## ANNEX I. DATA DESCRIPTION

## Source details

**Canonical URL:** [ANNEX I. DATA DESCRIPTION](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021276-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021276-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021276-print-pdf.pdf.json)

---

### Background and context
- ECCU membership: Antigua and Barbuda, Dominica, Grenada, St. Kitts and Nevis, Saint Lucia, St. Vincent and the Grenadines; plus U.K. territories Anguilla and Montserrat (territories excluded from empirical sovereign sample).
- COVID-19 impact: tourist arrivals plummeted by 70 percent and cruise ship travel completely halted, pushing the ECCU economy into a deep recession in 2020.
- 2020 outcomes: fiscal positions deteriorated sharply with public debt rising steeply; external accounts deteriorated, while official foreign reserve positions “held up relatively well” partly reflecting increased official financing.

### Currency boards, the ECCB, and the backing ratio
- ECCB established: October 1983; operates a quasi-currency board managing a common pool of reserves for the ECCU.
- Operational target: high foreign reserve cover—the “backing ratio” = ECCB foreign assets as percent of demand liabilities.
- Legal/operational thresholds:
  - ECCB Agreement Act (1983) minimum backing ratio: 60 percent.
  - Operational target: 80 percent.
- Historical practice: backing ratio maintained at 95–100 percent over the past two decades; ECCB limits credit extension to governments and banks to preserve backing.
- No observed twin crisis (currency crisis + systemic banking crisis) episode in the ECCU.

### Crisis definitions (Laeven and Valencia, 2020)
- Currency crises:
  - Large exchange rate depreciation (30 percent or more) within a year against the US dollar.
  - If two consecutive years feature large depreciations, the second year must be 10 percentage points larger than the previous year to count as a separate crisis.
- Banking crises:
  - Severe financial distress requiring significant policy interventions that can last multiple years.
  - Distress indicators: rise in NPLs, bank losses, bank runs, bank asset foreclosures.
  - Policy interventions include fiscal expenditure for bank nationalizations, liquidity support, government guarantees, asset freezes, bank holidays.
- Laeven and Valencia (2020) operational thresholds for banking crises (must satisfy both requirements listed):
  - 1. Significant financial distress: at least one of:
    - non-performing loans rise above 20 percent of total loans
    - bank closures of at least 20 percent of banking system assets
    - fiscal costs of restructuring the banking sector exceed 5 percent of GDP
  - 2. Significant policy interventions: at least three of:
    - deposit freezes and/or bank holidays
    - significant bank nationalizations
    - bank restructuring fiscal costs (at least 3 percent of GDP)
    - extensive liquidity support (at least 5 percent of deposits and liabilities to non-residents)
    - significant guarantees put in place
    - significant asset purchases (at least 5 percent of GDP)

### Dataset, sample coverage, and time horizon
- Original Laeven and Valencia (2020) sample: 165 countries from 1970 to 2017, including 26 small states.
- Expanded sample: authors add 14 extra small states (none with crises), producing 173 countries, including 40 small states; six ECCU members included; Anguilla and Montserrat excluded from sovereign sample.
- Common macroeconomic indicator availability: 1981–2017 for all countries; empirical analysis restricted to 1981–2017.

### Empirical strategy
- Model: binomial Logit estimating Pr(Crisis_i,t = 1) = e^{βX_i,t + ε_i,t} / (1+ e^{βX_i,t + ε_i,t}).
- Timing conventions:
  - Country-specific variables in X_i,t are one-year lags.
  - Global variables use the concurrent year value.
  - Banking crises: only the year in which a banking crisis begins is included; currency crises treated as one-year events.
- Estimation: Maximum Likelihood with robust standard errors.
- Baseline explanatory variables: credit/deposit ratio, banking sector net foreign assets (NFA) as share of GDP, growth rate of credit/GDP, real interest rate, M2/GDP ratio (M2/Reserves presented in results), growth rate of terms of trade, log real GDP per capita, real GDP growth.
- Expanded sets:
  - Fiscal: public debt/GDP ratio, fiscal balance/GDP ratio.
  - Financial: banking sector equity/GDP, banking sector assets/GDP, banking sector equity/assets, central bank’s share of foreign assets to total assets (“backing ratio”).
  - Exchange rate regime indicators: dollarized economy, currency board, currency union, pegged currency.
  - Global: world real GDP growth rate, global uncertainty index (VIX), US Federal Reserve funds rate, 10 year–3 month US government bond yield spread.
- Data sources summarized in Annex I (Haver Analytics, IMF WEO, IFS, Global Financial Data, WDI, MFS, Fed H.15, Cboe/VIX).

### Descriptive evidence (dynamics around crises)
- Banking crises (Figure 1):
  - Crisis-sample medians diverge markedly from non-crisis medians in years preceding banking crises for most variables.
  - Exceptions pre-event: real GDP growth and public debt/GDP exhibit similar medians across crisis and non-crisis groups.
  - Post-crisis: corrections in credit/deposit ratio, M2/reserves, credit/GDP growth; impacts on real GDP growth and public debt/GDP.
- Currency crises (Figure 2): similar systematic pre-crisis differences as banking crises.
- Global variables (Figure 3): VIX and world real GDP growth correlate with total number of crises in a year, notably during the 2008 GFC.

### Main empirical contributions and key findings
- Method: Logit model (building on Caggiano et al. (2016)) using Laeven and Valencia (2020) database, focusing on fixed exchange rate regimes and small states.
- Core result: Higher backing ratios are strongly and robustly associated with lower probabilities of banking and currency crises.
- Heterogeneity: the backing ratio’s association with lower crisis probability is quantitatively more important for fixed exchange rate regimes and small states.
- ECCU assessment: model implies the ECCU had low predicted probability of banking and currency crises in “normal times”; probabilities rose somewhat during global downturns (e.g., 2008 GFC), indicating vulnerability to exogenous shocks.
- Role of global and country-specific conditions: market uncertainty (VIX), world growth, and US monetary policy have predictive power for crisis likelihood.

### Econometric results — Banking crises (selected coefficients and fit)
- Selected coefficient estimates (robust standard errors in parentheses; significance: *** p<0.01, ** p<0.05, * p<0.1):
  - Credit/Deposit t-1: 0.00650*** (0.00174)
  - NFA/GDP t-1: -0.0104** (0.00451)
  - Credit/GDP Growth t-1: 0.0208*** (0.00557)
  - M2/Reserves t-1: 0.0129*** (0.00267)
  - ToT Growth t-1: 0.0174** (0.00859)
  - Inflation t-1: 0.0156*** (0.00373)
  - Log GDP per capita t-1: -0.1590 (0.103)
  - Real GDP growth t-1: -0.00778 (0.0288)
- Selected global variables:
  - Fed Funds Rate t: 0.132** (0.0567)
  - VIX t: 0.0908*** (0.0270)
  - World Real GDP Growth t: -0.284*** (0.104)
- Selected fiscal/financial variables:
  - Public Debt/GDP t-1: -0.0110*** (0.00424)
  - Foreign/Total CB Assets t-1 (backing ratio): -0.0107** (0.00453)
  - Bank Equity/GDP t-1: 0.0354* (0.0204) in some specifications
  - Euro (EMU membership) coefficients reported positive in some columns (e.g., 0.830* (0.466); up to 1.317*** (0.489)).
- Model fit and sample:
  - Observations: 3,867 (Column 1) down to 2,976 (Column 6).
  - Pseudo R-squared: 0.0886 (Column 1) to 0.210 (Column 6).
- Interpretation summary:
  - Positive associations with banking crisis probability: credit/deposit ratio, credit/GDP growth, inflation, M2/reserves, tighter US monetary policy (Fed Funds), global uncertainty (VIX).
  - Negative associations with banking crisis probability: banking sector NFA/GDP, world real GDP growth, backing ratio.
  - Public debt/GDP coefficient negative and significant in some specifications (noting potential sovereign–bank spillover puzzle).

### Econometric results — Currency crises (selected coefficients and fit)
- Selected coefficient estimates (significance markings as above):
  - NFA/GDP t-1: -0.0102*** (0.00343)
  - Real Int. Rate t-1: -0.00947** (0.00424)
  - Inflation t-1: 0.0153*** (0.00293)
  - Log GDP per capita t-1: -0.265*** (0.0811)
  - M2/Reserves t-1: 0.00098 (0.00360) — not significant baseline
- Global variables:
  - World Real GDP Growth t: -0.344*** (0.0974)
  - Spread 10y-3m US t: -0.302* (0.168) in Column 2
- Fiscal and financial variables:
  - Public Debt/GDP t-1: -0.0108*** (0.00398)
  - Overall Balance/GDP t-1: -0.0768** (0.0301) in Column 3
  - Foreign/Total CB Assets t-1 (backing ratio): -0.0228*** (0.00422)
  - Bank Assets/GDP t-1: -0.0137** (0.00578)
  - Peg t-1: -0.602** (0.262) in Column 1 (fixed exchange rate associated with lower chance of a currency crisis)
- Model fit and sample:
  - Observations: 4,057 (Column 1) down to 3,121 (Column 6).
  - Pseudo R-squared: 0.0639 (Column 1) to 0.176 (Column 6).
- Interpretation summary:
  - Currency crises more strongly linked to backing ratio, fiscal prudence (lower public debt and better overall balance), lower income (negative log GDP per capita coefficient), and world real GDP downturns.

### Heterogeneity: small states and fixed exchange rate countries
- Exchange rate regime heterogeneity (Table 4) — banking crises:
  - Pegged regimes more exposed to global uncertainty: Peg * VIX t: 0.140** (0.0563).
  - Backing ratio significant for pegged regimes: Peg * CB Foreign Assets Share t-1: -0.0261*** (0.00756); not significant for flexible regimes.
  - Exposure to world growth significant for both regimes with varying magnitudes.
- Exchange rate regime heterogeneity — currency crises:
  - Backing ratio negative and significant across regimes: Peg * CB Foreign Assets Share t-1: -0.0270*** (0.00966); No Peg coefficient also negative but smaller in some specs.
  - Peg t-1 effects on currency crises vary; baseline Peg t-1: -0.602** (0.262).
- Small-state heterogeneity (Table 5):
  - Backing ratio more strongly predictive for small states: Small State * CB Foreign Assets Share t-1: -0.0255* (0.0135) for banking crises.
  - Small states less linked to global growth in crisis predictability (coefficients less robust).
  - Not-small-state coefficients often strongly significant for world growth and backing ratio (e.g., Not Small State * World Growth t: -0.343*** (0.119); Not Small State * CB Foreign Assets Share t-1: -0.0107* (0.00548) in banking).
- Observations and fit (examples):
  - Banking Crisis regressions with regime heterogeneity: Observations 3,246; 3,149; 3,225. Pseudo R-squared 0.194; 0.210; 0.212.
  - Currency Crisis regressions with regime heterogeneity: Observations 3,401; 3,304; 3,123. Pseudo R-squared 0.119; 0.153; 0.155.
  - Banking Crisis regressions with small-state heterogeneity: Observations 3,117; 3,021; 3,097. Pseudo R-squared 0.192; 0.205; 0.201.
  - Currency Crisis regressions with small-state heterogeneity: Observations 3,277; 3,181; 3,004. Pseudo R-squared 0.109; 0.153; 0.154.

### Application to the ECCU — distributions and predicted probabilities
- Non-parametric densities (Figure 4):
  - Six key macro variables plotted with crisis vs non-crisis distributions; Smirnov tests reject equality of distributions at least at the 5 percent level for all variables.
  - ECCU 2019 backing ratios: all ECCU countries had backing ratios above 90 percent; crisis distribution shifted toward lower backing ratios.
  - ECCU 2019 credit/deposit ratio: ECCU mean just below the mode of the non-crisis distribution and in the lower tail of the crisis distribution.
- Predicted probabilities (Figure 6 overview; models use Columns 3 and 6 of Table 2 and Table 3):
  - In most years, crisis probabilities for ECCU are well below 1 percent and stable.
  - Short-lived spikes occur around global downturns (notably 2008 Great Financial Crisis).
  - Widening of cross-country range around 2008 indicates heterogeneity across ECCU members in vulnerability to global shocks.

### The importance of the backing ratio (interaction with global cycle)
- Backing ratio mitigates the impact of global cycle by reducing banking and currency crisis probabilities.
- Non-linearity in Logit creates interaction between backing ratios and external global variables.
- Figure 7 observation (currency crises example):
  - Lower backing ratios correspond to higher predicted probabilities (downward-sloped lines).
  - The negative gradient steepens when world real GDP growth is lower; crisis propensity more sensitive to global downturns when backing is low.
  - Quantitative example for currency crises: a fall in world real GDP growth from 0 percent to -3 percent raises the probability by less than 10 percentage points at full backing and by more than 20 points at a 60 percent backing ratio.
- Model experiments:
  - Excluding global variables (green dashed lines) removes spikes and yields a slight downward trend in predicted probabilities.
  - Excluding backing ratio (red dotted lines) increases mean predicted probabilities, especially during global downturns; in 2008 the probability without backing ratio was more than twice as high as baseline for both banking and currency crises.
- Conclusion: ECCU’s high backing ratio is particularly important for resilience during global slowdowns and market uncertainty.

### Policy recommendations and institutional observations
- Backing ratio is a key crisis predictor for small states and fixed exchange rate regimes.
- The negative association between backing ratio and crisis risk is larger in periods of low world real GDP growth, implying reserve-based support is especially relevant during global downturns.
- ECCU-specific institutional points:
  - The ECCB’s persistent high backing ratio is maintained by tightly containing credit provision to member governments.
  - Demand for central bank credit from banks has been limited as banks maintain large liquidity holdings.
  - Empirical analysis supports that the ECCB’s prudence in maintaining high backing ratios has contributed to currency and financial stability in the ECCU during turmoil.

### Data sources and selected variables used
- Data sources: Haver Analytics, IMF WEO database, IMF International Financial Statistics, Global Financial Data, World Bank World Development Indicators, IMF Monetary and Financial Statistics Database, Federal Reserve Board H.15, Cboe Global Markets, Inc.
- Selected variables used: Bank Assets, Bank Equity, Central Bank Foreign Assets, Credit, Credit/Deposits, Federal Funds Rate, GDP per Capita (log), Inflation (GDP deflator change), M2, Net Foreign Assets (NFA), Nominal GDP, Overall Balance, Public Debt, Real GDP Growth, Real Interest Rate, Foreign Reserves, Spread 10y – 3m US t, Terms of Trade, Total CB Assets, VIX.

### Annex excerpts — episodes and Appendix I (selected indicators)
- Annex III provides case studies of banking and currency crises in small states (selected episodes and reforms for countries such as Dominican Republic, Guyana, Guinea-Bissau, Haiti, Jamaica, Seychelles, Suriname, Trinidad and Tobago, Papua New Guinea).
- Appendix I — Small States Selected Indicators (extract; fields: 2018 GDP per capita (US$); Population (thousands) (as of 2018); FX Regime; Banking Crisis (starting year); Currency Crisis (starting year)). Select entries (values preserved exactly):
  - Antigua and Barbuda: 16,861; 95; Currency Board
  - Dominica: 7,081; 75; Currency Board
  - Grenada: 10,486; 111; Currency Board
  - St. Lucia: 11,557; 179; Currency Board
  - St. Kitts and Nevis: 19,270; 56; Currency Board
  - St. Vincent and the Grenadines: 7,354; 110; Currency Board
  - Guyana: 6,121; 782; Pegged; 1993 (Banking Crisis); 1987 (Currency Crisis)
  - Jamaica: 5,730; 2,731; Free float and flexible; 1996 (Banking Crisis); 1978, 1983, 1991 (Currency Crisis)
  - Suriname: 5,871; 590; Pegged; 1990, 1995, 2001, 2016 (Currency Crisis)
  - Trinidad and Tobago: 17,038; 1,390; Pegged; 1986 (Currency Crisis)
  - Seychelles: 16,143; 95; Free float and flexible; 2008 (Currency Crisis)
  - (Additional country entries and years included in the Appendix extract.)

*Source: ANNEX I. DATA DESCRIPTION, wpiea2021276-print-pdf*

### ANNEX I. DATA DESCRIPTION  _____________________________________________________ 22

### ANNEX I. DATA DESCRIPTION

### Background and context
- The Eastern Caribbean Currency Union (ECCU) comprises six independent counties: Antigua and Barbuda, Dominica, Grenada, St. Kitts and Nevis, Saint Lucia, and St. Vincent and the Grenadines, as well as two overseas territories of the United Kingdom: Anguilla and Montserrat.
- The COVID-19 pandemic starting in early 2020 led to tourist arrivals plummeting by 70 percent and cruise ship travel completely halted, pushing the ECCU economy into a deep recession in 2020.
- Fiscal positions deteriorated sharply in 2020 with public debt rising steeply; external accounts also deteriorated, although official foreign reserve positions held up relatively well, partly reflecting increased official financing.

### Currency boards, the ECCB, and the backing ratio
- The Eastern Caribbean Central Bank (ECCB) was established in October 1983 and operates a quasi-currency board arrangement managing a common pool of reserves for the ECCU.
- The ECCB’s operational target is a high level of foreign reserve cover—the “backing ratio” defined as the ECCB’s foreign assets as percent of its demand liabilities.
- Under the ECCB Agreement Act (1983), the ECCB must keep the currency “backing ratio” at a minimum of    60 percent, but operationally targets 80 percent.
- In practice, the backing ratio has been maintained at 95–100 percent over the past two decades, with the ECCB limiting credit extension to governments and banks to preserve the backing ratio.
- There has been no episode of a twin crisis (a currency crisis combined with a systemic banking crisis) in the ECCU.

### Definitions of crises (per Laeven and Valencia, 2020)
- Currency crises:
  - Defined as instances of a large exchange rate depreciation (30 percent or more) within a year against the US dollar.
  - If two consecutive years feature large depreciations, the depreciation in the second year must be 10 percentage points larger than that in the previous year to qualify as a separate instance of a crisis.
- Banking crises:
  - Defined as instances of severe financial distress in a country’s banking system, requiring significant policy interventions that can last for multiple years.
  - Distress indicators include a rise in NPLs, bank losses, bank runs, and bank asset foreclosures.
  - Policy interventions include fiscal expenditure for bank nationalizations, liquidity support, government guarantees, asset freezes, and bank holidays.

### Dataset, sample coverage, and time horizon
- Laeven and Valencia (2020)’s database covers currency and banking crises for 165 countries from 1970 to 2017, including 26 small states.
- The authors expand the sample by adding 14 extra small states (none of which includes a crisis), producing a sample of 173 countries, which include 40 small states, of which six countries are ECCU members.
- The U.K. territories of Anguilla and Montserrat are excluded from the empirical sample of sovereign ECCU countries.
- A common set of macroeconomic indicators is available only for 1981–2017 for all countries; thus, the empirical analysis time horizon is restricted to 1981–2017.

### Descriptive findings on crisis frequency and country characteristics (summary of Table 1)
- Statistical tests for differences in means suggest crisis frequency varies across country characteristics:
  - Banking crises are less common for:
    - Currency boards
    - Low-income countries
    - Small states
  - Currency crises are less common for:
    - Fixed exchange rates
    - Currency boards
    - Advanced economies
    - Countries with low-GDP volatility
  - Twin crises (currency + banking) are rare, and evidence suggests they are less common for:
    - Advanced economies
    - Small states
    - Economies with low GDP volatility

### Main empirical contributions and key findings
- Scope and method:
  - The paper runs a binomial Logit model (building on Caggiano et al. (2016)) using Laeven and Valencia (2020)’s crisis database to investigate determinants of banking and currency crises, with attention to fixed exchange rate regimes and small states.
- Main empirical findings:
  - Higher levels of the backing ratio are strongly and robustly associated with lower probabilities of banking and currency crises.
  - The relationship between backing ratio and lower crisis probability is quantitatively more important for fixed exchange rate regimes and small states.
  - The estimated model suggests the ECCU had a low predicted probability of banking and currency crises in “normal times.”
  - Model-implied crisis probability for the ECCU rose somewhat during global economic downturns, such as the 2008 Great Financial Crisis, indicating potential vulnerability to exogenous shocks.
  - Country-specific macroeconomic conditions and global conditions (including market uncertainty, growth, and interest rates) have predictive power for the likelihood of a crisis, broadly confirming prior literature.

*Source: ANNEX I. DATA DESCRIPTION, wpiea2021276-print-pdf*

### Annex I.

### Annex I — wpiea2021276-print-pdf

### Empirical strategy
- Model: binomial Logit estimating Pr(Crisis_i,t = 1) = e^{βX_i,t + ε_i,t} / (1+ e^{βX_i,t + ε_i,t}).
- Timing:
  - Country-specific variables in X_i,t are one-year lags.
  - Global variables use the concurrent year value.
  - Only the year in which a banking crisis begins is included for banking crisis analysis; currency crises are one-year events.
- Estimation: Maximum Likelihood with robust standard errors.
- Baseline explanatory variables: credit/deposit ratio, banking sector net foreign assets (NFA) as share of GDP, growth rate of credit/GDP, real interest rate, M2/GDP ratio (M2/Reserves presented in results), growth rate of terms of trade, log real GDP per capita, real GDP growth.
- Expanded sets:
  - Fiscal: public debt/GDP ratio, fiscal balance/GDP ratio.
  - Financial: banking sector equity/GDP, banking sector assets/GDP, banking sector equity/assets, central bank’s share of foreign assets to total assets (“backing ratio”).
  - Exchange rate regime: binary for dollarized economy, currency board, currency union, or pegged currency.
  - Global: world real GDP growth rate, global uncertainty index (VIX), US Federal Reserve’s funds rate, 10 year–3 month US government bond yield spread.
- Data sources summarized in Annex I.

### Descriptive evidence
- Dynamics around banking crises (Figure 1):
  - Crisis-sample medians are markedly apart from non-crisis medians in years preceding banking crises for most variables.
  - Exceptions prior to events: real GDP growth and public debt/GDP exhibit very similar medians across crisis and non-crisis groups.
  - Post-crisis: notable corrections in credit/deposit ratio, M2/reserves, credit/GDP growth; impacts on real GDP growth and public debt/GDP.
- Dynamics around currency crises (Figure 2):
  - Similar systematic pre-crisis differences as banking crises.
- Global variables (Figure 3):
  - VIX and world real GDP growth show correlation with total number of crises in a year, notably during 2008 GFC.

### Econometric results — Banking crises (Table 2)
- Key coefficients (robust standard errors in parentheses; significance: *** p<0.01, ** p<0.05, * p<0.1):
  - Credit/Deposit t-1: 0.00650*** (0.00174)
  - NFA/GDP t-1: -0.0104** (0.00451)
  - Credit/GDP Growth t-1: 0.0208*** (0.00557)
  - Real Int. Rate t-1: 0.00435 (0.00608)
  - M2/Reserves t-1: 0.0129*** (0.00267)
  - ToT Growth t-1: 0.0174** (0.00859)
  - Inflation t-1: 0.0156*** (0.00373)
  - Log GDP per capita t-1: -0.1590 (0.103)
  - Real GDP growth t-1: -0.00778 (0.0288)
- Selected global variables (Column 2):
  - Fed Funds Rate t: 0.132** (0.0567)
  - VIX t: 0.0908*** (0.0270)
  - World Real GDP Growth t: -0.284*** (0.104)
- Selected fiscal/financial variables (Columns 3–6):
  - Public Debt/GDP t-1: -0.0110*** (0.00424)
  - Foreign/Total CB Assets t-1: -0.0107** (0.00453)
  - Bank Equity/GDP t-1: 0.0354* (0.0204) in Column 3 (but not robust across all columns)
  - Euro (membership to European Monetary Union): 0.830* (0.466) in Column 1; up to 1.317*** (0.489) in Column 3
  - Peg t-1: -0.463 (0.316) in Column 5; non-significant when controlling for many variables.
- Model fit and sample:
  - Observations: 3,867 (Column 1) down to 2,976 (Column 6)
  - Pseudo R-squared: 0.0886 (Column 1) to 0.210 (Column 6)
- Interpretation:
  - Positive association with crisis probability: credit/deposit ratio, credit/GDP growth, inflation, M2/reserves, tighter US monetary policy (Fed Funds), global uncertainty (VIX).
  - Negative association with crisis probability: banking sector NFA/GDP, world real GDP growth, backing ratio (foreign/total CB assets).
  - Public debt/GDP coefficient negative and significant in some specifications (surprising given potential sovereign–bank spillovers).

### Econometric results — Currency crises (Table 3)
- Key coefficients (significance markings as above):
  - Credit/Deposit t-1: -0.00123 (0.00244) — not significant baseline
  - NFA/GDP t-1: -0.0102*** (0.00343)
  - Credit/GDP Growth t-1: -0.0101* (0.00520) in Column 1; positive and significant in some expanded specs
  - Real Int. Rate t-1: -0.00947** (0.00424)
  - Inflation t-1: 0.0153*** (0.00293)
  - Log GDP per capita t-1: -0.265*** (0.0811)
  - M2/Reserves t-1: 0.00098 (0.00360) — not significant baseline
- Global variables:
  - World Real GDP Growth t: -0.344*** (0.0974)
  - Spread 10y-3m US t: -0.302* (0.168) in Column 2
- Fiscal and financial variables:
  - Public Debt/GDP t-1: -0.0108*** (0.00398)
  - Overall Balance/GDP t-1: -0.0768** (0.0301) in Column 3
  - Foreign/Total CB Assets t-1 (backing ratio): -0.0228*** (0.00422)
  - Bank Assets/GDP t-1: -0.0137** (0.00578)
  - Peg t-1: -0.602** (0.262) in Column 1 (a fixed exchange rate is associated with lower chance of a currency crisis)
- Model fit and sample:
  - Observations: 4,057 (Column 1) down to 3,121 (Column 6)
  - Pseudo R-squared: 0.0639 (Column 1) to 0.176 (Column 6)
- Interpretation:
  - Currency crises more strongly linked to backing ratio, fiscal prudence (lower public debt and better overall balance), lower income (log GDP per capita negative coefficient), and world real GDP downturns.

### Heterogeneity: small states and fixed exchange rate countries (Tables 4–5)
- Exchange rate regime heterogeneity (Table 4):
  - For banking crises:
    - Pegged regimes more exposed to global uncertainty (Peg * VIX t: 0.140** (0.0563)).
    - Backing ratio significant for pegged regimes (Peg * CB Foreign Assets Share t-1: -0.0261*** (0.00756)); not significant for flexible regimes.
    - Exposure to world growth significant for both regimes (No Peg * World Growth t: -0.342** (0.148); Peg * World Growth t: -0.255 (0.161) with variations across columns).
  - For currency crises:
    - Backing ratio negative and significant across regimes (Peg * CB Foreign Assets Share t-1: -0.0270*** (0.00966); No Peg coefficient also negative but smaller magnitude in some specifications).
    - Peg effects on currency crises vary; Peg t-1 coefficients reported in earlier tables (e.g., Table 3 Peg t-1: -0.602** (0.262)).
  - Observations and fit:
    - Banking Crisis regressions: Observations 3,246; 3,149; 3,225. Pseudo R-squared 0.194; 0.210; 0.212.
    - Currency Crisis regressions: Observations 3,401; 3,304; 3,123. Pseudo R-squared 0.119; 0.153; 0.155.
- Small-state heterogeneity (Table 5):
  - Crises in small states:
    - Less associated with global growth (Small State * World Growth t: -0.0533 (0.310) in banking; results less robust).
    - More strongly associated with backing ratio (Small State * CB Foreign Assets Share t-1: -0.0255* (0.0135) for banking crises).
    - Not-small-state coefficients frequently highly significant for world growth and backing ratio (Not Small State * World Growth t: -0.343*** (0.119); Not Small State * CB Foreign Assets Share t-1: -0.0107* (0.00548) in banking).
  - Observations and fit:
    - Banking Crisis regressions: Observations 3,117; 3,021; 3,097. Pseudo R-squared 0.192; 0.205; 0.201.
    - Currency Crisis regressions: Observations 3,277; 3,181; 3,004. Pseudo R-squared 0.109; 0.153; 0.154.
  - Summary: backing ratio is generally a stronger predictor of crises for small states than for larger economies; small states appear less linked to global growth in crisis predictability.

### Application to the ECCU (section introduction and Figures 4–6 overview)
- Non-parametric densities (Figure 4):
  - Six key macro variables plotted with crisis vs non-crisis distributions.
  - ECCU 2019 values shown as shaded grey area; ECCU mean as vertical dash line.
  - For the six indicators, crisis and non-crisis distributions differ markedly; Smirnov tests reject equality of distributions at least at the 5 percent level for all variables.
  - ECCU position in 2019:
    - For credit/deposit ratio: ECCU mean just below the mode of the non-crisis distribution; ECCU mean in lower tail of crisis distribution.
    - For backing ratio: ECCU countries all had backing ratios above 90 percent; crisis distribution shifted toward lower backing ratios.
- Figure 5: similar non-parametric distributions for currency crises; ECCU largely in “safe” areas for the indicators shown.
- Predicted probabilities (Figure 6 referenced):
  - The estimated models (Columns 3 and 6 of Table 2 and Table 3) are used to plot predicted banking and currency crisis probabilities for ECCU countries over 1995–2018.
  - The models allow assessment of the main historical drivers of fluctuations in implied crisis probabilities for ECCU countries.

*IMF Working Paper — Annex I of "Assessing Banking and Currency Crisis Risk in Small States: An Application to the Eastern Caribbean Currency Union" (wpiea2021276-print-pdf).*

### 4. The solid black lines represent the mean probability across ECCU countries, while the shaded area

### wpiea2021276-print-pdf - 4. The solid black lines represent the mean probability across ECCU countries, while the shaded area

### Key findings on crisis probabilities and global conditions
- In most years, the crisis probabilities are well below 1 percent and stable over time.
- There are short-lived spikes in crisis probabilities, notably around the years of the GFC, suggesting global conditions induce substantial fluctuations in the likelihood of a crisis.
- The widening of the grey area around 2008 implies heterogeneity across ECCU countries in how global conditions amplify crisis chances.
- Excluding global variables from the Logit model (mean implied probabilities, green dashed lines) yields a slight downward trend in predicted probabilities and removes spikes around worldwide downturns.
- Excluding the backing ratio from the model (red dotted lines) increases the mean predicted probability for both banking and currency crises relative to the baseline:
  - During “tranquil times” (2003-2007 and 2010-2018) both probabilities are below 1 percent and the difference is small.
  - In 2008 the probability from the alternative specification without the backing ratio was more than twice as high as the baseline for both banking and currency crises.
- Conclusion: The ECCU’s high backing ratio may be particularly important to maintain macroeconomic stability during global slowdowns and spells of international market uncertainty.

### The importance of the backing ratio (interaction with global cycle)
- The high backing ratio mitigates the impact of the “global cycle” by reducing the probability of banking and currency crises.
- The Logit function’s non-linearity creates an interaction between domestic backing ratios and external global variables.
- Figure 7 observations (predicted probability by backing ratio and world real GDP growth):
  - Lower backing ratios are associated with higher predicted probabilities (downward-sloped lines).
  - The negative gradient of the lines becomes steeper for lower values of world real GDP growth, meaning crisis propensity is more susceptible to global downturns when backing is low.
  - Quantitatively for currency crises: a fall in world real GDP growth from 0 percent to -3 percent raises the probability by less than 10 percentage points at full backing and by more than 20 points at a 60 percent backing ratio.

### Model comparisons and implications
- Baseline Logit model (black solid line): mean predicted probability among ECCU countries with grey area showing min-max range across countries.
- Alternative model excluding backing ratio (red dotted): raises predicted probabilities, especially during global downturns.
- Alternative model excluding global variables (dashed green): produces a probability series without spikes at worldwide downturns.
- Implication: Both global conditions and the backing ratio are central determinants of predicted crisis risk; the backing ratio provides resilience particularly when global growth is weak.

### Policy recommendations and institutional observations
- The backing ratio is a key crisis predictor for small states and fixed exchange rate regimes.
- The negative association between backing ratio and crisis risk is larger in periods of low world real GDP growth; reserve-based support for the currency is more relevant during global downturns.
- For the ECCU specifically:
  - The ECCB has persistently maintained a high backing ratio by tightly containing credit provision to ECCU member governments.
  - Demand for central bank credit from banks has been quite limited as banks maintain a large amount of liquidity.
  - The empirical analysis supports that the ECCB’s prudence in maintaining high backing ratios has contributed to maintaining currency and financial stability in the ECCU even during times of turmoil.

### Crisis definitions (Laeven and Valencia (2020))
- Banking crises must satisfy two requirements:
  - 1. Significant financial distress in the banking system, if at least one of the following occurs:
    - non-performing loans rise above 20 percent of total loans
    - bank closures of at least 20 percent of banking system assets
    - fiscal costs of restructuring the banking sector exceed 5 percent of GDP
  - 2. Significant policy interventions if at least three of the following measures are applied:
    - deposit freezes and/or bank holidays
    - significant bank nationalizations
    - bank restructuring fiscal costs (at least 3 percent of GDP)
    - extensive liquidity support (at least 5 percent of deposits and liabilities to non-residents)
    - significant guarantees put in place
    - significant asset purchases (at least 5 percent of GDP)
- Currency crises must satisfy two criteria:
  - 1. A year-on-year depreciation against of the US dollar of 30 percent or more.
  - 2. A year-on-year depreciation against of the US dollar that is at least 10 percentage points higher than the rate of depreciation in the previous year.

### Data sources and selected variables used in the analysis
- Data sources: Haver Analytics, IMF WEO database, IMF International Financial Statistics, Global Financial Data, World Bank World Development Indicators, IMF Monetary and Financial Statistics Database, Federal Reserve Board H.15, Cboe Global Markets, Inc.
- Selected variables (as described in the source):
  - Bank Assets, Bank Equity, Central Bank (CB) Foreign Assets, Credit, Credit/Deposits
  - Federal Funds Rate, GDP per Capita (log), Inflation (GDP deflator change), M2
  - Net Foreign Assets (NFA), Nominal GDP, Overall Balance, Public Debt, Real GDP Growth
  - Real Interest Rate, Foreign Reserves, Spread 10y – 3m US t, Terms of Trade, Total CB Assets, VIX

*IMF Working Paper — Assessing Banking and Currency Crisis Risk in Small States: An Application to the Eastern Caribbean Currency Union*

### Annex III  . Banking and Currency Crises in Small

### Annex III  . Banking and Currency Crises in Small States

### Banking crises: episodes, causes, and end-of-crisis reforms
- Dominican Republic (2003-2004)
  - Nature/Cause of Crisis: Triggered by the collapse of one large bank (Baninter) in 2003, the banking crisis spread to two others (Bancredito and Banco Mercantil), with a rapid withdrawal of deposits. These banks were found to be undercapitalized as their rue level of assets and risks were hidden through the use of offshore banks and accounting manipulation.
  - End of Crisis/Reforms: The Central Bank stepped in to provide liquidity support, imposed additional capital requirements and new regulations, strengthened the regulatory framework, and improved transparency.
- Guyana (1993)
  - Nature/Cause of Crisis: Commercial lending became risky, while government owned banks were burdened with non-performing loans as directed credit programs had resulted in investments with low rates of return. By 1993, nonperforming loans rose to a unsustainably high level, and bank profitability declined due to increased provisions for bad loans and narrow opportunities to invest.
  - End of Crisis/Reforms: To quell the crisis, state-owned banks were privatized and merged, and fundamental changes were made to regulatory and legal frameworks through the Financial Institutions Act. Reserve and liquid asset requirements also helped to improve the effectiveness of monetary policy.
- Guinea Bissau (2014)
  - Nature/Cause of Crisis: In the wake of the 2012 coup and falling cashew prices in 2012-13, nonperforming loans ballooned in the banking system, and bank credit contracted heavily.
  - End of Crisis/Reforms: The two banks were required to fully provision nonperforming loans (leading to negative earnings and large drops in equity capital) and inject sufficient funds to meet capital requirements.
- Haiti (1994-1998)
  - Nature/Cause of Crisis: The banking system’s net domestic assets increased significantly in 1993-94 due to increased lending to the public sector. The Central Bank also registered considerable losses as the majority of its assets, represented by credit to the government, were nonperforming. Loosening fiscal and credit policy and political uncertainty contributed to soaring inflation, adding further pressure to bank deposits.
  - End of Crisis/Reforms: Upon the return of constitutional rule by the end of 1994, steps were taken to strengthen the banking system by modernizing the Central Bank, regaining control over monetary policy through the use of more indirect instruments, increasing competition, and improving regulation and supervision.
- Jamaica (1996-1998)
  - Nature/Cause of Crisis: The rapid expansion and emergence of large financial conglomerates resulted in a surge of credit to the private sector. Many commercial banks were not adequately capitalized, which resulted in impending insolvency, followed by a credit crunch and then the fall in profitability of the sector and increase in nonperforming loans. The crisis reached its peak when real estate and equity markets triggered illiquidity in the life insurance industry, which soon spread to affiliated banks whose depositors quickly moved funds into foreign banks.
  - End of Crisis/Reforms: In 1996, FINSAC, a government resolution agency, was established to resuscitate, reorganize, and consolidate the financial industry. Government recapitalized numerous troubled institutions.

### Currency crises: episodes, causes, and end-of-crisis reforms
- Dominican Republic (2003)
  - Nature/Cause of Crisis: The 2003 banking crisis led to a substantial depreciation of the peso, a sharp increase in inflation, a rise in debt, and a deceleration in GDP growth.
  - End of Crisis/Reforms: The exchange market was unified, and authorities were committed to a fully flexible exchange rate policy. Monetary policy was tightened, and fiscal consolidation measures were implemented, causing inflation to fall and the exchange rate to stabilize.
- Guyana (1987-1998)
  - Nature/Cause of Crisis: Under heavy state control, the economic growth started slowing in 1982 as a result of sharp contractions in the bauxite sector and the erosion of its export sector. External debt and arrears grew to unsustainable levels as foreign reserves dwindled due to large current account deficits, and domestic inflation shot up.
  - End of Crisis/Reforms: The informal, parallel exchange market became so large that by 1987, as an element of a multi-year economic restructuring program, the exchange rate was discretionally devalued by 56 percent to return external transactions to the formal markets.
- Haiti (2003)
  - Nature/Cause of Crisis: Between 2000 and 2003, economic growth was near zero, reflecting a difficult political situation, low private sector confidence and investment, sizeable fiscal expenditure overruns and revenue shortfalls, and shrinking external assistance. Due to a sharp rise in inflation, accompanied by falling international reserves, the gourde depreciated by over 60 percent.
  - End of Crisis/Reforms: A transition government restored macroeconomic stability by March 2004 and eliminated the liquidity surplus by halting central bank lending to the government. Inflation eventually declined, the exchange stabilized, and the rate of dollarization slowed.
- Jamaica (1983)
  - Nature/Cause of Crisis: In the early 1980s, falling commodity prices weakened terms of trade, world tourism slumped, and high global interest rates put pressure on debt services. Signs of exchange rate misalignment appeared as the balance of payments deficit lasted for many years. Credit to government reached to 400 percent of the monetary base. In 1982, the government removed sanctions on parallel foreign currency transactions, but pressure in the exchange system continued.
  - End of Crisis/Reforms: In November 1983, the official rate was allowed to devalue by 43 percent. The authorities introduced an exchange rate auction to replace the peg.
- Jamaica (1991)
  - Nature/Cause of Crisis: After a series of devaluations, the auction was suspended. Further devaluations and failure to contain inflation diminished the government’s credibility.
  - End of Crisis/Reforms: The exchange rate determination by the Central Bank was abandoned. The exchange rate subsequently fell by 300 percent in 1991.
- Papua New Guinea (1995)
  - Nature/Cause of Crisis: An increase in credit to the government led to a depletion of reserves which undermined the stability of the kina. A shift to a market determined exchange rate in late 1994 caused a sharp devaluation that continued into 1995 and inflation rose quickly.
  - End of Crisis/Reforms: A screen-based foreign exchange trading system was introduced later that year and tighter monetary policy quelled inflation pressure.
- Seychelles (2008)
  - Nature/Cause of Crisis: Seychelles had a balance of payments and public debt crisis after a period of expansionary fiscal and monetary policies, compounded by the global financial crisis. However, reforms to liberalize the economy were insufficient to address longstanding imbalances, which were exacerbated by fuel and food price shocks. This resulted in foreign exchange shortages, a rapid rise in inflation, current account deficits, and a nominal depreciation of 37.5 percent.
  - End of Crisis/Reforms: In November 2008, a liberalization of the exchange regime towards a float resulted in the elimination of restrictions on international transactions, after which the rupee stabilized and appreciated significantly.
- Suriname (1990-1995)
  - Nature/Cause of Crisis: Central bank financing of the fiscal deficit increased, but inflation only rose gradually, as prices were controlled for 80 percent of all goods and services. After the military government was replaced in free elections, the price controls were abandoned, and inflation soared to near hyperinflation. Fearing an outright devaluation as the parallel market rate diverged from the official one, the authorities implemented a multiple exchange system in 1992.
  - End of Crisis/Reforms: The authorities unified official and parallel foreign exchange markets in 1994. This slowed the depreciation in the parallel exchange rate. The Central Bank stepped up to absorb liquidity, which eventually helped stop exchange rate depreciation.
- Suriname (2000-2003)
  - Nature/Cause of Crisis: Economic activity weakened in 1999. By 2000, the fiscal position deteriorated due to an increase in election related expenses. Widened external and fiscal imbalances led to the removal of exchange rate bands and large currency devaluation (about 90 percent). Weakened macroeconomic policies following the stabilization effort resulted in further depreciations and higher inflation.
  - End of Crisis/Reforms: Government borrowing from the Central Bank was terminated, petroleum subsidies were eliminated, and electricity and water tariffs were increased. Eventually tighter fiscal and monetary policies helped stabilize exchange rate and inflation pressures in 2003.
- Suriname (2016)
  - Nature/Cause of Crisis: In 2015-16, the fall in gold and oil prices, and closure of the bauxite plant resulted in large GDP contraction, fiscal and current account deficits, and an uptick in unemployment. In an ambitious adjustment plan to cut deficits, build foreign reserves and curb monetary financing, authorities changed the de jure monetary regime to reserve money targeting in March 2016. But progress on numerous key policy items stalled, and with limited action to raise interest rates, bouts of exchange rate depreciation occurred. This resulted in a sharp rise in inflation and capital flight occurred.
  - End of Crisis/Reforms: Eventually, commodity prices rebounded, and large segments of the economy became dollarized.
- Trinidad and Tobago (1985-1986)
  - Nature/Cause of Crisis: The fall in oil production and prices starting in 1982 led to sustained GDP contraction. Initially, the government tried to reduce imports via exchange controls and licensing but was offset by fiscal expansion. Disequilibrium in the currency market was evident through the wide current account deficit, significant drawdowns in international reserves, the sharp rise in lending to the government, and capital flight. This all led to the emergence of a parallel foreign exchange market. The exchange rate was devalued by 50 percent in December 1995.
  - End of Crisis/Reforms: The authorities launched a stabilization and structural adjustment program in 1988 and moved to a flexible exchange rate regime.

### Appendix I. Small States Selected Indicators (extract)
- Table fields: 2018 GDP per capita (US$), Population (thousands) (as of 2018), FX Regime, Banking Crisis (starting year), Currency Crisis (starting year)
- Select country entries (preserve numeric values and text exactly as shown):
  - Antigua and Barbuda: 16,861; 95; Currency Board
  - Bahamas, The: 34,584; 377; Pegged
  - Barbados: 17,758; 286; Pegged
  - Belize: 4,813; 398; Pegged
  - Dominica: 7,081; 75; Currency Board
  - Grenada: 10,486; 111; Currency Board
  - Guyana: 6,121; 782; Pegged; 1993 (Banking Crisis); 1987 (Currency Crisis)
  - Jamaica: 5,730; 2,731; Free float and flexible; 1996 (Banking Crisis); 1978, 1983, 1991 (Currency Crisis)
  - St. Lucia: 11,557; 179; Currency Board
  - St. Kitts and Nevis: 19,270; 56; Currency Board
  - St. Vincent and the Grenadines: 7,354; 110; Currency Board
  - Suriname: 5,871; 590; Pegged; 1990, 1995, 2001, 2016 (Currency Crisis)
  - Trinidad and Tobago: 17,038; 1,390; Pegged; 1986 (Currency Crisis)
  - Bahrain: 25,051; 1,503; Pegged
  - Djibouti: 2,872; 1,049; Currency Board; 1991 (Currency Crisis)
  - Qatar: 66,422; 2,760; Pegged
  - Cyprus: 29,300; 864; Free float and flexible; 2011 (Currency Crisis)
  - Estonia: 23,181; 1,322; Free float and flexible; 1992 (Banking Crisis); 1992 (Currency Crisis)
  - Iceland: 75,260; 348; Free float and flexible; 2008 (Currency Crisis); 1975, 1981, 1989, 2008 (Banking/Currency episodes)
  - Malta: 31,282; 476; Free float and flexible
  - Montenegro: 8,855; 622; separate legal tender
  - Bhutan: 3,281; 735; Pegged
  - Brunei: 30,668; 442; Currency Board
  - Maldives: 14,477; 366; Pegged; 1975 (Currency Crisis)
  - Samoa: 4,198; 199; Pegged
  - Solomon Islands: 2,497; 627; Pegged
  - Timor-Leste: 1,230; 1,268; separate legal tender
  - Tonga: 4,836; 100; Pegged
  - Vanuatu: 3,256; 285; Pegged
  - Botswana: 8,290; 2,251; Pegged; 1984 (Currency Crisis)
  - Cabo Verde: 3,616; 544; Pegged; 1993 (Currency Crisis)
  - Comoros: 1,386; 851; Pegged; 1994 (Currency Crisis)
  - Equatorial Guinea: 10,106; 1,314; Pegged; 1983 (Banking Crisis); 1980, 1994 (Currency Crisis)
  - Gabon: 8,221; 2,053; Pegged; 1994 (Currency Crisis)
  - Gambia, The: 729; 2,280; Pegged; 1985, 2003 (Currency Crisis)
  - Guinea-Bissau: 866; 1,738; Pegged; 1995, 2014 (Banking/Currency episodes); 1980, 1994 (Currency Crisis)
  - Lesotho: 1,141; 2,034; Pegged; 1985, 2015 (Currency Crisis)
  - Mauritius: 11,206; 1,266; Free float and flexible
  - Namibia: 5,664; 2,414; Pegged; 1984, 2015 (Currency Crisis)
  - São Tomé and Príncipe: 1,989; 209; Pegged; 1992 (Banking Crisis); 1987, 1992, 1997 (Currency Crisis)
  - Seychelles: 16,143; 95; Free float and flexible; 2008 (Currency Crisis)

*Source: IMF Article IV, IMF Selected Issues, Laeven and Valencia (2020); table sourced to IMF WEO; Luc Laeven and Fabian Valencia (2018).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021276-print-pdf.pdf_