## 1. Exchange Rate Intervention (FXI), by Country From 1990–2018

## Source details

**Canonical URL:** [1. Exchange Rate Intervention (FXI), by Country From 1990–2018](https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022158-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2022/english/wpiea2022158-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2022/english/wpiea2022158-print-pdf.pdf.json)

---

### Introduction and motivation
- Exchange rate intervention used to smooth macroeconomic volatility, prevent financial vulnerabilities, and help attain inflation targets when conventional monetary policy is constrained.
- Recent literature emphasizes optimal policy mixes for small open economies with additional instruments (foreign exchange interventions and capital flow measures) improving tradeoffs in presence of real and financial frictions.
- Surveys (BIS 2004, 2015, 2019b; World Bank 2013) show many emerging market and small open advanced economy central banks prioritize stemming volatility over targeting a specific exchange rate; objectives also include smoothing commodity-price shocks and enhancing competitiveness.

### Methodological approach and data
- Sample and period:
  - 30 advanced- and emerging-market economies over 1990–2018 (sample reduced to 26 countries in some regressions).
  - Total observations reported in Appendix A: 3,353.
- Key series and frequency:
  - REER and USNER at monthly frequency; macro fundamentals at quarterly frequency.
  - FXI proxy: quarterly change in central banks’ Net Foreign Assets (NFA), adjusted for valuation effects (Appendix B).
- Cycle-specific misalignments:
  - Decompose equilibrium REER into short (1–4 years), medium (>4 and <10 years), and long (≥10 years) cycles using spectral regression, Fourier/Bandpass filtering and a peak/trough algorithm.
  - Baseline fundamentals vector X: log per capita income, net foreign assets (percent of GDP), log openness, government consumption (percent of GDP), log commodity terms of trade index.
  - Equilibrium and misalignment definitions:
    - er_{i,t}^f = β X_{i,t}^f + θ_i^f + ε_{i,t}^f
    - eer_{i,t}^f = α̂ + β̂ X_{i,t}^f + θ̂_i^f
    - mis_{i,t}^f = er_{i,t}^f − eer_{i,t}^f
  - Misalignments are cycle-specific and mean zero by construction.
- Baseline FXI effectiveness model (Equation 4) regresses ln(REER) on lagged REER, fundamentals X, controls Z, FXI, lagged misalignments mis_{i,t−1}^f, and interactions FXI × mis^f with country fixed effects; Newey-West corrected standard errors.

### First-stage fundamentals results (key coefficients, Table 1)
- Short, Medium, Long regressions — coefficients (significance as reported):
  - Income: 0.217***, 0.222***, 0.212***
  - Net foreign assets (NFA): -0.0130***, -0.0169***, -0.0161***
  - Trade openness: -0.163***, -0.168***, -0.125***
  - Government consumption: 0.0000860, -0.0000962, -0.00272***
  - Commodity terms of trade: -0.0912***, -0.0964***, -0.112***
- Observations: Short 2552, Medium 2552, Long 2260
- Countries: 30; R-squared: 0.52, 0.52, 0.55

### Baseline empirical findings (Table 2)
- Baseline sample: 26 countries; Observations: 1198 (Model Baseline), 1101 (FXI surprise specification).
- Selected coefficients (Column 1, Baseline):
  - Lag REER: -0.164***
  - Lag mis S: 0.468***
  - Lag mis M: 0.643**
  - Lag mis L: 0.810***
  - FXI: 0.000522**
  - FXI * mis S: -0.0146*
  - FXI * mis M: 0.0056
  - FXI * mis L: -0.0049*
  - R-squared: 0.92
- Economic interpretation (baseline):
  - For a short-run misalignment of 10 percent, a tenth-percentage-point-of-GDP (0.1 percent of GDP) FXI (purchases) is associated with a statistically significant 1.5 percent depreciation in the exchange rate (dependent variable is ln(REER); multiply coefficients by 100 to approximate percent change).
- FXI surprises (instrument, Column 2) — residual from estimated intervention rule:
  - Lag REER: -0.195***
  - FXI: -0.00226***
  - FXI * mis S: -0.0446**
  - FXI * mis M: -0.0259
  - FXI * mis L: -0.0081
  - Observations: 1101; R-squared: 0.90
  - Using FXI surprises increases magnitude and significance of FXI * mis S (threefold larger than baseline), indicating attenuation of endogeneity bias in baseline estimates.

### Endogeneity strategy and instruments
- FXI surprises constructed as residuals from:
  - FXI_{i,t} = β X_{i,t} + δ σ_{i,t}^{reer} + θ_i + ε_{i,t}^{fxi}
  - where σ_{i,t}^{reer} is the monthly variance of REER within the quarter.
- Alternative instrument: fitted FXI using global capital flows following Blanchard et al. (2015).

### Robustness checks (Table 3)
- Specifications:
  - (1) First differences
  - (2) Effective REER (equilibrium REER from eq.1 without filtering)
  - (3) Global capital flow instrument
- Key robust result: negative and statistically significant FXI * mis S across all three specifications.
- Representative coefficient (Column 1, First Difference):
  - Lag REER: 0.295***
  - FXI * mis S: -0.0362***
  - Observations: 1181; Countries: 26; R-squared: 0.39

### Persistence, size, and directionality of interventions
- Persistent interventions (Equation 6) use CFXI_{i,t−1} = sum_{j=1}^4 FXI_{i,t−j} (4-quarter cumulative FXI) and interactions.
  - Table 4 results:
    - Lag REER: 0.188***
    - FXI: 0.00294***
    - FXI * mis S: -0.0304**
    - CFXI: 0.000589**
    - FXI * CFXI: -0.00260**
    - Observations: 1207; Countries: 26; R-squared: 0.90
  - Interpretation: persistent interventions increase effectiveness. For every additional percent of CFXI, the exchange depreciates by a further .26 percent.
- Large interventions (Equation 7; Table 5):
  - Definition: large purchases/sales are FX interventions > 1 percent of GDP (dummy P/S).
  - Purchases model (Column 2):
    - FXI: 0.00378***
    - FXI * mis S: -0.221***
    - FXI * L Purch * mis S: -0.0296***
    - Observations: 1207; R-squared: 0.93
  - Sales model (Column 3):
    - FXI: 0.00301***
    - FXI * mis S: -0.0739***
    - Large sales: -0.00587*
    - FXI * L Sales * mis S: -0.0356***
    - Observations: 1207; R-squared: 0.91
  - Both large purchases and sales strengthen the countercyclical effect against short-cycle misalignments; FXI sales somewhat more effective in economic size (reported comparisons: 3.6 percent vs. 3.0 percent in text).

### Cross-sectional heterogeneity (Table 6)
- Regional and regime splits — FXI * mis S coefficients:
  - LATAM (5 countries, 211 observations): -0.0544***
  - SE Asia (9 countries, 159 observations): -0.0880***
  - Floaters (17 countries, 873 observations): -0.0116 (insignificant)
  - Interveners (13 countries, 228 observations): -0.0839***
- Findings:
  - Asian economies and countries classified as interveners show more negative and statistically significant FXI * mis S than floaters, indicating greater effectiveness in these groups.

### Market depth and liquidity (Table 7)
- Liquidity measure:
  - spread_{i,t} = (bid_{i,t} − offer_{i,t}) / er_{i,t} × 100 (percent)
  - Average spread reported: 10bps (0.1 percent); standard deviation: 20bps.
- Triple interaction results indicate less liquid markets amplify FX intervention effectiveness for short-run misalignments.
  - Table 7, Column 1 (Baseline):
    - FXI: 0.00110**
    - FXI * mis S: 0.0277
    - Bid-offer spread: -0.0462***
    - Spread * mis S: -1.825***
    - Spread * FXI * mis S: -0.400**
    - Observations: 1187; Countries: 26; R-squared: 0.91
  - Column 2 (FXI surprise):
    - FXI: 0.00122**
    - FXI * mis S: -0.104***
    - Spread * FXI * mis S: -0.710***
    - Observations: 1187; Countries: 26; R-squared: 0.94
- Economic example:
  - For a 1-percentage point increase in the bid-offer spread, a tenth-percentage-point-of-GDP FX purchase is associated with a statistically significant 4-percent larger depreciation when starting from a 10 percent short-run misalignment; using FXI surprises the differential increases to 7 percent.

### Main conclusions
- Effectiveness is cycle-dependent:
  - FX intervention is effective in leaning against short-cycle misalignments: FXI * mis S coefficients are negative and statistically significant across baseline, instrumented, and robustness specifications.
  - FX intervention is generally ineffective, or much less effective, against medium- and long-cycle misalignments (FXI * mis M often insignificant; FXI * mis L small and statistically fragile).
- Persistence, size, and one-sidedness increase effectiveness:
  - Persistent cumulative interventions (CFXI) and large interventions (> 1 percent of GDP) amplify the countercyclical impact against short-cycle misalignments.
  - Directional asymmetry observed: FX sales tend to be generally more effective than FX purchases in economic magnitude.
- Market conditions matter:
  - Interventions are more effective in shallow, illiquid FX markets (larger bid-offer spreads), consistent with interventions acting as liquidity buffers amid financial frictions.
- Cross-sectional differences:
  - Asian economies and regular interveners show stronger FXI effectiveness for short-cycle misalignments.
- Methodological contribution:
  - Band spectrum estimation and cycle-specific misalignment measures help identify when FX intervention is likely to work—particularly for short-run, finance-driven misalignments rather than long-run fundamentals-driven misalignments.

### Policy implications and recommendations
- Use FX intervention as a tool to smooth short-cycle exchange-rate volatility when:
  - Misalignments are short-cycle in nature.
  - Interventions are persistent and one-sided rather than isolated.
  - Interventions are sizable relative to GDP (larger interventions have greater effects).
  - Market conditions exhibit lower market depth (larger bid-offer spreads).
  - Employed in regions/regimes where evidence shows higher effectiveness (e.g., parts of Asia; regular interveners).
- Exercise caution in using FX intervention to counter medium- or long-cycle misalignments, which appear driven more by fundamentals and less responsive to intervention.
- Recommended further research and policy work:
  - Investigate mechanisms (portfolio balance vs signaling).
  - Determine optimal intervention tactics (sizes/types) and interactions with market conditions (depth/liquidity).
  - Examine communication strategies and the role of policy frameworks in enhancing intervention effectiveness.

*IMF Working Paper — Exchange-Rate Swings and Foreign Currency Intervention (section: 1. Exchange Rate Intervention (FXI), by Country From 1990–2018).*

### 1. Exchange Rate Intervention (FXI), by Country From 1990–2018 .........................................................

### 1. Exchange Rate Intervention (FXI), by Country From 1990–2018

### Introduction and motivation
- Exchange rate intervention has gained prominence as a policy instrument for smoothing macroeconomic volatility, preventing the buildup of financial vulnerabilities, and helping attain inflation targets when conventional monetary policy reaches limits.
- Recent theoretical and empirical work examines optimal policy mixes for small open economies, emphasizing that additional instruments (foreign exchange interventions and capital flow measures) can improve policy tradeoffs in the presence of real and financial frictions (see Basu et al., 2020; Adrian et al., 2020, 2021).
- Surveys (BIS 2004, 2015, 2019b; World Bank 2013) indicate many emerging market and small open advanced economy central banks prioritize stemming volatility over targeting a specific exchange rate; some also cite smoothing commodity-price shocks and enhancing competitiveness.

### Empirical and academic debate: challenges and recent progress
- Three key empirical challenges in assessing FX intervention effectiveness:
  - Endogeneity bias: authorities may intervene in response to exchange rate movements or capital flow cycles, making intervention correlated with other drivers of the exchange rate and biasing estimates.
  - Data limitations: historically limited high-quality cross-country data on interventions; recent improvements have eased but gaps remain.
  - Size and variation of interventions: past interventions were often small relative to current reserve buffers, potentially limiting detectability of effects.
- Academic responses have included short-run, high-frequency event studies to improve identification (Menkhoff 2010; Fatum and Hutchison, 2003; Fratzscher et al., 2019; Dominguez, 2003), but these often cannot assess longer-lasting, policy-relevant effects.
- Recent cross-country studies find evidence supporting intervention effectiveness:
  - Daude et al. (2016): interventions effective in influencing real exchange rate (quarterly data 2003-2011), with greater effectiveness when misalignment is larger.
  - Adler et al. (2019): persistent impacts with half-life of 12-23 months.
  - Blanchard et al. (2015): heavily intervening countries experience less exchange rate volatility.
  - Adler and Tovar (2014), Hofmann et al. (2019), Menkhoff et al. (2020): persistent impacts across various countries and frequencies.

### Theoretical context
- New theory emphasizes financial frictions and global intermediaries’ risk-bearing capacity as drivers of misalignments (Gabaix and Maggiori (2015); Maggiori (2022)).
- Short-run exchange-rate dynamics may be driven by private information and order flows (Bacchetta and van Wincoop (2006); Evans and Lyons (2002)), while macro fundamentals matter more at longer horizons (Mark, 1995).
- The “exchange determination puzzle”: drivers differ by horizon, implying intervention effectiveness may vary by the cyclical nature of misalignments.

### Methodological approach of this paper
- Uses an expanded cross-country dataset for 30 advanced- and emerging-market economies over 1990-2018 (constrained by macro fundamentals data availability).
- Introduces cycle-specific exchange rate misalignments: spectral regression methods estimate equilibrium real exchange rate at short-, medium-, and long-run cycles; misalignments are derived cycle by cycle.
- Empirical strategy:
  - Quarterly unbalanced panel estimation of FXI effectiveness for each cycle-specific misalignment.
  - Endogeneity addressed using FXI “surprises” (deviations from estimated policy rules) following Brandao-Marques et al. (2020).
  - Robustness: instrumenting with capital flows to other countries following Blanchard et al. (2015).
- Purpose: identify whether intervention effectiveness differs across short-, medium-, and long-cycle misalignments at policy-relevant horizons.

### Main empirical findings
- Effectiveness is cycle-dependent:
  - Interventions are effective when leaning against short-cycle misalignments.
  - No evidence of effectiveness against medium-cycle misalignments and long-cycle misalignments.
- Magnitude example:
  - For a short-cycle misalignment of 10 percent, a tenth-percentage-point-of-GDP exchange rate intervention is associated with a statistically significant percent change in the exchange rate ranging from 1.5 to 4.5 percent.
- Persistence and intervention style:
  - Persistent, one-sided interventions increase effectiveness relative to one-off interventions.
  - Larger interventions are associated with greater effectiveness.
- Directional asymmetry:
  - FX sales are generally more effective than FX purchases.
- Cross-sectional heterogeneity:
  - Effectiveness varies across regions: interventions in Asian economies are generally more effective.
  - Effectiveness varies across exchange rate regimes: interventions by regular interveners are more effective than by floaters.
- Market conditions:
  - Initial evidence indicates FX intervention is more effective when market depth is low.

### Interpretation and caveats
- Estimates should be viewed as lower bounds for true intervention effects because remaining endogeneity bias is likely to work against finding effectiveness.
- The distinction between financial and macroeconomic drivers of exchange-rate swings remains empirically challenging given limited comparable cross-country data on financial factors and balance sheets.

### Policy implications and recommendations (implicit from findings)
- FX intervention can be a useful tool for smoothing short-cycle exchange-rate volatility, particularly when:
  - Misalignments are short-cycle in nature.
  - Interventions are persistent and one-sided rather than isolated.
  - Interventions are sizable relative to GDP (larger interventions have greater effects).
  - Conducted in market conditions with lower market depth, and in regions/regimes where evidence shows higher effectiveness (e.g., parts of Asia, regular interveners).
- Caution about using FX intervention to counter medium- or long-cycle misalignments, which appear driven more by fundamentals and less responsive to intervention.

*IMF Working Paper — Exchange Rate-Swings and Foreign Currency Intervention (section: 1. Exchange Rate Intervention (FXI), by Country From 1990–2018).*

### Section III outlines our new approach to measuring exchange rate misalignments. Section IV outlines our

### Section III–V: New approach to measuring exchange rate misalignments; FXI effectiveness; Conclusions

### Data and sample
- Sample: 30 advanced and emerging market economies covering the period 1990–2018; sample size reduced to 26 countries in some regressions due to data on regressors.
- Exchange rate series:
  - Real effective exchange rate (REER) index and nominal bilateral exchange rate with the US dollar (USNER) at monthly frequency from IMF IFS.
  - Focus on trade-weighted REER and real bilateral exchange rate relative to the US dollar.
- Macroeconomic fundamentals (quarterly frequency): real GDP growth rates, general government fiscal deficit, monetary policy rates, CPI inflation, commodity terms of trade (monthly), GDP per capita in real USD.
- Exchange rate intervention proxy (FXI): quarterly change in central banks’ Net Foreign Assets (NFA), adjusted for valuation effects (constructed in Appendix B).
- Global and policy controls: NFA, exports (X), imports (M), gross/net capital inflows (GKI, NKI) from BoP/IIP; official reserves from COFER; global volatility proxied by S&P 100 Volatility Index (VXO); capital account openness from Fernández et al. (2016).

### New approach: cycle-specific misalignments (methodology)
- Core idea: decompose equilibrium exchange rate into short-, medium-, and long-cycles and estimate cycle-specific misalignments using spectral regression methods.
- Baseline relationship (Equation 1):
  - er_{i,t}^f = β X_{i,t}^f + θ_i^f + ε_{i,t}^f
  - er_{i,t}^f: real exchange rate for country i at time t at frequency f ∈ {s,m,l}
  - X: fundamentals vector — log per capita income, net foreign assets (percent of GDP), log openness (exports+imports percent of GDP), government consumption (percent of GDP), log commodity terms of trade index
  - θ_i^f: country fixed effects
- Frequency decomposition:
  - Use Fourier transformation / Butterworth high-pass filter without trend to obtain cyclical components.
  - Cycle identification via peak/trough algorithm (Bry and Boschan (1971) style) applied to real bilateral exchange rate vs. US dollar.
  - Calibrated durations:
    - Long-run cycle: cycles of duration of 10 years or more (filter out frequencies associated with cycles < 10 years).
    - Medium-run cycle: cycles greater than 4 years and less than 10 years.
    - Short-run cycle: cycles of 1 to 4 years.
  - Movements of less than a year treated as statistically unpredictable.

### First-stage (cycle-specific fundamentals regressions) — key coefficients (Table 1)
- Short, Medium, Long regressions (coefficients shown):
  - Income: 0.217***, 0.222***, 0.212***
  - Net foreign assets (NFA): -0.0130***, -0.0169***, -0.0161***
  - Trade openness: -0.163***, -0.168***, -0.125***
  - Government consumption: 0.0000860, -0.0000962, -0.00272***
  - Commodity terms of trade: -0.0912***, -0.0964***, -0.112***
- Observations: Short 2552, Medium 2552, Long 2260
- Countries: 30; R-squared: 0.52, 0.52, 0.55
- Notes: dependent variable is log(REER); Newey-West corrected standard errors up to 4 quarter lags.

### Defining misalignments
- Cycle-specific equilibrium REER (Equation 2):
  - eer_{i,t}^f = α̂ + β̂ X_{i,t}^f + θ̂_i^f
- Cycle-specific misalignment (Equation 3):
  - mis_{i,t}^f = er_{i,t}^f − eer_{i,t}^f
- Properties:
  - Misalignments are cycle-specific and mean zero by construction.
  - Frequency-domain estimates are theoretically unbiased if all variables are transformed using the same frequency band.

### Baseline FXI effectiveness model (Equation 4)
- Core specification regresses er_{i,t} on:
  - lagged REER, fundamentals X_{i,t}, policy/global controls Z_{i,t}, FXI_{i,t}, lagged misalignments mis_{i,t−1}^f, interactions FXI × mis^f, country fixed effects.
- Estimation: fixed effects with Newey-West corrected standard errors for heteroskedasticity and serial correlation.

### Baseline results (Table 2) — key findings
- Sample for baseline: 26 countries; Observations: 1198 (Model Baseline), 1101 (FXI surprise)
- Selected coefficients (Column 1, Baseline):
  - Lag REER: -0.164***
  - Lag mis S: 0.468***
  - Lag mis M: 0.643**
  - Lag mis L: 0.810***
  - FXI: 0.000522**
  - FXI * mis S: -0.0146*
  - FXI * mis M: 0.0056
  - FXI * mis L: -0.0049*
  - R-squared: 0.92
- Economic interpretation:
  - For a short-run misalignment of 10 percent, a tenth-percentage-point-of-GDP (0.1 percent of GDP) FXI (purchases) is associated with a statistically significant 1.5 percent depreciation in the exchange rate.
    - (Footnote clarifies: FXI measured in levels of foreign exchange purchases as percent of GDP; dependent variable is ln(REER); multiplying coefficients by 100 gives approximate percent change.)
- Column 2 (FXI surprise instrument — residual from intervention rule Equation 5):
  - Lag REER: -0.195***
  - FXI: -0.00226***
  - FXI * mis S: -0.0446**
  - FXI * mis M: -0.0259
  - FXI * mis L: -0.0081
  - Observations: 1101; R-squared: 0.90
  - Using FXI surprises increases magnitude and significance of FXI * mis S (threefold larger than baseline), suggesting attenuation of endogeneity bias.

### Addressing endogeneity
- FXI "surprises" estimated as residual from:
  - FXI_{i,t} = β X_{i,t} + δ σ_{i,t}^{reer} + θ_i + ε_{i,t}^{fxi}
  - σ_{i,t}^{reer}: monthly variance of REER within the quarter
- Alternative instrument explored: fitted FXI using global capital flows (Blanchard et al. (2015) style).

### Robustness checks (Table 3)
- Three specifications:
  - (1) First differences
  - (2) Effective REER (equilibrium REER from eq.1 without filtering)
  - (3) Global capital flow instrument
- Key robust result: negative and statistically significant FXI * mis S in all three specifications.
- Representative coefficients (Column 1, First Difference):
  - Lag REER: 0.295***
  - FXI * mis S: -0.0362***
  - Observations: 1181; Countries: 26; R-squared: 0.39

### Persistent, large, and one-sided interventions
- Persistent interventions model (Equation 6) introduces CFXI_{i,t−1} = sum_{j=1}^4 FXI_{i,t−j} (4-quarter cumulative FXI) and interactions.
- Table 4 (persistent interventions):
  - Lag REER: 0.188***
  - FXI: 0.00294***
  - FXI * mis S: -0.0304**
  - CFXI: 0.000589**
  - FXI * CFXI: -0.00260**
  - Observations: 1207; Countries: 26; R-squared: 0.90
- Interpretation: persistent interventions increase effectiveness. For every additional percent of CFXI, the exchange depreciates by a further .26 percent.
- Large interventions (Equation 7) — Table 5:
  - Definition: large purchases/sales are FX interventions > 1 percent of GDP (dummy P/S).
  - Purchases model (Column 2):
    - FXI: 0.00378***
    - FXI * mis S: -0.221***
    - FXI * L Purch * mis S: -0.0296***
    - Observations: 1207; R-squared: 0.93
  - Sales model (Column 3):
    - FXI: 0.00301***
    - FXI * mis S: -0.0739***
    - Large sales: -0.00587*
    - FXI * L Sales * mis S: -0.0356***
    - Observations: 1207; R-squared: 0.91
  - Both large purchases and sales strengthen the countercyclical effect for short-cycle misalignments; FXI sales somewhat more effective in economic size (3.6 percent vs. 3.0 percent in illustrative comparisons reported in text).

### Cross-sectional heterogeneity (Table 6)
- Regional and policy-framework splits:
  - LATAM (5 countries, 211 observations): FXI * mis S = -0.0544*** (Column 1)
  - SE Asia (9 countries, 159 observations): FXI * mis S = -0.0880*** (Column 2)
  - Floaters (17 countries, 873 observations): FXI * mis S = -0.0116 (Column 3)
  - Interveners (13 countries, 228 observations): FXI * mis S = -0.0839*** (Column 4)
- Findings:
  - Some evidence that Asian interventions are more impactful; interveners show more negative and statistically significant FXI * mis S than floaters.

### Market depth and liquidity (Table 7)
- Liquidity measure (spread, Equation 8):
  - spread_{i,t} = (bid_{i,t} − offer_{i,t}) / er_{i,t} × 100 (percent)
  - Average spread: 10bps (0.1 percent); standard deviation: 20bps.
- Triple interaction results:
  - Less liquid (larger bid-offer spread) markets are associated with greater effectiveness of FX intervention in leaning against short-run misalignments.
  - Table 7, Column 1 (Baseline):
    - FXI: 0.00110**
    - FXI * mis S: 0.0277
    - Bid-offer spread: -0.0462***
    - Spread * mis S: -1.825***
    - Spread * FXI * mis S: -0.400**
    - Observations: 1187; Countries: 26; R-squared: 0.91
  - Column 2 (FXI surprise):
    - FXI: 0.00122**
    - FXI * mis S: -0.104***
    - Spread * FXI * mis S: -0.710***
    - Observations: 1187; Countries: 26; R-squared: 0.94
- Economic example: For a 1-percentage point increase in the bid-offer spread, a tenth-percentage-point-of-GDP FX purchase is associated with a statistically significant 4-percent larger depreciation when starting from a 10 percent short-run misalignment; using FXI surprises the differential increases to 7 percent.

### Main conclusions and policy implications
- Central banks are effective at leaning against short-cycle misalignments:
  - FXI * mis S coefficients are negative and statistically significant across baseline, instrumented, and robustness specifications.
  - Sizeable economic effects: e.g., 0.1 percentage point of GDP FXI associated with ~1.5 percent depreciation for a 10 percent short-run misalignment (baseline); larger when using FXI surprises.
- FX interventions are generally ineffective or much less effective at leaning against medium- and long-cycle misalignments:
  - FXI * mis M often insignificant and occasionally of incorrect sign.
  - FXI * mis L small and statistically fragile.
- Persistent and one-sided interventions can increase effectiveness:
  - Cumulative interventions (CFXI) yield additional depreciation; FXI * CFXI interaction negative and significant.
  - Large purchases and sales (> 1 percent of GDP) amplify countercyclical effects on short-run misalignments; sales may be somewhat more effective economically.
- Market liquidity matters:
  - FX intervention more effective in shallow, illiquid FX markets (larger bid-offer spreads), consistent with interventions acting as liquidity buffers amid financial frictions.
- Cross-sectional differences:
  - Asian economies and countries that credibly intervene show stronger FXI effectiveness for short-cycle misalignments.
- Methodological contribution:
  - Band spectrum estimation and cycle-specific misalignment measures provide a new perspective on when FX intervention is likely to work—particularly for short-run, finance-driven misalignments rather than long-run, fundamentals-driven misalignments.
- Suggested directions for policy and research:
  - Interventions are more promising as a tool against short-run, finance-driven misalignments and when implemented persistently or at large scale.
  - Further research recommended on mechanisms (portfolio balance vs signaling), optimal intervention tactics (sizes/types), interaction with market conditions (depth/liquidity), and role of policy communication.

*International Monetary Fund — IMF Working Paper: Exchange-Rate Swings and Foreign Currency Intervention (Sections III–V as provided).*

### Appendix  A. Data Details

### Appendix A. Data Details

### Data sources (Table A.1)
- Real per capita GDP — IMF World Economic Outlook
- Real effective exchange rate index — IMF International Financial Statistics
- Real US bilateral exchange rate — IMF International Financial Statistics
- Net foreign assets — IMF Balance of Payments and International Investment Position; WB World Development Indicators
- Exports & Imports — IMF Balance of Payments
- Government consumption — IMF International Financial Statistics
- Commodity terms of trade (ToT) — IMF Commodity ToT (see Gruss and Kehaj (2019) for details)
- Policy rates — IMF International Financial Statistics
- VXO volatility index — St. Louis Federal Reserve FRED database
- Reserves — Composition of Official Foreign Exchange Reserves (COFER) database
- FX intervention — Central Banks websites and IMF staff estimates. NFA data: IMF Data Template on International Reserves and Foreign Currency Liquidity; Currency composition of reserves: COFER database.

### Country coverage (Table A.2)
- Total observations: 3,353
- Country list with Obs., De facto ER 2018, AREAR Classification, Intervener
  1. Australia — 113 — Free floating — AE —  
  2. Brazil — 113 — Floating — EM — Yes
  3. Canada — 112 — Free floating — AE —  
  4. Chile — 110 — Free floating — EM —  
  5. China — 15 — Stabilized arrangement — EM — Yes
  6. Colombia — 90 — Floating — EM —  
  7. Czech Republic — 94 — Floating — EM — Yes
  8. Denmark — 113 — Floating — AE —  
  9. Hong Kong SAR* — 105 — Floating — EM — Yes
  10. Hungary — 104 — Floating — EM —  
  11. India — 106 — Floating — EM — Yes
  12. Indonesia — 112 — Floating — EM — Yes
  13. Israel — 111 — Floating — AE —  
  14. Kazakhstan — 76 — Floating — EM — Yes
  15. Korea — 112 — Floating — AE — Yes
  16. Malaysia — 76 — Floating — EM — Yes
  17. Mexico — 113 — Free floating — EM —  
  18. New Zealand — 85 — Floating — AE —  
  19. Norway — 112 — Free floating — AE —  
  20. Peru — 112 — Floating — EM — Yes
  21. Philippines — 113 — Floating — EM — Yes
  22. Poland — 73 — Free floating — EM — Yes
  23. Romania — 108 — Stabilized arrangement — EM —  
  24. Russia — 69 — Free floating — EM — Yes
  25. Singapore — 94 — Crawl-like arrangement — AE —  
  26. South Africa — 113 — Floating — EM —  
  27. Sweden — 113 — Free floating — AE —  
  28. Switzerland — 113 — Floating — AE —  
  29. Thailand — 113 — Floating — EM — Yes
  30. Turkey — 113 — Floating — EM —  
- Note: Hong Kong SAR is a territory of the People’s Republic of China with its own currency.

### Summary statistics (Table A.3 — means 1990-2018)
Notes: Table reports the mean of each variable over the period 1990-2018. REER index; DSP per capita in real USD; government consumption, NFA, openness, as percent of GDP; reserves in percent of quarterly imports; FXI as FX purchases in percent of GDP; capital account openness index from Chinn-Ito; real policy rate in percentage points; commodity terms of trade as index.

- Australia — REER 86.6, USRER 1.2, GDP per capita 38,537, Gov’t cons -1.5, NFA -1.0, Reserves 0.9, FXI 0.0, Openness 9.9, Chinn-Ito 1.4, Real policy rate 3.3, Comm ToT 1.1
- Brazil — REER 83.9, USRER 1.5, GDP per capita 12,471, Gov’t cons -4.8, NFA -0.5, Reserves 4.4, FXI 0.2, Openness 5.6, Chinn-Ito -1.1, Real policy rate 7.0, Comm ToT 1.1
- Canada — REER 90.9, USRER 1.3, GDP per capita 39,972, Gov’t cons -2.8, NFA -0.4, Reserves 0.4, FXI 0.0, Openness 15.6, Chinn-Ito 2.3, Real policy rate 1.0, Comm ToT 1.1
- Chile — REER 106.2, USRER 386, GDP per capita 15,302, Gov’t cons 0.7, NFA -0.4, Reserves 1.8, FXI 0.3, Openness 15.3, Chinn-Ito -0.2, Real policy rate 0.7, Comm ToT 1.1
- China — REER 119.7, USRER 6.5, GDP per capita 5,221, Gov’t cons -1.4, NFA 0.8, Reserves 6.5, FXI 1.0, Openness 12.0, Chinn-Ito -1.3, Real policy rate 0.3, Comm ToT 1.5
- Colombia — REER 94.5, USRER 1,296, GDP per capita 10,265, Gov’t cons -2.0, NFA -0.7, Reserves 2.4, FXI 0.2, Openness 8.4, Chinn-Ito -1.0, Real policy rate 3.3, Comm ToT 0.9
- Czech Republic — REER 78.5, USRER 24.3, GDP per capita 31,216, Gov’t cons -3.0, NFA -0.4, Reserves 1.8, FXI 0.8, Openness 30.0, Chinn-Ito 1.7, Real policy rate 0.3, Comm ToT 1.1
- Denmark — REER 95.8, USRER 7.0, GDP per capita 45,774, Gov’t cons -0.9, NFA 0.3, Reserves 1.7, FXI 0.2, Openness 20.4, Chinn-Ito 1.8, Real policy rate 1.3, Comm ToT 1.1
- Hong Kong SAR* — REER 117.0, USRER 8.0, GDP per capita 38,303, Gov’t cons 1.5, NFA 1.9, Reserves 2.3, FXI 1.2, Openness 87.7, Chinn-Ito 2.3, Real policy rate 1.0, Comm ToT 0.9
- Hungary — REER 84.5, USRER 128, GDP per capita 21,535, Gov’t cons -4.7, NFA -0.6, Reserves 1.3, FXI 0.4, Openness 31.8, Chinn-Ito 0.6, Real policy rate 1.0, Comm ToT 1.2
- India — REER 33.5, USRER 2,917, GDP per capita -7.9, Gov’t cons -0.3, NFA 2.7, Reserves 0.3, FXI 7.5, Openness -1.2, Chinn-Ito -0.1, Real policy rate 1.3
- Indonesia — REER (blank), USRER (blank), GDP per capita 5,660, Gov’t cons 6,415, NFA -1.1, Reserves -0.2, FXI 2.5, Openness 0.2, Chinn-Ito 11.4, Real policy rate 1.4, Comm ToT 2.5, (additional 1.1 appears in table)
- Israel — REER 102.5, USRER 2.7, GDP per capita 28,459, Gov’t cons -3.2, NFA 0.1, Reserves 2.4, FXI 0.1, Openness 17.0, Chinn-Ito 0.6, Real policy rate 2.2, Comm ToT 1.2
- Kazakhstan — REER (blank), USRER 169, GDP per capita 16,748, Gov’t cons 1.9, NFA -0.7, Reserves 1.9, FXI 0.5, Openness 19.3, Chinn-Ito -1.2, Real policy rate 0.2, Comm ToT 0.9
- Korea — REER 124.4, USRER 877, GDP per capita 23,001, Gov’t cons 1.6, NFA 0.3, Reserves 2.1, FXI 0.3, Openness 17.3, Chinn-Ito -0.1, Real policy rate 0.7, Comm ToT 1.1
- Malaysia — REER 115.8, USRER 3.2, GDP per capita 15,781, Gov’t cons -1.6, NFA 1.6, Reserves 2.0, FXI 0.5, Openness 40.7, Chinn-Ito 0.7, Real policy rate 1.0, Comm ToT 1.0
- Mexico — REER 99.5, USRER 7.4, GDP per capita 16,965, Gov’t cons -2.5, NFA -0.6, Reserves 1.6, FXI 0.1, Openness 11.2, Chinn-Ito 0.5, Real policy rate 1.8, Comm ToT 0.8
- New Zealand — REER 94.0, USRER 1.5, GDP per capita 32,273, Gov’t cons -0.5, NFA -0.8, Reserves 1.3, FXI 0.2, Openness 14.6, Chinn-Ito 2.0, Real policy rate 2.2, Comm ToT 1.9
- Norway — REER 95.7, USRER 7.0, GDP per capita 51,553, Gov’t cons 7.3, NFA 1.3, Reserves 1.8, FXI 0.0, Openness 17.7, Chinn-Ito 1.4, Real policy rate 2.7, Comm ToT 0.8
- Peru — REER (blank), USRER 2.0, GDP per capita 8,150, Gov’t cons -0.4, NFA -0.7, Reserves 4.1, FXI 0.5, Openness 9.7, Chinn-Ito 1.2, Real policy rate 0.8, Comm ToT 1.1
- Philippines — REER 100.9, USRER 29.4, GDP per capita 5,225, Gov’t cons -1.1, NFA -0.3, Reserves 3.2, FXI 0.4, Openness 14.8, Chinn-Ito -0.5, Real policy rate 1.2, Comm ToT 1.1
- Poland — REER 88.4, USRER 2.1, GDP per capita 17,772, Gov’t cons -3.9, NFA -0.7, Reserves 1.6, FXI 0.2, Openness 20.8, Chinn-Ito -0.6, Real policy rate 2.9, Comm ToT 1.1
- Romania — REER 85.6, USRER 2.1, GDP per capita 16,978, Gov’t cons -2.9, NFA -1.0, Reserves 1.9, FXI 0.3, Openness 17.1, Chinn-Ito -0.1, Real policy rate -8.9, Comm ToT 1.1
- Russia — REER 80.8, USRER 30.6, GDP per capita 20,220, Gov’t cons 0.6, NFA 1.0, Reserves 3.8, FXI 0.6, Openness 12.6, Chinn-Ito 0.0, Real policy rate 1.6, Comm ToT 0.8
- Singapore — REER 100.2, USRER 1.9, GDP per capita 56,794, Gov’t cons 3.8, NFA 4.5, Reserves 2.1, FXI 1.2, Openness 89.8, Chinn-Ito 2.3, Real policy rate 0.1, Comm ToT 1.0
- South Africa — REER 110.9, USRER 5.8, GDP per capita 12,449, Gov’t cons -2.5, NFA -0.6, Reserves 1.9, FXI 0.1, Openness 12.5, Chinn-Ito -1.3, Real policy rate 3.3, Comm ToT 1.3
- Sweden — REER 115.9, USRER 7.7, GDP per capita 40,465, Gov’t cons -1.0, NFA 0.3, Reserves 0.9, FXI 0.1, Openness 17.9, Chinn-Ito 1.8, Real policy rate 0.6, Comm ToT 1.0
- Switzerland — REER 97.0, USRER 1.7, GDP per capita 59,500, Gov’t cons -0.3, NFA 2.5, Reserves 3.6, FXI 0.8, Openness 27.3, Chinn-Ito 2.3, Real policy rate 0.5, Comm ToT 1.1
- Thailand — REER (blank), USRER 29.2, GDP per capita 10,500, Gov’t cons -0.7, NFA 0.1, Reserves 2.9, FXI 0.5, Openness 24.6, Chinn-Ito -0.4, Real policy rate 0.2, Comm ToT 1.3
- Turkey — REER (blank), USRER 1.3, GDP per capita 16,548, Gov’t cons -3.9, NFA -0.5, Reserves 1.8, FXI 0.1, Openness 10.1, Chinn-Ito -0.7, Real policy rate 1.0, Comm ToT 1.3

### FX intervention proxy (Appendix B)
- FXI proxy definition (as in Brandao-Marques et al. (2020) and Adler et al. (2019)):
  퐹푋퐼_{푗,푡} = ∆푁퐹퐴_{푗,푡} − ∆푉푎푙_{푆푒푐_{푗,푡}} − ∆푉푎푙_{퐶푢푟퐷푒푝_{푗,푡}}.
- Concept: FXI approximates FX intervention with the change in the central bank’s Net Foreign Assets (NFA), adjusted for valuation changes and the currency composition.
- Data components:
  - ∆푁퐹퐴_{푗,푡} denotes change in NFA for country j and time t.
  - Breakdown to foreign securities (푆푒푐_{푗,푡,푐}) and foreign currency and deposits (퐶푢푟퐷푒푝_{푗,푡,푐}), where c stands for currency, from IMF’s Data Template on International Reserves and Foreign Currency Liquidity.
  - Currency composition of each asset class assumed to follow IMF COFER database.
  - Reserve currencies included: US dollar, Australian dollar, Canadian dollar, British pound, Japanese yen, Swiss Franc and Euro.
- Assumptions and adjustments:
  - Follow Dominguez (2012): securities are assumed to be all government securities; cash assumed to earn zero returns beyond exchange rate return.
  - Valuation adjustment: subtract from ∆NFA a COFER-weighted foreign currency total return index (using the US dollar as the numeraire); and subtract the total return of the government bond indexes of each reserve currency, multiplied by the weight of each currency and share of securities in the central bank’s NFA.
  - Formulas in text:
    ∆푉푎푙_{푆푒푐_{푗,푡}} = ∑_{c∈C} 푆푒푐_{푗,푡,푐} 푅_{푗,푡−1,푐}^{푆푒푐}
    ∆푉푎푙_{퐶푢푟퐷푒푝_{푗,푡}} = ∑_{c∈C} 퐶푢푟퐷푒푝_{푗,푡,푐} 푅_{푗,푡−1,푐}^{퐶푢푟}
- Scaling and validation:
  - Both actual FXI and proxy series scaled using respective country’s annual GDP.
  - Actual data correlated with proxy; correlation is high for emerging markets in particular, where actual intervention data are often lacking and proxy values are used.
- Footnotes:
  - 29: 푅_{푗,푡−1,푐}^{퐶푢푟} is the 3-month interbank rate, from Haver Analytics.
  - 30: 푅_{푗,푡−1,푐}^{푆푒푐} is the treasury’s total return index for the past three months, from Thomson Reuters Datastream.

### Supplementary exchange rate figures (Appendix C)
- Figures referenced:
  - Figure C.1. Peaks and Troughs in Real Exchange Rates
  - Figure C.2. Real Exchange Rate Spectral Decomposition

### Historical overview — evolving views on FX intervention, post-Bretton Woods (Appendix D)
- High-level narrative:
  - End of Bretton Woods (by March 1973) led to a system relying more on markets and less on government intervention, but governments remained reluctant to relinquish exchange rate control.
  - Mussa (1981): 1970s showed high real and nominal exchange rate volatility; central banks practiced “leaning against the wind” and considered exchange behavior a target of monetary policy.
  - 1980s: Rise of the Washington Consensus (Williamson (1989)) and skepticism about intervention effectiveness; notable unsuccessful interventions (Plaza and Louvre Accords) and doubts from academic research (Sarno and Taylor (2001)).
  - 1990s: Mixed experiences — some advanced economies adopted inflation targeting and avoided active intervention; crises (Tequila, Asian Financial Crisis) exposed risks of pegs and premature floating. Notion of “fear of floating” (Calvo and Reinhart (1999)) emerged; many Asian central banks moved to heavily managed regimes.
  - Late 1990s–2000s: Academic work in the 1990s found evidence that interventions could be effective at short horizons (Sarno and Taylor (2001)). The Great Financial Crisis and unconventional monetary policies among major advanced economies led to concerns about “currency wars” and prompted large, sustained interventions by emerging markets.
  - Policy responses after the crisis included adopting floors (Czech Republic, Switzerland), extensive intervention (Israel, South Korea), expanded authority to intervene (Sweden), and acceptance of higher-than-conventional reserve accumulation.
- Conclusion: Over five decades the intervention zeitgeist has, in many respects, come full circle: persistent practical challenges of responding to wide exchange rate swings remain, and intervention strategies are debated in terms of how best to promote stability rather than being dismissed as inconsistent with a well-functioning international monetary system. The paper uses longer time-series, a wider cross-section of countries, and an innovative empirical design to contribute to these discussions.

*Appendix A. Data Details.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022158-print-pdf.pdf_