## _wp12205

## Source details

**Canonical URL:** [_wp12205](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp12205.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp12205.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp12205.pdf.json)

---

### Introduction and contribution
- Examines reserves adequacy in emerging markets (EMs) and small islands (SIs).
- Three main contributions:
  - First paper to estimate observed cross-country preferences of reserves for 23 SIs, drawing on variants of the “buffer-stock model” and estimating OLS regression with fixed effects for SIs and comparing with 49 EMs.
  - First paper (to the authors’ knowledge) examining institutional, current and capital account motives for holding reserves allowing these factors to vary along the reserves’ holding distribution via quantile regressions; tests equality among slope coefficients of various quantile regressions and whether models at lower quantiles differ statistically from those at higher quantiles.
  - Develops a user-friendly operational metric for estimating reserve holdings in SIs building on the IMF (2011a, b) approach, tailored to economies with similar characteristics and likely BOP pressures.

### Main empirical findings
- Main drivers:
  - For SIs: import share and exchange rate volatility are key drivers of reserve holdings.
  - For EMs: import share and financial depth (broad money) are key drivers.
- Specific relationships and effect sizes (selected reported coefficients):
  - Full sample OLS (Observations = 729, Number of panels = 67, R-squared (OLS) = 0.1865):
    - d(Broad money/GDP) = 0.2945** (standard error 0.117)
    - Import-to-GDP = 0.0888*** (standard error 0.021)
    - NEER volatility = -0.0019** (standard error 0.001)
    - Short-term debt = 0.1716*** (standard error 0.030)
    - Constant = 0.0714*** (standard error 0.025)
  - Small islands (SIs) OLS (Observations = 247, Number of panels = 23, R-squared (OLS) = 0.2402):
    - d(Broad money/GDP) = 0.2241*** (standard error 0.086)
    - Import-to-GDP = 0.0645** (standard error 0.026)
    - NEER volatility = -0.0063*** (standard error 0.001)
    - Short-term debt = -0.3902*** (standard error 0.091)
    - Government Effectiveness = -0.0124* (standard error 0.007)
    - Constant = 0.1543*** (standard error 0.024)
  - Emerging markets (EMs) OLS (Observations = 482, Number of panels = 44, R-squared (OLS) = 0.2392):
    - d(Broad money/GDP) = 0.3640** (standard error 0.166)
    - Import-to-GDP = 0.1820*** (standard error 0.025)
    - NEER volatility = -0.0019*** (standard error 0.001)
    - Short-term debt = 0.1262*** (standard error 0.038)
    - Government Effectiveness = 0.0296*** (standard error 0.011)
- Additional empirical points:
  - Precautionary motives explain on average about 30 percent of the variation in reserve holdings.
  - Out-of-sample prediction since 2007 suggests reserves were slightly higher than levels suggested by fundamentals; the regression model does not track all individual country holdings well.

### Quantile regression results and heterogeneity
- Methodology:
  - Quantile regressions estimated at the 5th, 25th, 50th, 75th, and 95th quantiles.
  - Tests conducted for equality of slope coefficients across quantiles (F-tests, inter-quantile comparisons).
- Full sample (SIs + EMs) highlights:
  - Buffer-stock model predictions hold best for countries at the 50th quantile.
  - Short-term debt positive and rising to peak at the 95th quantile.
  - Change in broad money to GDP significant in all quantiles except tails.
  - Import-to-GDP strongly significant at most quantiles; coefficient size increases moving up the reserve distribution.
  - Government Effectiveness significant across many quantiles with varying sign; 95th quantile Government Effectiveness = 0.0649*** (standard error 0.023).
- SIs subsample:
  - Little significant variation across quantiles overall (F-tests could not reject equality across quantiles), but exceptions exist.
  - Short-term debt strongly significant and negative, peaking (in absolute terms) at the 75th quantile.
  - Broad money weakly significant at the 25th and 50th quantiles.
  - Government Effectiveness negative and significant at the 50th and 75th quantiles.
- EMs subsample:
  - Significant differences across quantiles (F-tests reject equality in several comparisons).
  - Short-term debt positive and significant, peaking at the 95th quantile.
  - Broad money significant in most quartiles except tail ends.
  - Institutions positive and strongly significant at the 5th, 75th, and 95th quantiles.
- Inter-quantile coefficient differences (selected, full sample):
  - d(Broad money/GDP): differences include -0.1738; -0.0212; -0.00530; 0.23910; 0.0386.
  - Import-to-GDP: differences include -0.0673***; -0.0124; 0.0465*; -0.0052; -0.1314**.
  - Short-term debt: differences include -0.0093; -0.0362; -0.0601*; -0.0348; -0.1409**.
  - Government Effectiveness: differences include 0.0241***; 0.0111*; -0.0108; -0.0746***; -0.0502**.
- Stylized cross-sample facts:
  - Largest reserve holders are disproportionately EMs (right-hand of distribution).
  - Example statistics: in 2009 median concessional debt to GDP was about 20 percent overall; for SIs with reserves at the 75th percentile concessional debt was about 84 percent, and about 11 percent for EMs.
  - Median multilateral debt about 25 percent overall; for SIs it ranged from 22 percent (economies with <25th percentile of reserves) to about 70 percent (those >75th percentile).

### A new reserve adequacy metric tailored to Small Islands (SIs)
- Metric design and rationale:
  - Core potential sources of balance-of-payments pressure for SIs: short-term debt on a remaining maturity basis, broad money, and exports of goods and services.
  - Exports of goods and services included to reflect SIs’ reliance on tourism and service exports.
  - EMP critical threshold set to 1 (instead of 1.5 or 3) to reflect higher volatility in SIs; using this yields about 30 EMP episodes between 1999 and 2010.
  - Bootstrapping used to obtain 10,000 random observations for each pressure source; the 10th percentile used as threshold; additional buffers: 5 percent general buffer and 25 percent buffer for export drains to fixed exchange regimes.
- Empirical distribution of 10th percentile flows during EMP events (percent):
  - Currency union: Exports 12.4, Broad money 12.6, Short-term debt 75.9
  - Fixed: Exports 10.5, Broad money 3.6, Short-term debt 89.1
  - Floating: Exports 4.8, Broad money 24.4, Short-term debt 35.2
- Proposed reserve-adequacy weights (metric formulas):
  - Currency union: 20% of exports + 20% of broad money + 80% of short-term debt
  - Fixed: 35% of exports + 10% of broad money + 95% of short-term debt
  - Floating: 10% of exports + 30% of broad money + 40% of short-term debt
- Comparative properties and sample status:
  - The proposed metric is more stringent than the IMF, 2011 metric, placing more weight on exports and short-term debt.
  - Based on end-2010, only 29 percent of the sample was within the proposed metric.

### Thresholds, predictive performance, and robustness
- Estimation approach:
  - Binary logit linking EMP events (dependent variable: 1 for crisis, 0 otherwise) to reserves divided by the proposed metric, controlling for government effectiveness and fiscal balance as a share of GDP.
  - Alternative regressions compared proposed metric to other metrics: three-month import coverage; broad money metric; Guidotti-Greenspan; maximum of three traditional metrics (Maxmetric); Lipschitz metric. Sample: annual data for 21 SIs from 1999 to 2010.
- Key predictive findings (logit results, Observations = 164):
  - Reserves/metric t-1 coefficients (selected): -3.771*** (1.361); -3.833*** (1.404); -4.565** (1.926); -3.797*** (1.446); -3.986*** (2.665); -2.812** (1.378).
  - Fiscal balance-to-GDP t-1 coefficients (selected): 0.0379* (0.0203); 0.0370* (0.0204); 0.0390* (0.0207); 0.0374* (0.0219); 0.0495** (0.0237).
  - Reserves/Lipschitz metric t-1 coefficients (selected): -5.279* (3.119); -2.771** (1.143).
- Crisis-probability thresholds and performance:
  - Crisis probabilities decline rapidly as reserves reach about 75-100 percent of the proposed metric; this range could be regarded as adequate for a typical country.
  - The probability of a crisis for countries that hold about 75 percent of reserves implied by this metric is less than 2 percent.
  - The reserves-to-proposed metric performs markedly better than alternative single or combined traditional metrics in reducing crisis probability (Table 11, models 1-6).
- Cross-country status (end-2010) using the proposed metric:
  - About 37 percent of the sample was within the proposed adequacy range.
  - Roughly 32 percent had reserves above the range.
  - About 42 percent had reserves below the range.
  - Of 14 countries with cover below 100 percent, six have currency unions and five have fixed exchange rate regimes.
- Caveats:
  - Country-specific characteristics, vulnerabilities, and degree of risk aversion may justify deviations from the metric-implied reserves.
  - Drivers of reserve changes during crisis periods (2008-10) may differ from the pre-crisis period; model forecasts may not fully capture crisis-period drivers.
  - Some reserve motives may be non-precautionary or reflect precautionary motives not captured by the model; ad hoc factors (currency swaps, increase in SDR allocations) may also affect observed reserves.

### Episodes of Exchange Market Pressure (EMP) — SIs, 1999–2010 (selected entries)
- EMP = 1 if crisis episode; 0 otherwise. Source: Author’s estimates.
- Examples of reported country-year EMP indicators:
  - Antigua and Barbuda: 1999 0.00, 2000 1.00, 2001 0.00, 2002 0.00, 2003 0.00, 2004 0.00, 2005 0.00, 2006 0.00, 2007 0.00, 2008 0.00, 2009 0.00, 2010 0.00
  - Bahamas, The: 1999 0.00, 2000 1.00, 2001 0.00, 2002 0.00, 2003 0.00, 2004 0.00, 2005 1.00, 2006 1.00, 2007 0.00, 2008 0.00, 2009 0.00, 2010 0.00
  - Barbados: 1999 1.00, 2000 0.00, 2001 0.00, 2002 0.00, 2003 0.00, 2004 0.00, 2005 0.00, 2006 0.00, 2007 0.00, 2008 0.00, 2009 0.00, 2010 0.00
  - Fiji: 1999 0.00, 2000 1.00, 2001 0.00, 2002 0.00, 2003 0.00, 2004 0.00, 2005 0.00, 2006 0.00, 2007 0.00, 2008 1.00, 2009 0.00, 2010 0.00
  - Jamaica: 1999 0.00, 2000 0.00, 2001 0.00, 2002 0.00, 2003 1.00, 2004 0.00, 2005 0.00, 2006 0.00, 2007 0.00, 2008 1.00, 2009 0.00, 2010 0.00
  - Mauritius: 1999 0.00, 2000 0.00, 2001 0.00, 2002 0.00, 2003 0.00, 2004 0.00, 2005 0.00, 2006 1.00, 2007 0.00, 2008 1.00, 2009 0.00, 2010 0.00
  - Sri Lanka: 1999 0.00, 2000 1.00, 2001 0.00, 2002 1.00, 2003 1.00, 2004 0.00, 2005 0.00, 2006 0.00, 2007 0.00, 2008 0.00, 2009 0.00, 2010 0.00
  - Tonga: 1999 1.00, 2000 0.00, 2001 0.00, 2002 0.00, 2003 1.00, 2004 0.00, 2005 0.00, 2006 0.00, 2007 0.00, 2008 0.00, 2009 0.00, 2010 0.00
  - Vanuatu: 1999 0.00, 2000 0.00, 2001 1.00, 2002 0.00, 2003 0.00, 2004 0.00, 2005 0.00, 2006 0.00, 2007 0.00, 2008 0.00, 2009 1.00, 2010 0.00
- Note: table contains blanks and ellipses for some country-year cells in the source.

### Conclusions and policy implications
- Cross-country heterogeneity:
  - Pronounced differences exist in determinants of reserve holdings across the distribution; EMs and SIs exhibit different reserve-demand elasticities and responses to short-term debt, broad money, and institutions.
  - For EMs, short-term debt and broad money are important positive determinants of reserves, consistent with precautionary insurance against rollover risk and deleveraging.
  - For SIs, short-term debt often associates with lower reserves (reflecting higher share of concessional debt); stronger institutions reduce the need for reserves in many SIs.
- Operational metric and policy guidance:
  - A user-friendly, operational metric for SIs is proposed (weighted function of broad money, short-term debt, and exports) with empirically derived performance.
  - Holding about 75-100 percent of this proposed metric reduces EMP crisis probability to less than 2 percent; the proposed metric outperforms traditional alternatives in predictive power.
  - Policymakers should consider country-specific risk profiles, debt structures (concessional vs. market debt), exposure to natural disasters, export concentration, and institutional quality when applying the metric and setting reserve targets.
- Areas for further research:
  - Investigation of debt-reserve dynamics in greater detail, particularly differences between concessional/multilateral and market short-term debt.
  - Examination of mutual insurance schemes and contingent financial instruments (e.g., Caribbean Catastrophic Risk Insurance Facility) and their impact on reserve adequacy.
  - Exploration of additional unobservable factors (rollover risk, market sentiment, risk aversion) that may explain residual variation in reserve holdings.

*Source: _wp12205 (IMF working paper content provided in the supplied PDF excerpt).*

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

### _wp12205 - References .............................................................................................................

### Introduction and contribution
- Examines reserves adequacy in emerging markets (EMs) and small islands (SIs).
- Three main contributions:
  - First paper to estimate observed cross-country preferences of reserves for 23 SIs, drawing on variants of the “buffer-stock model” and estimating OLS regression with fixed effects for SIs and comparing with 49 EMs.
  - First paper (to the authors’ knowledge) examining institutional, current and capital account motives for holding reserves allowing these factors to vary along the reserves’ holding distribution via quantile regressions; tests equality among slope coefficients of various quantile regressions and whether models at lower quantiles differ statistically from those at higher quantiles.
  - Develops a user-friendly operational metric for estimating reserve holdings in SIs building on the IMF (2011a, b) approach, tailored to economies with similar characteristics and likely BOP pressures.

### Key empirical findings
- Main drivers:
  - For SIs: import share and exchange rate volatility are key drivers of reserve holdings.
  - For EMs: import share and financial depth (broad money) are key drivers.
- Specific relationships observed:
  - Economies with higher import shares hold higher reserves (vulnerability to current account shocks).
  - SIs with more flexible exchange rate regimes tend to hold less reserves.
  - EMs: economies with higher broad money to GDP tend to hold higher reserves.
- Quantile regression findings:
  - Significant differences in cross-country preferences across the distribution of reserve holdings.
  - More significant heterogeneity within EMs than between EMs and SIs.
  - F-tests: for SIs, could not reject the null that models were similar across quantiles; for EMs, null was rejected in all quantiles except the 50th–75th interquantile regression.
  - Inter-quantile differences driven by elasticities of import share, short-term debt, and government effectiveness.
  - Short-term debt:
    - For EMs: positive effect on reserve holdings and increases across the reserve distribution.
    - For SIs: converse applies (short-term debt has the opposite effect).
  - Comparison to other literature:
    - Ghosh et al., 2012 found short-term debt has a large positive impact at the lower end for EMs but not at the positive tail.

### Reserve metric for SIs and crisis probability
- The SI reserve metric is a weighted function of broad money, short-term debt, and exports.
- Weights for SIs are higher than those implied by IMF (2011a, b) reflecting higher vulnerability to shocks.
- Empirical result: a reserve holding of about 75-100 percent of the metric reduces the probability of a crisis to less than 2 percent.
- Robustness tests suggest this metric performs better than others.
- Many SIs hold reserve levels either lower or significantly higher than suggested by the metric, raising questions about determinants of their reserve holdings.

### Modeling framework and data
- Reserve scaling and definition:
  - Scaled measure: r_{i,t} = R_{i,t} / GDP_{i,t} where R_{i,t} is international reserves and GDP_{i,t} is gross domestic product; both measured in US dollars.
  - Reserves exclude gold assets.
- Baseline panel model (notation preserved from source): equation (1) represents the baseline model with CrisisDum (EMP crisis dummy), time dummies t, and error term ϵ_{i,t}.
- Quantile regression extension: equation (2) modifies (1) to allow each elasticity coefficient to be a function of γ (the weight given to reserve policy); quantile regression method of Koenker and Bassett, 1978 is applied.
- Data and samples:
  - Panel model with fixed effects covering 23 SIs using annual data spanning 1999-2010.
  - Comparison models for 45 EMs spanning the same period in some analyses; elsewhere 49 EMs referenced for cross-country comparisons.
  - Out-of-sample prediction: models estimated using panel data for SIs from 1999 to 2007 with remaining years used for forecast comparison; exercise replicated for EMs.
- Variables and drivers considered (literature-informed grouping):
  - Current account vulnerabilities: imports, export earnings volatility, trade openness.
  - Capital account vulnerabilities: broad money, short-term debt, external financing exposures (FDI, portfolio flows).
  - Exchange rate regime: use of nominal effective exchange rate volatility to mitigate measurement problems from regime dummies.
  - Opportunity cost: expected negative correlation with reserves, though literature finds effect inconspicuous.
  - Controls: economic size (scaling), crisis episodes via an EMP index, institutions (government effectiveness, corruption), and other variables (tourism dependence, FDI volatility, interest rate differential).

### Empirical strategy and robustness
- Estimation methods:
  - Panel OLS with fixed effects; standard errors robust to within-cluster correlation (references to Froot, 1980; Williams, 2000; Wooldridge, 2002).
  - Quantile regressions to test varying elasticities across reserve distribution and to test the constant elasticity hypothesis.
- Model evaluation:
  - Focus on baseline models—model 2 for SIs and model 10 for EMs (as identified by the authors).
  - Out-of-sample forecast comparisons for average reserve growth and individual country holdings.
  - Robustness checks and comparative performance against alternative metrics using logit regressions (EMP event probability and crisis reduction performance).

*Source: _wp12205 - References .............................................................................................................*

### 0.2 percent in increase in the reserve-to-GDP ratio. This finding is significant and robust in

### _wp12205 - 0.2 percent in increase in the reserve-to-GDP ratio. This finding is significant and robust in

### Empirical determinants of reserve holdings (pooled, SIs, and EMs)
- A one percent acceleration in the broad money-to-GDP ratio for EMs is associated with about 0.2 percent increase in reserve-to-GDP holdings (model 10).
- For SIs, the change in broad money-to-GDP ratio is generally not statistically significant, except in one specification where it is associated with a 0.1 percent increase in reserve holdings (model 3) and in the OLS specification (model 1).
- Short-term debt to GDP:
  - Has a negative and weakly significant effect on reserve holdings in SIs (not robust across all specifications).
  - For EMs, it is positive and strongly significant: a one percent increase in short-term debt is associated with about 0.2 percent increase in short term debt (textual statement preserved as in source).
- Exchange rate flexibility:
  - Greater exchange rate flexibility is associated with lower reserve holdings in SIs but has no statistically significant impact on EMs except in the OLS specification.
  - Macroeconomic impact small: a one percent increase in flexibility is generally associated with a 0.01 percent point increase in the reserve-to-GDP ratio in SIs.
- Government effectiveness:
  - Stronger government effectiveness is associated with higher reserve holdings in SIs: a one percent increase in government effectiveness is associated with a 0.3 percent increase in reserve holdings in SIs (model 8).
  - No significant impact on EMs except in the OLS pooled specification (model 9).
- Precautionary motives explain on average about 30 percent of the variation in reserve holdings, though differences exist over time and across countries.
- Out-of-sample prediction since 2007 suggests reserves were slightly higher than levels suggested by fundamentals; the regression model does not track all individual country holdings well.

### Quantile regression results and heterogeneity across the reserve distribution
- Methodology:
  - Quantile regressions estimated at the 5th, 25th, 50th, 75th, and 95th quantiles (models eqn 3-7).
  - Findings include differences across distribution tails and evidence against constant elasticity for many variables (Tables 7-9, F-tests).
- Full sample (SIs + EMs):
  - Buffer-stock model predictions hold best for countries at the 50th quantile.
  - Short-term debt has a significant and positive impact on reserve holdings, rising to peak at the 95th quantile.
  - Change in broad money to GDP significant in all quantiles except tails.
  - Exchange rate flexibility significant only at the 50th quantile.
  - Import-to-GDP ratio strongly significant at most quantiles; coefficient size increases moving up the reserve distribution.
  - Government effectiveness significant across many quantiles, but sign varies (positive at tails, negative at 50th quantile).
- Small islands (SIs) subsample:
  - Little significant variation across quantiles overall (Table 8, F-test), but notable exceptions for exchange rate flexibility and short-term debt.
  - Short-term debt is strongly significant and has a large negative coefficient—peaking (in absolute terms) at the 75th quantile (Table 5). For higher-reserve SIs, increases in short-term debt lead them to hold lower reserves.
  - Broad money becomes significant for SIs (weakly) at the 25th and 50th quantiles.
  - Government effectiveness has a negative and significant impact on reserves for SIs at the 50th and 75th quantiles, suggesting stronger institutions reduce need for reserves.
- Emerging markets (EMs) subsample:
  - Significant differences across quantiles (Table 9, F-test) with large variation in coefficient size and sign as quantiles increase.
  - Short-term debt significant and positive, peaking at the 95th quantile (Table 6).
  - Broad money significant in most quartiles except tail ends.
  - Institutions have a positive and strongly significant effect at the 5th, 75th, and 95th quantiles.
- Stylized cross-sample facts:
  - Largest reserve holders are disproportionately EMs (right-hand of distribution).
  - Example: in 2009 median concessional debt to GDP was about 20 percent overall; for SIs with reserves at the 75th percentile concessional debt was about 84 percent, and about 11 percent for EMs.
  - Median multilateral debt about 25 percent overall; for SIs it ranged from 22 percent (economies with <25th percentile of reserves) to about 70 percent (those >75th percentile).

### A new reserve adequacy metric tailored to Small Islands (SIs)
- Rationale:
  - Existing metrics (extended Guidotti rule; Wijnholds and Kapteyn, 2001; Lipschitz et al., 2006; IMF 2011a,b) have limitations for SIs: arbitrary thresholds, lack of empirical backing, reliance on data not available for many SIs, and insufficient attention to SI-specific vulnerabilities (high trade openness, export concentration, natural disaster frequency, limited financial depth, concessional and multilateral debt structure).
- Metric design choices:
  - Core potential sources of balance-of-payments pressure for SIs: short-term debt on a remaining maturity basis, broad money, and exports of goods and services.
  - Exports of goods and services included (departure from IMF 2011b which focused on goods only), reflecting SIs’ reliance on tourism and service exports.
  - Thresholds for EMP episode detection modified: EMP critical threshold set to 1 (instead of 1.5 or 3) because SIs exhibit higher volatility; using this definition yields about 30 episodes of exchange market pressure between 1999 and 2010.
  - Bootstrapping used to obtain 10,000 random observations for each pressure source for robustness; 10th percentile used as threshold (adjusted for large outliers).
  - Additional buffer added: 5 percent general buffer; larger buffer of 25 percent for export drains to fixed exchange regimes.
- Empirical distribution of 10th percentile flows during EMP events (percent):
  - Currency union: Exports 12.4, Broad money 12.6, Short-term debt 75.9
  - Fixed: Exports 10.5, Broad money 3.6, Short-term debt 89.1
  - Floating: Exports 4.8, Broad money 24.4, Short-term debt 35.2
- Proposed reserve-adequacy weights (metric formulas):
  - Currency union: 20% of exports + 20% of broad money + 80% of short-term debt
  - Fixed: 35% of exports + 10% of broad money + 95% of short-term debt
  - Floating: 10% of exports + 30% of broad money + 40% of short-term debt
- Comparative properties:
  - The proposed metric is more stringent than the IMF, 2011 metric, placing more weight on exports and short-term debt.
  - Based on end-2010, only 29 percent of the sample was within the proposed metric.

### Thresholds, predictive performance, and robustness
- Estimation approach:
  - Binary logit model linking EMP events (dependent variable: 1 for crisis, 0 otherwise) to reserves divided by the proposed metric (reserves-to-proposed metric), controlling for government effectiveness and fiscal balance as a share of GDP.
  - Alternative regressions compare proposed metric to other metrics: three-month import coverage; broad money metric; Guidotti-Greenspan; maximum of three traditional metrics; Lipschitz metric. Sample: annual data for 21 SIs from 1999 to 2010 (includes Sri Lanka and excludes outliers such as Comoros).
- Key predictive findings:
  - The reserves-to-proposed metric ratio is strongly positively associated with a reduction in the probability of an EMP event.
  - Crisis probabilities rapidly decline as coverage reaches about 75-100 percent of the metric; this range could be regarded as adequate for a typical country.
  - The probability of a crisis for countries that hold about 75 percent of reserves implied by this metric is less than 2 percent.
  - The reserves-to-proposed metric performs markedly better than alternative single or combined traditional metrics in reducing crisis probability (Table 11, models 1-6).
  - The reserve-to-Lipschitz metric is the only traditional metric strongly significant with no other controls (model 11) but is weakly significant when including the proposed reserve-to-metric ratio (model 5).
- Cross-country status (end-2010):
  - About 37 percent of the sample was within the proposed adequacy range.
  - Roughly 32 percent had reserves above the range.
  - About 42 percent had reserves below the range.
  - Of 14 countries with cover below 100 percent, six have currency unions and five have fixed exchange rate regimes.
- Caveats:
  - Country-specific characteristics, vulnerabilities, and degree of risk aversion may justify deviations from the metric-implied reserves (for example, countries without short-term market debt may place no weight on that variable).
  - The drivers of reserve changes during crisis periods (2008-10) may differ from the pre-crisis period; model forecasts may not fully capture crisis-period drivers.
  - Some motives for holding reserves may be non-precautionary or reflect precautionary motives not captured by the model (e.g., risk aversion); ad hoc factors (currency swaps, increase in SDR allocations) may also affect observed reserves.

### Conclusions and policy implications
- Cross-country heterogeneity:
  - Pronounced differences exist in determinants of reserve holdings across the distribution; EMs and SIs exhibit different reserve-demand elasticities and responses to short-term debt, broad money, and institutions.
  - For EMs, short-term debt and broad money are important positive determinants of reserves, consistent with precautionary insurance against rollover risk and deleveraging.
  - For SIs, short-term debt often associates with lower reserves (reflecting higher share of concessional debt); stronger institutions reduce the need for reserves in many SIs.
- Operational metric and policy guidance:
  - A user-friendly, operational metric for SIs is proposed (weighted function of broad money, short-term debt, and exports) with empirically derived performance.
  - Holding about 75-100 percent of this proposed metric reduces EMP crisis probability to less than 2 percent, and the proposed metric outperforms traditional alternatives in predictive power.
  - Policymakers should consider country-specific risk profiles, debt structures (concessional vs. market debt), exposure to natural disasters, export concentration, and institutional quality when applying the metric and setting reserve targets.
- Areas for further research:
  - Investigation of debt-reserve dynamics in greater detail, particularly differences between concessional/multilateral and market short-term debt.
  - Examination of mutual insurance schemes and contingent financial instruments (e.g., Caribbean Catastrophic Risk Insurance Facility) and their impact on reserve adequacy.
  - Exploration of additional unobservable factors (rollover risk, market sentiment, risk aversion) that may explain residual variation in reserve holdings.

*Source: IMF working paper content provided in the supplied PDF excerpt.*

### References

### _wp12205 - References

### Key literature themes
- Extensive literature on international reserves, including motives for reserve accumulation, reserve adequacy, self-insurance, and links to capital flows and exchange rate regimes. Representative authors and works cited include: Aizenman; Marion; Bayoumi; Calvo; Reinhart; Obstfeld; Flood; Frenkel; Ghosh; Wijnholds; and IMF publications such as World Economic Outlook and “Assessing Reserve Adequacy.”
- Methodological references for econometric techniques and robustness: Koenker and Basett (1978); Koenker and Hallock (2001); Froot (1989); Williams (2000); Wooldridge (2002).

### Data, variables, and sample definitions
- Dependent variable:
  - Reserves: Stock of reserves as a share of GDP (data source: WEO, IMF).
- Key explanatory variables and rationales (data sources indicated):
  - Export volatility: Volatility of real exports receipts of goods and services (IFS, IMF).
  - Tourism volatility: Volatility of real travel receipts (IFS, IMF).
  - Remittance volatility: Volatility of real remittance receipts (IFS, IMF).
  - Aid volatility: Volatility of real aid receipts (IFS, IMF).
  - GDP pc growth: Real GDP per capita growth (WEO, IMF).
  - Import-to-GDP: Imports as a share of GDP (WEO, IMF).
  - Openness: Imports plus exports as a share of GDP (WEO, IMF).
  - Debt-to-GDP: Short-term debt as a share of GDP (WEO, IMF).
  - Money-to-GDP: Broad money as a share of GDP (WEO, IMF).
  - Interest rate: Interest differential with the US (WEO and IFS, IMF).
  - Government Effectiveness: Kaufmann and Kraay governance indicators (World Bank).
  - NEER volatility: Volatility of the nominal effective exchange rate (IMF).
  - Exchange rate dummy: Is 1 if flexible; 0 otherwise (AREAER, IMF).
  - Crisis dummy: Is 1 if crisis; 0 otherwise (constructed using EMP; authors’ estimates).
- Sample composition:
  - Full sample: Small islands and Emerging Market Economies; Number of panels = 67.
  - Small islands (SIs) sample: Number of panels = 23.
  - Emerging Markets (EMs) sample: Number of panels = 44.
- Appendix Table 1 lists country membership for SIs and EMs (not reproduced here per pipeline rules).

### Key empirical results and statistics (selected coefficients preserved exactly as reported)
- Appendix Table 4. Full sample (OLS and Quantile regression), Dependent variable: Reserve-to-GDP, Observations = 729, Number of panels = 67, R-squared (OLS) = 0.1865.
  - OLS (column 1):
    - d(Broad money/GDP) = 0.2945** (standard error 0.117)
    - Import-to-GDP = 0.0888*** (standard error 0.021)
    - NEER volatility = -0.0019** (standard error 0.001)
    - Short-term debt = 0.1716*** (standard error 0.030)
    - Crisis Dummy = 0.0080 (standard error 0.020)
    - Government Effectiveness = 0.0075 (standard error 0.007)
    - Constant = 0.0714*** (standard error 0.025)
  - Quantile results (selected):
    - 50th (median): d(Broad money/GDP) = 0.3050***; Import-to-GDP = 0.0709***; Short-term debt = 0.1130***.
    - 95th: Government Effectiveness = 0.0649*** (standard error 0.023).

- Appendix Table 5. Small islands (SIs): OLS and Quantile regressions, Dependent variable: Reserve-to-GDP, Observations = 247, Number of panels = 23, R-squared (OLS) = 0.2402.
  - OLS (column 1):
    - d(Broad money/GDP) = 0.2241*** (standard error 0.086)
    - Import-to-GDP = 0.0645** (standard error 0.026)
    - NEER volatility = -0.0063*** (standard error 0.001)
    - Short-term debt = -0.3902*** (standard error 0.091)
    - Government Effectiveness = -0.0124* (standard error 0.007)
    - Constant = 0.1543*** (standard error 0.024)
  - Quantile highlights:
    - 50th (median): d(Broad money/GDP) = 0.2654*; NEER volatility = -0.0026 (standard error 0.002); Short-term debt = -0.2780**.

- Appendix Table 6. Emerging markets (EMs): OLS and Quantile regressions, Dependent variable: Reserve-to-GDP, Observations = 482, Number of panels = 44, R-squared (OLS) = 0.2392.
  - OLS (column 1):
    - d(Broad money/GDP) = 0.3640** (standard error 0.166)
    - Import-to-GDP = 0.1820*** (standard error 0.025)
    - NEER volatility = -0.0019*** (standard error 0.001)
    - Short-term debt = 0.1262*** (standard error 0.038)
    - Government Effectiveness = 0.0296*** (standard error 0.011)
    - Constant = -0.0177 (standard error 0.037)
  - Quantile highlights:
    - 95th: d(Broad money/GDP) = 0.2764 (standard error 0.390); Short-term debt = 0.2708***.

- Appendix Table 7. Full sample: Inter-quantile regression results (Quantile differences and tests), Observations = 729.
  - Coefficient differences (5th-25th, 25th-50th, 50th-75th, 75th-95th, 5th-95th) for selected regressors:
    - d(Broad money/GDP): -0.1738; -0.0212; -0.00530; 0.23910; 0.0386 (standard errors in parentheses).
    - Import-to-GDP: -0.0673***; -0.0124; 0.0465*; -0.0052; -0.1314**.
    - Short-term debt: -0.0093; -0.0362; -0.0601*; -0.0348; -0.1409**.
    - Government Effectiveness: 0.0241***; 0.0111*; -0.0108; -0.0746***; -0.0502**.
  - Memo: F-test 1/ values and Prob>F:
    - F-test 1/ = 1.78** with Prob>F = 0.03 for 5th-25th comparison.
    - Other F-test values reported: 0.62 (Prob>F 0.87); 1.01 (Prob>F 0.44); 1.64** (Prob>F 0.05); 2.49*** (Prob>F 0.00).

- Appendix Tables 8 and 9. Inter-quantile regression results by sub-sample:
  - Small islands (Appendix Table 8), Observations = 247:
    - NEER volatility differences include 5th-95th = 0.0075** (standard error 0.004).
    - Government Effectiveness differences: 25th-50th = 0.0198**.
    - Memo F-tests show Prob>F values: 0.99, 0.69, 0.72, 0.67, 0.71 for the five quantile comparisons.
  - Emerging Markets (Appendix Table 9), Observations = 482:
    - Import-to-GDP: 25th-50th = -0.0990*** (standard error 0.027).
    - Short-term debt: 5th-25th = -0.0722*; 75th-95th = -0.2061**.
    - Government Effectiveness: 5th-25th = 0.0380***; 75th-95th = -0.1694***.
    - Memo F-tests: F-test values of 2.14*** (Prob>F 0.01), 1.50* (Prob>F 0.10), 0.99 (Prob>F 0.47), 1.53** (Prob>F 0.08), 2.69*** (Prob>F 0.00).

### Empirical patterns and implications (drawn from reported coefficients)
- Broad money growth (d(Broad money/GDP)) is positively associated with reserve-to-GDP in the full sample and sub-samples; reported OLS coefficients include 0.2945**, 0.2241*** (SIs), and 0.3640** (EMs).
- Import-to-GDP is positively associated with reserves across the full sample and EMs: 0.0888*** (full sample OLS) and 0.1820*** (EMs OLS).
- NEER volatility generally shows a negative association with reserve-to-GDP: -0.0019** (full sample OLS), -0.0063*** (SIs OLS), and -0.0019*** (EMs OLS).
- Short-term debt displays differing signs across sub-samples:
  - Full sample OLS: 0.1716***.
  - Small islands OLS: -0.3902***.
  - Emerging markets OLS: 0.1262***.
- Government Effectiveness exhibits varying relationships across quantiles and samples; notable coefficients include 0.0296*** (EMs OLS) and quantile-difference significance in the full sample (5th-25th = 0.0241***).

*Content unit: _wp12205 - References*

### Appendix Table 10.  SIs: Episodes of Exchange Market Pressure

### Appendix Table 10.  SIs: Episodes of Exchange Market Pressure

### Episodes of Exchange Market Pressure (EMP) — SIs, 1999–2010
- EMP = 1 if crisis episode; 0 otherwise.
- Source: Author’s estimates.
- Selected country-year EMP indicators as presented:
  - Antigua and Barbuda: 1999 0.00, 2000 1.00, 2001 0.00, 2002 0.00, 2003 0.00, 2004 0.00, 2005 0.00, 2006 0.00, 2007 0.00, 2008 0.00, 2009 0.00, 2010 0.00
  - Bahamas, The: 1999 0.00, 2000 1.00, 2001 0.00, 2002 0.00, 2003 0.00, 2004 0.00, 2005 1.00, 2006 1.00, 2007 0.00, 2008 0.00, 2009 0.00, 2010 0.00
  - Barbados: 1999 1.00, 2000 0.00, 2001 0.00, 2002 0.00, 2003 0.00, 2004 0.00, 2005 0.00, 2006 0.00, 2007 0.00, 2008 0.00, 2009 0.00, 2010 0.00
  - Belize: 1999 (blank), 2000 0.00, 2001 0.00, 2002 0.00, 2003 0.00, 2004 1.00, 2005 1.00, 2006 0.00, 2007 0.00, 2008 0.00, 2009 0.00, 2010 0.00
  - Cape Verde: 1999 (blank), 2000 0.00, 2001 0.00, 2002 0.00, 2003 0.00, 2004 0.00, 2005 0.00, 2006 0.00, 2007 0.00, 2008 0.00, 2009 0.00, 2010 (ellipsis)
  - Comoros: 1999 0.00, 2000 0.00, 2001 1.00, 2002 0.00, 2003 0.00, 2004 0.00, 2005 1.00, 2006 0.00, 2007 0.00, 2008 0.00, 2009 0.00, 2010 0.00
  - Dominica: 1999 0.00, 2000 0.00, 2001 0.00, 2002 0.00, 2003 0.00, 2004 1.00, 2005 0.00, 2006 0.00, 2007 0.00, 2008 0.00, 2009 0.00, 2010 0.00
  - Dominican Republic: 1999 0.00, 2000 0.00, 2001 0.00, 2002 0.00, 2003 1.00, 2004 0.00, 2005 0.00, 2006 0.00, 2007 0.00, 2008 0.00, 2009 0.00, 2010 0.00
  - Fiji: 1999 0.00, 2000 1.00, 2001 0.00, 2002 0.00, 2003 0.00, 2004 0.00, 2005 0.00, 2006 0.00, 2007 0.00, 2008 1.00, 2009 0.00, 2010 0.00
  - Grenada: 1999 0.00, 2000 0.00, 2001 0.00, 2002 0.00, 2003 0.00, 2004 0.00, 2005 1.00, 2006 0.00, 2007 0.00, 2008 0.00, 2009 0.00, 2010 0.00
  - Jamaica: 1999 0.00, 2000 0.00, 2001 0.00, 2002 0.00, 2003 1.00, 2004 0.00, 2005 0.00, 2006 0.00, 2007 0.00, 2008 1.00, 2009 0.00, 2010 0.00
  - Maldives: 1999 (blank), 2000 0.00, 2001 0.00, 2002 1.00, 2003 0.00, 2004 0.00, 2005 0.00, 2006 0.00, 2007 0.00, 2008 0.00, 2009 0.00, 2010 0.00
  - Mauritius: 1999 0.00, 2000 0.00, 2001 0.00, 2002 0.00, 2003 0.00, 2004 0.00, 2005 0.00, 2006 1.00, 2007 0.00, 2008 1.00, 2009 0.00, 2010 0.00
  - Samoa: entries include multiple blanks and 0.000 sequences for available years.
  - São Tomé & Príncipe: entries show 0.000 across listed years with ellipses for later columns.
  - Seychelles: 1999 0.00, 2000 0.00, 2001 0.00, 2002 0.00, 2003 0.00, 2004 0.00, 2005 0.00, 2006 0.00, 2007 0.00, 2008 1.00, 2009 0.00, 2010 0.00
  - Solomon Islands: 0.000 across years shown.
  - Sri Lanka: 1999 0.00, 2000 1.00, 2001 0.00, 2002 1.00, 2003 1.00, 2004 0.00, 2005 0.00, 2006 0.00, 2007 0.00, 2008 0.00, 2009 0.00, 2010 0.00
  - St. Kitts and Nevis: 1999 0.00, 2000 0.00, 2001 1.00, 2002 0.00, 2003 0.00, 2004 0.00, 2005 0.00, 2006 0.00, 2007 0.00, 2008 0.00, 2009 0.00, 2010 0.00
  - St. Lucia: 1999 0.00, 2000 0.00, 2001 0.00, 2002 0.00, 2003 0.00, 2004 1.00, 2005 0.00, 2006 0.00, 2007 0.00, 2008 0.00, 2009 0.00, 2010 0.00
  - St. Vincent & Grenadines: 1999 0.00, 2000 0.00, 2001 0.00, 2002 1.00, 2003 0.00, 2004 0.00, 2005 0.00, 2006 0.00, 2007 0.00, 2008 0.00, 2009 0.00, 2010 0.00
  - Tonga: 1999 1.00, 2000 0.00, 2001 0.00, 2002 0.00, 2003 1.00, 2004 0.00, 2005 0.00, 2006 0.00, 2007 0.00, 2008 0.00, 2009 0.00, 2010 0.00
  - Vanuatu: 1999 0.00, 2000 0.00, 2001 1.00, 2002 0.00, 2003 0.00, 2004 0.00, 2005 0.00, 2006 0.00, 2007 0.00, 2008 0.00, 2009 1.00, 2010 0.00
- Note: table contains blanks and ellipses for some country-year cells in the source.

### Observed patterns from Table 10
- Episodes are binary (1.00 crisis, 0.00 no crisis) and reported annually for 1999–2010.
- Several small island economies exhibit isolated EMP episodes across the period (examples: Bahamas, Barbados, Fiji, Jamaica, Mauritius, Sri Lanka, Tonga, Vanuatu).
- Some entries include missing values or ellipses; the dataset as presented is not fully populated for every country-year.

---

### Appendix Table 11.  Comparison on Various Reserve Adequacy Metrics: Logit Regression

### Regression setup and sample
- Dependent variable: EMP events.
- Observations: 164 (for all reported specifications).
- Standard errors in parentheses.
- Significance codes: *** p<0.01, ** p<0.05, * p<0.1.
- Source: Staff estimates.
- Note: "Maxmetric" equals the maximum of 100 percent of short-term debt, 20 percent of broad money, and three months of imports.

### Reported regression coefficients (selected)
- Reserves/metric t-1:
  - Column 1: -3.771*** (1.361)
  - Column 2: -3.833*** (1.404)
  - Column 3: -4.565** (1.926)
  - Column 4: -3.797*** (1.446)
  - Column 6: -3.986*** (2.665)
  - Column 7: -2.812** (1.378)
- Fiscal balance-to-GDP t-1:
  - Column 1: 0.0379* (0.0203)
  - Column 2: 0.0370* (0.0204)
  - Column 3: 0.0390* (0.0207)
  - Column 4: 0.0374* (0.0219)
  - Column 5: 0.0495** (0.0237)
  - Column 6: 0.0396* (0.0205)
- Government effectiveness t-1:
  - Column 1: 0.813 (1.435)
  - Column 2: 0.789 (1.435)
  - Column 3: 0.972 (1.466)
  - Column 4: 0.811 (1.432)
  - Column 5: 1.863 (1.561)
  - Column 6: 0.557 (1.536)
- Reserves/(CA deficit + s.term debt) metric t-1:
  - Column 7: 0.0385 (0.148)
  - Column 8: 0.0215 (0.0816)
- Reserves/broad money metric t-1:
  - Column 9: 0.421 (0.704)
  - Column 10: -0.934* (0.547)
- Reserves/imports metric t-1:
  - Column 11: -0.0832 (1.548)
  - Column 12: 0.500 (1.062)
- Reserves/Lipschitz metric t-1:
  - Column 1 (alternative): -5.279* (3.119)
  - Column 2 (alternative): -2.771** (1.143)
- Reserves/Maxmetric t-1:
  - Column 1: -0.529 (0.502)
  - Column 2: -0.268 (0.389)
- Constant terms (selected):
  - Column 1: -1.358 (1.687)
  - Column 4: -1.360 (1.684)
  - Column 8: -2.113*** (0.263)
  - Column 10: -1.088* (0.589)
  - Column 12: -2.007*** (0.272)

### Key regression insights implied by coefficients
- Negative and statistically significant coefficients on several Reserves/metric t-1 specifications (e.g., -3.771***, -3.833***, -4.565**) indicate higher reserves relative to the chosen metric are associated with lower probability of EMP events in those specifications.
- Fiscal balance-to-GDP t-1 shows small positive coefficients, some statistically significant at the 10 percent or 5 percent level (e.g., 0.0495**), indicating the fiscal balance variable enters positively in these specifications.
- Reserves/Lipschitz metric shows large negative coefficients in some specifications (e.g., -5.279*, -2.771**), suggesting sensitivity of EMP probability to this metric in those models.
- Some reserve metrics (Reserves/broad money, Reserves/imports, Reserves/Maxmetric) show mixed signs and significance across specifications.

---

### Figures and Empirical Diagnostics (figures referenced in source)
- Figure 1. EM and SIs: Traditional Metrics, 2000-2010
  - Metrics illustrated: Reserves in months of imports (benchmark: 3 months), Reserves in percent of broad money (benchmark: 20 percent), Reserves in percent of short-term debt (benchmark: 100 percent).
  - Series plotted: Emerging markets, Small islands, Benchmark lines.

- Figure 2. Sample Mean: Actual and Predicted Reserves, 2007-2010 (In percent of GDP)
  - Note: Out of sample prediction of baseline model (2 and 10) from Table A3.
  - Panels compare Actual Reserves and Predicted Reserves for Small Island Economies and Emerging Market Economies over 2005–2010.

- Figures 3–5. Comparison of OLS and Quantile Regression Coefficient Estimates
  - Figure 3: Full Sample.
  - Figure 4: SIs (Small Island Economies).
  - Figure 5: EM (Emerging Markets).
  - Coefficients displayed across quantiles (.05, .25, .5, .75, .95) for variables including d_fin, z, neer_v, stdebt, DV_EMP, kauf with scales and magnitudes varying by sample.

- Figure 6. EM and SIs: Concessional and Multilateral Debt, 1999-2009
  - Panels include:
    - Concessional debt in percent of external debt, for countries with less than 25th quantile of reserve-to-imports.
    - Median concessional debt in percent of external debt.
    - Concessional debt in percent of external debt, for countries with more than 75th quantile of reserve-to-imports.
    - Multilateral debt in percent of external debt, for countries with less than 25th quantile of reserve-to-imports.
    - Median multilateral debt in percent of external debt.
    - Multilateral debt in percent of external debt, for countries with more than 75th quantile of reserve-to-imports.

- Figure 7. SIs: Exchange Market Pressure Crisis Episodes
  - Panels show total number of EMP episodes and frequency of EMP episodes (in percent), by exchange rate regime (Currency Board Arrangement, Floating, Fixed) and year for 1999–2010.

- Figure 8. SIs: Exchange Market Pressure Crisis Episodes Triggers (in percent of total)
  - Triggers categorized by Natural disaster, Terms of trade, Capital flight, Political, Other; sample groups include Full sample, Currency unions, Fixed exchange rate regimes, Flexible exchange rate regimes.

- Figure 9. The Dominican Republic and Grenada: Balance of Payments Flows In Crisis Events (in percent of GDP)
  - Dominican Republic panel: 2002–2004 series with y-axis marks at -8, -3, 2, 7, 12.
  - Grenada panel: 2004–2006 series for Current account, FDI net, Other official inv. net, Other private inv. net, Capital account, Official portfolio inv. net, Private portfolio inv. net, Change in reserves.

- Figure 10. Distribution of Export, Broad Money, and Short-Term Debt
  - Histograms by exchange-rate regime (currency unions, fixed ER, flexible ER) for export drain, stermdebt/debt service, broad money.

- Figure 11. New Metric vs. Maximum of Traditional Metrics (in months of imports)
  - Comparison plotted using WEO and author estimates.

- Figure 12. EMP Event Probability (taking into account primary balance)
  - Panel displays up to the 40 percent probability of a crisis.
  - Horizontal axis: Reserves/Metric (in percent); vertical axis: probability from 0 to 40 percent.

- Figure 13. Reserves Against Risk-Weighted Metric (reserves as percent of metric, 2010)
  - Vertical scale shown from 0 to 200 with a labeled "Suggested adequacy range".

---

*Source: _wp12205 - Appendix Table 10. SIs: Episodes of Exchange Market Pressure (Author’s and Staff estimates, figures and tables as provided in source).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp12205.pdf_