## 1. Growth of Foreign Exchange Reserves

## Source details

**Canonical URL:** [1. Growth of Foreign Exchange Reserves](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1370.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1370.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1370.pdf.json)

---

### Overview
- Purpose: Identify determinants of foreign exchange intervention (INT) in emerging markets (EMs) with sufficiently flexible exchange rate regimes.
- Three main motives evaluated:
  - Smoothing short-run exchange rate fluctuations ("leaning against the wind").
  - Precautionary accumulation of foreign exchange reserves.
  - Competitiveness (mercantilist) accumulation.
- Approach: Hybrid panel model combining short-run exchange rate pressures and slower-moving macroeconomic variables to explain cross-country and time variation in intervention.

### Data, sample, and dependent variable
- Sample:
  - 15 EM cross-sections: Brazil, Colombia, Indonesia, India, Israel, Korea, Mexico, Peru, Philippines, Poland, Russia, South Africa, Taiwan, Thailand, Turkey.
  - Monthly data from 2001 to 2012 (sample period ends in February 2012).
  - Inclusion requires flexible exchange rates for entire observation period and available/estimable monthly intervention data.
- Dependent variable (INT):
  - Total monthly intervention = SInt + FInt, divided by an interpolated monthly GDP series.
  - Intervention expressed in U.S. dollars.
  - Summary statistics: mean(INT) = 2.4 percent of monthly GDP; median = 1.2 percent.
- Intervention data quality sources and treatments:
  - Published intervention data: Brazil, Colombia, India, Israel, Mexico, Peru, Russia (after 2007), Turkey.
  - Monthly change in reserves with valuation changes stripped out by central bank: Korea, Philippines, Poland, Taiwan (after October 2003), Thailand.
  - Valuation changes stripped out by author using known annual currency composition: Russia (before 2008), South Africa.
  - Valuation changes stripped out by author using unknown composition/COFER: Indonesia, Taiwan (before November 2003).
- Tables and appendices:
  - Tables 1–6 covering determinants, subperiods, reserve adequacy metrics, interactions, robustness, simultaneity-bias controls.
  - Appendix 1: Notes on Foreign Exchange Intervention Data Utilized in Model.

### Key independent variables and measurement
- Short-run exchange rate pressures:
  - Δe = monthly change in nominal U.S. dollar bilateral exchange rate.
  - Modeled using one-month lag Δe(-1) to reduce simultaneity bias.
  - VIX monthly change used as proxy for contemporaneous Δe in baseline.
- Medium-run exchange rate:
  - RER = percent change between current REER and its rolling five-year average; used with two-month lag RER(-2).
- Exports/GDP (EXGDP): monthly goods exports / GDP.
- Reserve adequacy dummies (based on IMF metric):
  - RAD1 = 1 if reserves > 150% of metric, else 0.
  - RAD2 = 1 if reserves < 100% of metric, else 0.
- Inflation (π): year-on-year change of monthly CPI.
- Other variables tested (not always in baseline): CDS, MANEX, MEUS, NEER/REER monthly changes, election/crisis dummies, policy rate, per capita GDP, population, seasonal dummies.

### Estimation technique
- Panel least squares with AR1 residuals and heteroskedasticity-corrected (White) standard errors.
- Cross-sections: 15 EMs; no country or time fixed effects.
- Crisis dummy: May 2008–Dec 2008 = 1.
- Baseline conceptual equation: INT = f(VIX, Δe(-1), RER(-2), EXGDP, RAD1, RAD2, π, crisis dummy, AR1 correction).

### Main empirical results — baseline and subgroup findings
- VIX:
  - A 10% m-o-m increase in the VIX (e.g., 20 to 22) is associated with a 0.32% of GDP decrease in INT (coefficient = -0.0320).
  - Coefficients similar and statistically significant for both non-commodity and commodity producers.
- Short-run exchange rate pressure (Δe(-1)):
  - Baseline (all countries): coefficient ≈ 0.117 (implied baseline mention).
  - For non-commodity producers: a 1% m-o-m appreciation of Δe(-1) is associated with a 0.16% of GDP increase in INT (coefficient ≈ 0.155 in some specifications).
  - Δe(-1) coefficient often not statistically significant for commodity producers.
- REER (RER(-2)):
  - Baseline coefficient = 0.0433 (implied): a REER(-2) that is 10% more appreciated than its five-year rolling average is associated with a 0.43% of GDP increase in INT for non-commodity producers.
  - RER(-2) not statistically significant for commodity producers in several specifications.
- Inflation (π):
  - For non-commodity exporters, a 1% increase in year-on-year CPI inflation is associated with a 0.1% of GDP decrease in INT for the full sample (baseline), rising to 0.7% of GDP decrease in the 2007-2012 period.
  - CPI coefficient generally not statistically significant for commodity exporters.
- Exports/GDP (EXGDP):
  - Baseline coefficient = 0.0782: a 10% increase in EXGDP is associated with a 0.78% of GDP increase in INT.
  - EXGDP statistically significant for both commodity and non-commodity exporters.
  - Illustrative comparison: Korea (EXGDP = 53%) estimated to purchase on average 1.6% of GDP more per month than Mexico (EXGDP = 32%) per baseline interpretation.
- Reserve adequacy dummies:
  - RAD1 (reserves > 150% of IMF metric):
    - Baseline (all countries) coefficient = 0.0158: months with RAD1 = 1 are associated with a 1.6% of GDP increase in net FX purchases per month for non-commodity exporters (other subsamples showed 0.0209).
  - RAD2 (reserves < 100% of IMF metric):
    - Baseline (all countries) coefficient = -0.00808: months with RAD2 = 1 are associated with a 0.6% of GDP decrease in net FX purchases per month for non-commodity exporters (other subsamples showed -0.0113).
  - Interpretation: EMs with already high reserve adequacy tend to continue accumulating; EMs with low adequacy do not necessarily "catch up" to the metric.
- Crisis dummy:
  - Crisis dummy coefficient (example baseline) = -0.0504, indicating reduced net FX purchases during May 2008–Dec 2008.
- AR1 and model fit:
  - AR1 coefficients ≈ 0.39–0.63 across specifications.
  - R² values typically between 0.20 and 0.44 depending on subsample/specification.

### What drives differences in rates of reserve accumulation across EMs
- Exchange rate variables have mean ≈ 0 and explain little of cross-country differences.
- EXGDP is the most important variable explaining cross-country differences in accumulation:
  - Example from baseline: EM with EXGDP = 60% expected to purchase on average 4% of GDP more net FX than an EM with EXGDP = 10%.
- Interactions:
  - EXGDP * RAD1: positive and statistically significant → EMs with higher reserve levels are more sensitive to exports in their intervention behavior (suggesting non-precautionary motive).
  - EXGDP * Δe(-1): positive and statistically significant → EMs with higher EXGDP are more sensitive to short-run exchange rate movements.
  - EXGDP * REER: not statistically significant.
- Alternative reserve adequacy metrics (Table 3):
  - Substitutes (Reserves/GDP, reserves/GDP relative to EM average, reserves/imports, IMF reserve adequacy metric lagged one month) produce positive and statistically significant coefficients consistent with baseline.
  - M2/GDP and short-term debt to GDP (STDG) produce positive, significant coefficients when substituted for EXGDP, though EXGDP remains the more robust indicator.
- Interpretation: Evidence supports coexistence of precautionary and competitiveness motives; persistent accumulation by EMs with high reserves is not fully consistent with a pure precautionary target calibrated to the IMF metric.

### Changing motivations over time (subperiod analysis)
- Subperiods compared: 2001–2006 vs. 2007–2012.
- Findings:
  - REER and CPI coefficients are much larger in 2007–2012 relative to 2001–2006 for non-commodity exporters, indicating increased sensitivity to medium-run exchange rate and inflation conditions.
  - Average level of the IMF reserve adequacy metric for sample EMs rises from 122% in 2001–06 to 168% in 2007–12.
  - RAD1 loses statistical significance over time as more EMs exceed the 150% threshold.
  - Interpretation: Later-period EM central banks may have become more sophisticated in conditional intervention based on REER and inflation.

### Robustness checks and endogeneity controls
- Intervention data robustness (Table 5):
  - Restricting sample to countries that publish actual intervention data (7 countries), or adding countries with valuation-adjusted reserve changes, yields broadly similar coefficients to full sample.
- Controls for Δe simultaneity bias:
  - Baseline uses VIX as proxy for contemporaneous Δe; excluding VIX barely changes coefficients/significance.
  - Two-stage least squares (TSLS) instrumenting Δe with change in VIX and Δe of other EMs:
    - TSLS estimates: coefficient on Δe = 0.584; coefficient on Δe(-1) = 0.315.
    - Together imply a 1% appreciation of the nominal U.S. dollar exchange rate results in a 0.90% of GDP increase in intervention over two months (0.584 + 0.315 ≈ 0.899, reported as 0.90% of GDP in text).

### Additional determinants tested (less robust or not in baseline)
- CDS: negative coefficient but highly correlated with inflation; interpretation ambiguous.
- Manufacturing exports (MANEX), commodity exports: less robust than total EXGDP.
- MEUS (manufacturing exports to U.S. / GDP): negative and significant in some tests but lost significance after removing Mexico.
- NEER/REER monthly changes: less robust than Δe.
- Election-period and post-crisis dummies: not statistically significant.
- Policy rate: behaved similarly to inflation but less robust.
- Per capita GDP and population: statistically insignificant.
- No evidence of seasonal patterns once main variables included.

### Conclusions and policy implications
- Empirical evidence supports joint operation of:
  - Short-run smoothing ("leaning against the wind"),
  - Precautionary reserve accumulation,
  - Competitiveness/mercantilist motives.
- Key takeaways:
  - Short-run exchange rate pressures (Δe) and global risk sentiment (VIX) influence monthly intervention.
  - Medium-run REER appreciation relative to a five-year average is associated with increased net FX purchases for non-commodity exporters.
  - Larger export sectors (EXGDP) correlate strongly with higher rates of reserve accumulation, reflecting both precautionary and competitiveness motives.
  - EMs with already high reserve adequacy tend to continue accumulating reserves; EMs with low adequacy do not necessarily accelerate accumulation toward recommended metrics.
- Suggested areas for further research:
  - Role of monetary policy framework quality in shaping intervention decisions.
  - Quantifying differences in central banks’ tools and sophistication in handling capital flows.
  - Testing the role of capital controls on foreign exchange intervention decisions.

### Appendix 1 — Notes on foreign exchange intervention data utilized in model (methodological highlights)
- Data quality hierarchy:
  - 1) FX intervention data directly from government authority.
  - 2) Change in reserves with valuation effects stripped out by authorities.
  - 3) Change in reserves with valuation effects stripped out using currency composition published by authorities.
  - 4) Change in reserves with valuation effects stripped out using a currency composition based on COFER.
- General construction rules:
  - Use SDDS reserve-template forward position data for forward market intervention when available.
  - Spot and forward transactions are added together to generate the total intervention series.
  - Principle: "best available" data prioritized over homogeneity across EMs or time periods.
- Country-level notes (selected):
  - Brazil, India, Mexico, Israel, Turkey, Colombia, Peru: intervention data available on central bank websites for entire period.
  - Korea, Philippines, Poland, Taiwan (after Oct 2003), Thailand: monthly change in reserves with valuation changes stripped out available in balance of payments data.
  - Russia: intervention data available from August 2008; valuation-adjusted reserves available Jan 2007–Jul 2008; prior adjustments based on annual currency composition.
  - Taiwan: valuation-adjusted reserve changes available until November 2003; prior adjustments used average currency composition (COFER); Taiwan does not submit SDDS reserve template; possible forward transactions not included.
  - South Africa, Indonesia: reserve stock changes used with valuation adjustments based on central bank reports or COFER.
- Modeling implication:
  - When direct intervention data unavailable, changes in reserve stocks are adjusted to remove valuation effects using the hierarchy above; combined spot and forward components produce total intervention series.

*Source: _wp1370 - 1. Growth of Foreign Exchange Reserves (IMF working paper).*

### 1. Growth of Foreign Exchange Reserves ................................................................................ 

### 1. Growth of Foreign Exchange Reserves

### Overview
- Section title: "1. Growth of Foreign Exchange Reserves"
- Appears beginning on page 5.

### Empirical content and tables
- Tables included in the section:
  - Table 1. Determinants of FX Intervention ........................................................................................ 19
  - Table 2. Time Subperiods ................................................................................................................. 20
  - Table 3. Other Reserve Adequacy Metrics ....................................................................................... 21
  - Table 4. Interactions between Independent Variables ...................................................................... 22
  - Table 5. FX Intervention Data Robustness Tests .............................................................................. 23
  - Table 6. Controls betweeen INT and ∆e Simultaneity Bias ............................................................. 24

### Appendices associated with the section
- Appendix 1. Notes on Foreign Exchange Intervention Data Utilized in Model ..................................... 25

### Key structural elements available for ingestion
- Regression output referenced under "Tables - Regression Output"
- Six numbered tables (Tables 1–6) covering determinants, subperiods, reserve adequacy metrics, interactions, robustness tests, and simultaneity-bias controls
- One appendix providing notes on foreign exchange intervention data used in the model

*Source: _wp1370 - 1. Growth of Foreign Exchange Reserves (pages and table listings as provided).*

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

### _wp1370 - References

### Introduction: purpose and approach
- Purpose: identify determinants of foreign exchange intervention in emerging markets (“EMs”) over the last decade for EMs with sufficiently flexible exchange rate regimes.
- Three main determinants evaluated:
  - Smoothing short-run exchange rate fluctuations (leaning against the wind).
  - Accumulation of foreign exchange reserves for precautionary purposes.
  - Accumulation of foreign exchange reserves for competitiveness reasons.
- Approach: hybrid panel model including short-run exchange rate pressures and slower-moving macroeconomic variables to explain why EMs intervene now and why intervention differs across EMs.

### Data: sample and dependent variable
- Sample: monthly data for 15 EM cross-sections: Brazil, Colombia, Indonesia, India, Israel, Korea, Mexico, Peru, Philippines, Poland, Russia, South Africa, Taiwan, Thailand, and Turkey.
- Sample period: monthly data from 2001 to 2012 (sample period ends in February 2012).
- Inclusion criterion: flexible exchange rates for entire observation period and availability/estimability of monthly intervention data.
- Dependent variable (INT): total monthly intervention = SInt + FInt, divided by an interpolated monthly GDP series; intervention expressed in U.S. dollars.
- Intervention data quality and sources:
  - Published intervention data: Brazil, Colombia, India, Israel, Mexico, Peru, Russia (after 2007), Turkey.
  - Monthly change in reserves with valuation changes stripped out by central bank: Korea, Philippines, Poland, Taiwan (after October 2003), Thailand.
  - Valuation changes stripped out by author with known annual currency composition: Russia (before 2008), South Africa.
  - Valuation changes stripped out by author with unknown composition: Indonesia, Taiwan (before November 2003).
- Summary statistics: mean(INT) for all EMs in sample is 2.4 percent of monthly GDP and median is 1.2 percent.

### Independent variables (key specifications and measurement)
- Short-run exchange rate pressures: Δe = monthly change in nominal U.S. dollar bilateral exchange rate; modeled using one-month lag Δe(-1) to reduce simultaneity bias; VIX monthly change used as proxy for current-month Δe in baseline.
- REER relative to 5-year rolling average (RER): RER = percent change between current REER and its rolling five-year average; used with a two-month lag RER(-2).
- Exports as a share of GDP (EXGDP): monthly goods exports / GDP, included to capture external trade sector size.
- Reserve adequacy dummies (based on IMF reserve adequacy metric):
  - RAD1 = 1 if reserves > 150% of the metric, else 0.
  - RAD2 = 1 if reserves < 100% of the metric, else 0.
- Inflation (π): year-on-year rate of change of monthly CPI.
- Other considered variables (tested but not in baseline): CDS, MANEX, MEUS, NEER/REER month-to-month changes, election/crisis dummies, policy rate, per capita GDP, population, seasonal dummies.

### Estimation technique
- Main model: panel least squares with AR1 residuals and heteroskedasticity-corrected (White) standard errors.
- Cross-sections: 15 EMs; no country or time fixed effects.
- Crisis dummy: May 2008–Dec 2008 equals 1.
- Baseline equation (conceptual): INT = f(VIX, Δe(-1), RER(-2), EXGDP, RAD1, RAD2, π, crisis dummy, AR1 correction).

### Main empirical results — baseline and subgroup findings
- VIX:
  - A 10% m-o-m increase in the VIX (e.g., 20 to 22) is associated with a 0.32% of GDP decrease in INT (coefficient = -0.0320).
  - Coefficients similar and statistically significant for both non-commodity and commodity producers.
- Short-run exchange rate pressure (Δe(-1)):
  - For non-commodity producers, a 1% m-o-m appreciation of Δe(-1) is associated with a 0.16% of GDP increase in INT (coefficient ≈ 0.155 in some specifications; baseline 0.117 for all countries).
  - Δe(-1) coefficient fairly stable across time periods; not statistically significant for commodity producers in several specifications.
- REER (RER(-2)):
  - For non-commodity producers, a REER(-2) that is 10% more appreciated than its five-year rolling average is associated with a 0.43% of GDP increase in INT (implied from coefficient 0.0433 in baseline).
  - REER(-2) not statistically significant for commodity producers in several specifications.
- Inflation (π):
  - For non-commodity exporters, a 1% increase in year-on-year CPI inflation is associated with a 0.1% of GDP decrease in INT for the full sample (baseline), rising to 0.7% of GDP decrease in the 2007-2012 period.
  - CPI coefficient generally not statistically significant for commodity exporters.
- Exports/GDP (EXGDP):
  - A 10% increase in EXGDP is associated with a 0.78% of GDP increase in INT (baseline coefficient 0.0782).
  - EXGDP is statistically significant for both commodity and non-commodity exporters.
  - Illustrative comparison: Korea (EXGDP = 53%) estimated to purchase on average 1.6% of GDP more per month than Mexico (EXGDP = 32%) per baseline interpretation.
- Reserve adequacy dummies:
  - RAD1 (reserves > 150% of IMF metric): months with RAD1 = 1 are associated with a 1.6% of GDP increase in net FX purchases per month for non-commodity exporters (baseline coefficient 0.0158 for all countries; 0.0209 in some subsamples).
  - RAD2 (reserves < 100% of IMF metric): months with RAD2 = 1 are associated with a 0.6% of GDP decrease in net FX purchases per month for non-commodity exporters (baseline coefficient -0.00808 for all countries; -0.0113 in some subsamples).
  - Interpretation: EMs with high observed reserve adequacy tend to keep accumulating reserves; EMs with low adequacy do not “catch up” to the metric.
- Crisis dummy:
  - Crisis dummy (May 2008–Dec 2008) has negative and statistically significant coefficients (e.g., -0.0504 in baseline), indicating reduced net FX purchases during crisis months.
- AR1 corrections and model fit:
  - AR1 coefficients around 0.39–0.63 across specifications; R² values typically between 0.20 and 0.44 depending on subsample/specification.

### What drives differences in rates of reserve accumulation?
- Exchange rate variables have mean ≈ 0 and explain little of cross-country differences.
- EXGDP is the most important variable explaining different rates of accumulation between EMs:
  - Example from baseline: EM with EXGDP = 60% expected to purchase on average 4% of GDP more net FX than an EM with EXGDP = 10%.
- Interactions:
  - EXGDP * RAD1: statistically significant positive coefficient → EMs with higher reserve levels are more sensitive to their export level regarding intervention (suggests non-precautionary motive).
  - EXGDP * Δe(-1): statistically significant positive coefficient → EMs with higher EXGDP are more sensitive to short-run exchange rate movements.
  - EXGDP * REER: not statistically significant.
- Alternative reserve adequacy metrics (Table 3):
  - Reserves/GDP, reserves/GDP relative to EM average, reserves/imports, and the IMF reserve adequacy metric (lagged one month) all have positive and statistically significant coefficients when substituted, consistent with baseline result that higher existing reserves associate with continued accumulation.
  - M2/GDP and short-term debt to GDP (STDG) produce positive, significant coefficients when used as substitutes for EXGDP, though EXGDP remains the more robust indicator.
- Interpretation: evidence points to both precautionary and competitiveness motives; positive EXGDP coefficient likely reflects a combination of precautionary demand (openness, financial development) and mercantilist/competitiveness motives. The persistence of accumulation among EMs with already high reserves is not well explained by a pure precautionary target calibrated to the IMF metric.

### Changing motivations over time
- Subperiod differences (2001-2006 vs. 2007-2012):
  - REER and CPI coefficients are much larger in 2007-2012 relative to 2001-2006 for non-commodity exporters, suggesting increased sensitivity to medium-run exchange rate and inflation conditions.
  - The average level of the IMF reserve adequacy metric for sample EMs rises from 122% in 2001-06 to 168% in 2007-12.
  - High reserve adequacy dummy loses statistical significance over time as more EMs exceed the 150% threshold.
  - Interpretation: later-period EM central banks may have become more sophisticated in intervention strategies, using intervention more conditionally on REER and inflation.

### Robustness checks and endogeneity controls
- Intervention data quality tests (Table 5):
  - Restricting sample to countries that publish actual intervention data (7 countries) or adding those that publish valuation-adjusted reserve changes (additional countries) yields broadly similar coefficients to full sample; main results robust to data quality variations.
- Controls for Δe simultaneity bias:
  - Baseline uses VIX as proxy for contemporaneous Δe; alternative specification excluding VIX yields barely changed coefficients and significance.
  - Two-stage least squares (TSLS) instrumenting Δe with change in VIX and Δe of other EMs:
    - TSLS estimates: coefficient on Δe = 0.584; Δe(-1) = 0.315.
    - Together imply a 1% appreciation of the nominal U.S. dollar exchange rate results in a 0.90% of GDP increase in intervention over two months (0.584 + 0.315 ≈ 0.899, reported as 0.90% of GDP in text).

### Additional considered determinants (not in final baseline or less robust)
- CDS: negative coefficient but high correlation with inflation; interpretation ambiguous.
- Manufacturing exports (MANEX), commodity exports: less robust than total EXGDP.
- MEUS (manufacturing exports to U.S. / GDP): negative and significant in some tests but lost significance after removing Mexico.
- NEER/REER monthly changes: less robust than Δe.
- Election-period and post-crisis dummy variables: not statistically significant.
- Policy rates behaved similarly to inflation but were less robust.
- Per capita GDP and total population: statistically insignificant.
- No evidence of a seasonal pattern once main variables included.

### Conclusions and implications
- Empirical evidence supports three determinants operating jointly: short-run smoothing (leaning against the wind), precautionary accumulation, and competitiveness/mercantilist motives.
- Key empirical takeaways:
  - Short-run exchange rate pressures (Δe) and global risk sentiment (VIX) influence monthly intervention.
  - Medium-run REER appreciation relative to a five-year average is associated with increased net FX purchases for non-commodity exporters.
  - Larger export sectors (EXGDP) correlate strongly with higher rates of reserve accumulation, reflecting both precautionary and competitiveness motives.
  - EMs with already high reserve adequacy tend to continue accumulating reserves; EMs with low adequacy do not necessarily accelerate accumulation to reach recommended adequacy metrics.
- Areas for further research:
  - Role of the type and quality of monetary policy framework in shaping intervention decisions.
  - Quantifying differences in central banks’ tools and sophistication in handling capital flows.
  - Testing the role of capital controls on foreign exchange intervention decisions.

*Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1370.pdf*

### Appendix 1: Notes on Foreign Exchange Intervention Data Utilized in Model

### Appendix 1: Notes on Foreign Exchange Intervention Data Utilized in Model

### Data quality hierarchy used
- 1) Foreign exchange intervention data directly from a respective government authority.
- 2) The change in foreign exchange reserves with valuation effects stripped out by the authorities.
- 3) The change in reserves with foreign exchange valuation effects stripped out using a currency composition of reserves published by the authorities.
- 4) The change in reserves with foreign exchange valuation effects stripped out using a currency composition based on COFER data.

### General data construction rules
- Unless otherwise indicated, forward position data available in the central bank’s SDDS public reserve template submission was used to calculate a central bank’s forward market intervention.
- Spot and forward transactions are added together to generate the total intervention series.
- Principle applied: “best available” data prioritized over homogeneity across EMs or time periods.

### Country-level notes on data sources and adjustments
- Brazil:
  - Data on foreign exchange intervention is available on the Brazilian central bank website for the entire period.
- India:
  - Data on foreign exchange intervention is available on the Indian central bank website for the entire period.
- Indonesia:
  - Monthly foreign exchange reserves stocks available on Indonesian central bank website.
  - Exchange rate valuation effects stripped out assuming currency composition is the same as average currency composition of reserves (based on the IMF’s COFER).
- Korea:
  - Monthly change in foreign exchange reserves with exchange rate valuation changes stripped out is available in Korea’s monthly balance of payments data.
  - Forward intervention data prior to 2005 is not available and thus is not included in the dataset.
- Mexico:
  - Data on foreign exchange intervention is available on the Mexican central bank website for the entire period.
- Philippines:
  - Monthly change in foreign exchange reserves with valuation changes stripped out is available in the Philippines’s monthly balance of payments data.
- Russia:
  - Data on foreign exchange intervention is available on the Russian central bank website from August 2008.
  - Monthly change in foreign exchange reserves with valuation changes stripped out is available in Russia’s monthly balance of payments data from January 2007 to July 2008.
  - Prior to January 2007, monthly foreign exchange reserves stocks are used and exchange rate valuation changes are stripped out based on the currency composition of reserves listed in the Russian central bank’s annual reports.
- Taiwan:
  - Change of foreign exchange reserves with valuation changes stripped out is available on the central bank website until November 2003.
  - Prior to November 2003, monthly foreign exchange reserves stocks are used and exchange rate valuation changes are stripped out based on the average currency composition of reserves (based on the IMF’s COFER).
  - Taiwan does not submit a reserve template to the IMF; possible forward transactions by Taiwan’s central bank are not included in the estimate of foreign exchange intervention.
- Thailand:
  - Monthly change in foreign exchange reserves with exchange rate valuation changes stripped out is available in Korea’s monthly balance of payments data.
- Poland:
  - Monthly change in foreign exchange reserves with exchange rate valuation changes stripped out is available in Poland’s monthly balance of payments data.
- Israel:
  - Data on foreign exchange intervention is available on the Israeli central bank website for the entire period.
- Turkey:
  - Data on foreign exchange intervention is available on the Turkish central bank website for the entire period.
- South Africa:
  - Monthly foreign exchange reserves stocks are available on the South African central bank website.
  - Exchange rate valuation changes are stripped out based on the currency composition of reserves listed in the South African central bank’s annual reports.
- Colombia:
  - Data on foreign exchange intervention is available on the Colombian central bank website for the entire period.
- Peru:
  - Data on foreign exchange intervention is available on the Peruvian central bank website for the entire period.

### Key methodological points for model inputs
- When direct intervention data from authorities is unavailable, changes in reserve stocks are adjusted to remove valuation effects using one of the hierarchy methods listed above.
- Use of SDDS reserve template forward positions is the standard approach to estimate forward interventions when available.
- Spot and forward components are combined to form total intervention series.

*Source: Appendix 1: Notes on Foreign Exchange Intervention Data Utilized in Model (IMF working paper).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1370.pdf_