## Annex II.A / Annex I — Policy-response estimation and country case study (Peru)

## Source details

**Canonical URL:** [Annex II.A / Annex I — Policy-response estimation and country case study (Peru)](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020010-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020010-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020010-print-pdf.pdf.json)

---

### A. Empirical strategy and distributional fitting
- Policy response estimates:
  - Annex I reports estimated sensitivities to the shock for each future policy response: ,ˆs fx αβ, ,ˆsi αβ and ,ˆse αβ, and corresponding estimates of the separate (mitigating or amplifying) impact of complementary policy tools: ,ˆsp fx αβ, ,ˆsp i αβ and ,ˆsp e αβ.
  - Short-term and medium-term sensitivities: Annex I, Table 1 captures future average responses over the next two quarters (h=2); Annex I, Table 2 reflects future average responses over the next two years (h=8).
- Distributional approach:
  - Empirical distributions of future policy responses (from quantile regressions) are fitted to a theoretical skewed-t probability distribution (Azzalini and Capitanio, 2003) by minimizing the sum of squared residuals between empirical percentile functions and the skewed-t distribution.
  - The skewed-t is characterized by four moments (mean, variance, skewness, kurtosis) and nests both normal and standard t-distribution.
  - For each of the 20 emerging markets in the sample, conditional forecast distributions quantify likelihoods and risks of future policy responses at each point in time and over a given horizon.

### B. Construction of the exogenous capital flow shock
- Rationale:
  - Policy actions can affect foreign investors’ decisions and portfolio flows, creating endogeneity in estimating policy response to capital flow shocks.
- Shock construction (following Blanchard et al. (2015)):
  - For each country j, the exogenous measure of aggregate gross portfolio flows is the sum of gross private portfolio inflows to all other countries in the sample, leaving out country j.
  - Scale the aggregate shock by the rolling 4-quarter sum of quarterly nominal HP-filtered GDP (in US dollars) of all other countries to smooth the denominator so high-frequency variation reflects portfolio inflows rather than business cycle volatility.
  - Focus on gross portfolio inflows (not total or net) because portfolio investments typically have shorter horizons and are more sensitive to shifts in global risk-aversion.
- Properties:
  - Each country-specific shock is close to identical across countries (small marginal contribution of any country to aggregate flows).
  - The constructed shocks are highly (and inversely) correlated with the VIX, indicating they proxy push factors and shifts in global risk aversion/fundamental uncertainty.

### C. Accounting for additional policy tools (MPMs and CFMs)
- Data and limitations:
  - Lack of comprehensive cross-country measures of intensity (rates) for MPMs and CFMs; available data report number and direction of policy actions by quarter.
  - Sources used: (1) IMF’s Macroprudential Policy Survey database (iMaPP) and (2) IMF’s 2019 Taxonomy of Capital Flow Management Measures (the Taxonomy).
  - iMaPP contains information on a wide range of MPMs; many are not relevant for the sample/type of shocks, so focus is on MPMs that target FX positions explicitly or implicitly.
- Focused policy instruments:
  - MPMs targeting FX positions include:
    - Reserve requirements for macroprudential purposes, differentiated by currencies or residency;
    - Limits on net or gross open FX positions;
    - Limits on FX exposures, FX funding, and currency mismatch regulations;
    - Limits on foreign currency (FC) lending, and rules/recommendations on FC loans.
  - CFMs classification (per the Taxonomy and Institutional View):
    - Residency-based measures (discriminate by residency).
    - Other CFMs (do not discriminate by residency but limit capital flows).
    - CFM/MPMs: measures that both limit capital flows and reduce systemic risks from such flows.
- Combined policy-action time series:
  - Construct AddPolicy (per country, per quarter) combining MPMs and CFMs:
    - Inflow_t CFM Δ, Outflow_t CFM Δ, and t MPM Δ enter AddPolicy via:
      - Inflow_t CFM Δ take value +1 for tightening/introduction of a CFM on inflows, -1 for loosening/removal in quarter t.
      - Outflow_t CFM Δ take value +1 for tightening/introduction of a CFM on outflows, -1 for loosening/removal in quarter t.
      - t MPM Δ take value +1 for tightening/introduction of a MPM, -1 for loosening/removal in quarter t.
    - AddPolicy defined as:
      - Inflow_t CFM Δ − Outflow_t CFM Δ + t MPM Δ  (equation (3) as given)
- Empirical implementation:
  - Interact the shock (Shock) with AddPolicy in each of the 3 equations in (2) and include AddPolicy as a regressor on its own to estimate ,ˆsp αβ· (the mitigating or amplifying impact on a future policy response, at percentile α, of implementing additional unconventional policy actions in the quarter the shock hits).
  - Note: Extensions could examine MPMs and CFMs separately or include other complementary policies (fiscal, macro-structural) in future work.

### III. Country-specific policy responses — Case study: Peru
- Context:
  - Peru adopted inflation targeting in 2002 but frequently intervened to smooth FX market volatility due to financial dollarization and currency mismatch vulnerabilities.
- Historical intervention episodes and magnitudes:
  - 2002–2007: net purchases of U.S. dollars totaling $20 bln or 27 percent of GDP (precautionary reserve accumulation).
  - June 2008–February 2009: $7.1 billion sold (about 6 percent of 2009 GDP) to avoid negative balance sheet effects.
  - 2013–2016: sol depreciated as much as 35 percent amid tapering, China slowdown, and commodity price declines.
  - Tapering period net FX sales of about $4.2 bln and $8.1 bln in 2014 and 2015, respectively.
- Conditional forecast distributions and “at-risk” monitoring:
  - Method: fit t-skewed distribution to predicted values from country-specific quantile regressions to compute conditional forecast distributions of FX intervention (FXI), policy rate changes, and exchange rate changes.
  - Reporting: for each quarter, actual average policy changes over the next two quarters (black solid line) and predicted downside tail risks at the 10th percentile (red dashed line) and upside tail risks at the 90th percentile (green dashed line) are estimated from in-sample predictive densities over Q1 2000-Q3 2018.
- Key findings from Peru case:
  - Predictive performance:
    - Realized policy changes generally fall within the predictive distribution’s 10th–90th percentile range for FXI, policy rate, and exchange rate changes; notable exceptions include:
      - Large FX purchase in 2007;
      - Extreme monetary loosening in 2009;
      - Strong depreciation pressures on the sol in mid-2018.
  - Tail-risk dynamics 2003–2007:
    - Likelihood of extreme FX purchases increased: in 2007, framework predicted a 10 percent chance that FX purchases over the following two quarters would average at least 3 percent of GDP compared to a forecast of at least 1 percent of GDP in 2003.
    - No obvious shift in likelihood of extreme exchange rate appreciations or policy rate changes over the same period (consistent “FXI-only” response to inflows).
  - Tail-risk dynamics 2011–2015:
    - During adverse external shocks, higher chances of extreme FX sales and slightly higher chances of extreme monetary policy loosening and extreme exchange rate depreciation — indicating asymmetry of policy actions over the cycle.
  - Short-term shifts between Q3 2016 and Q3 2018 (U.S. Fed tightening episode):
    - FXI:
      - Central estimate of expected intervention over next 2 quarters fell from a conditional mean of 0.5 percent of GDP to a conditional mean of -0.1 percent of GDP.
      - Probability of short-term FX sales “soared” between Q3 2016 and Q3 2018 from 21 to 46 percent of GDP (with a 5 percent chance that Peru’s FX sales over the next 2 quarters would be at least 1.6 percent of GDP as of Q3 2018, rather than by just 1.1 percent of GDP, as predicted 4 quarters earlier).
    - Policy rate changes:
      - Conditional forecast distribution virtually unchanged between Q3 2016 and Q3 2018 (future monetary policy stance not expected to change given prevailing conditions).
    - Exchange rate responses:
      - Short-term central outlook improved between Q3 2016 and Q3 2018: conditional mean moved from -0.38 percent (depreciation) to +0.13 percent (appreciation).
      - Probability of further depreciation receded from 70 to 55 percent.
      - Downside risks remain: 5 percent chance that the sol would depreciate over the next 2 quarters by at least 3.2 percent as of Q3 2018.
  - Additional note in source: "2.4 percent, as predicted 8 quarters earlier." (text appears in source alongside PDF identifier).

### Scenario analysis: adverse global portfolio flow shock and Peru’s short-term policy responses
- Shock simulated:
  - Fall of two times their standard deviation over the sample (approximately 0.46 percent of emerging markets’ trend GDP).
- Method:
  - Compare two-quarter-ahead forecast distributions of FX intervention (FXI), policy rate changes, and exchange rate changes under shock vs. 2018Q3 baseline.
- Key quantitative outcomes:
  - Probability of FX sales over the next two quarters rises from 46 to 80 percent under the outflow shock (alternative: green line; baseline: red line).
  - Downside (tail) risks to FX intervention increase; red shaded areas indicate 5th percentile downside risks in shocked scenarios.
  - Conditional distribution of interest rate changes remains virtually unchanged under the outflow shock.
  - Exchange rate is expected to depreciate slightly under the outflow shock.
- Asymmetry of responses:
  - A same-size increase in global portfolio inflows (0.46 percent of emerging markets’ trend GDP) produces smaller FXI and exchange rate responses than the symmetric outflow shock.
  - No interest rate response is predicted in either the inflow or outflow shock cases.

### Role of additional policies (MPMs and CFMs) in mitigating inflow shocks (Peru)
- Approach:
  - Decompose effects into (i) stand-alone effect from additional policies and (ii) interaction between additional policies and the external shock; analyze effects across percentiles α.
- Main findings for Peru:
  - Use of additional policy measures is associated with lower extreme FX sales (i.e., reductions in magnitude at the lowest percentiles of the predicted conditional distribution for FXI) in the near future.
  - The total mitigating impact vis-à-vis extreme FX purchases appears to be negligible.
  - Use of additional policy measures can make interest rate hikes more likely following an exogenous inflow shock of foreign capital.
  - Additional policy measures appear to mitigate impacts of external shocks on extreme exchange rate appreciation and depreciation in the near future, although this mitigation is not found to be statistically significant.
- Visualization notes:
  - Figure 9 presents standardized quantile regression coefficients for FXI, policy rate change, and exchange rate change: impacts of shock variable, additional policy variable, and their interaction are shown with one-standard deviation confidence bands.

### Cross-country evidence: heterogeneity, skewness, and fat tails in policy responses
- Framework:
  - Estimate each country’s predicted two-quarter-ahead responses to an exogenous, same-sized global capital flow shock and correlate those with country characteristics (slow-moving or infrequently measured).
- Definitions and metrics:
  - Median response: average 2-quarter ahead median (50th percentile) response, denoted 50.,sβ.
  - Skewness: difference between average 2-quarter ahead 90th and 10th percentile responses, denoted 9010,.,.ssββ−.
  - Fat tails: differences between 10th–50th (left tail) and 90th–50th (right tail) percentile responses relative to the median.
- Key descriptive findings:
  - Responses to capital flow shocks are highly heterogeneous across countries even after controlling for cyclical conditions and a same-sized shock.
  - Example magnitude: for FX intervention, the average 2-quarter ahead response to a portfolio inflow of 1 percentage point of GDP varies between purchases of FX of 2 percentage points of GDP to sales of FX of -1 percentage points of GDP.
  - Skewness patterns:
    - FXI median skewness close to zero, but several countries exhibit strongly negative skew (large left tails) for FX intervention.
    - Exchange rate and interest rate responses: median countries show negative skew; several countries have very large left-tail responses.
    - Extreme responses are larger when selling FX, when the exchange rate is depreciating, and when policy rates are being cut.
  - Fat tails:
    - For FXI, both left and right tail responses exceed the median in magnitude; the left tail is particularly large.
    - For exchange rates and interest rates, the left tail is generally thicker than the right tail.
  - Country examples:
    - Peru responds the most in terms of FX intervention, while its exchange rate barely moves.
    - South Africa exhibits large exchange rate movements without much intervention.
    - Turkey adjusts policy rates more vigorously.

### Linking policy responses to country characteristics (conceptual approach and candidate determinants)
- Objective:
  - Explore correlations between cross-country differences in predicted policy responses and country characteristics that are slow-moving or infrequently measured.
- Candidate determinants include:
  - Proxies for balance sheet vulnerabilities and buffers:
    - size of domestic banks’ net open FX position;
    - ratio of short-term debt in foreign currency to forex reserves;
    - structure of external assets and liabilities;
    - level of reserve adequacy.
  - Degree of trade and capital account openness.
  - Indicators of financial market depth:
    - index of overall financial market development;
    - degree of turnover in the local currency FX market;
    - ask-bid spread.
  - Structural indicators:
    - exchange rate regime;
    - share of trade invoiced in USD;
    - degree of Central Bank transparency.
- Methodological note:
  - Because these characteristics are slow-moving, they cannot be controlled for in equation (2) (they would be subsumed in the country-specific constant). The analysis therefore focuses on raw correlations between country characteristics and predicted responses without presuming causality.

### Annex II.A — Summary of Findings on Policy Responses and Country Characteristics
- Overview and Table 1 notes:
  - Table 1 summarizes findings across all characteristics for policy rate responses, forex intervention (FXI) and exchange rate responses.
  - Many empty cells indicate no statistically significant association between corresponding structural characteristics and policy responses.
  - Regressions analyze two-quarter ahead policy responses on country characteristics.
  - Statistical significance: relationships one-sided significant at the 10 percent significance level are highlighted; stars denote significance at * 10%; ** 5%; *** 1% significance level.
  - Relevant t-statistic computed for 18 degrees of freedom (20 countries – 2 estimated coefficients).
  - Percentiles used to define responses: 10th percentile (“Major Sales”), 50th percentile (“Normal Times”), 90th percentile (“Major Purchases”).
- Monetary Policy Rate Responses:
  - Main finding:
    - Monetary policy rate responses are mostly unrelated to country characteristics, reflecting similar monetary policy frameworks that respond primarily to domestic cyclical conditions.
  - Notable exception:
    - The extent to which a currency is overvalued is associated with looser monetary policy in “typical” and “extreme” loosening responses.
  - Interaction with additional policies:
    - In the presence of additional unconventional policy tools (e.g., MPMs or CFMs), interest rates tend to be set higher under inflow pressures and lower under outflow pressures.
    - Additional policies make monetary policy actions more sensitive to expected inflation developments: monetary policy is more likely to tighten in the subsequent half-year if inflation is expected to rise in the same period.
- FX Intervention (FXI) and Balance Sheet / Market Characteristics:
  - FX intervention correlates with balance sheet vulnerabilities, buffers, and average exchange rate overvaluation:
    - Longer net open FX positions and larger short-term debt in foreign currency relative to reserves are associated with larger forex reserve sales in the face of outflow pressures and with smaller intervention in the face of inflow pressures.
    - Larger forex reserve sales tend to be associated with lower foreign assets as a share of GDP.
    - Reserves are positively correlated with FX intervention in both directions.
  - Causality caveat: statistical correlations do not imply causality.
- Exchange Rate Responses:
  - Overall relationship between exchange rate responses and country characteristics is weaker than for FXI.
  - Two characteristics associated with different exchange rate responses:
    - Net International Investment Position (NIIP): countries with substantial foreign assets relative to foreign liabilities witness less exchange rate movement.
    - Degree of de facto exchange rate flexibility: more flexible currencies show higher appreciation/depreciation for the same shock.
  - Responses are fat-tailed and asymmetric:
    - For most countries, “extreme” responses are larger than median responses.
    - Responses to outflows tend to be larger than responses to inflows for exchange rate adjustment, interest rate changes and forex intervention.
- The Role of Additional Policy Tools (MPMs, CFMs):
  - Scope and caveats:
    - Only a subset of the 20-country sample implemented these additional policies.
    - Analysis focuses on statistically significant country-specific interactions of these tools with the exogenous shock—conclusions driven by the few countries using these tools effectively.
  - A. Additional Policy Tools and Monetary Policy:
    - Additional unconventional policy tools allow monetary policy to more squarely focus on domestic inflation stabilization.
    - In the presence of inflow pressures, interest rates are set higher when additional policies are deployed significantly; under outflow pressures, interest rates are set lower.
    - Monetary policy actions become more sensitive to expected inflation when countries use additional unconventional policies.
  - B. Asymmetric Impact of Additional Policies (FXI and Exchange Rate):
    - Additional tools seem to have asymmetric impacts: “extreme” FX sales or exchange rate depreciations are more affected than “extreme” FX purchases or exchange rate appreciations.
    - Asymmetry of FXI is driven by Malaysia’s responses; the asymmetry of exchange rate responses is broader across countries.
  - C. Effects of Additional Policies Over Time (2-quarter vs 2-year horizons):
    - Monetary policy independence and asymmetric use of additional policies remain when looking at average effects past the first 2 quarters and over 2 years.
    - At the 2-quarter ahead horizon, auxiliary policy tools barely affect non-resident capital inflows.
    - At the 2-year ahead horizon, these tools appear much more effective at moderating both extreme outflows and extreme inflows.
- Conclusions and Policy Implications:
  - Cross-country heterogeneity:
    - Policy responses to global capital flows are highly heterogeneous across 20 EMs over almost two decades, even for the same-sized exogenous shock.
  - Stylized empirical facts:
    - Policy responses are fat-tailed and asymmetric, with larger reactions to outflows than inflows.
    - FXI responses (typical and extreme) are positively related to balance sheet vulnerabilities and inversely related to forex market depth.
  - Role of complementary tools:
    - CFMs and MPMs targeting external balance sheet vulnerabilities can enhance monetary policy autonomy: bringing these instruments into play strengthens sensitivity of both “typical” and “extreme” monetary policy responses to expected domestic cyclical developments.
  - Non-normative stance and caveats:
    - The analysis is descriptive and not normative; it does not prescribe alternative policies nor quantify cost-benefit trade-offs.
    - The apparent short-term monetary policy space created by unconventional tools may hide longer-term unintended consequences or intertemporal trade-offs.
    - Additional research is needed to assess longer-term effects and potential welfare trade-offs.

### ANNEX I — Tables (selected notes and example coefficients)
- Table 1. Quantile Regressions Estimates—2-Quarters Ahead Predictions:
  - Note: Bold font indicates significance at 10 percent. The additional policy variable AddPolicy is also included as a stand-alone control in each of the regressions.
  - Example country-specific coefficients (Shock and Shock*AddPolicy) at quantiles 0.1, 0.5, 0.9 are reported for Exchange Rate, Interest Rate, and FXI. Selected examples:
    - Argentina — FXI (Shock): 2.13, 0.30, 0.25; Exchange Rate (Shock*AddPolicy): -0.10, -3.51, 7.46.
    - Peru — FXI (Shock): 1.73, 1.89, 2.18; Interest Rate (Shock): 1.23, 0.07, 0.22.
    - Malaysia — FXI (Shock): 2.15, 1.79, 3.43; FXI (Shock*AddPolicy): 7.23, -0.59, -7.33.
    - Thailand — Exchange Rate (Shock*AddPolicy): 8.69, 11.00, 14.74; Interest Rate (Shock*AddPolicy): 5.21, 5.38, 6.89.
- Table 2. Quantile Regressions Estimates—2-Years Ahead Predictions:
  - Note: Bold font indicates significance at 10 percent. The additional policy variable AddPolicy is also included as a stand-alone control.
  - Selected example coefficients:
    - Peru — FXI (Shock): 1.37, 0.93, 1.00; FXI (Shock*AddPolicy): 2.31, -0.01, 0.31.
    - Malaysia — FXI (Shock): 2.16, 1.67, -0.03; FXI (Shock*AddPolicy): -0.29, -3.31, -8.33.
    - Korea — FXI (Shock): -0.70, -0.43, -1.32 (2-year horizon).

### ANNEX II — Data (key descriptors)
- Countries in sample: Argentina, Brazil, Chile, China, Colombia, Hungary, India, Indonesia, Kazakhstan, Korea, Malaysia, Mexico, Peru, Philippines, Poland, Russia, South Africa, Thailand, Turkey, Ukraine.
- Capital flow shock (Shock_t):
  - Data: quarterly balance of payments data from 1995Q1 to 2018Q3 for 20 emerging markets, based on the IMF’s Financial Flow Analytics database.
  - Measurement: gross inflows in US dollars scaled by annual HP-filtered GDP sourced from the IMF’s World Economic Outlook dataset.
- Policy responses:
  - FX intervention: estimated from balance of payments reserve flow (IMF’s World Economic Outlook dataset) adjusted for income accrued to previous positions using IMF’s COFER database; normalized by annual HP-filtered GDP. Positive = increase in FX assets (FX purchase); negative = decline in FX assets (FX sale).
  - Policy interest rate: End-of-quarter domestic policy rates (nominal, percentage points) from Haver Analytics dataset.
  - Exchange rate: End-of-quarter effective nominal exchange rates from Haver Analytics dataset.
- Additional Policies:
  - CFMs: IMF’s 2019 Taxonomy of Capital Flow Management Measures.
  - MPMs: IMF’s Macroprudential Policy Survey database (iMaPP).
- Controls:
  - Growth rate of real GDP: IMF’s World Economic Outlook database.
  - Ratio of short-term foreign currency debt to forex reserves: IMF's ARA database.
  - International commodity price index: Haver Analytics dataset.

### ANNEX III — Robustness / Caveats (selected)
- Bilateral vs Effective Exchange Rates:
  - Finding: The bilateral exchange rate vis-à-vis the USD tends to be more responsive in either of the two tails (extreme depreciations and extreme appreciations) because other major economies’ currencies offset some of the USD movements.
- FXI Measure:
  - Rationale for estimates: many central banks do not publish FX operations, published data may be infrequent or inconsistent; estimated FXI is used for cross-country consistency.
  - Comparison with published data for countries that publish (Argentina, Brazil, Colombia, India, Mexico, Peru, Kazakhstan, Russia and Turkey) shows estimated spot interventions generally track published FXI closely; exceptions include Mexico and Turkey.

*Source: IMF staff estimates and text from Annex II.A and Annex I (wpiea2020010-print-pdf).*

### Annex II.A provides a list of countries and relevant data description. Annex I

### Annex II.A / Annex I — Policy-response estimation and country case study (Peru)

### A. Empirical strategy and distributional fitting
- Policy response estimates:
  - Annex I reports estimated sensitivities to the shock for each future policy response: ,ˆs fx αβ, ,ˆsi αβ and ,ˆse αβ, and corresponding estimates of the separate (mitigating or amplifying) impact of complementary policy tools: ,ˆsp fx αβ, ,ˆsp i αβ and ,ˆsp e αβ.
  - Short-term and medium-term sensitivities: Annex I, Table 1 captures future average responses over the next two quarters (h=2); Annex I, Table 2 reflects future average responses over the next two years (h=8).
- Distributional approach:
  - Empirical distributions of future policy responses (from quantile regressions) are fitted to a theoretical skewed-t probability distribution (Azzalini and Capitanio, 2003) by minimizing the sum of squared residuals between empirical percentile functions and the skewed-t distribution.
  - The skewed-t is characterized by four moments (mean, variance, skewness, kurtosis) and nests both normal and standard t-distribution.
  - For each of the 20 emerging markets in the sample, conditional forecast distributions quantify likelihoods and risks of future policy responses at each point in time and over a given horizon.

### B. Construction of the exogenous capital flow shock
- Rationale:
  - Policy actions can affect foreign investors’ decisions and portfolio flows, creating endogeneity in estimating policy response to capital flow shocks.
- Shock construction (following Blanchard et al. (2015)):
  - For each country j, the exogenous measure of aggregate gross portfolio flows is the sum of gross private portfolio inflows to all other countries in the sample, leaving out country j.
  - Scale the aggregate shock by the rolling 4-quarter sum of quarterly nominal HP-filtered GDP (in US dollars) of all other countries to smooth the denominator so high-frequency variation reflects portfolio inflows rather than business cycle volatility.
  - Focus on gross portfolio inflows (not total or net) because portfolio investments typically have shorter horizons and are more sensitive to shifts in global risk-aversion.
- Properties:
  - Each country-specific shock is close to identical across countries (small marginal contribution of any country to aggregate flows).
  - The constructed shocks are highly (and inversely) correlated with the VIX, indicating they proxy push factors and shifts in global risk aversion/fundamental uncertainty.

### C. Accounting for additional policy tools (MPMs and CFMs)
- Data and limitations:
  - Lack of comprehensive cross-country measures of intensity (rates) for MPMs and CFMs; available data report number and direction of policy actions by quarter.
  - Sources used: (1) IMF’s Macroprudential Policy Survey database (iMaPP) and (2) IMF’s 2019 Taxonomy of Capital Flow Management Measures (the Taxonomy).
  - iMaPP contains information on a wide range of MPMs; many are not relevant for the sample/type of shocks, so focus is on MPMs that target FX positions explicitly or implicitly.
- Focused policy instruments:
  - MPMs targeting FX positions include:
    - Reserve requirements for macroprudential purposes, differentiated by currencies or residency;
    - Limits on net or gross open FX positions;
    - Limits on FX exposures, FX funding, and currency mismatch regulations;
    - Limits on foreign currency (FC) lending, and rules/recommendations on FC loans.
  - CFMs classification (per the Taxonomy and Institutional View):
    - Residency-based measures (discriminate by residency).
    - Other CFMs (do not discriminate by residency but limit capital flows).
    - CFM/MPMs: measures that both limit capital flows and reduce systemic risks from such flows.
- Combined policy-action time series:
  - Construct AddPolicy (per country, per quarter) combining MPMs and CFMs:
    - Inflow_t CFM Δ, Outflow_t CFM Δ, and t MPM Δ enter AddPolicy via:
      - Inflow_t CFM Δ take value +1 for tightening/introduction of a CFM on inflows, -1 for loosening/removal in quarter t.
      - Outflow_t CFM Δ take value +1 for tightening/introduction of a CFM on outflows, -1 for loosening/removal in quarter t.
      - t MPM Δ take value +1 for tightening/introduction of a MPM, -1 for loosening/removal in quarter t.
    - AddPolicy defined as:
      - Inflow_t CFM Δ − Outflow_t CFM Δ + t MPM Δ  (equation (3) as given)
- Empirical implementation:
  - Interact the shock (Shock) with AddPolicy in each of the 3 equations in (2) and include AddPolicy as a regressor on its own to estimate ,ˆsp αβ· (the mitigating or amplifying impact on a future policy response, at percentile α, of implementing additional unconventional policy actions in the quarter the shock hits).
  - Note: Extensions could examine MPMs and CFMs separately or include other complementary policies (fiscal, macro-structural) in future work.

### III. Country-specific policy responses — Case study: Peru
- Context:
  - Peru adopted inflation targeting in 2002 but frequently intervened to smooth FX market volatility due to financial dollarization and currency mismatch vulnerabilities.
- Historical intervention episodes and magnitudes:
  - 2002–2007: net purchases of U.S. dollars totaling $20 bln or 27 percent of GDP (precautionary reserve accumulation).
  - June 2008–February 2009: $7.1 billion sold (about 6 percent of 2009 GDP) to avoid negative balance sheet effects.
  - 2013–2016: sol depreciated as much as 35 percent amid tapering, China slowdown, and commodity price declines.
  - Tapering period net FX sales of about $4.2 bln and $8.1 bln in 2014 and 2015, respectively.
- Conditional forecast distributions and “at-risk” monitoring:
  - Method: fit t-skewed distribution to predicted values from country-specific quantile regressions to compute conditional forecast distributions of FX intervention (FXI), policy rate changes, and exchange rate changes.
  - Reporting: for each quarter, actual average policy changes over the next two quarters (black solid line) and predicted downside tail risks at the 10th percentile (red dashed line) and upside tail risks at the 90th percentile (green dashed line) are estimated from in-sample predictive densities over Q1 2000-Q3 2018.
- Key findings from Peru case:
  - Predictive performance:
    - Realized policy changes generally fall within the predictive distribution’s 10th–90th percentile range for FXI, policy rate, and exchange rate changes; notable exceptions include:
      - Large FX purchase in 2007;
      - Extreme monetary loosening in 2009;
      - Strong depreciation pressures on the sol in mid-2018.
  - Tail-risk dynamics 2003–2007:
    - Likelihood of extreme FX purchases increased: in 2007, framework predicted a 10 percent chance that FX purchases over the following two quarters would average at least 3 percent of GDP compared to a forecast of at least 1 percent of GDP in 2003.
    - No obvious shift in likelihood of extreme exchange rate appreciations or policy rate changes over the same period (consistent “FXI-only” response to inflows).
  - Tail-risk dynamics 2011–2015:
    - During adverse external shocks, higher chances of extreme FX sales and slightly higher chances of extreme monetary policy loosening and extreme exchange rate depreciation — indicating asymmetry of policy actions over the cycle.
  - Short-term shifts between Q3 2016 and Q3 2018 (U.S. Fed tightening episode):
    - FXI:
      - Central estimate of expected intervention over next 2 quarters fell from a conditional mean of 0.5 percent of GDP to a conditional mean of -0.1 percent of GDP.
      - Probability of short-term FX sales “soared” between Q3 2016 and Q3 2018 from 21 to 46 percent of GDP (with a 5 percent chance that Peru’s FX sales over the next 2 quarters would be at least 1.6 percent of GDP as of Q3 2018, rather than by just 1.1 percent of GDP, as predicted 4 quarters earlier).
    - Policy rate changes:
      - Conditional forecast distribution virtually unchanged between Q3 2016 and Q3 2018 (future monetary policy stance not expected to change given prevailing conditions).
    - Exchange rate responses:
      - Short-term central outlook improved between Q3 2016 and Q3 2018: conditional mean moved from -0.38 percent (depreciation) to +0.13 percent (appreciation).
      - Probability of further depreciation receded from 70 to 55 percent.
      - Downside risks remain: 5 percent chance that the sol would depreciate over the next 2 quarters by at least 3.2 percent as of Q3 2018.

*Source: IMF staff estimates and text from the provided Annex II.A / Annex I content.*

### 2.4 percent, as predicted 8 quarters earlier.

### wpiea2020010-print-pdf - 2.4 percent, as predicted 8 quarters earlier.

### Scenario analysis: adverse global portfolio flow shock and Peru’s short-term policy responses
- Shock simulated: fall of two times their standard deviation over the sample (approximately 0.46 percent of emerging markets’ trend GDP).
- Method: compare two-quarter-ahead forecast distributions of FX intervention (FXI), policy rate changes, and exchange rate changes under shock vs. 2018Q3 baseline.
- Key quantitative outcomes:
  - Probability of FX sales over the next two quarters rises from 46 to 80 percent under the outflow shock (alternative: green line; baseline: red line).
  - Downside (tail) risks to FX intervention increase; red shaded areas indicate 5th percentile downside risks in shocked scenarios.
  - Conditional distribution of interest rate changes remains virtually unchanged under the outflow shock.
  - Exchange rate is expected to depreciate slightly under the outflow shock.
- Asymmetry of responses:
  - A same-size increase in global portfolio inflows (0.46 percent of emerging markets’ trend GDP) produces smaller FXI and exchange rate responses than the symmetric outflow shock.
  - No interest rate response is predicted in either the inflow or outflow shock cases.

### Role of additional policies (MPMs and CFMs) in mitigating inflow shocks
- Approach: decompose effects into (i) stand-alone effect from additional policies and (ii) interaction between additional policies and the external shock; analyze effects across percentiles α.
- Main findings for Peru:
  - Use of additional policy measures is associated with lower extreme FX sales (i.e., reductions in magnitude at the lowest percentiles of the predicted conditional distribution for FXI) in the near future.
  - The total mitigating impact vis-à-vis extreme FX purchases appears to be negligible.
  - Use of additional policy measures can make interest rate hikes more likely following an exogenous inflow shock of foreign capital.
  - Additional policy measures appear to mitigate impacts of external shocks on extreme exchange rate appreciation and depreciation in the near future, although this mitigation is not found to be statistically significant.
- Visualization notes:
  - Figure 9 presents standardized quantile regression coefficients for FXI, policy rate change, and exchange rate change: impacts of shock variable, additional policy variable, and their interaction are shown with one-standard deviation confidence bands.

### Cross-country evidence: heterogeneity, skewness, and fat tails in policy responses
- Framework: estimate each country’s predicted two-quarter-ahead responses to an exogenous, same-sized global capital flow shock and correlate those with country characteristics (slow-moving or infrequently measured).
- Definitions and metrics:
  - Median response: average 2-quarter ahead median (50th percentile) response, denoted 50.,sβ.
  - Skewness: difference between average 2-quarter ahead 90th and 10th percentile responses, denoted 9010,.,.ssββ−.
  - Fat tails: differences between 10th–50th (left tail) and 90th–50th (right tail) percentile responses relative to the median.
- Key descriptive findings:
  - Responses to capital flow shocks are highly heterogeneous across countries even after controlling for cyclical conditions and a same-sized shock.
  - Example magnitude: for FX intervention, the average 2-quarter ahead response to a portfolio inflow of 1 percentage point of GDP varies between purchases of FX of 2 percentage points of GDP to sales of FX of -1 percentage points of GDP.
  - Skewness patterns:
    - FXI median skewness close to zero, but several countries exhibit strongly negative skew (large left tails) for FX intervention.
    - Exchange rate and interest rate responses: median countries show negative skew; several countries have very large left-tail responses.
    - Extreme responses are larger when selling FX, when the exchange rate is depreciating, and when policy rates are being cut.
  - Fat tails:
    - For FXI, both left and right tail responses exceed the median in magnitude; the left tail is particularly large.
    - For exchange rates and interest rates, the left tail is generally thicker than the right tail.
  - Country examples (country-level detail):
    - Peru responds the most in terms of FX intervention, while its exchange rate barely moves.
    - South Africa exhibits large exchange rate movements without much intervention.
    - Turkey adjusts policy rates more vigorously.

### Linking policy responses to country characteristics (conceptual approach and candidate determinants)
- Objective: explore correlations between cross-country differences in predicted policy responses and country characteristics that are slow-moving or infrequently measured.
- Candidate determinants considered include:
  - Proxies for balance sheet vulnerabilities and buffers:
    - size of domestic banks’ net open FX position;
    - ratio of short-term debt in foreign currency to forex reserves;
    - structure of external assets and liabilities;
    - level of reserve adequacy.
  - Degree of trade and capital account openness.
  - Indicators of financial market depth:
    - index of overall financial market development;
    - degree of turnover in the local currency FX market;
    - ask-bid spread.
  - Structural indicators:
    - exchange rate regime;
    - share of trade invoiced in USD;
    - degree of Central Bank transparency.
- Methodological note:
  - Because these characteristics are slow-moving, they cannot be controlled for in equation (2) (they would be subsumed in the country-specific constant). The analysis therefore focuses on raw correlations between country characteristics and predicted responses without presuming causality.

*Source: Haver, Bloomberg LP, and IMF Staff Estimates (contents excerpted from the provided PDF).*

### Annex II.A., with corresponding sources.

### Annex II.A. Summary of Findings on Policy Responses and Country Characteristics

### Overview and Table 1 notes
- Table 1 summarizes findings across all characteristics for policy rate responses, forex intervention (FXI) and exchange rate responses.
- Many empty cells indicate no statistically significant association between corresponding structural characteristics and policy responses.
- Regressions analyze two-quarter ahead policy responses on country characteristics.
- Statistical significance: relationships one-sided significant at the 10 percent significance level are highlighted; stars denote significance at * 10%; ** 5%; *** 1% significance level.
- Relevant t-statistic computed for 18 degrees of freedom (20 countries – 2 estimated coefficients).
- Percentiles used to define responses: 10th percentile (“Major Sales”), 50th percentile (“Normal Times”), 90th percentile (“Major Purchases”).

### Monetary Policy Rate Responses
- Main finding:
  - Monetary policy rate responses are mostly unrelated to country characteristics, reflecting similar monetary policy frameworks that respond primarily to domestic cyclical conditions.
- Notable exception:
  - The extent to which a currency is overvalued is associated with looser monetary policy in “typical” and “extreme” loosening responses.
  - Interpretation: countries may be wary of using monetary policy if it risks pushing their exchange rate further out of equilibrium.
- Interaction with additional policies:
  - In the presence of additional unconventional policy tools (e.g., MPMs or CFMs), interest rates tend to be set higher under inflow pressures and lower under outflow pressures.
  - Additional policies make monetary policy actions more sensitive to expected inflation developments: monetary policy is more likely to tighten in the subsequent half-year if inflation is expected to rise in the same period.
  - Evidence holds for both median (“typical”) and extreme (10th/90th percentile) responses.

### FX Intervention (FXI) and Balance Sheet / Market Characteristics
- FX intervention correlates with balance sheet vulnerabilities, buffers, and average exchange rate overvaluation:
  - Longer net open FX positions (or shorter local currency positions) and larger short-term debt in foreign currency relative to reserves:
    - Associated with larger forex reserve sales in the face of outflow pressures.
    - Associated with smaller intervention in the face of inflow pressures.
  - Larger forex reserve sales tend to be associated with lower foreign assets as a share of GDP.
    - Possible interpretation: FX intervention may substitute for a lack of a domestic investor base able to repatriate capital during non-resident outflows.
  - Reserves are positively correlated with FX intervention in both directions:
    - Correlation with purchases may reflect reverse causality (buy more → more reserves).
    - Correlation with sales may reflect that interventionist countries hold higher reserve stocks and engage in two-sided intervention, or that a precautionary floor on reserves limits sales.
- Causality caveat:
  - Statistical correlations do not imply causality. Interventionist policies may have contributed to persistent shallowness in FX markets; balance sheet mismatches may have been more persistent absent exchange rate flexibility.

### Exchange Rate Responses
- Overall relationship between exchange rate responses and country characteristics is weaker than for FXI.
  - Implies no systematically different exchange rate responses between interveners and non-interveners for most characteristics.
- Two characteristics associated with different exchange rate responses:
  - Net International Investment Position (NIIP): countries with substantial foreign assets relative to foreign liabilities witness less exchange rate movement than peers.
    - Possible explanations: resident portfolio flows offset non-resident flows; marginal investor perceives such countries as less risky, demanding lower risk premia during shocks.
  - Degree of de facto exchange rate flexibility: more flexible currencies show higher appreciation/depreciation for the same shock (tautological if measured correctly).
- Responses are fat-tailed and asymmetric:
  - For most countries, “extreme” responses are larger than median responses.
  - Responses to outflows tend to be larger than responses to inflows for exchange rate adjustment, interest rate changes and forex intervention.

### The Role of Additional Policy Tools (MPMs, CFMs)
- Scope and caveats:
  - Only a subset of the 20-country sample implemented these additional policies.
  - Analysis focuses on statistically significant country-specific interactions of these tools with the exogenous shock—conclusions driven by the few countries using these tools effectively.
  - Potential longer-term unintended consequences of unconventional tools are acknowledged but not quantified here.

A. Additional Policy Tools and Monetary Policy
- Key findings:
  - Additional unconventional policy tools allow monetary policy to more squarely focus on domestic inflation stabilization.
  - In the presence of inflow pressures, interest rates are set higher when additional policies are deployed significantly; under outflow pressures, interest rates are set lower.
  - Monetary policy actions become more sensitive to expected inflation when countries use additional unconventional policies:
    - The slope of cross-country fitted line of future inflation responses on future monetary policy actions becomes more positively sloped when additional tools are used.
  - This finding holds for both “typical” and “extreme” responses to inflows and outflows.
- Interpretation:
  - MPMs/CFMs may help break the dilemma that tighter monetary policy in the face of inflows can attract further inflows; they can enable monetary policy to prioritize inflation objectives.

B. Asymmetric Impact of Additional Policies (FXI and Exchange Rate)
- Key findings:
  - Additional tools seem to have asymmetric impacts: “extreme” FX sales or exchange rate depreciations are more affected than “extreme” FX purchases or exchange rate appreciations.
  - When facing outflows, FX sales are larger when additional policy measures are active; the opposite pattern for inflows is not clearly discernible.
  - Exchange rates tend to depreciate more in outflow episodes when auxiliary tools are deployed; no clear pattern for inflows.
  - The asymmetry of FXI is driven by Malaysia’s responses; the asymmetry of exchange rate responses is broader across countries.
- Interpretation:
  - Deployment of additional policies may reflect a few countries’ concern about states of the world where “extreme” sales and depreciations are the natural responses.

C. Effects of Additional Policies Over Time (2-quarter vs 2-year horizons)
- Key findings:
  - Monetary policy independence and asymmetric use of additional policies remain when looking at average effects past the first 2 quarters and over 2 years.
  - Country-specific flows:
    - At the 2-quarter ahead horizon, auxiliary policy tools barely affect non-resident capital inflows.
    - At the 2-year ahead horizon, these tools appear much more effective at moderating both extreme outflows and extreme inflows.
  - Estimation replaces dependent variable in equation (2) with non-resident capital inflows to study horizon effects.
- Interpretation:
  - Additional unconventional tools may improve policy trade-offs in the short term and affect quantities (flows themselves) over the medium term.

### Conclusions and Policy Implications
- Cross-country heterogeneity:
  - Policy responses to global capital flows are highly heterogeneous across 20 EMs over almost two decades, even for the same-sized exogenous shock.
- Stylized empirical facts:
  - Policy responses are fat-tailed and asymmetric, with larger reactions to outflows than inflows.
  - FXI responses (typical and extreme) are positively related to balance sheet vulnerabilities and inversely related to forex market depth.
- Role of complementary tools:
  - CFMs and MPMs targeting external balance sheet vulnerabilities can enhance monetary policy autonomy: bringing these instruments into play strengthens sensitivity of both “typical” and “extreme” monetary policy responses to expected domestic cyclical developments.
- Non-normative stance and caveats:
  - The analysis is descriptive and not normative; it does not prescribe alternative policies nor quantify cost-benefit trade-offs.
  - The apparent short-term monetary policy space created by unconventional tools may hide longer-term unintended consequences or intertemporal trade-offs.
  - Additional research is needed to assess longer-term effects and potential welfare trade-offs.

*Source: IMF Staff Estimates, Annex II.A., wpiea2020010-print-pdf - Annex II.A., with corresponding sources.*

### ANNEX I – TABLES

### ANNEX I – TABLES

### Table 1. Quantile Regressions Estimates—2-Quarters Ahead Predictions
- Note: Bold font indicates significance at 10 percent. The additional policy variable AddPolicy is also included as a stand-alone control in each of the regressions. Comprehensive results and goodness of fit statistics are available upon request.
- Quantiles (columns): 0.1, 0.5, 0.9 for Exchange Rate; 0.1, 0.5, 0.9 for Interest Rate; 0.1, 0.5, 0.9 for FXI.

- Argentina
  - Exchange Rate (Shock): 0.17, -2.17, -1.23
  - Interest Rate (Shock): -0.03, -0.44, -5.53
  - FXI (Shock): 2.13, 0.30, 0.25
  - Exchange Rate (Shock*AddPolicy): -0.10, -3.51, 7.46
  - Interest Rate (Shock*AddPolicy): -0.45, -0.14, 2.92
  - FXI (Shock*AddPolicy): 0.76, 1.23, 1.92

- Brazil
  - Exchange Rate (Shock): -4.64, -7.16, -10.41
  - Interest Rate (Shock): 0.41, -0.92, 0.00
  - FXI (Shock): 0.32, 1.14, 1.27
  - Exchange Rate (Shock*AddPolicy): -1.44, 3.69, 8.85
  - Interest Rate (Shock*AddPolicy): 0.12, -0.52, 0.37
  - FXI (Shock*AddPolicy): -0.83, -0.82, -1.14

- Chile
  - Exchange Rate (Shock): -1.38, -1.45, -4.07
  - Interest Rate (Shock): 1.82, 0.73, 0.41
  - FXI (Shock): -0.26, -0.11, -1.11
  - Exchange Rate (Shock*AddPolicy): 0.00, 0.00, 0.00
  - Interest Rate (Shock*AddPolicy): 0.00, 0.00, 0.00
  - FXI (Shock*AddPolicy): 0.00, 0.00, 0.00

- China
  - Exchange Rate (Shock): 0.72, -0.13, 0.79
  - Interest Rate (Shock): 0.37, 0.08, 0.12
  - FXI (Shock): 0.19, 0.82, 1.66
  - Exchange Rate (Shock*AddPolicy): 3.46, 0.80, 4.36
  - Interest Rate (Shock*AddPolicy): 0.60, 0.06, -0.06
  - FXI (Shock*AddPolicy): 1.62, -0.21, -0.81

- Colombia
  - Exchange Rate (Shock): -5.42, -2.66, -6.33
  - Interest Rate (Shock): 1.21, 0.20, -0.35
  - FXI (Shock): 0.20, 0.38, 0.56
  - Exchange Rate (Shock*AddPolicy): 0.00, 0.00, 0.00
  - Interest Rate (Shock*AddPolicy): 0.00, 0.00, 0.00
  - FXI (Shock*AddPolicy): 0.00, 0.00, 0.00

- Hungary
  - Exchange Rate (Shock): -0.97, -0.41, -2.51
  - Interest Rate (Shock): 0.44, -0.19, -0.07
  - FXI (Shock): -0.15, -0.34, 0.00
  - Exchange Rate (Shock*AddPolicy): -3.26, 3.34, 8.15
  - Interest Rate (Shock*AddPolicy): -2.53, -0.74, 0.36
  - FXI (Shock*AddPolicy): -1.24, 1.34, 4.35

- Indonesia
  - Exchange Rate (Shock): -1.88, 0.26, -0.59
  - Interest Rate (Shock): 0.88, 0.46, 0.76
  - FXI (Shock): -0.07, 0.30, 1.11
  - Exchange Rate (Shock*AddPolicy): -1.00, -2.82, -0.25
  - Interest Rate (Shock*AddPolicy): -0.11, 0.10, -0.23
  - FXI (Shock*AddPolicy): 0.52, 0.34, -0.03

- India
  - Exchange Rate (Shock): -1.99, -0.57, -0.77
  - Interest Rate (Shock): 0.81, 0.35, 0.41
  - FXI (Shock): 0.52, 0.13, 0.37
  - Exchange Rate (Shock*AddPolicy): -0.37, -1.81, -3.39
  - Interest Rate (Shock*AddPolicy): 0.63, 0.06, 0.15
  - FXI (Shock*AddPolicy): 0.17, 0.62, 1.54

- Kazakstan
  - Exchange Rate (Shock): 7.13, 2.90, 7.00
  - Interest Rate (Shock): 0.41, 0.32, -1.50
  - FXI (Shock): -0.43, 0.39, 1.23
  - Exchange Rate (Shock*AddPolicy): 0.00, 0.00, 0.00
  - Interest Rate (Shock*AddPolicy): 0.00, 0.00, 0.00
  - FXI (Shock*AddPolicy): 0.00, 0.00, 0.00

- Korea
  - Exchange Rate (Shock): 0.10, -2.51, -1.25
  - Interest Rate (Shock): 0.54, 0.45, 0.31
  - FXI (Shock): 0.62, 0.50, 0.63
  - Exchange Rate (Shock*AddPolicy): -7.87, -1.42, -2.07
  - Interest Rate (Shock*AddPolicy): -0.39, -0.18, 0.12
  - FXI (Shock*AddPolicy): 0.41, -0.09, 2.64

- Mexico
  - Exchange Rate (Shock): -0.84, -4.82, -5.75
  - Interest Rate (Shock): 0.88, -0.01, -0.42
  - FXI (Shock): 0.73, 0.62, 0.63
  - Exchange Rate (Shock*AddPolicy): 0.00, 0.00, 0.00
  - Interest Rate (Shock*AddPolicy): 0.00, 0.00, 0.00
  - FXI (Shock*AddPolicy): 0.00, 0.00, 0.00

- Malaysia
  - Exchange Rate (Shock): -0.61, -2.06, -2.16
  - Interest Rate (Shock): 0.53, 0.21, 0.24
  - FXI (Shock): 2.15, 1.79, 3.43
  - Exchange Rate (Shock*AddPolicy): 5.26, 2.71, -0.18
  - Interest Rate (Shock*AddPolicy): 0.13, -0.10, -0.55
  - FXI (Shock*AddPolicy): 7.23, -0.59, -7.33

- Philippines
  - Exchange Rate (Shock): -2.18, -2.41, -1.75
  - Interest Rate (Shock): 0.39, 0.13, -0.16
  - FXI (Shock): 2.82, 1.68, 2.54
  - Exchange Rate (Shock*AddPolicy): 0.00, 0.00, 0.00
  - Interest Rate (Shock*AddPolicy): 0.00, 0.00, 0.00
  - FXI (Shock*AddPolicy): 0.00, 0.00, 0.00

- Peru
  - Exchange Rate (Shock): 0.88, 0.48, 1.07
  - Interest Rate (Shock): 1.23, 0.07, 0.22
  - FXI (Shock): 1.73, 1.89, 2.18
  - Exchange Rate (Shock*AddPolicy): 0.49, -0.09, 0.30
  - Interest Rate (Shock*AddPolicy): -0.16, 0.49, 0.64
  - FXI (Shock*AddPolicy): 0.33, 0.03, 0.23

- Poland
  - Exchange Rate (Shock): -0.96, -3.21, -4.08
  - Interest Rate (Shock): 0.89, 0.27, 0.26
  - FXI (Shock): 0.89, 0.80, 0.23
  - Exchange Rate (Shock*AddPolicy): 3.45, 4.32, 0.01
  - Interest Rate (Shock*AddPolicy): 0.31, -0.24, -0.15
  - FXI (Shock*AddPolicy): 0.70, 0.90, -0.53

- Russia
  - Exchange Rate (Shock): -0.31, -2.70, -5.50
  - Interest Rate (Shock): -0.14, 0.08, 1.16
  - FXI (Shock): 1.96, 0.35, 0.90
  - Exchange Rate (Shock*AddPolicy): -2.46, -2.38, 0.31
  - Interest Rate (Shock*AddPolicy): 1.29, 1.24, 2.51
  - FXI (Shock*AddPolicy): -1.35, -3.53, -1.44

- Thailand
  - Exchange Rate (Shock): -1.14, -0.94, 0.02
  - Interest Rate (Shock): 0.68, 0.30, 0.57
  - FXI (Shock): 1.41, 0.68, 0.42
  - Exchange Rate (Shock*AddPolicy): 8.69, 11.00, 14.74
  - Interest Rate (Shock*AddPolicy): 5.21, 5.38, 6.89
  - FXI (Shock*AddPolicy): 5.79, 3.05, -0.77

- Turkey
  - Exchange Rate (Shock): -2.63, -3.31, -6.36
  - Interest Rate (Shock): 2.91, 1.38, -0.33
  - FXI (Shock): -0.01, 0.60, 0.33
  - Exchange Rate (Shock*AddPolicy): -6.06, -3.87, 4.10
  - Interest Rate (Shock*AddPolicy): -4.35, 0.71, 4.65
  - FXI (Shock*AddPolicy): -0.28, 0.03, 1.26

- Ukraine
  - Exchange Rate (Shock): 1.11, -0.77, -7.62
  - Interest Rate (Shock): 0.61, 0.69, 1.75
  - FXI (Shock): 2.52, -1.15, 0.52
  - Exchange Rate (Shock*AddPolicy): 0.23, 0.08, 1.66
  - Interest Rate (Shock*AddPolicy): -0.57, 1.07, 0.37
  - FXI (Shock*AddPolicy): 0.71, -0.38, -0.84

- South Africa
  - Exchange Rate (Shock): 3.51, -4.68, -4.26
  - Interest Rate (Shock): 0.91, -0.47, -0.19
  - FXI (Shock): 0.11, 0.25, 0.71
  - Exchange Rate (Shock*AddPolicy): 0.00, 0.00, 0.00
  - Interest Rate (Shock*AddPolicy): 0.00, 0.00, 0.00
  - FXI (Shock*AddPolicy): 0.00, 0.00, 0.00

### Table 2. Quantile Regressions Estimates—2-Years Ahead Predictions
- Note: Bold font indicates significance at 10 percent. The additional policy variable AddPolicy is also included as a stand-alone control in each of the regressions. Comprehensive results and goodness of fit statistics are available upon request.
- Quantiles (columns): 0.1, 0.5, 0.9 for Exchange Rate; 0.1, 0.5, 0.9 for Interest Rate; 0.1, 0.5, 0.9 for FXI.

- Argentina
  - Exchange Rate (Shock): -2.35, -0.60, -7.51
  - Interest Rate (Shock): -0.03, -0.44, -5.53
  - FXI (Shock): 0.92, 0.00, 0.53
  - Exchange Rate (Shock*AddPolicy): -1.76, -2.52, 2.51
  - Interest Rate (Shock*AddPolicy): -0.45, -0.14, 2.92
  - FXI (Shock*AddPolicy): 1.48, 0.77, 1.75

- Brazil
  - Exchange Rate (Shock): 1.02, -1.12, -5.57
  - Interest Rate (Shock): 0.41, -0.92, 0.00
  - FXI (Shock): -0.14, 1.14, 0.23
  - Exchange Rate (Shock*AddPolicy): -1.69, 2.19, 4.22
  - Interest Rate (Shock*AddPolicy): 0.12, -0.52, 0.37
  - FXI (Shock*AddPolicy): 0.27, -0.96, 0.33

- Chile
  - Exchange Rate (Shock): 0.35, -0.02, -0.71
  - Interest Rate (Shock): 1.82, 0.73, 0.41
  - FXI (Shock): 0.28, -0.01, 0.46
  - Exchange Rate (Shock*AddPolicy): 0.00, 0.00, 0.00
  - Interest Rate (Shock*AddPolicy): 0.00, 0.00, 0.00
  - FXI (Shock*AddPolicy): 0.00, 0.00, 0.00

- China
  - Exchange Rate (Shock): -0.87, -0.30, 0.05
  - Interest Rate (Shock): 0.37, 0.08, 0.12
  - FXI (Shock): 0.03, 0.11, -0.05
  - Exchange Rate (Shock*AddPolicy): -0.55, 0.10, 7.76
  - Interest Rate (Shock*AddPolicy): 0.60, 0.06, -0.06
  - FXI (Shock*AddPolicy): 1.61, 0.05, -0.43

- Colombia
  - Exchange Rate (Shock): -1.25, -0.98, -4.93
  - Interest Rate (Shock): 1.21, 0.20, -0.35
  - FXI (Shock): 0.09, 0.22, 0.21
  - Exchange Rate (Shock*AddPolicy): 0.00, 0.00, 0.00
  - Interest Rate (Shock*AddPolicy): 0.00, 0.00, 0.00
  - FXI (Shock*AddPolicy): 0.00, 0.00, 0.00

- Hungary
  - Exchange Rate (Shock): 1.88, 0.40, -0.55
  - Interest Rate (Shock): 0.44, -0.19, -0.07
  - FXI (Shock): -0.58, -0.28, -0.17
  - Exchange Rate (Shock*AddPolicy): -1.19, 1.01, 3.40
  - Interest Rate (Shock*AddPolicy): -2.53, -0.74, 0.36
  - FXI (Shock*AddPolicy): -0.92, 0.32, 1.77

- Indonesia
  - Exchange Rate (Shock): -0.31, 0.80, 1.85
  - Interest Rate (Shock): 0.88, 0.46, 0.76
  - FXI (Shock): 0.11, 0.05, 0.28
  - Exchange Rate (Shock*AddPolicy): 0.01, -0.83, -0.97
  - Interest Rate (Shock*AddPolicy): -0.11, 0.10, -0.23
  - FXI (Shock*AddPolicy): -0.21, -0.25, -0.28

- India
  - Exchange Rate (Shock): -0.31, 0.67, 1.19
  - Interest Rate (Shock): 0.81, 0.35, 0.41
  - FXI (Shock): -0.13, -0.21, -0.49
  - Exchange Rate (Shock*AddPolicy): 0.17, 0.41, 0.16
  - Interest Rate (Shock*AddPolicy): 0.63, 0.06, 0.15
  - FXI (Shock*AddPolicy): 0.55, -0.15, -0.38

- Kazakstan
  - Exchange Rate (Shock): 0.17, 1.65, 5.07
  - Interest Rate (Shock): 0.41, 0.32, -1.50
  - FXI (Shock): -0.07, -0.70, -0.05
  - Exchange Rate (Shock*AddPolicy): 0.00, 0.00, 0.00
  - Interest Rate (Shock*AddPolicy): 0.00, 0.00, 0.00
  - FXI (Shock*AddPolicy): 0.00, 0.00, 0.00

- Korea
  - Exchange Rate (Shock): -0.20, -0.30, 2.30
  - Interest Rate (Shock): 0.54, 0.45, 0.31
  - FXI (Shock): -0.70, -0.43, -1.32
  - Exchange Rate (Shock*AddPolicy): -3.30, 0.37, -3.89
  - Interest Rate (Shock*AddPolicy): -0.39, -0.18, 0.12
  - FXI (Shock*AddPolicy): 0.59, 0.19, 1.83

- Mexico
  - Exchange Rate (Shock): 0.18, -0.73, -0.71
  - Interest Rate (Shock): 0.88, -0.01, -0.42
  - FXI (Shock): 0.05, 0.14, 0.25
  - Exchange Rate (Shock*AddPolicy): 0.00, 0.00, 0.00
  - Interest Rate (Shock*AddPolicy): 0.00, 0.00, 0.00
  - FXI (Shock*AddPolicy): 0.00, 0.00, 0.00

- Malaysia
  - Exchange Rate (Shock): -0.31, -0.28, -1.22
  - Interest Rate (Shock): 0.53, 0.21, 0.24
  - FXI (Shock): 2.16, 1.67, -0.03
  - Exchange Rate (Shock*AddPolicy): 2.54, 0.68, -0.91
  - Interest Rate (Shock*AddPolicy): 0.13, -0.10, -0.55
  - FXI (Shock*AddPolicy): -0.29, -3.31, -8.33

- Philippines
  - Exchange Rate (Shock): -1.02, -1.83, -2.19
  - Interest Rate (Shock): 0.39, 0.13, -0.16
  - FXI (Shock): 0.96, 0.87, 0.56
  - Exchange Rate (Shock*AddPolicy): 0.00, 0.00, 0.00
  - Interest Rate (Shock*AddPolicy): 0.00, 0.00, 0.00
  - FXI (Shock*AddPolicy): 0.00, 0.00, 0.00

- Peru
  - Exchange Rate (Shock): -0.09, 0.10, 0.20
  - Interest Rate (Shock): 1.23, 0.07, 0.22
  - FXI (Shock): 1.37, 0.93, 1.00
  - Exchange Rate (Shock*AddPolicy): 0.02, -0.24, -0.45
  - Interest Rate (Shock*AddPolicy): -0.16, 0.49, 0.64
  - FXI (Shock*AddPolicy): 2.31, -0.01, 0.31

- Poland
  - Exchange Rate (Shock): -1.55, -0.51, -1.50
  - Interest Rate (Shock): 0.89, 0.27, 0.26
  - FXI (Shock): 0.05, 0.12, 0.08
  - Exchange Rate (Shock*AddPolicy): 1.02, 2.04, 1.71
  - Interest Rate (Shock*AddPolicy): 0.31, -0.24, -0.15
  - FXI (Shock*AddPolicy): -0.21, 0.05, 0.20

- Russia
  - Exchange Rate (Shock): 0.70, 0.15, -1.27
  - Interest Rate (Shock): -0.14, 0.08, 1.16
  - FXI (Shock): 0.93, -0.54, -0.37
  - Exchange Rate (Shock*AddPolicy): 1.00, 2.70, -4.55
  - Interest Rate (Shock*AddPolicy): 1.29, 1.24, 2.51
  - FXI (Shock*AddPolicy): 0.56, -2.71, -0.36

- Thailand
  - Exchange Rate (Shock): -0.01, -0.06, 0.44
  - Interest Rate (Shock): 0.68, 0.30, 0.57
  - FXI (Shock): 0.18, 0.24, -0.57
  - Exchange Rate (Shock*AddPolicy): -1.51, 1.86, 3.88
  - Interest Rate (Shock*AddPolicy): 2.21, 5.38, 6.89
  - FXI (Shock*AddPolicy): -2.65, -2.67, -1.17

- Turkey
  - Exchange Rate (Shock): 0.11, -0.72, 2.31
  - Interest Rate (Shock): 2.91, 1.38, -0.33
  - FXI (Shock): -0.37, 0.06, 0.41
  - Exchange Rate (Shock*AddPolicy): -3.26, 0.21, 5.62
  - Interest Rate (Shock*AddPolicy): -4.35, 0.71, 4.65
  - FXI (Shock*AddPolicy): -0.65, -0.26, 0.54

- Ukraine
  - Exchange Rate (Shock): 0.74, 1.73, -0.32
  - Interest Rate (Shock): 0.61, 0.69, 1.75
  - FXI (Shock): 0.30, -1.78, -1.25
  - Exchange Rate (Shock*AddPolicy): -2.49, -1.87, -0.13
  - Interest Rate (Shock*AddPolicy): -0.57, 1.07, 0.37
  - FXI (Shock*AddPolicy): 0.37, 0.55, -0.47

- South Africa
  - Exchange Rate (Shock): 4.18, 3.59, -0.38
  - Interest Rate (Shock): 0.91, -0.47, -0.19
  - FXI (Shock): -0.04, -0.06, -0.02
  - Exchange Rate (Shock*AddPolicy): 0.00, 0.00, 0.00
  - Interest Rate (Shock*AddPolicy): 0.00, 0.00, 0.00
  - FXI (Shock*AddPolicy): 0.00, 0.00, 0.00

### ANNEX II – DATA

### A. Data used in quantile local projections, equation (2)
- Countries: Argentina, Brazil, Chile, China, Colombia, Hungary, India, Indonesia, Kazakhstan, Korea, Malaysia, Mexico, Peru, Philippines, Poland, Russia, South Africa, Thailand, Turkey, Ukraine.
- Capital flow shock (Shock_t)
  - Focus: gross portfolio inflows to EMs (net non-resident purchases of EM assets).
  - Data: quarterly balance of payments data from 1995Q1 to 2018Q3 for 20 emerging markets, based on the IMF’s Financial Flow Analytics database.
  - Measurement: gross inflows in US dollars scaled by annual HP-filtered GDP sourced from the IMF’s World Economic Outlook dataset.
  - Shock construction: as described in sub-section II.B (in source).
- Policy responses
  - FX intervention: estimated from balance of payments reserve flow (IMF’s World Economic Outlook dataset) adjusted for income accrued to previous positions using IMF’s COFER database (currency and maturity composition) and long-term and short-term yields; normalized by annual HP-filtered GDP. Positive = increase in FX assets (FX purchase); negative = decline in FX assets (FX sale). Estimates track published FX intervention for countries that publish (see Annex III.B).
  - Policy interest rate: End-of-quarter domestic policy rates (nominal, percentage points) from Haver Analytics dataset.
  - Exchange rate: End-of-quarter effective nominal exchange rates from Haver Analytics dataset.
- Additional Policies
  - CFMs: IMF’s 2019 Taxonomy of Capital Flow Management Measures.
  - MPMs: IMF’s Macroprudential Policy Survey database (iMaPP) as described in Alam et al. (2019).
- Controls (Controls_{t−1})
  - Growth rate of real GDP: IMF’s World Economic Outlook database.
  - Ratio of short-term foreign currency debt to forex reserves: IMF's Assessing Reserve Adequacy (ARA) database.
  - International commodity price index: Haver Analytics dataset.
  - Additional Policies: as above.

### B. List of Country Characteristics (data sources and measures)
- Portfolio investment assets, liabilities and assets to liabilities, Net International Investment Position Accounts, World Economic Outlook dataset.
- Trade openness: sum of nominal import and export as a share of nominal domestic GDP, World Economic Outlook.
- Capital openness: Chinn-Ito Index, computed using IMF’s AREAER database.
- Financial Market development index: from "Rethinking Financial Deepening: Stability and Growth in Emerging Markets" SDN 2015 MCM/SPR (comprehensive index capturing FI and FM; depth, access, efficiency).
- FX market depth: average daily turnover in FX derivatives market (gross value of all new deals), based on Triennial Central Bank Survey of Foreign Exchange and OTC Derivatives Markets.
- Short-term debt in foreign currency as percent of forex reserves: forex reserves (USD) and external debt in foreign currency from IMF's ARA database and World Bank’s Quarterly External Debt Statistics database; forex reserves percent of ARA metric from IMF's ARA database.
- Net open position in foreign exchange to capital of domestic banking sector: Financial Soundness Indicators, Last Updated: September 2018, Core FSIs for Deposit Takers.
- REER overvaluation: estimated as a residual from fitting a trend through the data.
- De facto Exchange rate Flexibility: Annual Report on Exchange Arrangements and Exchange Restrictions (classification based on IMF staff analysis).
- USD invoicing: data reference in source to Gopinath (2016) "The International Price System."
- Normalizations: several variables normalized by nominal USD GDP (World Economic Outlook) or by reserves (IMF's ARA database).

### ANNEX III – ROBUSTNESS/CAVEATS

### A. Bilateral vs Effective Exchange Rates
- In the paper, responses of effective exchange rate changes were explored; this subsection discusses results when using bilateral exchange rate changes.
- Figure AIII.1 compares average two-quarter ahead extreme responses of bilateral vs effective exchange rates for the two tails (10th and 90th percentiles).
- Finding: The bilateral exchange rate vis-à-vis the USD tends to be more responsive in either of the two tails (extreme depreciations and extreme appreciations). This occurs because other major economies’ currencies offset some of the USD movements.

### B. FXI Measure
- The paper uses estimates of FX intervention rather than published intervention data because:
  - Many central banks in the sample do not publish FX operations.
  - Published data may be infrequent, too aggregated, or not go back far enough.
  - Definitions and coverage of published interventions are not necessarily consistent across countries.
- To ensure cross-country consistency, the analysis relies only on estimated FXI data rather than mixing published and estimated series.
- Method: estimates of FXI are based on BOP data after adjusting official foreign exchange reserve flows for interest receipts.
- Figure AIII.2 compares estimates with published data on forex intervention for countries that publish intervention data (Argentina, Brazil, Colombia, India, Mexico, Peru, Kazakhstan, Russia and Turkey).
  - Result: In most cases, estimated spot interventions track published FXI closely while non-spot interventions do not (typically not included in published numbers).
  - Exceptions: Mexico and Turkey show discrepancies likely arising from different definitions of FX interventions across published and estimated series.
  - Conclusion: The comparison provides confidence that the estimates are reasonably close to actuals.

*Source: ANNEX I – TABLES; ANNEX II – DATA; ANNEX III – ROBUSTNESS/CAVEATS (from the provided IMF content).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020010-print-pdf.pdf_