## wpiea2019279-print-pdf - Section V, we focus on two country cases to show how our methodology can be tailored to

## Source details

**Canonical URL:** [wpiea2019279-print-pdf - Section V, we focus on two country cases to show how our methodology can be tailored to](https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019279-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2019/wpiea2019279-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2019/wpiea2019279-print-pdf.pdf.json)

---

### A Risk Management Framework for Capital Flows — Key concepts
- Stylized representation of shocks and policy actions:
  - Initial predicted median capital flows: 2 percent of GDP.
  - Post-shock median falls to 1.5 percent of GDP in the example.
  - Policy action example raises median from 1.5 to 1.8 percent of GDP and reduces left-tail mass (probability of net outflows).
- Two quantifications of risk:
  - Probability that capital flows fall below a threshold (natural threshold: zero, distinguishing inflows from outflows).
  - Capital flows at risk (CaR): the amount of outflows reached or exceeded at a given probability (Adrian, Boyarchenko, and Giannone (2019) use the 5th percentile). Country examples use CaR at both the 5th and the 10th percentile.

### Econometric approach — Overview and specification
- Two-step estimation procedure:
  - Step 1: Estimate future flows with quantile regression across α = 0.05, 0.1, 0.15, ..., 0.95 to obtain predicted quantiles of average gross portfolio inflows (percent of GDP) over chosen horizons.
    - Short term: h=2, j=1 → average inflows in the first and second quarters ahead.
    - Medium term: h=8, j=3 → average quarterly portfolio inflows over quarters 5–8 ahead.
  - Step 2: Fit empirical distribution of predicted quantiles to a skewed-t probability distribution (Azzalini and Capitanio (2003)) characterized by mean, variance, skewness, and kurtosis using a minimum distance estimator and Azzalini (2019) algorithm.
- General quantile regression specification (percentile α):
  - ŷ_{i,t+h−j : t+h | t}^{α} = δ_{i}^{α} + β_{1}^{α} G_{t} + β_{2}^{α} DG_{i,t} + β_{3}^{α} P_{i,t} + β_{4}^{α} G_{t} × P_{i,t} + ε_{it}
  - Variables defined in model:
    - ŷ_{i,t+h−j : t+h | t}^{α}: average gross portfolio inflows (percent of GDP) to country i at percentile α between quarters t+h-j and t+h.
    - G_{t}: global “push” factors (U.S. corporate BBB spread or BBB yield, U.S. sovereign 10-year yield, DXY index). Some specs control for commodity prices and global growth.
    - DG_{i,t}: country-specific “pull” factors (e.g., domestic year-on-year GDP growth, short-term external debt to FX reserves).
    - P_{i,t}: domestic policy frameworks, structural characteristics, and policy actions (financial sector development, capital account openness, exchange rate regime, rule of law, central bank transparency; and policy actions: monetary policy, macroprudential policies, FX interventions (FXIs), capital flow management measures (CFMs)).
    - Interactions G_{t} × P_{i,t} used to assess mitigation of global shocks by domestic characteristics/policies.
- Endogeneity handling:
  - Policy actions estimated via country-by-country policy functions; residuals from first-stage regressions used as policy shocks.
  - CFMs and macroprudential shocks constructed as the sum of residuals in current quarter t and three lags (t-1, t-2, t-3) to capture infrequent, persistent actions.
- Estimation details:
  - Algorithmic method for unbalanced panels (Koenker and d’Orey (1987, 1994); Koenker (2005)).
  - Bootstrapped standard errors clustered at country level and corrected for serial correlation.

### Data, sample, and summary statistics
- Quarterly data period: 1996: Q4 to 2018: Q4.
- Country sample: 35 emerging market and developing countries (unbalanced panel).
- Key summary statistics (from Table 1 for subsample used in policy actions analysis):
  - Gross portfolio inflows: Min -24.112; Median 1.149; Max 36.411 (percent of GDP).
  - FX intervention (% of GDP): Min -12.8; Median 0.020; Max 148.5.
  - Macroprudential measures (index): tightening actions 10; easing actions 33 (per-country summary over sample period).
  - CFMs: CFMs introduced 1356; CFMs abolished 1356.
- Cross-country heterogeneity examples:
  - One standard deviation of portfolio flows: Malaysia ≈ 8 percent of GDP; Colombia ≈ 2 percent of GDP.
  - Policy usage: countries mostly tightened macroprudential tools; CFMs were tightened or abolished with similar frequency.
- Data sources noted:
  - Dependent variable: gross portfolio inflows as percent of GDP (IMF Financial Flow Analytics database).
  - Structural variables: World Bank WDI and WGI, Transparency International, IMF AREAER, IMF Financial Development Index, Chinn-Ito Index.
  - Policy variables: policy rates (IFS), FX interventions (central bank reports, FRED, COFER; valuation-adjusted proxies where needed), macroprudential actions (iMaPP database), CFMs (AREAER broad restrictiveness index).

### Aggregate flows to emerging markets — Main findings
- Simplified aggregate specification (excluding China) for portfolio debt inflows to EMs:
  - ŷ_{t+h−j : t+h | t}^{α} = δ^{α} + β_{1}^{α} G_{t} + β_{2}^{α} Gdp_{t,h} + y_{t−1} + ε_{t}
  - Global factors with predictive power: U.S. corporate BBB spread, U.S. sovereign 10-year yield, DXY index.
- Key empirical findings:
  - Higher U.S. interest rates and a stronger U.S. dollar are associated with weaker future inflows in both short and medium term.
  - Investor risk aversion (BBB spread) disproportionately explains large outflows and, to a lesser extent, surges: higher coefficient estimates at lowest and highest percentiles.
  - Coefficient on global risk appetite (BBB spread) exhibits sign reversal over horizons:
    - Median quantile coefficient negative and significant in the current quarter.
    - Turns positive and significant about 3–5 quarters into the future — consistent with mean-reversion in risk appetite.
  - Outlook change example:
    - Predicted medium-term distribution using information up to 2018: Q4 (for 2020: Q1–2020: Q4) shown in black.
    - Updated distribution using information as of 2019: Q2 (for 2020: Q3–2021: Q2) shown in green/red.
    - Outlook improved from 2018: Q4 → 2019: Q2: distribution shifted right and probability of outflows declined (red tail reduced), driven mainly by lower U.S. interest rates (10-year Treasury yield fell markedly over this period).

### Panel data analysis — Policy and structural interactions
- Parsimonious panel specification uses BBB yield as G_{t}:
  - ŷ_{i,t+h−j : t+h | t}^{α} = δ_{i}^{α} + β_{1}^{α} BBB_{t} + β_{2}^{α} DG_{i,t} + β_{3}^{α} Gcyc_{t} + β_{4}^{α} P_{i,t} + β_{5}^{α} BBB_{t} × P_{i,t} + β_{6}^{α} FI_{i,t} + β_{7}^{α} BBB_{t} × FI_{i,t} + ε_{it}
  - Gcyc_{t}: U.S. GDP growth (detrended, average over last four quarters).
  - FI_{i,t}: financial segmentation/integration indicator (available for 18 countries only).
- Theoretical expectations and empirical interpretations:
  - β_{1}^{α} expected negative (higher BBB yield → lower portfolio inflows, at least short term).
  - β_{5}^{α} positive if policy/characteristic mitigates negative impact of higher BBB yield.
  - Short-term external debt: may increase short-term inflows (financing needs) but reduce medium-term flows via sustainability concerns — important in low percentiles.
  - Financial market depth: associated with larger median short-term flows; ambiguous effect on tails.
  - Capital account openness: encourages larger inflows but could lead to larger outflows during risk-off.
  - Financial integration: greater integration could either worsen outflows (easier exit) or reduce withdrawal incentives (investor familiarity); interaction with BBB capture moderation.

### Policy trade-offs, signaling, and empirical horizons
- Policy trade-offs:
  - Introduction of outflow restrictions might prevent capital from flowing out, but might also reduce new inflows when the tide turns.
  - A CFM designed to stem inflows may reduce surges and reduce vulnerability to large outflows by preventing buildup of domestic imbalances.
  - Introduction of CFMs might have negative signaling effects and cause larger foreign capital withdrawals, especially if not comprehensive or unenforceable.
- Empirical horizons considered:
  - Short-term: average quarterly portfolio inflows over the next two quarters as dependent variable.
  - Medium-term: average quarterly portfolio inflows over quarters 5–8 ahead.

### Key empirical findings: global shocks, structural characteristics, and portfolio flows (exact quantitative examples)
- Financial market depth and tail outcomes:
  - Comparing countries with shallow financial markets to countries with more developed markets: conditional lower-tail capital outflows increase from 0.3 to 1.8 percent of GDP, and the 95th percentile value (corresponding to capital flows “surges”) rises from 4 to 8.75 percent of GDP.
- Exchange rate regime flexibility:
  - Short term: more flexible regimes are associated with higher probabilities of large in- and outflows following global shocks.
  - Medium term: only the positive effect (a higher probability of rebounds) persists.
- Capital account openness:
  - On average, more open capital accounts experience larger short-term inflows.
  - After an adverse global shock, countries with more open capital accounts face fewer large inflows; likelihood of large outflows remains unchanged.
- Legal and governance indicators:
  - High-quality legal framework and low perceived corruption associated with higher median medium-term portfolio inflows conditional on an increase in the BBB yield today, and fewer large in- and outflows.
  - Central bank transparency: medium-term median inflows estimated at 1.6 percent of GDP for countries ranked at the 80th percentile in central bank transparency, and 1.2 percent of GDP for countries at the bottom 20th percentile. No significant short-term effect reported.
- Macroprudential interaction:
  - Positive and highly significant coefficients on the interaction term with the BBB yield for top percentiles (α=90,95) imply that macroprudential tightening can help reduce the likelihood of portfolio flow surges following a period of very lax global conditions; effect is quantitatively small.

### Policy actions and quantitative impact (exact figures)
- FX intervention (reserves sales):
  - Sales of reserves associated with smaller likelihood of very large outflows in the quarters immediately after a global shock.
  - Quantitative example: an unexpected sale of FX reserves of 1.4 percent of GDP (corresponding to a two standard deviation shock in the sample) is associated with a reduction in the probability of outflows from around 35 percent to 29 percent.
  - CaR at the 5th percentile improves from -4.5 percent of GDP to -3.5 percent of GDP with such intervention.
- Capital flow measures (CFMs):
  - A tightening of CFMs in response to an adverse global shock is associated with larger outflows in the short term (exacerbating downside tail risk), but not later on.
  - Quantitatively: after a rise in the BBB yield, capital outflows at the lowest 5th percentile equal -4.5 percent of GDP without CFM tightening and -5.0 percent of GDP with a two-standard deviation tightening; difference persists at the 10th percentile (-2.75 percent versus -3.25 percent) but vanishes at higher percentiles. Overall probability of outflows remains broadly unchanged.
  - Potential caveats: results may reflect reverse causality or insufficiently comprehensive/enforceable controls.
- Monetary and macroprudential policy:
  - Monetary policy and macroprudential actions do not seem to affect the short-term outlook for portfolio inflows.
  - Some evidence that macroprudential tools mitigate risk of very large inflows in the medium term (top percentiles), though effect size is small.

### Country-level illustrations: Chile and Turkey (selected quantitative details)
- Chile:
  - Short-term drivers: global financial conditions (commodity prices) key drivers; higher international commodity prices strongly associated with higher likelihood of very large inflows in the short term.
  - Short-term CaR (10th percentile) evolution: tumbled from +2.8 percent of GDP at end-2012 to -2 percent of GDP by 2018: Q3 (implying a 10 percent chance that over the next 2 quarters there would be portfolio outflows of at least 2 percent of GDP as of 2018: Q3).
  - Scenario under two-standard-deviation increase in U.S. BBB spread: conditional distribution shifts leftward (worsening CaR at the 5th percentile and reducing likelihood of very large outflows in the right tail).
  - Policy effects: short term FX intervention appears significant in stemming high inflows; significance vanishes within two years. Monetary policy actions do not appear to mitigate medium-term risks.
- Turkey:
  - Short-term drivers: domestic factors (balance-sheet vulnerabilities) relatively more important; higher short-term external debt to foreign reserves strongly associated with reduced likelihood of surges.
  - Short-term CaR (10th percentile) as of 2018: Q3: around -4.5 percent of GDP (implying a 10 percent chance of outflows of at least 4.5 percent of GDP over next 2 quarters).
  - Between 2017: Q3 and 2018: Q3, mode of forecast distribution remained ~3 percent of GDP but skewness increased; probability of short-term outflows rose from 26 to 34 percent; worst 5 percent outcome moved from -5.7 percent to -8 percent of GDP.
  - Scenario under two-standard-deviation rise in U.S. BBB spread: upside risks sharply revised down; downside risks (CaR at 5th percentile) remain almost unchanged.
  - Policy effects: tightening in MPMs associated with smaller likelihood of very large outflows immediately after a global shock and higher inflows across distribution in medium term; FX intervention largely insignificant; monetary policy and CFM changes do not appear to mitigate medium-term risks.

### Robustness checks and construction notes
- Policy shock construction:
  - Monetary policy and FXI shocks: residuals from country-by-country first-stage regressions.
  - CFMs and macroprudential shocks: sum of residuals in current quarter t and three lags (t-1, t-2, t-3).
- Robustness:
  - Regressions on full sample of 35 economies (not controlling for financial integration) broadly in line with preferred results, sometimes less significant.
  - Allowing for cross-section correlation of standard errors (clustering at year level) leaves majority of results unchanged; interaction of BBB yield with macroprudential policy loses medium-term significance in some specifications.
  - Results robust to including lag of dependent variable.

### Conclusions and policy implications (explicit policy-relevant conclusions)
- The CaR methodology predicts the entire future probability distribution of capital flows to emerging markets conditional on current domestic structural characteristics, policies, and global shocks.
- Main policy-relevant conclusions (exact findings preserved):
  - Structural characteristics, policy frameworks, and policy actions shape the response of portfolio inflows differently across horizons and the distribution (median vs tails).
  - FX interventions mitigate downside tail risks in the short term but have limited impact on median future flows and limited persistence.
  - Tightening CFMs in response to adverse global shocks can exacerbate downside tail severity (larger outflows if they occur) while leaving median flows unchanged.
  - Macroprudential policies show limited effectiveness for outflow protection but can modestly reduce the probability of very large inflow surges in the medium term.
  - Monetary policy shows little evidence of mitigating portfolio inflow or outflow risks driven by global shocks.
- Suggested research extensions:
  - Role of fiscal policies.
  - Differential effects on bank lending and FDI.
  - Effects of combining policies.
  - Higher-frequency fund flow data.
  - Bilateral flow patterns.
  - Role of multiple simultaneous policy actions.

*Source: IMF staff calculations; content from wpiea2019279-print-pdf (Section V and surrounding sections).*

### Section V, we focus on two country cases to show how our methodology can be tailored to

### wpiea2019279-print-pdf - Section V, we focus on two country cases to show how our methodology can be tailored to 

### A Risk Management Framework for Capital Flows — Key concepts
- Stylized representation:
  - Initial predicted median capital flows: 2 percent of GDP.
  - Post-shock median falls to 1.5 percent of GDP in the example.
  - Policy action example raises median from 1.5 to 1.8 percent of GDP and reduces left-tail mass (probability of net outflows).
- Two quantifications of risk:
  - Probability that capital flows fall below a threshold (natural threshold: zero, distinguishing inflows from outflows).
  - Capital flows at risk (CaR): the amount of outflows reached or exceeded at a given probability (Adrian, Boyarchenko, and Giannone (2019) use the 5th percentile). Country examples use CaR at both the 5th and the 10th percentile.

### Econometric approach — Overview
- Two-step procedure:
  1. Estimate future flows with quantile regression across α = 0.05, 0.1, 0.15, ..., 0.95 to obtain predicted quantiles of average gross portfolio inflows (percent of GDP) over chosen horizons.
     - Short term: h=2, j=1 → average inflows in the first and second quarters ahead.
     - Medium term: h=8, j=3 → average quarterly portfolio inflows over quarters 5–8 ahead.
  2. Fit empirical distribution of predicted quantiles to a skewed-t probability distribution (Azzalini and Capitanio (2003)) characterized by mean, variance, skewness, and kurtosis using a minimum distance estimator and Azzalini (2019) algorithm.
- Focus and rationale:
  - Focus on gross capital flows, specifically non-resident portfolio flows (“gross inflows”).
  - Component focus: portfolio debt and equity flows (most relevant for policy due to volatility and external sensitivity).
  - Exclusions/less emphasis: foreign direct investment (less affected by considered drivers); banking flows (“other flows”) have been dwarfed by portfolio debt flows in the post-crisis period.

### Model specification and variables (cross-country panel)
- General quantile specification (for percentile α):
  - ŷ_{i,t+h−j : t+h | t}^{α} = δ_{i}^{α} + β_{1}^{α} G_{t} + β_{2}^{α} DG_{i,t} + β_{3}^{α} P_{i,t} + β_{4}^{α} G_{t} × P_{i,t} + ε_{it}
  - ŷ_{i,t+h−j : t+h | t}^{α} is average gross portfolio inflows (percent of GDP) to country i at percentile α between quarters t+h-j and t+h.
  - G_{t}: global “push” factors (U.S. corporate BBB spread or BBB yield, U.S. sovereign 10-year yield, DXY index). Some specs control for commodity prices and global growth.
  - DG_{i,t}: country-specific “pull” factors (e.g., domestic year-on-year GDP growth, short-term external debt to FX reserves).
  - P_{i,t}: domestic policy frameworks, structural characteristics, and policy actions (financial sector development, capital account openness, exchange rate regime, rule of law, central bank transparency; and policy actions: monetary policy, macroprudential policies, FX interventions (FXIs), capital flow management measures (CFMs)).
  - Interactions G_{t} × P_{i,t} used to assess mitigation of global shocks by domestic characteristics/policies.
- Endogeneity handling:
  - Policy actions estimated via country-by-country policy functions; residuals from first-stage regressions used as policy shocks (approach similar to Brandao et al. (forthcoming) and Forbes and Klein (2015)); Appendix describes first-stage regressions in detail.
- Estimation details:
  - Algorithmic method for unbalanced panels (Koenker and d’Orey (1987, 1994); Koenker (2005)).
  - Bootstrapped standard errors clustered at country level and corrected for serial correlation.

### Data and sample
- Quarterly data period: 1996: Q4 to 2018: Q4.
- Country sample: 35 emerging market and developing countries (unbalanced panel). Appendix provides data descriptions.
- Summary statistics highlights (from Table 1 for subsample used in policy actions analysis):
  - Gross portfolio inflows: Min -24.112; Median 1.149; Max 36.411 (percent of GDP).
  - FX intervention (% of GDP): Min -12.8; Median 0.020; Max 148.5.
  - Macroprudential measures (index): tightening actions 10; easing actions 33 (per-country summary over sample period).
  - CFMs: CFMs introduced 1356; CFMs abolished 1356 (note: preserved exactly as in source table layout).
  - Policy rate (post GFC): summary statistics cover the post-global-financial-crisis period.
- Cross-country heterogeneity:
  - One standard deviation of portfolio flows: Malaysia ≈ 8 percent of GDP; Colombia ≈ 2 percent of GDP.
  - Policy usage: countries mostly tightened macroprudential tools; CFMs were tightened or abolished with similar frequency.

### Aggregate flows to emerging markets — Findings
- Simplified aggregate specification (excluding China) for portfolio debt inflows to EMs:
  - ŷ_{t+h−j : t+h | t}^{α} = δ^{α} + β_{1}^{α} G_{t} + β_{2}^{α} Gdp_{t,h} + y_{t−1} + ε_{t}
  - Included global factors with predictive power: U.S. corporate BBB spread, U.S. sovereign 10-year yield, DXY index.
  - Domestic control: aggregate EM real GDP growth and lagged dependent variable.
- Key empirical findings:
  - Higher U.S. interest rates and a stronger U.S. dollar are associated with weaker future inflows in both short and medium term.
  - Investor risk aversion (BBB spread) disproportionately explains large outflows and, to a lesser extent, surges: higher coefficient estimates at lowest and highest percentiles (Figure 3).
  - Coefficient on global risk appetite (BBB spread) exhibits a sign reversal over horizons:
    - Median quantile coefficient negative and significant in the current quarter.
    - Turns positive and significant about 3–5 quarters into the future — consistent with mean-reversion in risk appetite.
  - Medium-term distribution comparisons:
    - Predicted medium-term distribution using information up to 2018: Q4 (for 2020: Q1–2020: Q4) shown in black.
    - Updated distribution using information as of 2019: Q2 (for 2020: Q3–2021: Q2) shown in green/red.
    - Outlook improved from 2018: Q4 → 2019: Q2: distribution shifted right and probability of outflows declined (red tail reduced), driven mainly by lower U.S. interest rates (10-year Treasury yield fell markedly over this period).
- Interpretation:
  - External factors play an important role in predicting aggregate portfolio flows to EMs.
  - Downside and upside risks to flows vary considerably over time with external conditions.

### Panel data analysis — Specification and policy/structural roles
- Parsimonious focus: single global measure — U.S. corporate BBB yield used as G_{t} in eq. (3):
  - ŷ_{i,t+h−j : t+h | t}^{α} = δ_{i}^{α} + β_{1}^{α} BBB_{t} + β_{2}^{α} DG_{i,t} + β_{3}^{α} Gcyc_{t} + β_{4}^{α} P_{i,t} + β_{5}^{α} BBB_{t} × P_{i,t} + β_{6}^{α} FI_{i,t} + β_{7}^{α} BBB_{t} × FI_{i,t} + ε_{it}
  - Gcyc_{t}: U.S. GDP growth (detrended, average over last four quarters).
  - FI_{i,t}: integration with global financial markets (financial segmentation indicator, available for 18 countries only).
- Expectations on coefficients:
  - β_{1}^{α} expected negative (higher BBB yield → lower portfolio inflows, at least short term).
  - β_{3}^{α} sign ambiguous depending on timing of global cycle.
  - β_{5}^{α} positive if policy/characteristic mitigates negative impact of higher BBB yield on inflows at percentile α.
- Domestic controls included in all regressions:
  - Year-on-year GDP growth in current quarter.
  - GDP per capita.
  - Short-term external-debt-to-reserves ratio.
  - Financial market depth.
  - Capital account openness.
- Theoretical expectations and heterogeneity:
  - Short-term external debt may increase short-term inflows (financing needs) but reduce medium-term flows via sustainability concerns — especially important in low percentiles (tail events).
  - Deeper financial markets: expected to be associated with larger median short-term flows; ambiguous effect on tails (could mitigate price impact of outflows or enable faster investor exit).
  - Capital account openness: encourages larger inflows but could lead to larger outflows during risk-off.
  - Financial integration (FI_{i,t}):
    - Sample reduces from 35 to 18 countries when FI included.
    - Greater integration could make investor exit easier (worsen outflows) or reduce withdrawal incentives (investor familiarity).
    - Interaction BBB_{t} × FI_{i,t} included to capture this moderating effect.
- Policy variables of interest and expected impacts:
  - FX sales and monetary policy tightening: expected to mitigate negative impact of global shocks on portfolio inflows.
  - Macroprudential policies: expected to primarily reduce likelihood/size of surges; may also increase investor confidence in adverse shocks by strengthening resilience.
  - CFMs: policymakers hope CFMs dampen gross inflows and outflows; effects are investigated empirically.

*Source: IMF staff calculations; content from wpiea2019279-print-pdf (Section V and surrounding sections).*

### introduction of outflow restrictions might prevent capital from flowing out from a country,

### wpiea2019279-print-pdf - introduction of outflow restrictions might prevent capital from flowing out from a country

### Policy trade-offs and signaling
- Introduction of outflow restrictions might prevent capital from flowing out from a country, but it might also reduce new inflows when the tide turns.
- A CFM designed to stem inflows may reduce surges, while also reducing a country’s vulnerability to large outflows since it may help prevent the buildup of domestic financial imbalances.
- The introduction of CFMs might have a negative signaling effect and cause larger foreign capital withdrawals, especially if the capital controls are not sufficiently comprehensive or are operationally unenforceable.

### Empirical approach and horizons
- Regressions include country fixed effects.
- Two horizons for future portfolio inflows are considered:
  - Short-term: average quarterly portfolio inflows over the next two quarters as the dependent variable.
  - Medium-term: average quarterly portfolio inflows over quarters 5–8 ahead.

### Baseline results
- Baseline specifications without policy variable interactions:
  - An increase in the BBB has a statistically significant and negative impact on short-term portfolio inflows across different quantiles.
  - The impact remains mostly negative in the medium term, but is statistically significant only for low quantiles.
- Subsequent analysis focuses on the role domestic factors play in shaping the post-shock distributions.

### Structural characteristics and policy frameworks (methods and visualization)
- Role of structural characteristics and policy frameworks examined in response to a rise in the BBB yield (Tables A8 and A9).
- Total effect quantified as 훽4훼 + 훽5훼BBB_t and illustrated in Figure 6:
  - Compares distributions of predicted portfolio inflows over the next two quarters following a one-standard-deviation increase in the U.S. BBB yield today (around 160 basis points).
  - Distributions when domestic policy and structural variables are set at different levels:
    - Red (solid): countries with weak structural characteristics (set at the 20th percentile in the country sample) or policy action set to zero.
    - Blue (dashed): countries with high structural characteristics (set at the 80th percentile) or policy action equal to two standard deviations (the shocks correspond to FX sales, a monetary policy tightening and a CFM tightening).

### Short-term role of financial market depth (empirical finding)
- In the short term, greater financial market depth increases the likelihood of a rebound in capital flows after an adverse shock to global financial conditions.
- The effects are visible mostly for the upper side of the predicted distribution.

### Other variable associations (footnote summary)
- U.S. GDP growth is negatively and statistically significant associated with short- and medium-term portfolio inflows.
- Higher GDP per capita lowers portfolio inflows, although it is mostly insignificant.
- In the short term, higher FX debt relative to reserves implies higher inflows, consistent with larger financing needs of more indebted countries; in the medium term the effect becomes negative.
- Financial market depth is positively and statistically significant associated with portfolio inflows in the short term.

*Source: wpiea2019279-print-pdf (IMF staff calculations as presented in the supplied content).*

### 0.3 to 1.8 percent of GDP, and the 95

### wpiea2019279-print-pdf - 0.3 to 1.8 percent of GDP, and the 95

### Key empirical findings: global shocks, structural characteristics, and portfolio flows
- Comparing countries with shallow financial markets to countries with more developed markets: conditional lower-tail capital outflows increase from 0.3 to 1.8 percent of GDP, and the 95th percentile value (corresponding to capital flows “surges”) rises from 4 to 8.75 percent of GDP.  
- Exchange rate regime flexibility:
  - Short term: more flexible regimes are associated with higher probabilities of large in- and outflows following global shocks (higher short-term volatility).
  - Medium term: only the positive effect (a higher probability of rebounds) persists.  
- Capital account openness:
  - On average, more open capital accounts experience larger short-term inflows.
  - After an adverse global shock, countries with more open capital accounts face fewer large inflows; likelihood of large outflows remains unchanged.  
- Legal and governance indicators:
  - High-quality legal framework and low perceived corruption are associated with higher median medium-term portfolio inflows conditional on an increase in the BBB yield today, and fewer large in- and outflows.
  - Central bank transparency: medium-term median inflows are estimated at 1.6 percent of GDP for countries ranked at the 80th percentile in central bank transparency, and 1.2 percent of GDP for countries at the bottom 20th percentile. No significant short-term effect reported.  
- Macroprudential interaction: positive and highly significant coefficients on the interaction term with the BBB yield for top percentiles (α=90,95) imply that macroprudential tightening can help reduce the likelihood of portfolio flow surges following a period of very lax global conditions; effect is quantitatively small.

### Policy actions and quantitative impact
- FX intervention (reserves sales):
  - Sales of reserves are associated with a smaller likelihood of very large outflows in the quarters immediately after a global shock.
  - Quantitative example: an unexpected sale of FX reserves of 1.4 percent of GDP (corresponding to a two standard deviation shock in the sample) is associated with a reduction in the probability of outflows from around 35 percent to 29 percent.
  - CaR at the 5th percentile improves from -4.5 percent of GDP to -3.5 percent of GDP with such intervention.  
- Capital flow measures (CFMs):
  - A tightening of CFMs in response to an adverse global shock is associated with larger outflows in the short term (exacerbating downside tail risk), but not later on.
  - Quantitatively: after a rise in the BBB yield, capital outflows at the lowest 5th percentile equal -4.5 percent of GDP without CFM tightening and -5.0 percent of GDP with a two-standard deviation tightening; difference persists at the 10th percentile (-2.75 percent versus -3.25 percent) but vanishes at higher percentiles. Overall probability of outflows remains broadly unchanged.
  - Potential caveats: results may reflect reverse causality or insufficiently comprehensive/enforceable controls.  
- Monetary and macroprudential policy:
  - Monetary policy and macroprudential actions do not seem to affect the short-term outlook for portfolio inflows.
  - Some evidence that macroprudential tools mitigate risk of very large inflows in the medium term (top percentiles), though effect size is small.

### Country-level illustrations and scenarios
- Framework: country-specific quantile regressions with lagged domestic GDP growth and lagged short-term external debt to FX reserves; global factors include U.S. corporate BBB spread, U.S. 10-year Treasury yield, U.S. dollar DXY index, and global commodity prices. Short-term horizon: average inflows h=2 (two quarters); medium term: h=8 (4-quarter average 5–8 quarters ahead).
- Chile:
  - Short-term drivers: global financial conditions (commodity prices) are key drivers; higher international commodity prices are strongly associated with higher likelihood of very large inflows in the short term.
  - Short-term CaR (10th percentile) evolution: tumbled from +2.8 percent of GDP at end-2012 to -2 percent of GDP by 2018: Q3 (implying a 10 percent chance that over the next 2 quarters there would be portfolio outflows of at least 2 percent of GDP as of 2018: Q3).
  - Scenario: under a two-standard-deviation increase in the U.S. BBB spread, Chile’s short-term conditional distribution shifts leftward (worsening CaR at the 5th percentile and reducing likelihood of very large outflows in the right tail).
  - Policy effects: short term FX intervention appears significant in stemming high inflows; significance vanishes within two years. Monetary policy actions do not appear to mitigate risks to portfolio inflows in the medium term.
- Turkey:
  - Short-term drivers: domestic factors (balance-sheet vulnerabilities) are relatively more important; higher short-term external debt to foreign reserves strongly associated with reduced likelihood of surges.
  - Short-term CaR (10th percentile) as of 2018: Q3: around -4.5 percent of GDP (implying a 10 percent chance of outflows of at least 4.5 percent of GDP over next 2 quarters).
  - Between 2017: Q3 and 2018: Q3, mode of forecast distribution remained ~3 percent of GDP but skewness increased; probability of short-term outflows rose from 26 to 34 percent; worst 5 percent outcome moved from -5.7 percent to -8 percent of GDP.
  - Scenario: under a two-standard-deviation rise in the U.S. BBB spread, upside risks sharply revised down; downside risks (CaR at 5th percentile) remain almost unchanged.
  - Policy effects: tightening in MPMs associated with smaller likelihood of very large outflows immediately after a global shock and higher inflows across distribution in medium term; FX intervention largely insignificant; monetary policy and CFM changes do not appear to mitigate medium-term risks.

### Robustness, data, and construction notes
- Sample and data:
  - Quarterly data from 1996: Q4 to 2018: Q4 for 35 emerging market and developing countries.
  - Dependent variable: gross portfolio inflows as percent of GDP (IMF Financial Flow Analytics database).
  - Short-term prediction horizon: h=2 (average of next two quarters); medium-term: h=8 (average over quarters 5–8 ahead).
  - Structural variables from World Bank WDI and WGI, Transparency International, IMF AREAER, IMF Financial Development Index (market depth), Chinn-Ito Index for capital account openness.
  - Policy variables: policy rates (IFS), FX interventions (central bank reports, FRED, COFER; valuation-adjusted proxies where needed), macroprudential actions (iMaPP database), CFMs (AREAER broad restrictiveness index).
- Policy shock construction:
  - Monetary policy and FXI shocks: residuals from country-by-country first-stage regressions.
  - CFMs and macroprudential shocks: sum of residuals in current quarter t and three lags (t-1, t-2, t-3) to capture infrequent, persistent actions.
- Robustness checks:
  - Regressions on the full sample of 35 economies (not controlling for financial integration) broadly in line with preferred results, sometimes less significant.
  - Allowing for cross-section correlation of standard errors (clustering at year level) leaves majority of results unchanged; interaction of BBB yield with macroprudential policy loses medium-term significance in some specifications.
  - Results robust to including lag of dependent variable.

### Conclusions and policy implications
- The CaR methodology predicts the entire future probability distribution of capital flows to emerging markets conditional on current domestic structural characteristics, policies, and global shocks.
- Main policy-relevant conclusions:
  - Structural characteristics, policy frameworks, and policy actions shape the response of portfolio inflows differently across horizons and the distribution (median vs tails).
  - FX interventions mitigate downside tail risks in the short term but have limited impact on median future flows and limited persistence.
  - Tightening CFMs in response to adverse global shocks can exacerbate downside tail severity (larger outflows if they occur) while leaving median flows unchanged.
  - Macroprudential policies show limited effectiveness for outflow protection but can modestly reduce the probability of very large inflow surges in the medium term.
  - Monetary policy shows little evidence of mitigating portfolio inflow or outflow risks driven by global shocks.
- Research extensions suggested: role of fiscal policies; differential effects on bank lending and FDI; effects of combining policies; higher-frequency fund flow data; bilateral flow patterns; role of multiple simultaneous policy actions.

*Source: IMF staff calculations and analysis as presented in the provided content unit.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019279-print-pdf.pdf_