## _wp04189

## Source details

**Canonical URL:** [_wp04189](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2004/_wp04189.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2004/_wp04189.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2004/_wp04189.pdf.json)

---

### Background and purpose
- Purpose: investigate whether holding higher reserves may reduce the volatility of the real exchange rate independently of the influence of the choice of exchange rate regime or of the role of foreign exchange intervention.
- Relevance: at the end of 2002, as many as 25 out of 31 emerging market economies could be classified as having freely floating or managed floating exchange regimes.
- Framing: real exchange rate volatility is a policy concern in many emerging market economies; the paper tests for a third benefit of reserves—lowering real exchange rate volatility—alongside reduced crisis likelihood and lower external borrowing costs.

### Data and sample
- Sample: annual observations of 28 emerging market countries for 1994–2002 (primary sample); secondary sample of 22 countries for 1986–2002.
- 28-country list (used in the study): Argentina, Brazil, Chile, China, Colombia, Czech Republic, Egypt, Hungary, India, Indonesia, Jordan, Korea, Malaysia, Mexico, Morocco, Nigeria, Pakistan, Peru, Philippines, Poland, Russia, Slovak Republic, South Africa, Sri Lanka, Thailand, Turkey, Venezuela, and Zimbabwe.
- Regime classification in sample:
  - Fixers: China, Egypt, Hungary, India, Jordan, Morocco, Pakistan, Peru, and Sri Lanka.
  - Floaters: Czech Republic, Indonesia, Korea, Mexico, Nigeria, Philippines, Russia, South Africa, Thailand, Turkey, Venezuela, and Zimbabwe.
- Key variable definitions:
  - vdreer: Standard deviation of monthly changes of the natural logarithm of the real effective exchange rate (REER).
  - resstd: Ratio of reserve assets to short-term debt on a remaining maturity basis.
  - Other reserve measures: resi (reserves-to-imports), resy (reserves-to-GDP).
  - invexpRS / invexpRI / invexpRY: inverse of the exponential of resstd, resi, resy, respectively.
  - logresstd_L: lagged natural log of resstd (lagged one period in all regressions).
  - interv: proxy for foreign exchange intervention (standard deviation of monthly difference in the ratio of reserves to GDP).

### Descriptive evidence (1994–2002, 28 EMs)
- Visual and descriptive patterns:
  - Casual plot (Figure 1, Panel A) indicates a negative relationship between real exchange rate volatility and the ratio of reserves to short-term debt for 1994–2002.
  - Relationship is non-linear: benefits of holding reserves for lowering volatility diminish at higher reserve levels.
  - Specific thresholds:
    - All incidences of observed annual volatility higher than 5 percent were associated with reserve coverage ratios below 2.
    - Most observations of high volatility were observed in the 0-1 range.
  - Robustness: Panel B of Figure 1 shows the negative relationship persists when outliers (observations > 2 standard deviations from the mean) are removed.
- Main descriptive conclusion:
  - Strong descriptive evidence of a non-linear negative relationship between the ratio of reserves to short-term debt and REER volatility in the 1994–2002 sample.

### Methodology and identification logic
- Dependent variable: y = vdreer (annual standard deviation of monthly log changes of the REER).
- Linear model: y = α + β X + ε.
- Econometric tools to address endogeneity and country effects:
  - Lagging reserve adequacy by one period.
  - System-generalized methods of moments (diff-GMM / system-GMM) using lags > 2.
  - Identification through heteroscedasticity (IH) estimator (Rigobon (2003)).
  - Inclusion of exchange rate regime dummy (exrdum) based on Reinhart and Rogoff (2002).
  - Inclusion of variables influencing regime choice and an intervention proxy (interv).
- Interpretation logic:
  - If reserve adequacy remains significant after controlling for fundamentals, policy actions, and sentiment variables, the effect is likely via market sentiment rather than direct intervention.

### Main empirical results and robustness
- Reserve adequacy and volatility (1994–2002, 28 EMs):
  - Evidence that reduced REER volatility is associated with higher reserves relative to short-term debt:
    - Positive coefficient on invexpRS and negative coefficient on logrestd_L reported in multiple specifications.
  - Robustness:
    - Effect persists after endogeneity controls (system-GMM, IH), lagging, and inclusion of regime-influencing variables.
    - Effect mainly present for floaters:
      - invexpRSD or logrestd_LD (interaction with floating dummy) coefficients are positive and significant while whole-sample invexpRS or logrestd_L often lose significance.
      - Excluding fixers or including only countries floating at least half the time increases the effect size.
    - Effect does not appear to operate through measured foreign exchange intervention:
      - invexpRS or logrestd_L remain significant when interv is included; interv_ reported as 0.049 (z = 1.48) in one specification.
  - Sample-period sensitivity:
    - Using 1986–2002 sample, effect is virtually absent.
    - Using 1994–2002 sample, effect is robust.
    - Difference not due to country composition; 1994–2002 effect persists with the smaller country set available for 1986–2002.
- Other robust determinants (across specifications):
  - Growth (gro): negative and significant (e.g., gro_ -0.208 (z = 4.06)*** in Table 2 Main).
  - Fiscal balance (fis): positive and significant in many specifications (e.g., fis_ 0.082 (z = 2.07)** in Table 2 Main).
  - Monetary volatility (mon): highly significant (e.g., mon_ 0.775 (z = 2.88)*** in Table 2 Main); a 1 percentage point increase in mon would increase exchange rate volatility by 0.7-0.8 percentage-points (or by about 20 percent of average sample volatility in 2002).
  - External debt (extfin): positive and significant (e.g., extfin_ 0.060 (z = 3.54)*** in Table 2 Main).
  - Country size (rpppwgt): positive and significant (e.g., rpppwgt_ 0.001 (z = 2.75)*** in Table 2 Main).
  - Financial deepness (findeap): generally not significant in these regressions.
- Econometric diagnostics (system-GMM):
  - Hansen test: does not reject validity of over-identifying restrictions.
  - Arellano-Bond test: does not reject no second-order autocorrelation.
  - Both tests generally yield probability-values way above the 10 percent threshold.

### Selected key regression estimates (exact reported estimates)
- Table 2 (1994–2002, 28 EMs; Observations 252; No. of countries 28) — Main specification:
  - invexpRS: 0.016 (z = 2.18)**  
  - invexpRSD: 0.031 (z = 2.62)***  
  - gro_: -0.208 (z = 4.06)***  
  - fis_: 0.082 (z = 2.07)**  
  - mon_: 0.775 (z = 2.88)***  
  - exrdum_: 0.007 (z = 1.89)*  
  - extfin_: 0.060 (z = 3.54)***  
  - rpppwgt_: 0.001 (z = 2.75)***  
  - Interv (when included): interv_ 0.049 (z = 1.48)
- Table 3 (1994–2002, sample changes):
  - invexpRS (No fixers): 0.021 (z = 1.79)*  
  - invexpRS (Only floaters): -0.002 (z = 0.10)  
  - gro_: around -0.190 to -0.205 (significant at 1% in multiple samples)  
  - fis_: e.g., 0.103 (z = 3.22) in some subsamples  
  - mon_: e.g., 0.987 (z = 3.02) significant at 1%  
  - extfin_: e.g., 0.083 (z = 6.10) significant
- Table 4 & Table 8 (other reserve measures):
  - invexpRI: Main 0.012 (t = 0.87) — not significant in Main  
  - invexpRY: Main 0.111 (t = 2.06)**; invexpRYD: 0.049 (t = 0.87)  
  - logresy_L: -0.008 (t = 2.32)** in one specification; logresy_LD: -0.005 (t = 0.92)  
  - logresi_L: -0.003 (t = 1.23) in Main
- Longer sample (1986–2002, 22 EMs; Tables 5 and 9):
  - invexpRS: 0.006 (t = 1.23) Main; invexpRSD: 0.018 (t = 1.18)  
  - gro_: -0.180 (t = 3.57)*** in Main  
  - mon_: 0.511 (t = 3.02)*** in Main  
  - exrdum_: 0.018 (t = 4.55)*** in Main (1986–2002)  
  - extfin_: 0.027 (t = 1.99)** in Main
- Table 6 & Table 7 (lagged log resstd and first differences):
  - logresstd_L: -0.004 (t = 1.66)* in Main (1994–2002)  
  - logresstd_LD: -0.008 (t = 2.64)*** in Main for first difference in one specification
- Table 10 (IH estimations, 1994–2002):
  - invexpRS: 0.025 *** (IH main)  
  - With intervention proxy: 0.033 ***
- Table 11 (OLS and Fixed Effects, 1994–2002, 28 EMs; Observations 252):
  - OLS: invexpRS 0.042 (t = 5.01)***; logresstd_L -0.010 (t = 4.84)***; gro_ -0.224 (t = 5.30)***; mon_ 0.581 (t = 3.89)***; extfin_ 0.058 (t = 4.47)***; rpppwgt_ 0.001 (t = 4.12)***. R-squared 0.12.
  - FE: invexpRS 0.069 (t = 3.89)***; logresstd_L -0.016 (t = 3.49)***; gro_ -0.193 (t = 4.52)***; mon_ 0.759 (t = 3.32)***; extfin_ 0.091 (t = 5.91)***; rpppwgt_ 0.000 (t = 0.10). R-squared 0.32.
- Observations and sample sizes:
  - Typical panel regressions: Observations 252, No. of countries 28 (1994–2002 main sample).
  - Longer-sample regressions: Observations 374, No. of countries 22 (1986–2002 sample).
- Statistical significance notation:
  - * Significant at 10%; ** significant at 5%; *** significant at 1%.

### Quantification, non-linearity, and magnitudes (Box 1 illustrations)
- Non-linearity:
  - The higher the level of reserves, the smaller the effect of a given increase in reserves on REER variability.
  - For any level of the reserve ratio, the effect of a decrease is larger than the effect of an equal-magnitude increase.
  - Logarithmic specification implies a flatter (smaller) impact than inverse-exponential.
- Illustrative scenario for the average country in 1994–2002:
  - Halving reserves-to-short-term-debt from 170 percent to 85 percent:
    - Exponential specification (coefficient ≈ 0.02 for invexpS): increase in REER volatility by almost 20 percent; probability of a change in REER larger than two times its initial standard deviation would double from 5 to 10 percent.
    - Logarithmic specification (coefficient ≈ -0.004 for logresstd_L): effect almost half that of exponential specification.
  - For floaters:
    - Appropriate coefficient about 0.04 (sum of invexpS and invexpSD), indicating effect twice as large in coefficient terms.
    - Average reserve ratio for floaters is 100 percent and exchange rate volatility is 4.4 percent.
    - When these values are taken into account, the effect for floaters of halving the reserve ratio turns out similar to the whole-sample illustration above.
  - Asymmetry: increasing reserve ratio from 170 percent to 255 percent would imply a decline in volatility by less than 10 percent.

### Policy implications and conclusions
- Main empirical conclusions:
  - Strong evidence of a negative non-linear effect of reserves on short-term REER volatility for emerging market countries in 1994–2002.
  - Effect is robust to endogeneity controls (system-GMM, IH), lagging, and inclusion of regime-choice variables.
  - Effect is not large in magnitude: halving reserves-to-short-term-debt for the average country in 1994–2002 could increase REER volatility by up to 20 percent, nearly doubling the probability of very large swings.
  - Effect appears mainly associated with floaters, though overall impact is similar when accounting for sample differences.
- Channel and further research:
  - Effect does not appear to operate via observed use of reserves for intervention (as measured); may operate through signaling potential for intervention or increased market “comfort level.”
  - Further research needed to identify channels, including better proxies for exchange market intervention and policy intentions.
- Policy considerations:
  - Other policy actions (credible, stability-oriented monetary policy and external debt management) may be at least as powerful as reserve accumulation in reducing REER volatility.
  - Results do not constitute unequivocal advocacy for increasing reserves:
    - Benefits mainly confined to countries with relatively low reserves.
    - Benefits (reduced volatility, lower interest rates, reduced crisis risk) must be weighed against costs of holding reserves.

### Key descriptive statistics (Table 1: 1994–2002, 28 EMs)
- vdreer: Mean 0.028 SD 0.032 Between SD 0.017 Within SD 0.028
- resstd: Mean 1.668 SD 1.836 Between SD 1.629 Within SD 0.899
- resi: Mean 5.470 SD 2.822 Between SD 2.464 Within SD 1.444
- resy: Mean 0.145 SD 0.087 Between SD 0.077 Within SD 0.043
- gro: Mean 0.032 SD 0.044 Between SD 0.021 Within SD 0.039
- fis: Mean -0.038 SD 0.038 Between SD 0.026 Within SD 0.028
- infl: Mean 0.148 SD 0.271 Between SD 0.150 Within SD 0.226
- tot: Mean 0.050 SD 0.065 Between SD 0.041 Within SD 0.051
- tropen: Mean 0.691 SD 0.414 Between SD 0.405 Within SD 0.111
- mon: Mean 0.022 SD 0.017 Between SD 0.012 Within SD 0.013
- exrdum: Mean 0.377 SD 0.486 Between SD 0.332 Within SD 0.360
- findeap: Mean 0.555 SD 0.325 Between SD 0.320 Within SD 0.078
- extfin: Mean 0.511 SD 0.230 Between SD 0.193 Within SD 0.131
- rppppc: Mean 62.079 SD 35.900 Between SD 36.043 Within SD 5.578
- rpppwgt: Mean 4.976 SD 8.225 Between SD 8.183 Within SD 1.677
- interv: Mean 0.075 SD 0.062 Between SD 0.039 Within SD 0.049

*Source: _wp04189 (selected excerpts).*

### 1. Descriptives: 1994–2002, 28 Emerging Markets.................................................................19

### 1. Descriptives: 1994–2002, 28 Emerging Markets.................................................................19

### Background and purpose
- Purpose: investigate whether holding higher reserves may reduce the volatility of the real exchange rate independently of the influence of the choice of exchange rate regime or of the role of foreign exchange intervention.
- Relevance: at the end of 2002, as many as 25 out of 31 emerging market economies could be classified as having freely floating or managed floating exchange regimes.
- Framing: real exchange rate volatility is a policy concern in many emerging market economies; existing literature emphasizes reserves’ role in reducing currency-crisis likelihood and external borrowing costs, and this paper tests for a third benefit—lowering real exchange rate volatility.

### Data definitions (as used in figures and analysis)
- Reserve adequacy (resstd): measured as the ratio of gross reserves to short-term debt.
- Volatility of the real exchange rate (vdeer): measured as the standard deviation of monthly changes in the natural logarithm of the real effective exchange rate.
- Sample: annual observations of 28 emerging market countries for 1994–2002.

### Descriptive patterns and visual evidence
- Casual plot observation (Figure 1, Panel A) indicates a negative relationship between real exchange rate volatility and the ratio of reserves to short-term debt for 1994–2002.
- Relationship appears non-linear: benefits of holding reserves for lowering volatility diminish at higher reserve levels.
- Specific descriptive thresholds observed:
  - All incidences of observed annual volatility higher than 5 percent were associated with reserve coverage ratios below 2.
  - Most observations of high volatility were observed in the 0-1 range.
- Robustness of pattern:
  - Panel B of Figure 1 shows the negative relationship persists when “outliers” (observations that deviated by more than 2 standard deviations from the mean) are removed.

### Main descriptive conclusion
- There is strong descriptive evidence of a non-linear negative relationship between the ratio of reserves to short-term debt and real effective exchange rate volatility in the 1994–2002 sample of 28 emerging market countries.

### Links to broader empirical context (from the source)
- The paper situates this potential third benefit of reserves alongside established findings that higher reserve adequacy:
  - Reduces the likelihood of currency crises or “sudden stops.”
  - Is associated with lower external borrowing costs via improved confidence and sovereign credit ratings.
- The analysis acknowledges potential endogeneity from exchange rate regime choice and considers foreign exchange intervention; the descriptive patterns motivate formal econometric tests reported elsewhere in the paper.

*Source: _wp04189 - 1. Descriptives: 1994–2002, 28 Emerging Markets (extracted content).*

### conclusions.

### conclusions.

### Key findings on exchange rate volatility
- There is much more volatility of exchange rates than can be explained by the volatility of macroeconomic fundamentals, even in models that assume sticky prices, sluggish money adjustment, and time-varying risk premia.
- Country differences in exchange rate volatility are linked to macroeconomic variables, capital control measures, foreign exchange regime, and market structure.
- Balance sheet factors matter, including the size of external debt, domestic market liquidity, country size, and level of development.
- The choice of exchange rate regime can reduce currency volatility, given that prices are normally stickier than the exchange rate.

### Insights from related studies
- Devereux and Lane (2003)
  - Theoretical model: choice of exchange rate flexibility (volatility of nominal bilateral rates) based on optimal currency area considerations with credit constraints.
  - Empirical test: monthly data averages from January 1995 to September 2000 for a large cross-section of industrial and developing countries.
  - Variables: bilateral trade as a share of GDP, standard deviation of bilateral growth rate differentials, log of the product of two countries’ GDP, size of the domestic financial sector, bilateral external debt, and GDP per capita in PPP terms.
  - Finding: all these variables play a significant role in determining bilateral volatility of the exchange rate.
- Canales-Kriljenko and Habermeier (1999)
  - Cross-section of 85 developing and transition economies.
  - Nominal exchange rate volatility measured by the standard deviation of daily percentage changes in the nominal effective exchange rate during 2001.
  - Model selection algorithm results: most important macroeconomic determinants of nominal exchange rate volatility were inflation, real GDP growth, fiscal deficit (in percent of GDP), and trade openness.
  - Reserve adequacy measures were not chosen by the model selection algorithm and were not retained.

### Our approach: objective and identification logic
- Objective: investigate the role of reserve adequacy in reducing volatility of the real effective exchange rate, controlling for volatility of fundamentals, policy actions to stabilize the exchange rate, and variables influencing market sentiment.
- Interpretation logic:
  - Reserve levels may reflect policy choice (capacity to intervene) and market sentiment (signal of ability to intervene or lower crisis probability).
  - If reserve adequacy remains significant after controlling for fundamentals, policy actions, and sentiment variables, the effect is likely via market sentiment rather than policy intervention.

### Methodology
- Dependent variable y: annual observations of real effective exchange rate volatility (vdreer) defined as the annual standard deviation of the monthly changes of the natural logarithm of the real effective exchange rate.
- Linear model estimated: y = α + β X + ε
- Econometric tools used to address endogeneity and country-specific effects:
  - Lagging reserve adequacy indicators by one period.
  - System-generalized methods of moments (diff-GMM / system-GMM) using lags of levels and first differences (all available lags larger than 2 allowed).
  - Identification through heteroscedasticity (IH) estimator developed by Rigobon (2003).
  - Inclusion of a dummy for exchange rate regime (exdrum) based on Reinhart and Rogoff (2002).
  - Inclusion of variables that influence regime choice and, in some specifications, a proxy for foreign exchange intervention (interv).

### Sample
- Focus: emerging market economies.
- Primary sample: 28 countries for the 1994–2002 period.
- Secondary sample: 22 countries for the 1986–2002 period (excluding the transition economies and Colombia).
- Country lists and classification notes (as used in the study):
  - The 28-country list includes Argentina, Brazil, Chile, China, Colombia, Czech Republic, Egypt, Hungary, India, Indonesia, Jordan, Korea, Malaysia, Mexico, Morocco, Nigeria, Pakistan, Peru, Philippines, Poland, Russia, Slovak Republic, South Africa, Sri Lanka, Thailand, Turkey, Venezuela, and Zimbabwe.
  - Fixers: China, Egypt, Hungary, India, Jordan, Morocco, Pakistan, Peru, and Sri Lanka.
  - Floaters: Czech Republic, Indonesia, Korea, Mexico, Nigeria, Philippines, Russia, South Africa, Thailand, Turkey, Venezuela, and Zimbabwe.

### Choice and construction of key variables
- Endogenous variable: vdreer (annual standard deviation of monthly log changes of the REER).
- Reserve adequacy measures:
  - Primary: ratio of reserve assets to short-term debt on a remaining maturity basis (resstd).
  - Alternatives: reserve-assets-to-imports (resi) and reserve-assets-to-GDP (resy).
  - Non-linear transformation applied: inverse of the exponential yielding invexprS, invexprI, and invexprY.
- Control variable sets:
  - Macroeconomic: growth (gro), inflation (infl), fiscal balance to GDP ratio (fis).
  - Volatility-of-fundamentals: monetary volatility (mon), terms of trade volatility proxied by the absolute annual change (tot).
  - Regime-related: exchange rate dummy (exdrum), trade openness (tropen), domestic financial deepness (findeap), external financial exposure (extfin), economic size (rpppwgt), level of economic development (rppppc).
  - Foreign exchange intervention proxy: interv (annual volatility of monthly changes in reserves).

### Main empirical results and robustness
- Reserve adequacy and volatility
  - Strong evidence that reduced volatility of the real effective exchange rate is associated with a higher level of reserves relative to short-term debt, as indicated by:
    - positive coefficient on invexprS
    - negative coefficient on logrestd_L
  - Robustness:
    - Result persists after controlling for endogeneity via inclusion of regime-influencing variables and using system-GMM and IH estimation.
    - Effect appears mainly present for floaters:
      - When invexprSD or logrestd_LD (interaction with floating dummy) are entered, their coefficients are positive and significant while invexprS or logrestd_L lose significance.
      - Excluding fixers or including only countries floating at least half the time increases the effect size.
    - The impact does not appear to operate through measured foreign exchange intervention:
      - invexprS or logrestd_L remain significant when the intervention proxy (interv) is included.
      - Caveat: the proxy may be limited and intervention is characteristic of fixed regimes which are already controlled for.
  - Sample period sensitivity:
    - The impact of reserve adequacy on currency volatility appears to be a recent phenomenon:
      - Using the 1986–2002 sample, the effect is virtually absent (Tables 5 and 9).
      - Using the 1994–2002 sample, the effect is very robust.
      - The difference is not due to country composition; the 1994–2002 effect persists even with the smaller country set available for 1986–2002.

- Other robust determinants
  - Five regressors are robustly significant across specifications: growth (gro), fiscal balance (fis), monetary volatility (mon), external debt (extfin), and country size (rpppwgt).
  - Monetary volatility (mon) is highly significant, highlighting the importance of monetary policy for short-term real exchange rate volatility.
  - Financial deepness (findeap) is generally not significant in these regressions, differing from some earlier studies.
  - Fiscal balance coefficient sign is puzzling: a positive sign implies that a higher deficit can be expected to result in lower volatility of the real effective exchange rate.

### Econometric diagnostics
- For system-GMM regressions:
  - Hansen test: does not reject the null that over-identifying restrictions are valid.
  - Arellano-Bond test: does not reject the null that there is no second-order autocorrelation.
  - Both tests generally yield probability-values way above the 10 percent threshold.

*Source: _wp04189 - conclusions.*

### Box 1. How to Interpret Tables 2-10

### Box 1. How to Interpret Tables 2-10

### Interpretation of reserve indicator transformations and coefficients
- Tables 2-5 and 10 use the inverse of the exponential of the reserve measure (invexprS); Tables 6-9 use a (natural) logarithmic transformation of the reserve measure (logrestd_L).
- Reserve measures are multiplied by the flexible exchange rate dummy when presenting a D at the end of the name (e.g., invexprSD, logrestd_LD).
- Key sign interpretations:
  - A positive coefficient on invexprS implies a negative non-linear effect of the reserves to short-term debt ratio on exchange rate volatility (because invexprS declines when the reserve ratio increases).
  - The same effect is captured by a negative logrestd_L.
  - A positive coefficient for invexprSD (invexprS multiplied by the exchange rate dummy) indicates a “larger” effect of the reserves to short-term debt ratio under the floating exchange rate regime. The same effect is captured by a negative logrestd_LD.

### Table headings and sample treatments (Tables 2–9)
- “Main” — main specification.
- “Only sig.” — retention of only variables that are significant in the main specification.
- “No outliers” — elimination of countries that appeared as outliers with high leverage (to ensure results are not driven by a few observations), as visible in Figures 2 and 3 for regressions with fixed effects in the two samples.
- “W/ interv” — inclusion of the intervention proxy.
- “No fixers” — elimination of countries that were pure fixers (exchange rate dummy always equal to 0) throughout the sample.
- “Only floaters” — retention of only countries that were floating at least half of the time (exchange rate dummy always larger than 0.5, on average).
- “86 sample” — retention only of the 22 countries that are present in the 1986–2002 sample (to ensure comparability of results across the two samples).

### Table 10 and IH estimations
- Table 10 reports IH estimations for regressions with all control variables and with only significant ones for both samples, reporting only the coefficient estimate on invexprS.
- Due to the nature of the methodology, the exchange rate dummy was dropped from the set of variables (even when included, the results do not vary significantly).

### Quantifying the estimated effects (non-linearity and magnitudes)
- The relationship between reserve adequacy and short-term volatility of the real exchange rate is non-linear:
  - The higher the level of reserves, the smaller the effect of a given increase in reserves on real exchange rate variability.
  - For any level of the reserve ratio, the effect of any increase is smaller than the effect of a decrease.
  - The inverse-exponential and logarithmic specifications indicate different magnitudes; the logarithmic specification’s flatter shape implies a somewhat smaller impact.
- Illustration for the average country in the 1994–2002 sample halving its reserves-to-short-term-debt ratio from 170 percent to 85 percent:
  - Using the exponential specification with a coefficient of 0.02 (roughly the one for invexprS in the main regressions in Table 2):
    - The country would experience an increase in exchange rate volatility (standard deviation of monthly changes in the real exchange rate) by almost 20 percent.
    - This implies the probability of a change in the real exchange rate larger than two times its initial standard deviation would double from 5 to 10 percent.
  - Using the logarithmic specification with a coefficient of negative 0.004 (roughly the one for logresstd_L in the main regressions in Table 6):
    - The effect is almost half that with the exponential specification.
- For floaters:
  - Appropriate coefficient is about 0.04 (the sum of invexprS and invexprsd, in Tables 2 and 3), indicating an effect twice as large in coefficient terms.
  - Average reserve ratio for floaters is 100 percent and exchange rate volatility is 4.4 percent.
  - When these values are taken into account, the effect for floaters of halving the reserve ratio turns out to be similar to the one described above for the whole sample.
- Note on non-linearity asymmetry:
  - Reflecting non-linearity, the effect of an increase in the reserve ratio of equal magnitude (i.e., from 170 percent to 255 percent) would imply a decline in volatility by less than 10 percent.

### Significant control-variable effects (average implications)
- A 1 percentage point increase in monetary volatility would increase exchange rate volatility by 0.7-0.8 percentage-points (or by about 20 percent of average sample volatility in 2002). This suggests that three quarters of monetary volatility is translated into exchange rate volatility.
  - It is noted that the effect is larger (close to 1) when pure fixers are excluded from the sample (Table 2, columns 1-4).
- A 1 percentage point increase in real GDP growth would reduce volatility by 0.2 percentage points (or by about 6 percent).
- A 10 percentage point increase in external debt relative to GDP growth would increase volatility by 0.6 percentage points (or by about 20 percent).

### Methodological notes (standard methods and robustness)
- Ordinary least squares (OLS) and fixed-effect estimation methods were also presented (Table 11) in bivariate and multivariate setups; these are considered less reliable than system-GMM because they do not control for endogeneity.
  - Significant coefficient on the reserves variable in bivariate OLS confirms what is visible in Figure 1.
  - Bivariate fixed-effect regressions indicate the result cannot be solely explained by country-specific constant terms.
  - Multivariate OLS and fixed-effect regressions show the result is robust to controlling for other determinants of exchange rate volatility, including the exchange rate regime.
- Fixed-effects and difference-GMM estimations were conducted for most specifications in Tables 2 through 9 (not reported for brevity) and again found a very robust effect of reserves on real exchange rate volatility.
  - These methods were less consistent in terms of whether the effect applies to fixers or floaters or whether it changes over time.

### Conclusions and policy implications
- Main empirical findings:
  - Strong evidence of a negative non-linear effect of reserves on the short-term volatility of the real effective exchange rate for a sample of emerging market countries.
  - The effect is robust to endogeneity controls, including system-GMM, IH estimation, lagging reserve adequacy measures, and inclusion of variables influencing or capturing the choice of exchange rate regime.
  - The effect is not particularly large: halving the reserves-to-short-term-debt ratio for the average country in the 1994–2002 sample could increase exchange rate volatility by up to 20 percent, nearly doubling the probability of very large swings.
  - The effect appears mainly associated with floaters; estimated coefficients are higher for floaters, and allowing a different slope for floaters makes the floater coefficient strongly significant while the whole-sample coefficient loses significance. However, when accounting for sample differences in reserve ratio and exchange rate volatility, the overall impact of changing reserves on volatility appears similar for floaters and the whole sample.
- Channel and further research:
  - The effect does not appear to operate via observed use of reserves to intervene in the foreign exchange market.
  - The effect may operate through signaling a greater potential for foreign exchange intervention in a crisis situation or a generally increased “comfort level.”
  - Further research is needed to identify channels (for example, a proxy for exchange market intervention that better captures policy intentions).
- Policy considerations:
  - Other policy actions may be at least as powerful as reserve accumulation in reducing real exchange rate volatility: credible and stability-oriented monetary policy and external debt management.
  - The results should not be interpreted as unequivocal advocacy for increasing reserve levels:
    - Benefits are mainly confined to countries with relatively low reserves.
    - Benefits (reduced currency volatility, lower interest rates, reduced crisis risk, more predictable business environment) need to be weighed against costs of holding reserves.

*Source: Box 1. How to Interpret Tables 2-10 (excerpt).*

### References

### _wp04189 - References

### Variable definitions and measurement
- vdreer: Standard deviation of monthly changes of the natural logarithm of the real effective exchange rate (REER) (source: IMF Internal Notice System).
- Reserve variables (lagged one period in all regressions)
  - resstd: The ratio of reserve assets to short-term debt on a remaining maturity basis. Short-term debt for Korea is based on staff estimates. Short-term debt for South Africa includes the forward book liabilities of the central bank.
  - resi: The ratio of reserve assets to average monthly imports of goods and nonfactor services (GNFS).
  - resy: The ratio of reserve assets to nominal GDP.
  - invexpRS / invexpRI / invexpRY: Inverse of the exponential of the above three measures, respectively.
- Variables reflecting macroeconomic factors and volatility of fundamentals
  - fis: The fiscal balance as ratio of nominal GDP.
  - infl: The CPI inflation rate (annual percent change).
  - gro: Real GDP growth (annual percent change).
  - mon: Annual standard deviation of monthly changes of broad money (source: IMF International Financial Statistics).
  - tot: The absolute value of the difference in annual growth rates of GNFS import and export price deflators.
- Variables that affect the choice of exchange rate regime
  - tropen: Trade openness, measured as the ratio of the sum of annual GFNS exports and imports to nominal GDP.
  - findeap: An indicator domestic financial deepness: the ratio of broad money to nominal GDP.
  - extfin: An indicator external financial exposure: the ratio of total external debt to GDP.
  - rppppc: An indicator of the country’s level of economic development: per capita constant-price GDP measured at the purchasing power parity (PPP) exchange rate.
  - rpppwgt: A size indicator: GDP measured at PPP exchange rates and deflated by the U.S. GDP deflator.
- Exchange rate regime (dummy)
  - exrdum: A dummy describing the exchange rate regime: floating, managed float, or freely falling = 1; de facto peg or crawling peg = 0. Based on Reinhart and Rogoff (2002) dataset, which is available until 2001. Datapoints for 2002 are equal to 2001, with the exception of Venezuela, which was classified as a 1.
- Foreign exchange intervention proxy
  - interv: A proxy of foreign exchange intervention. The standard deviation of monthly difference in the ratio of reserves to GDP.

### Descriptive statistics (Table 1: 1994–2002, 28 Emerging Markets)
- vdreer: Mean 0.028 SD 0.032 Between SD 0.017 Within SD 0.028
- resstd: Mean 1.668 SD 1.836 Between SD 1.629 Within SD 0.899
- resi: Mean 5.470 SD 2.822 Between SD 2.464 Within SD 1.444
- resy: Mean 0.145 SD 0.087 Between SD 0.077 Within SD 0.043
- gro: Mean 0.032 SD 0.044 Between SD 0.021 Within SD 0.039
- fis: Mean -0.038 SD 0.038 Between SD 0.026 Within SD 0.028
- infl: Mean 0.148 SD 0.271 Between SD 0.150 Within SD 0.226
- tot: Mean 0.050 SD 0.065 Between SD 0.041 Within SD 0.051
- tropen: Mean 0.691 SD 0.414 Between SD 0.405 Within SD 0.111
- mon: Mean 0.022 SD 0.017 Between SD 0.012 Within SD 0.013
- exrdum: Mean 0.377 SD 0.486 Between SD 0.332 Within SD 0.360
- findeap: Mean 0.555 SD 0.325 Between SD 0.320 Within SD 0.078
- extfin: Mean 0.511 SD 0.230 Between SD 0.193 Within SD 0.131
- rppppc: Mean 62.079 SD 35.900 Between SD 36.043 Within SD 5.578
- rpppwgt: Mean 4.976 SD 8.225 Between SD 8.183 Within SD 1.677
- interv: Mean 0.075 SD 0.062 Between SD 0.039 Within SD 0.049

### Key regression estimates and robustness (selected results from Tables)
- Table 2 (Volatility of Growth in REER, 1994–2002, 28 EMs) — Main specification (observations 252; No. of countries 28)
  - invexpRS: 0.016 (z = 2.18)**  
  - invexpRSD: 0.031 (z = 2.62)***  
  - gro_: -0.208 (z = 4.06)***  
  - fis_: 0.082 (z = 2.07)**  
  - mon_: 0.775 (z = 2.88)***  
  - exrdum_: 0.007 (z = 1.89)*  
  - extfin_: 0.060 (z = 3.54)***  
  - rpppwgt_: 0.001 (z = 2.75)***  
  - Interv (when included in specification): interv_ 0.049 (z = 1.48)
- Table 3 (Sample changes, 1994–2002)
  - invexpRS (No fixers): 0.021 (z = 1.79)*; invexpRS (Only floaters): -0.002 (z = 0.10)
  - gro_: ranges around -0.190 to -0.205 with significance at 1% in multiple samples
  - fis_: positive and significant in many subsamples (e.g., 0.103, z = 3.22)  
  - mon_: ranges (e.g., 0.987, z = 3.02) significant at 1%
  - extfin_: large positive and significant across subsamples (e.g., 0.083, z = 6.10)  
  - rppppc_: significant in some subsamples (e.g., 0.000 with z = 2.05)
- Table 4 & Table 8 (Other reserve measures: invexpRI, invexpRY, logresi_L, logresy_L)
  - invexpRI: Main 0.012 (t = 0.87) — not significant in Main
  - invexpRY: Main 0.111 (t = 2.06)**; invexpRYD: 0.049 (t = 0.87)
  - logresy_L: -0.008 (t = 2.32)** in one specification; logresy_LD: -0.005 (t = 0.92)
  - logresi_L: -0.003 (t = 1.23) in Main
- Table 5 and Table 9 (Longer sample 1986–2002, 22 EMs)
  - invexpRS: 0.006 (t = 1.23) Main; invexpRSD: 0.018 (t = 1.18)
  - gro_: -0.180 (t = 3.57)*** in Main specifications
  - mon_: 0.511 (t = 3.02)*** in Main
  - exrdum_: 0.018 (t = 4.55)*** in Main (1986–2002)
  - extfin_: 0.027 (t = 1.99)** in Main
- Table 6 & Table 7 (Lagged log of resstd and first differences)
  - logresstd_L: -0.004 (t = 1.66)* in Main (1994–2002)
  - logresstd_LD: -0.008 (t = 2.64)*** in Main for first difference in one specification
  - In subsamples, logresstd_L and logresstd_LD show negative coefficients with statistical significance in several columns
- Table 10 (Identification Through Heteroscedasticity (IH): 1994–2002)
  - invexpRS: 0.025  ***  (reported significance at 1 percent)
  - With intervention proxy: 0.033  ***
  - Note annotations: 1/ "Main" regression includes all control variables (apart from exchange rate dummy). 2/ "Signif. control" regression excludes usual insignificant control variable (and exchange rate dummy). 3/ “With intervention proxy” includes the “interv_” variable.
- Table 11 (Volatility of Growth in REER, OLS and Fixed Effects: 1994–2002, 28 EMs; Observations 252)
  - OLS: invexpRS 0.042 (t = 5.01)***; logresstd_L -0.010 (t = 4.84)***; gro_ -0.224 (t = 5.30)***; mon_ 0.581 (t = 3.89)***; extfin_ 0.058 (t = 4.47)***; rpppwgt_ 0.001 (t = 4.12)***. R-squared reported 0.12 in first OLS.
  - FE: invexpRS 0.069 (t = 3.89)***; logresstd_L -0.016 (t = 3.49)***; gro_ -0.193 (t = 4.52)***; mon_ 0.759 (t = 3.32)***; extfin_ 0.091 (t = 5.91)***; rpppwgt_ 0.000 (t = 0.10). R-squared reported 0.32 in first FE.
  - Note: Absolute value of t-statistics in parentheses. REER, real effective exchange rate; OLS, ordinary least squares; FE, fixed-effects.
- Observations and sample sizes (selected)
  - Typical panel regressions: Observations 252, No. of countries 28 (1994–2002 main sample).
  - Longer-sample regressions: Observations 374, No. of countries 22 (1986–2002 sample).

### Statistical significance notation (as presented)
- * Significant at 10%; ** significant at 5%; *** significant at 1%.
- Note: Robust z statistics (or t-statistics in some tables) are reported in parentheses next to coefficient estimates.
- Note: REER = real effective exchange rate.

### Figures and diagnostics
- Figures 2 and 3: Outliers versus leverage residuals for 1994–2002 and 1986–2002 respectively (charts plotting leverage against normalized residual squared for country panels; country labels included, axis tick marks shown).

*Source: _wp04189 - References (extracted content).*

### REFERENCES

### REFERENCES

### Citations
- Bartolini, Leonardo, and Gordon M. Bodnar, 1996, “Are Exchange Rates Excessively Volatile? And What Does ‘Excessive Volatile’ Mean, Anyway?”  IMF Staff Papers, Vol. 43, pp. 72–96.
- Calvo, Guillermo, and Carmen M. Reinhart, 1999, “When Capital Inflows Come to a Sudden Stop: Consequences and Policy Options,” in P. Kenen and A. Swoboda, Key Issues in Reform of the International Monetary and Financial System, (Washington: International Monetary Fund).
- Canales-Kriljenko, Jorge and Karl Habermeier, 1999, “Factors Affecting Exchange Rate Volatility,” in Exchange Arrangements and Foreign Exchange Markets; Developments and Issues, World Economic and Financial Surveys, International Monetary Fund.
- Caramazza,  F.,  L.  A.  Ricci,  and  R.  Salgado,  2004,  "International  Financial  Contagion  in  Currency Crises", Journal of International Money and Finance, Vol. 23, pp. 51–70.
- Christofides, C., C. Mulder, and A. Tiffin, 2003, “The Link Between Adherence to International Standards of Good Practice, Foreign Exchange Spreads, and Ratings,” IMF Working Paper 03/74 (Washington: International Monetary Fund).
- Clark, P., A. Sadikov, N. Tamirisia, S-J. Wei, and L. Zeng, 2003, "Exchange Rate Volatility and Trade Flows", forthcoming, International Monetary Fund.
- Devereux, Michael B., and Charles Engel, 2002, “Exchange Rate Pass-Through, Exchange Rate Volatility, and Exchange Rate Disconnect,” Journal of Monetary Economics (July), pp. 913–40.
- Devereux, Michael B. and Philip R. Lane, 2003, “Understanding Bilateral Exchange Rate Volatility,” Journal of International Economics, Vol. 50 , pp. 109–132.
- Dornbusch, Rudiger, 1976, “Expectations and Exchange Rate Dynamics,” Journal of Political Economy, Vol. 84, No. 6, pp. 1161–76.
- Hasan, Shahriar and Myles Wallace, 1996, “Real Exchange Rate Volatility and Exchange Rate Regimes: Evidence from Long-Term Data,” Economic Letters, Vol. 52 (July), pp. 67–73.
- Hausmann, Ricardo, Ugo Panizza, and Ernesto Stein, 2001, “Why do Countries Float the Way They Float?” Journal of Development Economics, Vol. 66, pp. 387–414.
- International Monetary Fund, 1998 (and other years), World Economic Outlook (Washington, October and May).
- Jonsson, Gunnar, 2001, “The Risk Premium in South African Long-term Interest Rates,” South Africa—Recent Economic Developments, IMF, SM/01/79.
- Lee, Jaewoo, 2004, “Insurance Value of International Reserves: An Option-pricing Approach,” forthcoming, International Monetary Fund.
- Levy-Yeyati, Eduardo, and Federico Sturzeneger, 2002, “Classifying Exchange Rate Regimes: Deeds vs. Words,” forthcoming, European Economic Review.
- Mulder, Christian, Roberto Perrelli, and Manuel Rocha, 2002, “The Role of Corporate, Legal and Macroeconomic Balance Sheet Indicators in Crisis Detection and Prevention,” IMF Working Paper 02/59 (Washington: International Monetary Fund).
- Reinhart, Carmen M. and Kenneth S. Rogoff, 2002, “the Modern History of Exchange Rate Arrangements: A Reinterpretation,” NBER Working Paper No. 8963 (Cambridge: National Bureau of Economic Research).
- Rigobon, Roberto, 2003, “Identification through Heteroskedasticity,” Review of Economic Statistics, Vol. 85, No. 4, pp. 777–792.
- Rose, Andrew K., 1994, “Are Exchange Rates Macroeconomic Phenomena?” Federal Reserve Bank of San Francisco Economic Review, pp. 19–30.

*Source: _wp04189 - REFERENCES*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2004/_wp04189.pdf_