## 2.1 Portfolio flows (and related empirical sections)

## Source details

**Canonical URL:** [2.1 Portfolio flows (and related empirical sections)](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021033-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021033-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021033-print-pdf.pdf.json)

---

### Data, coverage, and measurement
- Data source: weekly portfolio flows from EPFR Global; flows measured as the US$ net value of purchases and redemptions into investment funds.
- EPFR fund domicile: most funds covered are domiciled in DMs.
- Sample coverage: includes 21 DMs and 16 EMs over January 2014 to May 2020.
- EPFR industry coverage (as of September 2020): funds reporting weekly data covered about 74 percent and 70 percent of the global investment equity and bond fund industry, respectively.
- Sample end: second week of May 2020 (sample ends in the second week of May due to a change in the availability of the fiscal spending data obtained from the IMF, after which the IMF switched from weekly to biweekly updates for some countries).
- Interpretation: EPFR flows are interpreted as a proxy for investment fund flows (not the universe of portfolio flows). At the fund level, EPFR flows fully account for valuation effects due to asset returns and exchange rate movements.

### Stylized facts on net portfolio flows during the COVID episode
- Magnitude: portfolio outflows during the pandemic were of a historically large magnitude, markedly exceeding those experienced during earlier episodes (taper tantrum and global financial crisis), with particularly large bond outflows.
- Timing and reversal:
  - Net portfolio flows reversed sharply both in EMs and DMs.
  - DMs saw flows recovering less than two months into the pandemic.
  - Net flows to EMs continued to decline for a longer period.
- Normalization note: the magnitude of the reversal in bond flows was historically unprecedented when measured in US$ but not when normalized by portfolio allocation.
- Cross-country heterogeneity:
  - At the height of the crisis, the standard deviation of net portfolio flows as a share of allocation was 1.1 percent.
  - The distribution of net flows across countries during COVID was wider than during the taper tantrum and similar to the global financial crisis.
- Geographic progression and heterogeneity:
  - Asia-Pacific (both DMs and EMs) started to experience outflows in mid-February (earlier than other regions).
  - Emerging Africa (South Africa only in sample) and Europe saw net flows declining from the last week of February 2020 onward.
  - EMs in the Americas and the Middle East, and other DMs, saw initial outflows by the first week of March—one week before WHO characterized COVID as a pandemic.
  - Outflows from Asia-Pacific were lower than outflows from other regions such as the Americas; this pattern suggests that measures to control the pandemic or mitigate economic fallout might have affected investors’ allocation decisions.
  - Large outflows from the Middle East might also reflect the impact of the pandemic on the oil price collapse.

### Episode timing (figure notes)
- Episode start and end dates (weeks following start date shown on x-axis in Figure 2):
  - COVID: 2020/02/26-2020/04/29
  - Taper tantrum: 2013/05/22-2013/08/14
  - Global financial crisis: 2008/10/08-2008/12/31

### COVID-related controls and policy measures (data sources and variables)
- COVID cases: collected from Haver Analytics; New COVID Cases variable is new domestic infections scaled by population (per capita) and standardized by its sample standard deviation (non-zero values used when computing the standard deviation).
- Lockdown stringency: obtained from Hale et al. (2020); index ranges 0 to 100 and is standardized for analysis.
- Fiscal stimulus: discretionary fiscal spending in response to the pandemic from the IMF’s Survey on Country Responses; measured as percent of GDP.
- Monetary policy: policy rates from Haver Analytics; Policy Rate defined as the policy rate differential to the U.S. Fed Funds rate.
- Sovereign risk: sovereign CDS spreads from Bloomberg.
- Other controls: stock market returns from Haver Analytics.
- Notes on heterogeneity and timing:
  - COVID cases (per 1,000 population) were more heterogeneous for DMs than EMs in March and April; as cases picked up, EMs showed wide dispersion.
  - Lockdown stringency was relatively lenient until end-March for both DMs and EMs and then increased sharply in April with response becoming more homogeneous.
  - Fiscal spending (first-round stimulus) largely concentrated up to 5 percent of GDP, but some countries spent more sizeable amounts.
  - Policy rate differentials were relatively homogeneous for DMs and more scattered for EMs.

### Summary statistics of COVID cases and policy measures (observations over 2020/01/22-2020/05/13)
- Developed markets (N = 357)
  - COVID-19 cases (per 1000 pop.): Mean 0.7; St. Dev. 1.2; Min 0.0; Median 0.15; Max 5.0
  - Lockdown stringency (0 to 100): Mean 44.3; St. Dev. 33.2; Min 0.0; Median 45.8; Max 96.0
  - Fiscal spending (percent of GDP) (N = 100): Mean 4.6; St. Dev. 3.1; Min 1.1; Median 3.9; Max 11.2
  - Policy rate differential (pp): Mean −0.6; St. Dev. 0.8; Min −2.4; Median −0.1; Max 1.6
- Emerging markets (N = 272)
  - COVID-19 cases (per 1000 pop.): Mean 0.2; St. Dev. 0.8; Min 0.0; Median 0.0; Max 19.0
  - Lockdown stringency (0 to 100): Mean 48.1; St. Dev. 35.7; Min 0.0; Median 56.9; Max 97.0
  - Fiscal spending (percent of GDP) (N = 76): Mean 3.5; St. Dev. 2.6; Min 0.2; Median 2.7; Max 9.8
  - Policy rate differential (pp): Mean 2.8; St. Dev. 2.9; Min −1.6; Median 2.4; Max 11.4

### Empirical approach: objectives and identification
- Objective: empirically evaluate the impact of countries’ exposure to the pandemic and key policy measures (lockdown stringency, fiscal stimulus, policy rate differential) on capital markets (net portfolio flows and sovereign CDS spreads).
- Frequency and method: weekly frequency; local projections approach (Jordà, 2005) to estimate contemporaneous and dynamic cumulative effects (horizons h = 0 and cumulative h = 0,...,3).
- Dependent variables: net portfolio flows (scaled by pre-COVID allocation and standardized by its pre-COVID sample standard deviation) and sovereign CDS spreads.
- Scaling and expectations:
  - Portfolio flows scaled by pre-COVID allocation (allocation value reported for week of February 26, 2020).
  - E_{t−1}{Y_{c,t+h}} term captures the expected component of Y_{c,t+h} using one lag of portfolio flows, sovereign CDS spreads, stock market returns, interest rate differential with the U.S., total domestic COVID-19 cases, and log portfolio allocation.
- Fixed effects and controls:
  - Country fixed effect η_c; week fixed effects η_DMs,t and η_EMs,t to allow heterogeneous global shocks across DMs and EMs; country×month fixed effect η_c×η_m to control for slow-moving domestic factors.
- Endogeneity and identification strategy:
  - Identification rests on high frequency of data, rapid adjustment of portfolio flows to new information, and comprehensive controls and fixed effects.
  - New COVID cases treated as exogenous conditional on the specification, aided by inclusion of lagged flows, stock returns, and country×month fixed effects.
  - Lockdown, fiscal, and monetary measures assumed contemporaneously exogenous due to decision lags; dynamic endogeneity mitigated by lagged controls and fixed effects.
  - Conservative specification includes country×month fixed effects to control for institutional variations, capital controls, and slow-moving domestic shocks.

### Empirical model (baseline regression, local projections)
- Baseline equation (local projections for horizon h):
  - Y_{c,t+h} = E_{t−1}{Y_{c,t+h}} + β^h_1 New COVID Cases_{c,t} + β^h_2 Lockdown_{c,t} + β^h_3 Fiscal Stimulus_{c,t} + β^h_4 Policy Rate_{c,t} + β^h_5 isCOVID_{c,t} + β^h_6 Policy Rate_{c,t} × isCOVID_{c,t} + η_c + η_DMs,t+h + η_EMs,t+h + η_c×η_m + e_{c,t+h}
- Horizons analyzed: contemporaneous h = 0 and cumulative over h = 0,...,3 (monthly cumulative effect computed as ∑_{h=0}^3 Y_{c,t+h}).
- isCOVID_{c,t}: dummy = 1 when number of COVID cases > 0 in country-week; 0 otherwise.
- Rationale for CDS focus: sovereign CDS spreads used as a measure of risk given dominant role of sovereign borrowing in EMs.

### Model scaling and dynamics
- Net portfolio flows scaled by pre-COVID allocation to ensure estimates are not driven by market size differences; ratio further scaled by its pre-COVID sample standard deviation.
- Expectation term E_{t−1}{Y_{c,t+h}} uses one lag of selected financial and COVID variables to capture predictable component and help mitigate endogeneity.

### Preview of results (section summary)
- Heterogeneity in countries’ exposure to COVID-19 and related policy measures significantly influenced the dynamics of portfolio flows and CDS spreads during the COVID episode.
- Results are analyzed by:
  - DMs vs EMs,
  - bond vs equity flows,
  - foreign vs domestic investors,
  - interactions between domestic COVID-related factors and global factors,
  - role of pre-existing macroeconomic fundamentals.

---

### 4.1 COVID-19 related domestic factors as drivers of portfolio flows

### Empirical strategy and identification (foreign-domiciled flows focus)
- Analysis focuses on foreign-domiciled flows (funds domiciled outside the recipient country).
- Preferred (most conservative) specification controls for country fixed effects; week fixed effects (global factors) with two independent time trends for DMs and EMs; country×month fixed effects; expected net portfolio flows (function of lagged flows and other lagged determinants, including sovereign risk and domestic stock prices).
- Two key observations:
  - Estimated coefficients on domestic factors change considerably once global time fixed effects are included, indicating a common global component in COVID-related domestic variables.
  - Adjusted R2 rises substantially when adding country×month fixed effects, pointing to the importance of pull (domestic) factors in explaining contemporaneous variations in portfolio flows.

### COVID cases: effects and heterogeneity
- Aggregate results (columns 5–6, table 2):
  - Higher new domestic COVID cases per capita: insignificant effect on portfolio flows upon impact (h = 0, column 5).
  - Over one-month horizon (cum(h=0:3), column 6): a one standard deviation increase in new domestic COVID cases per capita associated with a 1.13 standard deviation increase in foreign net portfolio flows (as a ratio of total allocation).
- Heterogeneity by market and asset class:
  - The positive average effect driven mainly by bond flows in EMs.
  - Equity flows experienced a cumulative decline in both EMs and DMs.
  - Net bond flows declined upon impact of the COVID shock, indicating an initial perception of higher COVID cases as bad news.
- Sovereign CDS spread response (table 4):
  - A one standard deviation increase in new COVID cases per capita associated with a significant increase in growth of CDS spreads by 0.18 percent in EMs and 0.03 percent in DMs over the one-month horizon.
  - Interpretation: increased demand for financing (larger supply of bonds) in affected economies was a dominant force behind cumulative increase in bond flows; bond flows increased together with a rise in the cost of risk.

### Lockdown measures: effects and drivers
- Aggregate sample (columns 5–6, table 2):
  - A one standard deviation change in lockdown stringency led to an increase in cumulative net portfolio flows by 0.32 standard deviations over a one-month horizon in the average country, after an initial decline.
- Market heterogeneity:
  - The cumulative one-month increase driven by EMs; response in DMs was insignificant (columns 7–8, table 2).
  - China was the main determinant of the strong positive cumulative effect in EMs, being an early adopter of relatively strict lockdowns that controlled the pandemic sooner and contributed to investor confidence.
- Asset-class response (table 3):
  - Increase in foreign fund flows into EMs after the initial decline dominated by bond flows (reallocation to safety).
  - Equity flows to EMs experienced a significant decline.
- CDS spreads (table 4):
  - A stricter lockdown associated with lower CDS spreads over the one-month horizon in both EMs and DMs.
  - Interpretation: stricter lockdowns triggered increased demand from EM investors, especially for safer assets.

### Discretionary fiscal stimulus: magnitude and composition
- Aggregate effect (columns 5–6, table 2):
  - Fiscal stimulus: initially negative impact, but over one month, increases in fiscal spending perceived positively, attracting portfolio flows.
- Magnitude and heterogeneity:
  - For every one percent increase in fiscal spending as a share of GDP:
    - Net flows increased by 1.65 standard deviations in EMs.
    - Net flows increased by 0.33 standard deviations in DMs.
  - Effect was strong for both bond and equity flows and primarily driven by EMs (table 3).
- CDS spreads:
  - Response in sovereign CDS spreads was not significant over the one-month horizon (table 4), providing no clear indication on supply vs. demand drivers of cumulative flow increases.
- Mechanical vs. non-mechanical channels:
  - Many governments financed stimulus through international borrowing; estimated increase in flows might partially reflect a mechanical effect.
  - Evidence that non-mechanical response, driven by improved economic outlook, played an important role: net equity flows increased in response to stimulus, not only net bond flows.

### Discretionary monetary policy (short-term policy rate): average and episode-specific effects
- Average effect over full sample (table 2, columns 5–6):
  - Changes in the policy rate relative to the U.S. do not play a statistically significant role in dynamics of portfolio flows on average.
- Heterogeneity and COVID episode effects:
  - Insignificance over full sample driven by EMs (columns 7–8, table 2).
  - During the COVID episode:
    - A hundred basis points cut in the interest rate led to a cumulative decline in net portfolio flows by 0.21 standard deviations in EMs (no significant immediate impact).
    - In DMs, a hundred basis points cut was associated with an increase in net flows by 0.64 standard deviations upon impact, suggesting an initial reassuring effect; cumulatively, however, a rate cut led to a decline in net portfolio flows in DMs, with the cumulative response larger than in EMs.
- Asset-class drivers (table 3):
  - In EMs, the cumulative decline in total net portfolio flows after a rate cut was entirely driven by bond flows (consistent with interest rate parity).
  - In DMs, both bond and equity flows responded negatively to a rate cut, with larger elasticity for equity flows.
- CDS returns and interpretation (table 4 and text):
  - Expansionary EM monetary policy associated with an increase in CDS returns during the COVID episode: a 3.14 percent increase for every hundred basis point cut in the policy rate.
  - CDS returns increased largely upon impact.
  - Findings inconsistent with classical credit channel; may reflect an information channel where aggressive rate cuts temporarily fueled market concerns about the expected fallout from the pandemic.

### Response of domestic-domiciled portfolio flows (contrast with foreign-domiciled)
- Domestic vs. foreign investor behavior:
  - Domestic investors have more vulnerable balance sheets to idiosyncratic shocks and are more likely to hold local-currency assets; foreign investors allocate more to foreign-denominated assets.
  - Overall finding: domestic and foreign investors responded differently to the pandemic and policy measures, with domestic flows generally lower relative to foreign flows.
- COVID cases:
  - Foreign investors eventually increased holdings in response to higher new COVID cases (potentially meeting domestic financing needs); domestic investors showed a far more muted positive response.
- Lockdown measures:
  - Cumulatively, both foreign- and domestic-domiciled investors increased allocations after lockdowns.
  - Only domestic funds experienced a negative initial response; cumulative positive response smaller for domestic funds relative to foreign funds.
  - Suggests negative effects of the economic shutdown weighed more for domestic investors due to greater exposure to domestic shocks.
- Fiscal stimulus:
  - Foreign-domiciled fund flows increased in response to fiscal stimulus; domestic-domiciled flows declined over the one-month horizon.
  - The contemporaneous positive impact of fiscal stimulus on portfolio flows was weaker for domestic-domiciled flows.
- Monetary policy:
  - Negative response of portfolio flows to monetary policy cuts was considerably stronger for domestic investors; domestic allocations fell more in response to a policy rate cut during the COVID episode.
  - Potential explanation: negative impact of monetary easing on the domestic currency exerted additional pressure on domestic funds.

---

### 4.3 Policy measures and the global shock

### Interaction of global shock (VIX) with policy measures
- Global shock measured using log(VIX).
- Estimated coefficients for log(VIX) (Table 6):
  - log(VIX) t = −3.26 (∗∗∗)
  - log(VIX) t (cum h=0:3) = −5.12 (∗∗)
  - log(VIX) t (alt specification h=0) = −1.72 (∗∗∗)
  - log(VIX) t (alt specification cum h=0:3) = −2.30 (∗∗∗)
- Interaction of isCOVID with log(VIX):
  - isCOVID c,t × log(VIX) t = −2.26 (∗∗∗)
  - isCOVID c,t × log(VIX) t (cum) = −7.06 (∗∗∗)

### Lockdown stringency interactions
- Lockdown measures mitigated the adverse impact of the global shock on portfolio flows over the one-month horizon but did not change sensitivity upon impact.
- Quantitative interaction result:
  - Between any two economies that differed in lockdown intensity by one standard deviation, a one percent increase in VIX was associated with an increase in net flows by 6.77 standard deviations in the country with the more stringent lockdown.
- Estimated interactions (Table 6):
  - Lockdown stringency c,t−1 × log(VIX) t = 0.13 (SE 0.24)
  - Lockdown stringency c,t−1 × log(VIX) t (cum) = 6.77 (∗∗∗) (SE 1.12)

### Fiscal stimulus interactions
- Discretionary fiscal spending played a significant role in mitigating the impact of the VIX contemporaneously and exacerbating it over the one-month horizon.
- Quantitative interpretation:
  - Between two countries that differed in fiscal spending by one percent of GDP, a one percent increase in VIX was associated contemporaneously with a 1.00 standard deviation increase in net flows in the country with higher spending.
  - The cumulative effect was negative: net flows were 0.95 standard deviations lower in the country with higher fiscal spending.
- Estimated interactions (Table 6):
  - Fiscal stimulus c,t−1 × log(VIX) t = 1.00 (∗∗∗) (SE 0.12)
  - Fiscal stimulus c,t−1 × log(VIX) t (cum) = −0.95 (∗∗) (SE 0.49)

### Monetary policy interactions
- Expansionary monetary policy actions attenuated the negative impact of the global shock both upon impact and over the one-month horizon.
- Quantitative interaction:
  - Between two countries that differed in their interest rate differential to the U.S. by 100 basis points, a 100 percent increase in VIX was contemporaneously associated with higher net flows by 0.27 standard deviations in the country with the lower rate.
  - The estimated cumulative effect was 0.25 standard deviations higher for the country with the lower interest rate.
- Estimated interactions (Table 6):
  - Policy rate c,t−1 × log(VIX) t = −0.27 (∗∗∗) (SE 0.07)
  - Policy rate c,t−1 × log(VIX) t (cum) = −0.25 (∗∗) (SE 0.11)

### Net interpretation of policy–global shock interactions
- Lockdowns: mitigated cumulative adverse VIX effects (large positive interaction at cum horizon).
- Fiscal stimulus: contemporaneous mitigation but subsequent cumulative amplification of adverse VIX-driven reallocation.
- Monetary easing: consistent attenuation of negative VIX impact both contemporaneously and cumulatively.

---

### 4.4 Pre-COVID macroeconomic conditions and policy space

### Sovereign default risk (CDS spread)
- Higher sovereign default risk (higher CDS spread) led to stronger falls in portfolio flows:
  - For two countries that differ in their CDS spread by one percent, net flows were lower by 0.15 standard deviations upon impact and by 0.47 standard deviations cumulatively in the country with the higher default risk.
- Estimated interaction (Table 7):
  - log(CDS spread) c × isCOVID c,t = −0.15 (∗) (SE 0.08)
  - log(CDS spread) c × isCOVID c,t (cum) = −0.47 (∗∗) (SE 0.21)

### Public debt-to-GDP ratio
- Countries with higher pre-COVID public debt-to-GDP experienced larger net portfolio flows during the pandemic:
  - For two countries that differ in debt-to-GDP by one percentage point, net flows were higher by 0.02 standard deviations cumulatively in the country with the higher debt level.
- Estimated interaction (Table 7):
  - Debt to GDP ratio c × isCOVID c,t = 0.00 (∗) (SE 0.00)
  - Debt to GDP ratio c × isCOVID c,t (cum) = 0.01 (∗) (SE 0.00)

### Reserves and trade openness
- Higher reserves-to-GDP mitigated the contemporaneous decline in flows:
  - Reserves to GDP ratio c × isCOVID c,t = 0.06 (∗) (SE 0.03)
  - Reserves effect cumulatively = 0.09 (SE 0.06) (not starred)
- Trade openness supported flows:
  - For two countries that differ in trade-to-GDP by one percentage point, the more open economy saw larger net portfolio flows by 0.01 standard deviations contemporaneously and by 0.03 standard deviations cumulatively.
- Estimated interaction (Table 7):
  - Trade to GDP ratio c × isCOVID c,t = 0.01 (∗∗∗) (SE 0.00)
  - Trade to GDP ratio c × isCOVID c,t (cum) = 0.03 (∗∗∗) (SE 0.01)

### Fiscal balance and current account
- Conditioning on CDS spreads, government deficit (fiscal balance to GDP ratio) did not significantly affect the magnitude of portfolio flows.
- Current account to GDP ratio had no significant effect (estimates near −0.01 contemporaneous and cumulatively, SEs 0.02 and 0.06).

---

### 4.5 Historical contribution of domestic COVID factors and global shocks (March 4 to May 13, 2020)

### Method summary
- Computed contemporaneous historical contribution of each factor by:
  1. Extracting the shock component of each factor orthogonal to other covariates (residuals from regressions).
  2. Multiplying the residual shocks by estimated elasticities (local projections).
- The exercise focuses on contemporaneous contributions and is suggestive; it abstracts from lagged effects and estimation uncertainty.

### Key historical contributions (emerging markets)
- Global factor (VIX) dominated: its contemporaneous effect was negative and orders of magnitude larger than domestic COVID-related factors, especially in the early and most uncertain phase.
- Domestic COVID infection shocks:
  - Contemporaneous elasticity was negative; contribution was positive early in the sample because countries with slower virus spread (negative COVID surprises) maintained higher flows.
  - In the second half of the sample, as cases grew, positive COVID surprises prompted lower portfolio flows.
- Lockdown stringency:
  - During the height of capital outflows in EMs, stricter lockdowns had a notably positive contemporaneous contribution to flows.
  - Later loosening of lockdowns produced negative contemporaneous contributions in the second half of the sample.
- Monetary policy (policy rate):
  - Contribution was similar across EMs, less dispersed, and largest in early April.
  - Contribution was quite negative during aggressive rate cuts as flows searched for yield.
- Fiscal stimulus:
  - Contribution was positive for the majority of countries but skewed: the distribution had a long right tail, with much smaller magnitude in the lower two quartiles.
- Figure 6 (descriptive): plots median and 25th/75th percentiles of contemporaneous contributions for each factor across weeks between March 4 and May 13, 2020 (weekly flows normalized by pre-COVID standard deviation).

---

### 5 Conclusion and policy implications

### Key empirical findings
- Capital market dynamics were influenced by both global push factors and domestic pandemic severity and policy responses, particularly in EMs.
- Higher domestic COVID cases led to a cumulative increase in flows, particularly in EMs, reflecting widening financing needs.
- Lockdown stringency and fiscal stimulus generally stimulated portfolio investment, with stronger responses in EMs.
- Monetary policy cuts were associated with lower flows, particularly in DMs, consistent with yield-seeking behavior.

### Policy implications
- Public health measures (stringent lockdowns) were rewarded by markets; prioritizing public health can be supported by market responses.
- Countercyclical fiscal policy has an important role: the positive and sizeable contemporaneous effect of fiscal stimulus on portfolio flows suggests building fiscal buffers in good times and maintaining financing access during bad times can help EMs shield against capital flow volatility.
- Monetary accommodation had adverse effects on portfolio flows; coordination of macro policy and attention to fiscal space and sovereign risk dynamics is important.

### Broader note
- While global shocks explained much of the movement in flows, country-specific policy and institutional factors mediated the net historical effects, leading to heterogeneity across countries.

*Source: wpiea2021033-print-pdf*

### 2.1  Portfolio flows

### 2.1  Portfolio flows

### Data, coverage, and measurement
- Data source: weekly portfolio flows from EPFR Global; flows measured as the US$ net value of purchases and redemptions into investment funds.
- EPFR fund domicile: most funds covered are domiciled in DMs.
- Sample coverage: includes 21 DMs and 16 EMs over January 2014 to May 2020.
- EPFR industry coverage (as of September 2020): funds reporting weekly data covered about 74 percent and 70 percent of the global investment equity and bond fund industry, respectively.
- Sample end: second week of May 2020 (sample ends in the second week of May due to a change in the availability of the fiscal spending data obtained from the IMF, after which the IMF switched from weekly to biweekly updates for some countries).
- Interpretation: EPFR flows are interpreted as a proxy for investment fund flows (not the universe of portfolio flows). At the fund level, EPFR flows fully account for valuation effects due to asset returns and exchange rate movements.

### Stylized facts on net portfolio flows during the COVID episode
- Magnitude: portfolio outflows during the pandemic were of a historically large magnitude, markedly exceeding those experienced during earlier episodes (taper tantrum and global financial crisis), with particularly large bond outflows.
- Timing and reversal:
  - Net portfolio flows reversed sharply both in EMs and DMs.
  - DMs saw flows recovering less than two months into the pandemic.
  - Net flows to EMs continued to decline for a longer period.
- Normalization note: the magnitude of the reversal in bond flows was historically unprecedented when measured in US$ but not when normalized by portfolio allocation.
- Cross-country heterogeneity:
  - At the height of the crisis, the standard deviation of net portfolio flows as a share of allocation was 1.1 percent.
  - The distribution of net flows across countries during COVID was wider than during the taper tantrum and similar to the global financial crisis.
- Geographic progression and heterogeneity:
  - Asia-Pacific (both DMs and EMs) started to experience outflows in mid-February (earlier than other regions).
  - Emerging Africa (South Africa only in sample) and Europe saw net flows declining from the last week of February 2020 onward.
  - EMs in the Americas and the Middle East, and other DMs, saw initial outflows by the first week of March—one week before WHO characterized COVID as a pandemic.
  - Outflows from Asia-Pacific were lower than outflows from other regions such as the Americas; this pattern suggests that measures to control the pandemic or mitigate economic fallout might have affected investors’ allocation decisions.
  - Large outflows from the Middle East might also reflect the impact of the pandemic on the oil price collapse.

### Episode timing (figure notes)
- Episode start and end dates (weeks following start date shown on x-axis in Figure 2):
  - COVID: 2020/02/26-2020/04/29
  - Taper tantrum: 2013/05/22-2013/08/14
  - Global financial crisis: 2008/10/08-2008/12/31

### COVID-related controls and policy measures (data sources and variables)
- COVID cases: collected from Haver Analytics; New COVID Cases variable is new domestic infections scaled by population (per capita) and standardized by its sample standard deviation (non-zero values used when computing the standard deviation).
- Lockdown stringency: obtained from Hale et al. (2020); index ranges 0 to 100 and is standardized for analysis.
- Fiscal stimulus: discretionary fiscal spending in response to the pandemic from the IMF’s Survey on Country Responses; measured as percent of GDP.
- Monetary policy: policy rates from Haver Analytics; Policy Rate defined as the policy rate differential to the U.S. Fed Funds rate.
- Sovereign risk: sovereign CDS spreads from Bloomberg.
- Other controls: stock market returns from Haver Analytics.
- Notes on heterogeneity and timing:
  - COVID cases (per 1,000 population) were more heterogeneous for DMs than EMs in March and April; as cases picked up, EMs showed wide dispersion.
  - Lockdown stringency was relatively lenient until end-March for both DMs and EMs and then increased sharply in April with response becoming more homogeneous.
  - Fiscal spending (first-round stimulus) largely concentrated up to 5 percent of GDP, but some countries spent more sizeable amounts.
  - Policy rate differentials were relatively homogeneous for DMs and more scattered for EMs.

### Summary statistics of COVID cases and policy measures (observations over 2020/01/22-2020/05/13)
- Developed markets (N = 357)
  - COVID-19 cases (per 1000 pop.): Mean 0.7; St. Dev. 1.2; Min 0.0; Median 0.15; Max 5.0
  - Lockdown stringency (0 to 100): Mean 44.3; St. Dev. 33.2; Min 0.0; Median 45.8; Max 96.0
  - Fiscal spending (percent of GDP) (N = 100): Mean 4.6; St. Dev. 3.1; Min 1.1; Median 3.9; Max 11.2
  - Policy rate differential (pp): Mean −0.6; St. Dev. 0.8; Min −2.4; Median −0.1; Max 1.6
- Emerging markets (N = 272)
  - COVID-19 cases (per 1000 pop.): Mean 0.2; St. Dev. 0.8; Min 0.0; Median 0.0; Max 19.0
  - Lockdown stringency (0 to 100): Mean 48.1; St. Dev. 35.7; Min 0.0; Median 56.9; Max 97.0
  - Fiscal spending (percent of GDP) (N = 76): Mean 3.5; St. Dev. 2.6; Min 0.2; Median 2.7; Max 9.8
  - Policy rate differential (pp): Mean 2.8; St. Dev. 2.9; Min −1.6; Median 2.4; Max 11.4

### Empirical approach: objectives and identification
- Objective: empirically evaluate the impact of countries’ exposure to the pandemic and key policy measures (lockdown stringency, fiscal stimulus, policy rate differential) on capital markets (net portfolio flows and sovereign CDS spreads).
- Frequency and method: weekly frequency; local projections approach (Jordà, 2005) to estimate contemporaneous and dynamic cumulative effects (horizons h = 0 and cumulative h = 0,...,3).
- Dependent variables: net portfolio flows (scaled by pre-COVID allocation and standardized by its pre-COVID sample standard deviation) and sovereign CDS spreads.
- Scaling and expectations:
  - Portfolio flows scaled by pre-COVID allocation (allocation value reported for week of February 26, 2020).
  - E_{t−1}{Y_{c,t+h}} term captures the expected component of Y_{c,t+h} using one lag of portfolio flows, sovereign CDS spreads, stock market returns, interest rate differential with the U.S., total domestic COVID-19 cases, and log portfolio allocation.
- Fixed effects and controls:
  - Country fixed effect η_c; week fixed effects η_DMs,t and η_EMs,t to allow heterogeneous global shocks across DMs and EMs; country×month fixed effect η_c×η_m to control for slow-moving domestic factors.
- Endogeneity and identification strategy:
  - Identification rests on high frequency of data, rapid adjustment of portfolio flows to new information, and comprehensive controls and fixed effects.
  - New COVID cases treated as exogenous conditional on the specification, aided by inclusion of lagged flows, stock returns, and country×month fixed effects.
  - Lockdown, fiscal, and monetary measures assumed contemporaneously exogenous due to decision lags; dynamic endogeneity mitigated by lagged controls and fixed effects.
  - Conservative specification includes country×month fixed effects to control for institutional variations, capital controls, and slow-moving domestic shocks.

### Empirical model (baseline regression, local projections)
- Baseline equation (local projections for horizon h):
  - Y_{c,t+h} = E_{t−1}{Y_{c,t+h}} + β^h_1 New COVID Cases_{c,t} + β^h_2 Lockdown_{c,t} + β^h_3 Fiscal Stimulus_{c,t} + β^h_4 Policy Rate_{c,t} + β^h_5 isCOVID_{c,t} + β^h_6 Policy Rate_{c,t} × isCOVID_{c,t} + η_c + η_DMs,t+h + η_EMs,t+h + η_c×η_m + e_{c,t+h}
- Horizons analyzed: contemporaneous h = 0 and cumulative over h = 0,...,3 (monthly cumulative effect computed as ∑_{h=0}^3 Y_{c,t+h}).
- isCOVID_{c,t}: dummy = 1 when number of COVID cases > 0 in country-week; 0 otherwise.
- Rationale for CDS focus: sovereign CDS spreads used as a measure of risk given dominant role of sovereign borrowing in EMs.

### Model scaling and dynamics
- Net portfolio flows scaled by pre-COVID allocation to ensure estimates are not driven by market size differences; ratio further scaled by its pre-COVID sample standard deviation.
- Expectation term E_{t−1}{Y_{c,t+h}} uses one lag of selected financial and COVID variables to capture predictable component and help mitigate endogeneity.

### Preview of results (section summary)
- Heterogeneity in countries’ exposure to COVID-19 and related policy measures significantly influenced the dynamics of portfolio flows and CDS spreads during the COVID episode.
- Results are analyzed by:
  - DMs vs EMs,
  - bond vs equity flows,
  - foreign vs domestic investors,
  - interactions between domestic COVID-related factors and global factors,
  - role of pre-existing macroeconomic fundamentals.

*Source: wpiea2021033-print-pdf - 2.1  Portfolio flows*

### 4.1  COVID-19 related domestic factors as drivers of portfolio flows

### 4.1  COVID-19 related domestic factors as drivers of portfolio flows

### Empirical strategy and identification
- Analysis focuses on foreign-domiciled flows (funds domiciled outside the recipient country).
- Preferred (most conservative) specification controls for:
  - country fixed effects,
  - week fixed effects (global factors) with two independent time trends for DMs and EMs,
  - country×month fixed effects (country-level slow-moving variations),
  - expected net portfolio flows (function of lagged flows and other lagged determinants, including sovereign risk and domestic stock prices).
- Two key observations from comparing specifications:
  - Estimated coefficients on domestic factors change considerably once global time fixed effects are included, indicating a common global component in COVID-related domestic variables.
  - Adjusted R2 rises substantially when adding country×month fixed effects, pointing to the importance of pull (domestic) factors in explaining contemporaneous variations in portfolio flows.

### COVID-19 cases
- Aggregate results (columns 5–6, table 2):
  - Higher new domestic COVID cases per capita: insignificant effect on portfolio flows upon impact (h = 0, column 5).
  - Over one-month horizon (cum(h=0:3), column 6): a one standard deviation increase in new domestic COVID cases per capita associated with a 1.13 standard deviation increase in foreign net portfolio flows (as a ratio of total allocation).
- Heterogeneity by market and asset class:
  - The positive average effect driven mainly by bond flows in EMs.
  - Equity flows experienced a cumulative decline in both EMs and DMs.
  - Net bond flows declined upon impact of the COVID shock, indicating an initial perception of higher COVID cases as bad news.
- Sovereign CDS spread response (table 4):
  - A one standard deviation increase in new COVID cases per capita associated with a significant increase in growth of CDS spreads by 0.18 percent in EMs and 0.03 percent in DMs over the one-month horizon.
  - Interpretation: increased demand for financing (larger supply of bonds) in affected economies was a dominant force behind cumulative increase in bond flows; bond flows increased together with a rise in the cost of risk.

### Lockdown measures
- Aggregate sample (columns 5–6, table 2):
  - A one standard deviation change in lockdown stringency led to an increase in cumulative net portfolio flows by 0.32 standard deviations over a one-month horizon in the average country, after an initial decline.
- Market heterogeneity:
  - The cumulative one-month increase driven by EMs; response in DMs was insignificant (columns 7–8, table 2).
  - China was the main determinant of the strong positive cumulative effect in EMs, being an early adopter of relatively strict lockdowns that controlled the pandemic sooner and contributed to investor confidence.
- Asset-class response (table 3):
  - Increase in foreign fund flows into EMs after the initial decline dominated by bond flows (reallocation to safety).
  - Equity flows to EMs experienced a significant decline.
- CDS spreads (table 4):
  - A stricter lockdown associated with lower CDS spreads over the one-month horizon in both EMs and DMs.
  - Interpretation: stricter lockdowns triggered increased demand from EM investors, especially for safer assets.

### Discretionary fiscal stimulus
- Aggregate effect (columns 5–6, table 2):
  - Fiscal stimulus: initially negative impact, but over one month, increases in fiscal spending perceived positively, attracting portfolio flows.
- Magnitude and heterogeneity:
  - For every one percent increase in fiscal spending as a share of GDP:
    - Net flows increased by 1.65 standard deviations in EMs.
    - Net flows increased by 0.33 standard deviations in DMs.
  - Effect was strong for both bond and equity flows and primarily driven by EMs (table 3).
- CDS spreads:
  - Response in sovereign CDS spreads was not significant over the one-month horizon (table 4), providing no clear indication on supply vs. demand drivers of cumulative flow increases.
- Mechanical vs. non-mechanical channels:
  - Many governments financed stimulus through international borrowing; estimated increase in flows might partially reflect a mechanical effect.
  - Evidence that non-mechanical response, driven by improved economic outlook, played an important role: net equity flows increased in response to stimulus, not only net bond flows.

### Discretionary monetary policy (short-term policy rate)
- Average effect over full sample (table 2, columns 5–6):
  - Changes in the policy rate relative to the U.S. do not play a statistically significant role in dynamics of portfolio flows on average.
- Heterogeneity and COVID episode effects:
  - Insignificance over full sample driven by EMs (columns 7–8, table 2).
  - During the COVID episode:
    - A hundred basis points cut in the interest rate led to a cumulative decline in net portfolio flows by 0.21 standard deviations in EMs (no significant immediate impact).
    - In DMs, a hundred basis points cut was associated with an increase in net flows by 0.64 standard deviations upon impact, suggesting an initial reassuring effect; cumulatively, however, a rate cut led to a decline in net portfolio flows in DMs, with the cumulative response larger than in EMs.
- Asset-class drivers (table 3):
  - In EMs, the cumulative decline in total net portfolio flows after a rate cut was entirely driven by bond flows (consistent with interest rate parity).
  - In DMs, both bond and equity flows responded negatively to a rate cut, with larger elasticity for equity flows.
- CDS returns and interpretation (table 4 and text):
  - Expansionary EM monetary policy associated with an increase in CDS returns during the COVID episode: a 3.14 percent increase for every hundred basis point cut in the policy rate.
  - CDS returns increased largely upon impact.
  - Findings inconsistent with classical credit channel; may reflect an information channel where aggressive rate cuts temporarily fueled market concerns about the expected fallout from the pandemic.

### Response of domestic-domiciled portfolio flows
- Domestic vs. foreign investor behavior:
  - Domestic investors have more vulnerable balance sheets to idiosyncratic shocks and are more likely to hold local-currency assets; foreign investors allocate more to foreign-denominated assets.
  - Overall finding: domestic and foreign investors responded differently to the pandemic and policy measures, with domestic flows generally lower relative to foreign flows.
- COVID cases:
  - Foreign investors eventually increased holdings in response to higher new COVID cases (potentially meeting domestic financing needs); domestic investors showed a far more muted positive response.
- Lockdown measures:
  - Cumulatively, both foreign- and domestic-domiciled investors increased allocations after lockdowns.
  - Only domestic funds experienced a negative initial response; cumulative positive response smaller for domestic funds relative to foreign funds.
  - Suggests negative effects of the economic shutdown weighed more for domestic investors due to greater exposure to domestic shocks.
- Fiscal stimulus:
  - Foreign-domiciled fund flows increased in response to fiscal stimulus; domestic-domiciled flows declined over the one-month horizon.
  - The contemporaneous positive impact of fiscal stimulus on portfolio flows was weaker for domestic-domiciled flows.
- Monetary policy:
  - Negative response of portfolio flows to monetary policy cuts was considerably stronger for domestic investors; domestic allocations fell more in response to a policy rate cut during the COVID episode.
  - Potential explanation: negative impact of monetary easing on the domestic currency exerted additional pressure on domestic funds.

*Source: wpiea2021033-print-pdf - 4.1  COVID-19 related domestic factors as drivers of portfolio flows*

### 4.3  Policy measures and the global shock

### 4.3  Policy measures and the global shock

### Interaction of global shock (VIX) with policy measures
- The analysis uses log(VIX) as a continuous market-based measure of the evolving intensity of the global shock.
- Estimated coefficients for log(VIX) (Table 6):
  - log(VIX) t = −3.26 (∗∗∗)
  - log(VIX) t (cum h=0:3) = −5.12 (∗∗)
  - log(VIX) t (alt specification h=0) = −1.72 (∗∗∗)
  - log(VIX) t (alt specification cum h=0:3) = −2.30 (∗∗∗)
- The interaction of isCOVID with log(VIX):
  - isCOVID c,t × log(VIX) t = −2.26 (∗∗∗)
  - isCOVID c,t × log(VIX) t (cum) = −7.06 (∗∗∗)

### Lockdown stringency
- Lockdown measures mitigated the adverse impact of the global shock on portfolio flows over the one-month horizon but did not change sensitivity upon impact.
- Between any two economies that differed in lockdown intensity by one standard deviation, a one percent increase in VIX was associated with an increase in net flows by 6.77 standard deviations in the country with the more stringent lockdown.
- Estimated interactions (Table 6):
  - Lockdown stringency c,t−1 × log(VIX) t = 0.13 (SE 0.24)
  - Lockdown stringency c,t−1 × log(VIX) t (cum) = 6.77 (∗∗∗) (SE 1.12)

### Fiscal stimulus
- Discretionary fiscal spending played a significant role in mitigating the impact of the VIX contemporaneously and exacerbating it over the one-month horizon.
- Between two countries that differed in fiscal spending by one percent of GDP, a one percent increase in VIX was associated contemporaneously with a 1.00 standard deviation increase in net flows in the country with higher spending.
- The cumulative effect was negative: net flows were 0.95 standard deviations lower in the country with higher fiscal spending.
- Estimated interactions (Table 6):
  - Fiscal stimulus c,t−1 × log(VIX) t = 1.00 (∗∗∗) (SE 0.12)
  - Fiscal stimulus c,t−1 × log(VIX) t (cum) = −0.95 (∗∗) (SE 0.49)

### Monetary policy (policy rate)
- Expansionary monetary policy actions attenuated the negative impact of the global shock both upon impact and over the one-month horizon.
- Between two countries that differed in their interest rate differential to the U.S. by 100 basis points, a 100 percent increase in VIX was contemporaneously associated with higher net flows by 0.27 standard deviations in the country with the lower rate.
- The estimated cumulative effect was 0.25 standard deviations higher for the country with the lower interest rate.
- Estimated interactions (Table 6):
  - Policy rate c,t−1 × log(VIX) t = −0.27 (∗∗∗) (SE 0.07)
  - Policy rate c,t−1 × log(VIX) t (cum) = −0.25 (∗∗) (SE 0.11)

### Net interpretation of policy–global shock interactions
- Lockdowns: mitigated cumulative adverse VIX effects (large positive interaction at cum horizon).
- Fiscal stimulus: contemporaneous mitigation but subsequent cumulative amplification of adverse VIX-driven reallocation.
- Monetary easing: consistent attenuation of negative VIX impact both contemporaneously and cumulatively.

---

### 4.4  Pre-COVID macroeconomic conditions and policy space

### Sovereign default risk (CDS spread)
- Higher sovereign default risk (higher CDS spread) led to stronger falls in portfolio flows:
  - For two countries that differ in their CDS spread by one percent, net flows were lower by 0.15 standard deviations upon impact and by 0.47 standard deviations cumulatively in the country with the higher default risk.
  - Estimated interaction (Table 7):
    - log(CDS spread) c × isCOVID c,t = −0.15 (∗) (SE 0.08)
    - log(CDS spread) c × isCOVID c,t (cum) = −0.47 (∗∗) (SE 0.21)

### Public debt-to-GDP ratio
- Countries with higher pre-COVID public debt-to-GDP experienced larger net portfolio flows during the pandemic:
  - For two countries that differ in debt-to-GDP by one percentage point, net flows were higher by 0.02 standard deviations cumulatively in the country with the higher debt level.
  - Estimated interaction (Table 7):
    - Debt to GDP ratio c × isCOVID c,t = 0.00 (∗) (SE 0.00)
    - Debt to GDP ratio c × isCOVID c,t (cum) = 0.01 (∗) (SE 0.00)

### Reserves and trade openness
- Higher reserves-to-GDP mitigated the contemporaneous decline in flows:
  - Reserves to GDP ratio c × isCOVID c,t = 0.06 (∗) (SE 0.03)
  - Reserves effect cumulatively = 0.09 (SE 0.06) (not starred)
- Trade openness supported flows:
  - For two countries that differ in trade-to-GDP by one percentage point, the more open economy saw larger net portfolio flows by 0.01 standard deviations contemporaneously and by 0.03 standard deviations cumulatively.
  - Estimated interaction (Table 7):
    - Trade to GDP ratio c × isCOVID c,t = 0.01 (∗∗∗) (SE 0.00)
    - Trade to GDP ratio c × isCOVID c,t (cum) = 0.03 (∗∗∗) (SE 0.01)

### Fiscal balance and current account
- Conditioning on CDS spreads, government deficit (fiscal balance to GDP ratio) did not significantly affect the magnitude of portfolio flows.
- Current account to GDP ratio had no significant effect (estimates near −0.01 contemporaneous and cumulatively, SEs 0.02 and 0.06).

---

### 4.5  Historical contribution of domestic COVID factors and global shocks (March 4 to May 13, 2020)

### Method summary
- Computed contemporaneous historical contribution of each factor by:
  1. Extracting the shock component of each factor orthogonal to other covariates (residuals from regressions).
  2. Multiplying the residual shocks by estimated elasticities (local projections).
- The exercise focuses on contemporaneous contributions and is suggestive; it abstracts from lagged effects and estimation uncertainty.

### Key historical contributions (emerging markets)
- Global factor (VIX) dominated: its contemporaneous effect was negative and orders of magnitude larger than domestic COVID-related factors, especially in the early and most uncertain phase.
- Domestic COVID infection shocks:
  - Contemporaneous elasticity was negative; contribution was positive early in the sample because countries with slower virus spread (negative COVID surprises) maintained higher flows.
  - In the second half of the sample, as cases grew, positive COVID surprises prompted lower portfolio flows.
- Lockdown stringency:
  - During the height of capital outflows in EMs, stricter lockdowns had a notably positive contemporaneous contribution to flows.
  - Later loosening of lockdowns produced negative contemporaneous contributions in the second half of the sample.
- Monetary policy (policy rate):
  - Contribution was similar across EMs, less dispersed, and largest in early April.
  - Contribution was quite negative during aggressive rate cuts as flows searched for yield.
- Fiscal stimulus:
  - Contribution was positive for the majority of countries but skewed: the distribution had a long right tail, with much smaller magnitude in the lower two quartiles.
- Figure 6 (descriptive): plots median and 25th/75th percentiles of contemporaneous contributions for each factor across weeks between March 4 and May 13, 2020 (weekly flows normalized by pre-COVID standard deviation).

---

### 5  Conclusion and policy implications
- Capital market dynamics were influenced by both global push factors and domestic pandemic severity and policy responses, particularly in EMs.
- Key empirical findings:
  - Higher domestic COVID cases led to a cumulative increase in flows, particularly in EMs, reflecting widening financing needs.
  - Lockdown stringency and fiscal stimulus generally stimulated portfolio investment, with stronger responses in EMs.
  - Monetary policy cuts were associated with lower flows, particularly in DMs, consistent with yield-seeking behavior.
- Policy implications:
  - Public health measures (stringent lockdowns) were rewarded by markets; prioritizing public health can be supported by market responses.
  - Countercyclical fiscal policy has an important role: the positive and sizeable contemporaneous effect of fiscal stimulus on portfolio flows suggests building fiscal buffers in good times and maintaining financing access during bad times can help EMs shield against capital flow volatility.
  - Monetary accommodation had adverse effects on portfolio flows; coordination of macro policy and attention to fiscal space and sovereign risk dynamics is important.
- Broader note:
  - While global shocks explained much of the movement in flows, country-specific policy and institutional factors mediated the net historical effects, leading to heterogeneity across countries.

*Source: wpiea2021033-print-pdf - 4.3  Policy measures and the global shock*

### References

### References

### References
- Alfaro, Laura et al. (2020).Aggregate and Firm-Level Stock Returns During Pandemics, in Real Time. Covid Economics: Vetted and Real-Time Papers, CEPR.
- Avdjiev, Stefan et al. (2020). “The shifting drivers of global liquidity”.Journal of International Economics125, p. 103324.
- Berk, Jonathan B and Richard C Green (2004). “Mutual fund flows and performance in rational markets”.Journal of political economy112.6, pp. 1269–1295.
- Broner, Fernando et al. (2013). “Gross capital flows: Dynamics and crises”.Journal of Monetary Economics60.1, pp. 113 –133.
- Caballero, Ricardo J and Alp Simsek (2020). “A Model of Fickle Capital Flows and Retrenchment”. Journal of Political Economy128.6, pp. 2288–2328.
- Çakmakli, Cem et al. (2020).COVID-19 and Emerging Markets: An Epidemiological Model with International Production Networks and Capital Flows. IMF Working Paper 20/133.
- Chari, Anusha, Karlye Dilts Stedman, and Christian T. Lundblad (2020).Capital Flows in Risky Times: Risk-On/Risk-Off and Emerging Market Tail Risk. NBER Working Paper 27927.
- Davis, Scott D., Giorgio Valente, and Eric van Wincoop (2019).Global Capital Flows Cycle: Impact on Gross and Net Flows. NBER Working Paper 25721.
- Esteves, Rui and Nathan Sussman (2020).The Corona Spreads. Graduate Institute of International and Development Studies.
- Forbes, Kristin and Francis Warnock (2011). “Capital Flow Waves: Surges, Stops, Flight and Re-trenchment”.Journal of International Economics88.2, pp. 235–251.
- Fratzscher, Marcel (2012). “Capital flows, push versus pull factors and the global financial crisis”. Journal of International Economics88.2, pp. 341–356.
- Fratzscher, Marcel, Marco Lo Duca, and Roland Straub (2018). “On the international spillovers of US quantitative easing”.The Economic Journal128.608, pp. 330–377.
- Gilchrist, Simon and Egon Zakrajšek (2012). “Credit spreads and business cycle fluctuations”. American Economic Review102.4, pp. 1692–1720.
- Hale, Thomas et al. (2020).Oxford COVID-19 Government Response Tracker. Blavatnik School of Government.
- Jordà, Òscar (2005). “Estimation and inference of impulse responses by local projections”.American Economic Review95.1, pp. 161–182.
- Jotikasthira, Chotibhak, Christian Lundblad, and Tarun Ramadorai (2012). “Asset Fire Sales and Purchases and the International Transmission of Funding Shocks”.The Journal of Finance67.6, pp. 2015–2050.
- Kalemli-Özcan, Şebnem (2019).U.S. Monetary Policy and International Risk Spillovers. Prepared for the Jackson Hole Economic Policy Symposium, Federal Reserve Bank of Cansas City.
- Kilian, Lutz and Helmut Lütkepohl (2017).Structural vector autoregressive analysis. Cambridge University Press.
- Koepke, Robin and Simon Paetzold (2020).Capital Flow Data—A Guide For Empirical Analysis and Real-Time Tracking. IMF Working Paper 20/171.
- Maggiori, Matteo, Brent Neiman, and Jesse Schreger (2020). “International Currencies and Capital Allocation”.Journal of Political Economy128.6, pp. 2019–2066.
- Miao, Yanlian and Malika Pant (2012).Coincident Indicators of Capital Flows. IMF Working Paper 12/55.
- Milesi-Ferretti, Gian-Maria et al. (2011). “The great retrenchment: international capital flows during the global financial crisis”.Economic Policy26.66, pp. 285–342.
- Nakamura, Emi and Jón Steinsson (2018). “High-frequency identification of monetary non-neutrality: the information effect”.The Quarterly Journal of Economics133.3, pp. 1283–1330.
- Nickell, Stephen (1981). “Biases in dynamic models with fixed effects”.Econometrica: Journal of the Econometric Society, pp. 1417–1426.
- Rey, Hélène (2015).Dilemma not trilemma: the global financial cycle and monetary policy independence. National Bureau of Economic Research.
- Sarno, Lucio, Ilias Tsiakas, and Barbara Ulloa (2016). “What drives international portfolio flows?” Journal of International Money and Finance60, pp. 53–72.

### Appendix to Capital Markets, COVID-19, and Policy Measures

#### A  Data
- Table A.1: Country classification

- Developed markets
  - AustraliaAustriaBelgiumCanada
  - DenmarkFinlandFranceGermany
  - Hong Kong SAR  IrelandIsraelItaly
  - JapanNetherlands   New ZealandNorway
  - PortugalSpainSwedenSwitzerland
  - United Kingdom

- Emerging markets
  - BrazilChinaCzech Republic  Greece
  - IndonesiaMalaysiaMexicoPakistan
  - PeruPhilippines    PolandQatar
  - Saudi ArabiaSouth Africa  ThailandTurkey

- Notes: Countries are categorized according to the 2020 MSCI developed and emerging market classification.

#### B  Additional figures
- Figure B.1: Cumulative equity and bond flows (percent of allocation)
- Notes: x-axis shows weeks after the start date for the following episodes: COVID: 2020/02/26-2020/04/29; taper tantrum: 2013/05/22-2013/08/14; global financial crisis: 2008/10/08-2008/12/31. Flows are normalized by the allocation at the start of each episode.

*Content derived from wpiea2021033-print-pdf - References*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021033-print-pdf.pdf_