ECCU.
  - Following the global financial crisis, credit terms and conditions tightened and banks restricted access to credit to focus on strengthening balance sheets.
  - The reduction in credit supply, combined with weak demand, resulted in a contraction in credit to the private sector beginning in early 2013 that persists.
  - The erosion of bank asset quality and associated contraction in credit have likely reinforced the economic contraction and contributed to adverse macro-financial feedback loops in the region.

### III. Key findings (determinants and macroeconomic links) reported in Section 1
- Determinants of NPLs in the ECCU reflect both macroeconomic and bank-specific factors:
  - Macroeconomic drivers: prolonged recession following the global financial crisis and slow pace of economic recovery reduced borrowers’ income and capacity to repay.
  - Bank-specific drivers: banks with stronger profitability and lower exposure to construction, tourism, and household (personal) loans tend to have lower NPLs.
  - Ownership effect: evidence indicates foreign owned banks systematically have lower NPLs than domestic banks, pointing to meaningful institutional differences across bank practices that affect asset quality.
- Macroeconomic feedbacks:
  - Strong macrofinancial feedback loops exist in the ECCU: elevated NPLs reduce credit supply, which depresses economic activity, which in turn can further weaken asset quality.
  - Strengthened asset quality is important to reverse negative feedback loops and support sustained economic growth; conversely, stronger economic growth is imperative to strengthen asset quality and financial stability.

### IV. Policy implications emphasized in Section 1
- Strengthen financial infrastructure to improve credit risk management and speed NPL resolution:
  - Establish credit-information-sharing institutions (e.g., credit bureau, credit-rating agencies).
  - Improve foreclosure and bankruptcy resolution frameworks and reduce collateral resolution periods.
  - Develop orderly prior pledge (lien) registries and deepen property markets to value collateral at market value.
- Support ECCB-led measures to clean up bank balance sheets and reduce financial risk:
  - Operationalize the ECAMC to purchase NPLs and expedite collection of NPLs or their collateral.
  - Enact regional foreclosure legislation and appraisal guidelines to reduce time, cost, and uncertainty in collateral valuation.
  - Continue strengthening regulation and supervision, including risk-based supervision under new regional legislation.
- Address bank-level vulnerabilities:
  - Improve bank profitability and risk management to enable adequate provisioning.
  - Monitor and limit excessive exposure to volatile sectors such as construction and tourism and to unsecured household lending.

*Source: WP/16/229 — Non-Performing Loans in the ECCU: Determinants and Macroeconomic Impact (Section 1).*

### Section 2

### _wp16229 - Section 2

### Nonperforming Loans and Economic Activity
- High NPLs continue to impede private sector access to credit, as credit growth has remained negative since early 2013.
- The economic recovery in the ECCU after the global financial crisis has occurred despite the continued contraction in credit.
- Empirical correlations illustrated in the source:
  - Negative relationship between NPL Ratio (NPLs in percent of total loans) and Credit Growth (percent, year-over-year) for 1996-2015.
  - Negative relationship between Credit Growth and Real GDP Growth (percent, year-over-year) for 1996-2015.
  - Negative relationship between NPL Ratio and bank profitability (Return on Average Assets = Net Profit before Taxes/Average Assets) for 1996-2015.

### Literature Review — Determinants of Non-Performing Loans
- Determinants identified in the literature include global and domestic macroeconomic indicators, bank-level indicators, and institutional indicators.
- Institutional indicators (e.g., credit bureaus coverage, foreclosure and bankruptcy frameworks, property rights enforcement, specialized courts) are noted as difficult to measure and largely unexplored in cross-country comparisons.

Macroeconomic indicators
- GDP growth: commonly negative correlation with NPLs (anti-cyclical properties of NPLs).
- Unemployment: strong positive relationship with NPLs in several studies.
- Inflation: ambiguous effect — may increase NPLs if wages are sticky, or reduce NPLs if real debt service declines.
- Interest rates: changing interest rates directly affect borrowers’ lending capacity, particularly when variable-rate loans are significant.
- Credit growth: higher credit growth increases credit risk and is often associated with higher NPLs.
- Exchange rate depreciation: ambiguous effect — can increase NPLs in economies with large foreign-currency lending, or reduce NPLs by improving exporters’ debt-servicing.
- Asset prices (house prices, stock indices): may influence NPLs through wealth effects; effect of stock indices is not obvious.

Bank-specific indicators
- Capital adequacy (equity-to-assets): often negatively correlated with NPLs (moral hazard hypothesis).
- Profitability (ROE, ROA): higher profitability often associated with lower NPLs.
- Efficiency measures (cost-to-income, expense-to-asset), bank size, net interest margin, credit growth, portfolio composition: found to have explanatory power in various studies.
- Excessive lending (loans-to-assets): tends to correlate with higher NPLs.

Global variables
- VIX (global financial volatility) used as proxy for global risk; higher VIX associated with higher NPLs in some studies.
- Oil prices: effects differ between oil exporters and importers.

### Macroeconomic Spill-overs of NPLs
- Many studies use panel VAR approaches to analyze feedbacks between banking sector asset quality and economic performance.
- Mechanisms:
  - Real economy → NPLs: weaker macro conditions reduce borrowers’ capacity to repay.
  - NPLs → real economy: mainly through the credit supply channel; non-credit channels (e.g., debt overhang effects on investment) also suggested.
- Cross-study approaches:
  - Single-country, bank-level studies (e.g., Love and Ariss (2013) on Egypt).
  - Cross-country, aggregate country-level studies (e.g., Nkusu (2011), Klein (2013)).
  - Cross-country, bank-level panel studies (e.g., Espinoza and Prasad (2010) for GCC banks).
- Key cross-study findings:
  - Deterioration in macro environment (slower growth, higher unemployment, falling asset prices) associated with rising NPLs.
  - NPLs have been found to reduce credit, inflation, and real GDP growth, and to raise unemployment in some studies.

### This paper’s contribution
- Uses a unique dataset of bank-specific and country aggregate series for independent ECCU economies spanning 1996 to 2015.
- Quarterly, bank-level granularity with universal coverage of all banks operating in the independent ECCU countries across the sample period.
- Employs panel VAR on annual country-aggregate data to use low-frequency macro variables while retaining cross-country perspective.

### Data and Methodology
- Dataset: quarterly bank-level panel spanning 1996Q1-2015Q4.
- Sample size and composition:
  - 2,359 observations
  - 34 banks
  - six countries: Antigua and Barbuda, Dominica, Grenada, St. Kitts and Nevis, St. Lucia, and St. Vincent and the Grenadines
  - 80 quarters (1996Q1-2015Q4)
  - Of 34 banks: 12 domestic banks, 6 subsidiaries of foreign banks, and 15 branches of foreign banks.
  - Foreign banks predominantly branches and subsidiaries of Canadian banks; a bank from Trinidad and Tobago operates a subsidiary in Grenada.
  - As of end-2015, foreign banks held 47 percent of total banking system assets (or 109 percent of GDP) and 40 percent of total deposits in the banking system (or 79 percent of GDP).
- Dependent variable transformation:
  - The logit transformation of the NPL ratio (ݏܮܲܰ
௜,௝.௧) is used to ensure the dependent variable spans the interval ሾെ∞,൅∞ሿ and is distributed symmetrically, and that predicted values are non-negative and between 0 and 1.
- Model specification elements:
  - Dependent variable explained by its lag, global variables (݈ܾܽ݋݈ܩ
௧), country-specific variables (ݕݎݐ݊ݑ݋ܥ
௜,௧), and bank-level variables (݇݊ܽܤ
௜,௝,௧).
  - Controls include individual country effects and a foreign bank dummy (ܦܾ݇݊ܽ݊݃݅݁ݎ݋ܨ
௝) and country dummy (ܦݕݎݐ݊ݑ݋ܥ
௜).
  - Institutional factors (e.g., quality of bank supervision and financial regulation) are not included as explanatory variables given harmonized classification of NPLs across the currency union under ECCB supervision.
- Estimation techniques:
  - Fixed effects to control for unobserved heterogeneity across banks.
  - Random effects to allow inclusion of country-specific and foreign-bank dummy variables.
  - Generalized Method of Moments with instrumental variables (GMM-IV) to control for endogeneity of the lagged dependent variable and bank-specific endogenous variables; global and country-specific variables treated as strictly exogenous.
  - Lagged values of endogenous variables and bank-specific dependent variables treated as instruments in GMM-IV.
  - Rationale: Because T = 80, the Nickell (1981) dynamic panel bias (≈ 1/T) is relatively small (1.25 percent); system GMM is not used because it requires the number of cross-sections to exceed the time dimension, so GMM-IV is employed instead.
- Ownership impact:
  - The impact of bank ownership structure is tested by interacting the foreign bank dummy with other explanatory variables in the benchmark model to assess differential responses of domestic and foreign banks’ NPLs to macroeconomic and bank-level changes.
  - Interaction specifications were also estimated with GMM-IV.

### Results
- Baseline model findings:
  - Macroeconomic developments are important determinants of bank asset quality.
  - Both global and country-specific macro developments affect NPLs in the ECCU.
  - Stronger growth in advanced economies lowers NPLs in the ECCU, consistent with high openness of the small ECCU economies.
  - Tourism growth (used as proxy for domestic activity due to lack of quarterly GDP and unemployment data) is found to increase NPLs, possibly reflecting the inherent riskiness of lending to the tourism sector; this result holds even when controlling for lending concentration to tourism.
  - Results are robust to fixed effects, random effects, and GMM-IV estimation.
- Bank-specific findings:
  - Higher bank profitability (return on assets) is found to lower NPLs in baseline estimation; however, this relationship does not hold when return on assets endogeneity is controlled for in the GMM-IV estimation.
  - Loan portfolio composition matters: higher concentration of lending to households and the construction sectors is associated with worse bank asset quality.
  - Lagged credit growth is found to result in lower NPLs in this sample, suggesting that the impact of credit growth on the denominator outweighs the impact on the numerator of the NPL ratio.
  - Contrary to expectations and some prior studies, expense-to-income ratio, loans-to-assets, loans-to-deposits, and interest rates were not found to have a significant impact on bank asset quality in the ECCU sample.

*Source: _wp16229 - Section 2*

### Section 3

### _wp16229 - Section 3

### Key empirical findings on determinants of NPLs
- NPLs exhibit relatively high auto-correlation, indicating that a shock to NPLs could have a prolonged effect on the ECCU banking system.
- The global financial crisis is associated with a structural increase in NPLs: the 2009-2015 dummy is positive and significant (2009-2015 Dummy = 0.192***, 0.144***, 0.0951***, 0.186***, 0.139***, 0.105*** in various specifications).
- Lagged NPLs (NPLs t-1) display strong persistence across specifications:
  - Examples of estimated coefficients: 0.828***, 0.788***, 0.858***, 0.835***, 0.917***, 0.901***, 0.803***, 0.771***, 0.834***, 0.816***, 0.900***, 0.894*** (standard errors reported in the source).
- Tourism growth has a consistent positive association with NPLs in lagged form:
  - Tourism Growth t-2 coefficients include 0.00224***, 0.00283***, 0.00217***, 0.00260***, 0.00205**, 0.00234**, 0.00238***, 0.00290***, 0.00239***, 0.00280***, 0.00219**, 0.00247** (with standard errors in the source).
- Credit to the private sector (lagged) tends to reduce NPLs in several specifications:
  - Credit to the Private Sector t-1 coefficients include -0.00172***, -0.00134**, -0.00146***, -0.00123**, and in some GMM-IV specifications -0.00351*, -0.00333*.
- Return on Assets (ROA) shows mixed results:
  - Examples: -0.0369**, -0.0158, -0.0381***, -0.0257*; some GMM-IV estimates are not significant (e.g., 0.0405, 0.0851).
- Sectoral loan composition matters in some specifications:
  - Construction Loans/Total Loans t-1: coefficients include 0.0148, -0.000277, 0.0276***, 0.0246***, 0.0320**, 0.0318**.
  - Household Loans/Total Loans t-1: coefficients include 0.162***, 0.162***, 0.0491*, 0.0350, 0.0541**, 0.0439*.

### Ownership structure and differences between domestic and foreign banks
- Foreign bank dummy is negative and significant across baseline specifications:
  - Examples: Foreign Banks' Dummy = -0.0905***, -0.106***, -0.0502**, -0.0611***, -0.103***, -0.117***, -0.0666***, -0.0752***.
- Interaction analysis (Tables 2 and 3) indicates:
  - Determinants of NPLs are broadly comparable across domestic and foreign banks, with some differences in sensitivity.
  - Asset quality of foreign banks appears more responsive to macroeconomic developments and to banks’ profitability.
    - In interaction results, Return on Assets t-1 for foreign banks shows a negative adjustment: "Foreign banks -0.0502*" (Table 2 interaction term).
  - Foreign banks’ asset quality is more responsive to the concentration of lending to households; domestic banks’ asset quality is more responsive to construction lending—consistent with differing lending concentrations by bank type; however, differences are often not statistically significant (Table 3 interaction terms reported, e.g., Construction Loans/Total Loans * Foreign bank's dummy = 0.0612 (Table 3) with standard errors provided in source).
- The point estimates for ROA differences between bank types are not statistically different (Table 3).

### Panel estimation summary statistics and model fit
- Sample sizes and fit (selected reported values):
  - Observations: 2,418; 2,415; 2,361; 2,356 across various specifications.
  - R-squared reported in some specifications: 0.732, 0.740, 0.823, 0.825, 0.734, 0.741, 0.823, 0.823.
  - Number of Bank_code entries shown as 35353535 and 34343434 (as reported in the source table).
- Significance notation used in tables: *** p<0.01, ** p<0.05, * p<0.1.
- Robust standard errors are reported in parentheses in all tables.

### Panel VAR approach to dynamics and macroeconomic feedback
- Model specification:
  - Vector Y_i,t includes five endogenous variables in levels growth rates: ∆npl_i,t (growth in level NPLs), ∆credit_i,t (growth in credit to the private sector), ∆FDI_i,t (growth in foreign direct investment series), ∆GDP_i,t (real GDP growth), and ∆CPI_i,t (average annual CPI inflation in percent).
  - Exogenous covariates X_i,t include advanced economies’ real GDP growth and a natural disaster dummy (value = one for the year a disaster occurred, zero otherwise).
  - Fixed effects u_i capture country-specific effects; idiosyncratic errors e_it are included.
- Estimation details:
  - Forward mean-differencing (Helmert procedure) is used to remove panel fixed effects while preserving orthogonality between lagged regressors and transformed variables.
  - Lagged regressors are used as instruments and coefficients are estimated by GMM methodology.
  - 300 Monte Carlo simulations are used to generate confidence intervals for impulse responses; impulse responses are orthogonalized.
  - Orthogonal shocks are identified using Cholesky decomposition with ordering: NPLs first, then credit growth, FDI growth, GDP growth, and CPI inflation.
  - Assumptions: FDI growth, GDP growth, and inflation affect delinquent loans only with a lag; non-performing loans have contemporaneous effects on economic activity largely through credit.
  - Results are broadly robust to alternative ordering of variables.
- Data coverage:
  - Panel estimated for 6 independent ECCU economies over the period 1997 to 2015.
  - Total observations for panel VAR: 114.
  - Annual frequency used due to availability of macroeconomic indicators.
- Descriptive statistics (NPL-related):
  - Distribution of NPLs to total loans clustered around median of 7.9 percent of total loans with standard deviation of 4.5 percentage points.
  - Median growth rate of non-performing loans is about 8.8 percent, with a standard deviation of about 36 percent.
  - Higher NPLs toward the end of the period largely reflect the global financial crisis, which increased loan delinquency rates across the ECCU.

### Economic interpretation and expected sign of macro variables
- Real GDP growth: higher real GDP growth is expected to raise incomes and lower delinquency rates (negative relationship with NPL growth).
- FDI inflows: higher FDI, often concentrated in tourism, stimulate activity and raise incomes, leading to lower NPLs.
- Inflation: ambiguous effect on NPLs
  - Higher inflation can erode the real value of borrowers’ debt service and lower NPLs (negative sign).
  - Alternatively, if wages do not keep up with rising inflation, borrowers’ ability to repay declines and NPLs rise (positive sign).

*Source: _wp16229 - Section 3 (selected tables and text excerpts).*

### Section 4

### _wp16229 - Section 4

### Correlations, stationarity, and cointegration
- Simple correlations in the sample indicate:
  - NPLs are negatively correlated with inflation, FDI, and real GDP growth rates.
  - Correlation with credit growth remains positive.
- Unit-root and integration testing:
  - Preference was given to the Fisher-ADF and PP unit root tests (do not require balanced samples; based on overall test statistic generated using individual unit-root tests).
  - Results indicate that all panel VAR variables are stationary of order I(0).
  - Johansen’s trace and maximum-eigenvalue tests support the presence of cointegrating relationships in the models, providing basis for implementation of panel VAR analysis.

### Panel VAR specifications and data
- Annual and quarterly models estimated:
  - Quarterly models estimated separately for agricultural, tourism, construction, manufacturing, trade, and personal loans sectors.
  - Quarterly models period: 2005Q1-2015Q4.
  - Total of 263 quarterly observations.
- Endogenous variables in quarterly models:
  - First difference of the NPL ratios by sector.
  - Credit growth by sector.
  - Real GDP growth.
  - CPI inflation.
- Exogenous variables in quarterly models:
  - Real GDP growth of advanced economies.
  - Natural disasters dummy.
- Quarterly real GDP proxy:
  - Quarterly real GDP estimates imputed using annual real GDP for the ECCU economies and quarterly real GDP data for the United States, based on close correlation of real GDP for the ECCU and the US.
- NPL definitions:
  - Baseline annual model includes growth rate of NPLs.
  - Alternative specification uses first difference of NPL-to-total loans ratio (used in some figures and quarterly models).
  - Quarterly models include NPLs defined as the first difference of the sector-specific NPL-to-total loans ratio (preference given to this specification because small sector sizes introduce excessive volatility when growth rates are applied).

### Main empirical findings (Impulse Response Functions)
- A shock to NPL growth has implications for economic activity and the credit cycle.
- Aggregate (annual) model results:
  - A deterioration in asset quality has a negative effect on real GDP growth, CPI inflation, and FDI growth, but the results of the aggregate model are statistically significant only for FDI growth.
  - A deterioration in asset quality leads to a decline in credit, although the response is not statistically significant in aggregate.
- Sector-specific (quarterly) model results:
  - Lower NPLs in the agricultural and construction sectors may result in a significant effect on real GDP growth.
  - Deterioration in asset quality leads to a statistically significant decline in credit to tourism, agriculture, construction, and manufacturing industries.
  - The negative relationship between asset quality deterioration and credit is much weaker for personal loans and trade industry.
- Macroeconomic performance effects on asset quality:
  - Stronger economic performance leads to a statistically significant decline in NPL growth.
  - Baseline model: an increase in 1 percentage point in real GDP growth, holding other factors constant, leads to a decline in NPL growth by about 1.8 percent.
  - The result appears driven largely by the personal and tourism industries, which broadly comprise the majority of NPLs (about 61 percent at end-2015).
- FDI and feedback loops:
  - Stronger economic activity leads to stronger FDI (not statistically significant in baseline), and a boost to FDI leads to a significant increase in GDP growth.
  - The positive effect of FDI on economic growth is considerably stronger than that of domestic credit.
- Persistence of credit shocks:
  - The persistent nature of credit growth implies that a shock to credit growth in the initial period lingers longer than in the case of other variables.

### Variance decomposition results
- Over a 5-year horizon:
  - About 7 percent of NPLs is explained by economic performance variables.
  - NPL growth explains about 1 percent of real GDP growth over the medium term (aggregate).
  - In complementary quarterly models, the portion of real GDP growth explained by NPL growth is estimated to be marginally higher (under 4 percent) for agricultural and construction sectors.

### Sectoral heterogeneity
- Impact by sector varies markedly:
  - Tourism, agriculture, manufacturing, and construction: deterioration of asset quality → significant and prolonged decline in credit growth.
  - Personal loans and trade: weaker negative credit response to asset quality deterioration.
  - Improvement in asset quality has potential to boost real GDP growth particularly through agricultural and construction industries.
  - Real GDP growth reduces NPL growth notably in personal and tourism sectors (likely via lower unemployment and increased disposable income, though labor market data are excluded).

### Conclusions and policy recommendations
- Drivers of NPL deterioration:
  - Attributed to both macroeconomic conditions and bank-specific factors in the ECCU.
  - More profitable banks and banks with lower exposure to volatile construction and tourism sectors and household loans tend to have lower NPLs.
  - Evidence that foreign owned banks systematically have lower NPLs than domestic banks, possibly reflecting institutional differences, risk management practices, NPL recovery strategies, and scale of operations.
- Policy priorities to strengthen asset quality and break negative macro-financial feedback loops:
  - Operationalize the ECAMC established by the region as a priority to facilitate rapid workout of bad assets and purchase NPLs to clean up bank balance sheets.
  - Continue to modernize foreclosure laws to support resolution of NPLs.
  - Enhance supervision and regulation:
    - Implement new regional legislative framework for bank regulation, supervision and resolution.
    - Complement framework with improvements in supervisory capacity to ensure proper classification of loans, improve collateral valuation, and strengthen on-and off-site supervision.
  - Establish a regional credit bureau swiftly to support banks’ assessment of credit risks and support a revival of bank credit.
  - Pursue structural reforms to boost growth, as stronger economic growth is imperative to improvement in asset quality.

*Source: IMF staff estimates and calculations (extracted from _wp16229 - Section 4).*

### Section 5

### _wp16229 - Section 5

### Data and Sources
- Variable descriptors, frequency, time coverage, units, transformations, and sources:
  - Non-performing loans
    - Annual; 1997-2015; Percent; Year-on-year growth rate; ECCB
    - Annual; 1997-2015; Percent; First difference of NPL ratio; ECCB
    - Quarterly; 2005Q1-2015Q4; Unit; ECCB
  - By sector (Quarterly; 2005Q1-2015Q4; Percent; Annual difference of NPL ratio; ECCB)
    - Tourism
    - Agriculture
    - Construction
    - Personal loans
    - Manufacturing
    - Trade
  - Credit to private sector
    - Annual; 1997-2015; Percent; Year-on-year growth rate; ECCB
    - By sector (Quarterly; 2005Q1-2015Q4; Percent; Year-on-year growth rate; ECCB)
      - Tourism
      - Agriculture
      - Construction
      - Personal loans
      - Manufacturing
      - Trade
  - Total Loans
    - Quarterly; 2005Q1-2015Q4; Unit; ECCB
  - Return on Assets
    - Quarterly; 2005Q1-2015Q4; Percent; ECCB
  - FDI inflows
    - Annual; 1997-2015; Percent; Year-on-year growth rate; ECCB; IMF, World Economic Outlook
  - Real GDP
    - Annual; 1997-2015; Percent; Year-on-year growth rate; ECCB; IMF, World Economic Outlook
    - Quarterly; 2005Q1-2015Q4; Percent; Year-on-year growth rate; ECCB; IMF, World Economic Outlook
    - Note: Quarterly real GDP estimates are imputed using annual real GDP for the ECCU economies and quarterly real GDP data for the United States, given the close correlation of real GDP for the ECCU and the US.
  - CPI inflation
    - Annual; 1997-2015; Percent; Year-on-year growth rate; ECCB; IMF, World Economic Outlook
    - Quarterly; 2005Q1-2015Q4; Percent; Year-on-year growth rate; ECCB; IMF, World Economic Outlook
  - Advanced economies real GDP
    - Annual; 1997-2015; Percent; Year-on-year growth rate; IMF, World Economic Outlook
    - Quarterly; 2005Q1-2015Q4; Percent; Year-on-year growth rate; IMF, World Economic Outlook
  - US real GDP
    - Quarterly; 2005Q1-2015Q4; Percent; Year-on-year growth rate; IMF, World Economic Outlook
  - Tourism Growth
    - Quarterly; 2005Q1-2015Q4; Percent; Year-on-year growth rate; Caribbean Tourism Organization
  - Natural disasters dummy
    - Annual; 1997-2015; Unit; Natural disaster occurrence =1; EM-DAT
  - Foreign bank dummy
    - Quarterly; 2005Q1-2015Q4; Unit; Foreign bank =1; ECCB

### Panel VAR: Descriptive Statistics (Annual Data Model: 1997-2015)
- Panel coverage and notes:
  - 120 observations for annual data for overall specification; 480 observations for quarterly data for sector specifications.
- Non-performing loan, % of total loans (percentiles, average, standard deviation, observations)
  - Panel (overall): 10th = 4.3; 50th = 7.9; 90th = 15.3; Average = 8.9; Standard deviation = 4.5; Observations = 114
  - Antigua and Barbuda: 10th = 5.7; 50th = 10.4; 90th = 14.3; Average = 10.4; Standard deviation = 3.3; Observations = 19
  - Dominica: 10th = 5.2; 50th = 8.7; 90th = 22.0; Average = 10.9; Standard deviation = 5.6; Observations = 19
  - Grenada: 10th = 2.9; 50th = 5.8; 90th = 13.8; Average = 6.7; Standard deviation = 3.6; Observations = 19
  - Saint Lucia: 10th = 6.1; 50th = 11.3; 90th = 18.2; Average = 11.8; Standard deviation = 4.8; Observations = 19
  - St Kitts and Nevis: 10th = 4.1; 50th = 4.8; 90th = 13.3; Average = 6.9; Standard deviation = 3.8; Observations = 19
  - St Vincent and the Grenadines: 10th = 3.9; 50th = 5.6; 90th = 9.4; Average = 6.4; Standard deviation = 2.2; Observations = 19
- Non-performing loan growth, percent
  - Panel (overall): 10th = -23.4; 50th = 8.8; 90th = 57.1; Average = 14.4; Standard deviation = 36.3; Observations = 114
  - Antigua and Barbuda: 10th = -45.9; 50th = 17.7; 90th = 65.3; Average = 18.0; Standard deviation = 53.0; Observations = 19
  - Dominica: 10th = -23.7; 50th = 8.3; 90th = 76.1; Average = 14.2; Standard deviation = 37.9; Observations = 19
  - Grenada: 10th = -35.2; 50th = 10.4; 90th = 76.3; Average = 10.7; Standard deviation = 32.3; Observations = 19
  - Saint Lucia: 10th = -15.0; 50th = 8.1; 90th = 51.5; Average = 13.8; Standard deviation = 24.6; Observations = 19
  - St Kitts and Nevis: 10th = -27.1; 50th = 14.5; 90th = 76.9; Average = 15.3; Standard deviation = 35.6; Observations = 19
  - St Vincent and the Grenadines: 10th = -11.5; 50th = 8.4; 90th = 68.6; Average = 14.7; Standard deviation = 32.8; Observations = 19
- Credit growth, percent
  - Panel (overall): 10th = -2.8; 50th = 4.9; 90th = 13.8; Average = 5.6; Standard deviation = 6.8; Observations = 114
  - Antigua and Barbuda: 10th = -5.2; 50th = 5.0; 90th = 17.3; Average = 4.9; Standard deviation = 8.7; Observations = 19
  - Dominica: 10th = -2.8; 50th = 6.0; 90th = 9.5; Average = 4.2; Standard deviation = 4.5; Observations = 19
  - Grenada: 10th = -5.1; 50th = 4.6; 90th = 17.1; Average = 6.3; Standard deviation = 7.4; Observations = 19
  - Saint Lucia: 10th = -6.7; 50th = 6.2; 90th = 22.9; Average = 6.7; Standard deviation = 9.1; Observations = 19
  - St Kitts and Nevis: 10th = -1.1; 50th = 5.9; 90th = 12.6; Average = 6.2; Standard deviation = 5.1; Observations = 19
  - St Vincent and the Grenadines: 10th = 1.0; 50th = 3.5; 90th = 13.8; Average = 5.2; Standard deviation = 4.6; Observations = 19
- FDI growth, percent
  - Panel (overall): 10th = -39.0; 50th = 2.0; 90th = 104.2; Average = 22.4; Standard deviation = 69.4; Observations = 114
  - Antigua and Barbuda: 10th = -51.7; 50th = 9.1; 90th = 152.5; Average = 30.3; Standard deviation = 73.9; Observations = 19
  - Dominica: 10th = -42.6; 50th = 17.5; 90th = 103.6; Average = 17.0; Standard deviation = 56.9; Observations = 19
  - Grenada: 10th = -41.1; 50th = -7.3; 90th = 97.5; Average = 23.1; Standard deviation = 72.9; Observations = 19
  - Saint Lucia: 10th = -35.4; 50th = -0.3; 90th = 199.0; Average = 33.2; Standard deviation = 107.1; Observations = 19
  - St Kitts and Nevis: 10th = -26.5; 50th = 8.9; 90th = 66.7; Average = 13.5; Standard deviation = 35.2; Observations = 19
  - St Vincent and the Grenadines: 10th = -39.0; 50th = 1.7; 90th = 116.8; Average = 17.1; Standard deviation = 56.3; Observations = 19
- Real GDP growth, percent
  - Panel (overall): 10th = -2.3; 50th = 2.3; 90th = 7.1; Average = 2.3; Standard deviation = 4.1; Observations = 114
  - Antigua and Barbuda: 10th = -8.5; 50th = 3.4; 90th = 7.2; Average = 2.4; Standard deviation = 5.5; Observations = 19
  - Dominica: 10th = -2.8; 50th = 0.7; 90th = 6.4; Average = 1.7; Standard deviation = 3.1; Observations = 19
  - Grenada: 10th = -4.0; 50th = 3.4; 90th = 11.8; Average = 3.2; Standard deviation = 5.2; Observations = 19
  - Saint Lucia: 10th = -1.7; 50th = 0.5; 90th = 8.2; Average = 1.5; Standard deviation = 3.3; Observations = 19
  - St Kitts and Nevis: 10th = -3.8; 50th = 3.9; 90th = 7.3; Average = 2.9; Standard deviation = 4.1; Observations = 19
  - St Vincent and the Grenadines: 10th = -2.0; 50th = 2.1; 90th = 6.1; Average = 2.3; Standard deviation = 2.7; Observations = 19
- Inflation, percent
  - Panel (overall): 10th = 0.0; 50th = 2.0; 90th = 4.8; Average = 2.2; Standard deviation = 2.2; Observations = 114
  - Antigua and Barbuda: 10th = -0.2; 50th = 1.9; 90th = 3.5; Average = 1.9; Standard deviation = 1.4; Observations = 19
  - Dominica: 10th = 0.0; 50th = 1.4; 90th = 3.2; Average = 1.6; Standard deviation = 1.6; Observations = 19
  - Grenada: 10th = -0.8; 50th = 2.1; 90th = 4.3; Average = 2.0; Standard deviation = 2.1; Observations = 19
  - Saint Lucia: 10th = -0.3; 50th = 2.8; 90th = 5.3; Average = 2.5; Standard deviation = 1.9; Observations = 19
  - St Kitts and Nevis: 10th = 0.7; 50th = 2.2; 90th = 8.5; Average = 3.1; Standard deviation = 2.8; Observations = 19
  - St Vincent and the Grenadines: 10th = 0.1; 50th = 1.0; 90th = 7.0; Average = 2.0; Standard deviation = 2.7; Observations = 19
- Sources for descriptive statistics: ECCB, National Authorities, and IMF staff estimates and calculations.

### Panel VAR: Johansen Fisher Panel Cointegration Test
- Note: Figures indicate the selected number of cointegration relations at the 5% significance level.
- Test specifications and selected numbers of cointegration relations (at 5%):
  - NPL growth, Credit growth, FDI growth, Real GDP growth, Inflation:
    - No intercept, No trend in CE: 3
    - Intercept, No trend in CE: 2
    - No intercept, No trend in CE (alternative column): 2
    - Intercept, Trend in CE: 3
    - Intercept, Trend in CE (alternative column): 3
  - Assumptions on cointegrating test specification listed:
    - No trend in data
    - Linear trend in data
    - Quadratic trend in data

### References (selected as listed)
- Abrigo M., Love I., (2015). Estimation of Panel Vector Autoregression in Stata: a Package of Programs.
- Beck, R., Jakubik, P., Piloiu, A. (2015). Key Determinants of Non-Performing Loans: New Evidence from a Global Sample.
- Belgrave, A.; Guy, K.; Jackman, M. (2012). Industry Specific Shocks and Non-Performing Loans in Barbados. The Review of Finance and Banking. Volume 04, Issue 2, p.123-133.
- Espinoza, R., & Prasad, A. (2010, October). Nonperforming Loans in the GCC Banking System and their Macroeconomic Effects. International Monetary Fund.
- Holtz-Eakin, D., W. Newey, and H.S. Rosen. 1988. "Estimating vector auto-regressions with panel data." Econometrica 56: 1371-95.
- International Monetary Fund. (2016). Eastern Caribbean Currency Union: Staff Report for the 2016 Common Policies Discussion.
- Klein, N. (2013). Non-performing Loans in CESEE: Determinants and Impact on Macroeconomic Performance. IMF Working Paper, 01-27.
- Louzis, D. P., Vouldis, A. T., & Metaxas, V. L. (2011, October 13). Macroeconomic and bank-specific determinants of non-performing loans in Greece: A comparative study of mortgage, business and consumer loan portfolios. Journal of Banking and Finance, 01-16.
- Love, I., & Ariss, R. T. (2013). Macro-Financial Linkages in Egypt: A Panel Analysis of Economic Shocks and Loan Portfolio Quality. IMF Working Paper, 01-40.
- Makri, V., Tsagkanos, A., & Bellas, A. (2014). Determinants of Non-Performing Loans: The Case of Eurozone. (2), pp. 193-206.
- Nkusu, M. (2011). Non-performing Loans and Macrofinancial Vulnerabilities in Advanced Economies. IMF WP/11/161.
- Skarica, B. (2014). Determinants of Non-Performing Loans in Central and Eastern European Countries. Financial Theory and Practice, 38(1), pp. 37-59.
- Sorge, M. (2004). Stress-testing Financial Systems: An Overview of Current Methodologies. BIS Working Papers, 01-41.
- Tan, T. B. (2012). Determinants of Credit Growth and Interest Margins in the Philippines and Asia. IMF Working Paper, 01-25.

*Source: _wp16229 - Section 5 (PDF).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16229.pdf_