## ch2annex

## Source details

**Canonical URL:** [ch2annex](https://www.imf.org/-/media/files/publications/gfsr/2026/april/english/ch2annex.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/gfsr/2026/april/english/ch2annex.pdf.md)
- [Structured JSON version](/-/media/files/publications/gfsr/2026/april/english/ch2annex.pdf.json)

---

### External portfolio debt and equity liabilities — overview, data, and measurement
- Purpose:
  - Quantifies how portfolio debt and equity liability flows to emerging markets respond to global risk factors.
  - Compares sensitivities before and after the global financial crisis (GFC).
  - Contrasts NBFI (non-bank financial intermediation) flows with banking flows and flows to advanced economies.
- Data, sample, and measurement:
  - Data sources: IMF Balance of Payments (PIP), BIS LBS, EPFR, Bank for International Settlements, International Debt Statistics, and IMF staff calculations.
  - Baseline sample: 20 emerging markets (ARG, BGR, BRA, CHL, CHN, COL, HUN, IDN, IND, MEX, MYS, PER, PHL, POL, ROU, RUS, THA, TUR, UKR, ZAF).
  - Expanded sample: 41 emerging markets (adds AGO, ALB, BLR, BOL, CRI, DOM, ECU, EGY, GHA, GTM, JAM, JOR, KAZ, LBN, LKA, MAR, MKD, PAK, TUN, URY, VNM).
  - Sample period: 2003Q2-2025Q1 (post-GFC comparisons primarily 2009Q2 onward).
  - Flows normalized by country c’s seasonally-adjusted quarterly GDP at t-1 and standardized by flow type.
  - Global risk factors: VIX (baseline) and Risk-on/Risk-off (RORO); also evaluated MOVE, World Uncertainty index, and US dollar index (DXY).
  - Banking flows: FX- and break-adjusted cross-border claims in BIS LBS.
  - Investment-fund flows proxied by EPFR country-level fund flows (mutual funds and ETFs).

### Summary statistics (2003Q2–2025Q1; percent of lagged GDP unless noted)
- Banking inflows: count 1740; mean 0.614; sd 3.982; min -41.583; p25 -1.140; p50 0.352; p75 2.009; max 29.838.
- Overall portfolio debt liability flows: count 1740; mean 0.992; sd 3.048; min -19.212; p25 -0.499; p50 0.464; p75 2.152; max 19.792.
- Overall portfolio equity liability flows: count 1740; mean 0.231; sd 1.351; min -21.728; p25 -0.202; p50 0.064; p75 0.558; max 10.323.
- Investment fund debt inflows: count 1664; mean 0.233; sd 1.385; min -7.151; p25 -0.283; p50 0.084; p75 0.724; max 8.840.
- Investment fund equity inflows: count 1740; mean 0.082; sd 0.375; min -1.635; p25 -0.080; p50 0.010; p75 0.197; max 2.552.
- Distributional notes:
  - Portfolio debt liability flows average 1 percent of GDP (upper quartile > 2 percent; max up to 20 percent).
  - Portfolio equity flows average 0.2 percent of GDP (upper quartile ~0.5 percent; max up to 10 percent).
  - Banking flows average 0.6 percent of GDP (can reach up to 30 percent of GDP).
- Cross-flow standard deviations: banking sd 4.0; portfolio debt sd 3.0; portfolio equity sd 1.4.

### Baseline empirical approach and identification
- Baseline regression (Equation 2.3.1): sensitivity of standardized flows Yc,t^k to global risk factor (VIX) and post-2009Q2 dummy D_ppp_t, with country-year fixed effects η_c,t(y).
- Identification from quarterly variation in global risk factor.
- Inference: Driscoll-Kraay standard errors to account for cross-sectional dependence and serial correlation.

### Main empirical findings: portfolio vs banking flows and role of NBFIs
- Post-GFC relative sensitivity:
  - Sensitivity of portfolio debt inflows to emerging markets increased relative to banking flows in the post-2009Q2 period.
  - Portfolio equity liability flows show a similar post-GFC pattern, though differences are not always statistically significant.
- Magnitudes (selected reported comparisons):
  - Using RORO in the post-GFC period, a one-standard-deviation increase in RORO → decline of about 0.3 standard deviations in portfolio debt flows (comparable to VIX effect).
  - Corresponding response of portfolio equity inflows to RORO ≈ 0.3 standard deviations, exceeding sensitivity to the VIX.
  - For investment funds, a one-standard-deviation increase in RORO → declines of about 0.9 (portfolio debt) and 0.5 (portfolio equity) standard deviations in the post-GFC period.
- Comparison with banking flows:
  - Pre-GFC: portfolio debt flows less sensitive to VIX than banking inflows.
  - Post-GFC: portfolio debt sensitivity increased relative to banking inflows, becoming statistically similar.
  - Using RORO: both portfolio debt and equity liability flows significantly more sensitive to global risk than banking flows.
  - Banking flows’ sensitivity to global risk declines significantly post-GFC and becomes statistically indistinguishable from zero in several specifications.

### Robustness checks
- Alternative risk measures (RORO, MOVE, World Uncertainty, DXY) yield qualitatively similar results; RORO often shows larger sensitivities for portfolio and investment-fund flows than the VIX.
- Expanded sample (41 EMs): sensitivity of portfolio debt flows to a one-standard-deviation VIX is smaller (~1¼ standard deviations) than baseline; portfolio equity and banking flow sensitivities broadly similar.
- Global and country controls (Equation 2.3.2) attenuate but do not eliminate portfolio debt sensitivity to VIX; banking flows’ sensitivity declines and becomes insignificant.
- Specification comparing flow types jointly (Equation 2.3.3) confirms greater post-GFC sensitivity of portfolio debt relative to banking flows.
- Results robust to excluding China, India, Brazil/China/India/Mexico, and across regional subsamples.

---

### Investment-fund (NBFI) flows — sensitivity, nonlinearity, and investor types
- Investment-fund (EPFR) flows:
  - More sensitive to global risk than overall portfolio debt liability flows in the post-2009Q2 period.
  - Baseline: one-standard-deviation increase in RORO → declines of about 0.9 (debt) and 0.5 (equity) standard deviations in investment-fund flows post-GFC.
- EPFR caveats:
  - Captures mutual funds and ETFs; excludes sovereign wealth funds, pension funds outside mutual vehicles, insurance companies, and banks’ trading books.
  - Fund domicile may not equal ultimate investor residence.
- Nonlinear effects (Equation 2.3.5; evaluated 2009Q2-2025Q1, 20 EMs):
  - Portfolio debt flows relatively unresponsive when VIX below its historical mean (~20).
  - Sensitivity increases as VIX rises; strong reaction at higher VIX levels.
  - Portfolio debt flows exhibit non-linear response to RORO as well.
- Quantile (capital-flows-at-risk) approach (2.3.6):
  - Post-GFC, portfolio debt flows intermediated by investment funds exhibit pronounced non-linear effects.
  - Increases in global risk have stronger effects during large outflows and strongest during periods of large inflows.

### Country policies and vulnerabilities (interaction specification 2.3.7; 20 EMs, 2009Q2-2025Q1)
- Additional sensitivity of portfolio debt inflows to a one-standard-deviation VIX increase is larger in countries with weaker fundamentals vs stronger fundamentals.
- Among institutional quality components, government stability is the primary driver of differential sensitivity.
- Even some countries with stronger fundamentals may see lower inflows with higher VIX.
- Specific numeric association: a one-standard-deviation increase in the VIX is associated with a one-standard-deviation decline in investment-fund portfolio debt flows (statements preserved as in source).
- Using RORO yields similar qualitative patterns; stronger institutions and external buffers mute responses; higher public debt associated with greater sensitivity.
- Exchange rate regime classification based on Ilzetzki, Reinhart and Rogoff (2021).

### Which NBFI investor types are most sensitive? — FactSet Ownership coverage
- FactSet captures roughly 65 percent of total portfolio equity positions intermediated by non-bank financial institutions in the PIP.
- For debt securities, coverage reached close to 25 percent in 2024.
- For investment funds: up to 70 percent coverage for equity holdings and more than 30 percent for debt holdings.
- Emerging market securities holdings of NBFIs satisfying sample restrictions account for 74 percent of total in the data.
- Coverage by investor type (selected 2024:Q4 holdings in trillions of US dollars):
  - Mutual fund (Advanced economies): Number nonresident 7,703; Number resident 7,482; Holdings nonresident 17.11; Holdings resident 37.57.
  - Mutual fund (Emerging markets): Number nonresident 4,357; Number resident 1,258; Holdings nonresident 2.02; Holdings resident 1.91.
  - Hedge fund (Advanced economies): Number nonresident 2,893; Number resident 2,889; Holdings nonresident 0.84; Holdings resident 1.79.
  - Hedge fund (Emerging markets): Number nonresident 1,523; Number resident 74; Holdings nonresident 0.06; Holdings resident 0.01.
  - Pension fund (Advanced economies): Number nonresident 208; Number resident 162; Holdings nonresident 0.90; Holdings resident 1.12.
  - Pension fund (Emerging markets): Number nonresident 135; Number resident 49; Holdings nonresident 0.08; Holdings resident 0.11.
  - Insurance company (Advanced economies): Number nonresident 167; Number resident 158; Holdings nonresident 0.05; Holdings resident 0.10.
  - Insurance company (Emerging markets): Number nonresident 96; Number resident 78; Holdings nonresident 0.01; Holdings resident 0.24.
  - Other nonbanks (Advanced economies): Number nonresident 140; Number resident 153; Holdings nonresident 1.38; Holdings resident 0.08.
  - Other nonbanks (Emerging markets): Number nonresident 70; Number resident 21; Holdings nonresident 0.14; Holdings resident 0.05.

---

### Emerging market security holdings of nonbanks — methodology and key empirical findings
- Main outcome: change in the share of a security held by an investor, defined as the change in holdings not attributable to valuation changes (equation (2.4.1)); change defined only when H...t−1 > 0 and S...t > 0.
- Baseline specification (2.4.3): sensitivity of nonresident relative to resident holdings to global risk factor (GRF), with issuer-security-type-time (γf,s,t) and issuer-investor-security-type (δf,j,s) fixed effects; coefficient β measures relative sensitivity.
- Flightiness measure (2.4.5): share of holdings by investors whose statistically significant βj is below threshold β̄ (β̄ = 0 for securities; β̄ = 20th percentile at country level).
- Portfolio rebalancing specification (2.4.6): examines rotations across corporate/sovereign bond/equity using asset-class controls and fixed effects.
- Sample restrictions:
  - Issuer-investor pair must have at least 25 quarters of holdings.
  - Only investors with emerging market security holdings > USD 10 million included.

Key empirical findings
- Nonresident NBFIs reduce emerging market investments more than resident NBFIs when global risk aversion rises.
  - Differential response significant for emerging market economies; not significant for advanced-economy investments (advanced-economy sample excludes Japan, Switzerland, United States).
  - When VIX is elevated (top decile), nonresident reductions in EM holdings more pronounced.
- Sensitivity to US dollar:
  - A one-standard-deviation appreciation of the US dollar → decrease of 1.4 percentage points in nonresident emerging market investments — substantially larger than effect from analogous VIX increase.
- Flighty investors:
  - 167 nonresident NBFIs with statistically significant negative βj at 5 percent: 120 mutual funds and 32 hedge funds.
  - These flighty investors account for 23 percent of emerging market bond holdings and 12 percent of equity holdings of NBFIs in the sample.
- State dependence and market depth:
  - Nonresident EM investments more sensitive to VIX when it increases from a low level than when it decreases from a high level (differences not always statistically significant).
  - Nonresident reductions larger in countries with shallower financial markets (private debt securities to GDP proxy).
- Recipient vulnerabilities amplify flighty investor responses:
  - Flightier nonresident investors cut EM investments more after VIX increases in countries with lower international reserves, weaker institutions, and higher sovereign credit risk.
  - Higher foreign exchange reserves substantially moderate fall in nonresident investments, particularly in countries with managed exchange rates.
  - Differences between high and low vulnerability groups statistically significant for institutional quality and reserve adequacy in managed exchange rate countries.

Robustness
- More saturated fixed effects (time–security) yield similar β estimates; advanced-economy coefficient slightly larger in absolute value and more significant.
- Leave-one-country-out exercise: estimates tightly concentrated around baseline; excluding China and India produces largest/smallest absolute estimates but no material economic difference.
- Aggregation to investor parent level used to focus heterogeneity across NBFI types.

---

### Change in growth of nonresident securities and portfolio rebalancing patterns
- Regression estimates measure changes in growth of valuation-adjusted EM securities holdings of nonresident NBFIs relative to resident NBFIs for a one-standard-deviation increase in VIX.
- Conditional sensitivities:
  - Sensitivity varies with prior-quarter VIX (low/high relative to mean), financial depth (private debt securities to GDP below/above median), and reserve adequacy/institutional quality/sovereign credit risk (low/high relative to medians).
  - Flighty nonresident NBFIs defined as those with statistically significant negative sensitivity at 5 percent.
- Portfolio rebalancing when volatility rises:
  - Nonresident NBFIs holding both equity and debt of an issuer tend to rebalance from debt to equity.
  - Across EMs, nonresident investors rotate from corporate to sovereign bonds.
  - Issuers with both local and foreign currency bonds: nonresident investors decrease portfolio share of foreign currency bonds more than local currency bonds.
  - Interpretation: corporate bonds and foreign-currency denomination are more prone to being sold by nonresident NBFIs during elevated volatility.

Numeric axis/marker values shown in figures (preserved)
- Axis/marker values appearing in figures: -2.4, -2.0, -1.6, -1.2, -0.8, -0.4, 0.0, 0.4, 0.8; additional: -3.0, -2.5, -2.0, -1.5, -1.0, -0.5, 0.0, 0.5; extended: -4.5, -4.0, -3.5, -3.0, -2.5, -2.0, -1.5, -1.0, -0.5, 0.0.
- Figure labels/conditions shown: Low VIX, High VIX; Low financial depth, High financial depth; Low, High (Reserve adequacy, Institutional quality, Sovereign credit risk); Low reserve adequacy, High reserve adequacy; Managed exchange rate, Floating exchange rate.

---

### Mutual funds/ETFs — fund–security panel, synthetic leverage, and flow decomposition
- Data and coverage:
  - Quarterly panel 2013q2-2025q2; ~4k funds, 12k securities from 52 emerging countries, ~5 million fund–security–time observations.
  - Securities roughly equally split between equities, corporate bonds, sovereign bonds.
  - One standard deviation of VIX in quarterly sample ≈ 5.22.
  - Filters: funds that invest in at least one security issued in IMF Vulnerability Exercise EMs; fund–issuer pairs observed ≥25 quarters.
- Flow construction (intensive margin, valuation-adjusted):
  - Security-level flows isolate net trading by adjusting for contemporaneous price movements (formula text preserved in source).
  - Investor flows defined with G_j,t the return of the fund (formula text preserved).
- Synthetic leverage/hedging (Fricke (2025) methodology):
  - Return gap Δ_jt = G_jt − G_jt^H decomposed into passive, active, and contemporaneous residual components.
  - Regression (Δ_jt − Δ_jt^Active) = α_j + β̄_j ⋅ G_jt^H + controls + error; β̄_j interpretation: positive → synthetic leverage; negative → synthetic hedging (significance at 10 percent).
- Descriptive statistics (selected, preserved):
  - Quarterly fund–security flows median 0.2% of holdings; interdecile roughly −19% to +27%.
  - Median holdings USD 0.7 million; right tail > USD 8 million.
  - Median security-level tenure: 12 quarters; issuer-level tenure: 33 quarters.
  - Fund size (Million USD): p10 30.7; p25 87.9; p50 312.65; p75 1079.37; p90 3339.37.
  - End-investor flow (% of lagged AUM): p10 −3.61; p25 −1.34; p50 −0.08; p75 1.13; p90 4.09.
  - Online Annex Table 2.4.2 selected: Flow (thousand USD) p10 −346.6; p25 −43.55; p50 56.09; p75 53.17; p90 357.51. Flow (% of previous holding) p10 −18.59; p25 −6.4; p50 0.22; p75 8.52; p90 26.84. Holding value (USD, thousands) p10 72; p25 215.43; p50 730; p75 2611.6; p90 8356.
- Fund-type composition and trends:
  - Actively managed mutual funds dominate by assets and counts; passive vehicles (ETFs and passive mutual funds) have steadily gained share, more pronounced in holdings than in counts.

Empirical specification and clustering
- Baseline holdings regression:
  - FLOW_irt = τ1 ⋅ GGRF_t × Type_d_j + μ_f,p,t + ξ_f,p,j + ε_ijt
  - GGRF_t = standardized quarterly VIX.
  - Type_d_j: fund characteristics (foreign vs domestic, passive vs active, ETF vs mutual fund, institutional vs retail investor base).
  - Fixed effects absorb issuer–security-type–time and issuer–investor–security-type relationships.
  - Two-way cluster-robust standard errors by fund and security.

Decomposition of fund sensitivity to global risk (channels)
- Three channels isolated:
  1. End-investor behavior (end-investor redemptions).
  2. Fund structure and liquidity management (pass-through of redemptions into asset sales).
  3. Portfolio rebalancing independent of end-investor flows.
- Estimation steps preserve formulae and relations (β1, β2, β3, γ1 definitions preserved).
- Key decomposition findings:
  - Passive funds retrench more sharply from EM assets than active funds due to direct passthrough of investor outflows and stress-induced rebalancing.
  - ETFs (mostly passive) show strong passthrough and somewhat more stress-sensitive investor base than mutual funds.
  - Institutional funds experience larger investor redemptions during global stress; retail funds show high passthrough from investor flows to asset sales.
  - Synthetically hedged funds allocate larger share to EM sovereign bonds; synthetically leveraged funds invest mostly in EM equities.
- Overall conclusion: heterogeneity in fund sensitivity arises from combinations of vehicle design, investor base, and portfolio-management constraints.

Robustness
- Alternative fixed-effect structures (fund × security and security × time) yield robust estimates, indicating differential behavior across fund types holding same securities.

---

### Decomposition of non-residents’ EM exposure, issuance, and firm-level impacts
- Flow sensitivity decomposition:
  - Decomposes funds’ EM holdings sensitivity to a one–standard-deviation VIX increase into end-investor flow sensitivity, passthrough, portfolio rebalancing, and residual; dependent variables reported in percentage points relative to initial holdings.
- Fund-type differences:
  - EM-targeted funds significantly more sensitive to VIX than EM-invested funds.
  - Drivers for targeted funds: stronger stress-induced rebalancing, more risk-responsive investor base, greater equity concentration.
  - ETFs and passive retail funds: sensitivities driven by high passthrough and portfolio rebalancing.
  - Institutional active mutual funds: sensitivities driven by procyclical redemptions and limited smoothing ability.
- Impacts on international bond issuance and spreads:
  - Flighty investors defined as top quintile of nonresident investors with highest VIX sensitivity (separately for sovereign and corporate markets).
  - Share of flighty investors ≈ 1 percent on average across EM sovereign and corporate bond markets; up to 14 percent in some countries.
  - Estimation captures effect of a one-percentage-point increase in flighty investor share on issuance volume (log) and bond spreads (basis points); effect during stress periods given by β̂1+β̂3.
  - No substitution toward domestic bond markets or syndicated loans for firms relying more on flighty investors.
- Impacts on firms’ deleveraging and investment:
  - For firms with greater reliance on flighty investors, constrained access to international bond markets during stress leads to deleveraging and reduced investment (dependent variables: total debt or capital expenditure scaled by total assets).
  - Firm-level controls include log total assets, log cash holdings, Tobin’s q, leverage ratio, ROA; variables winsorized at one percent.
- Chinese issuers:
  - Chinese issuers ~10 percent of sovereign bond observations and ~60 percent of corporate bond sample; excluding China does not materially affect baseline results.
- Maturity restriction:
  - Sample restricted to bonds with duration up to ten years; including longer maturities yields qualitatively similar but statistically insignificant results.

Growth-at-Risk (GaR) extension
- GaR model includes flighty investor measure and interactions with VIX; uses panel of 18 EMs (2013q4–2024q1); standard errors bootstrapped.
- Regressors include realized GDP growth, country financial conditions index, and flighty investor base interactions; quantiles τ = 0.05,...,0.95 and horizons h = 1,...,12.

---

### Cross-border stablecoin flows to emerging markets — data, stylized facts, and regressions
- Data and methodology (Chainalysis):
  - On-chain transfers between wallets allocated to exchanges; exchanges assigned to countries via web-traffic patterns.
  - Assets covered: BTC, ETH, USDT, USDC, and others; period January 2019 to February 2025 at daily frequency for bilateral gross and net flows.
  - Aggregation to country-level monthly gross and net inflows; exclude observations classified as “defi” and domestic flows.
- Stylized facts:
  - Large share of EM-related cross-border stablecoin activity reflects within-group and across-group flows (latter mainly with advanced economies); LICs account for very small fraction of gross volume.
  - Asia largest share of cross-border stablecoin activity, then Europe and the Americas; Oceania and Africa limited.
  - Nominally, Russia, India, Türkiye, and Brazil among largest recipients of gross stablecoin inflows in 2024, comparable to largest advanced economy recipients.
  - Net recipients in 2024: Russia, Türkiye, Vietnam, Korea, Ukraine.
- Panel regression for gross stablecoin inflows (monthly; 18 EMs; Jan 2019–Feb 2025):
  - Dependent variable: sum of USDC and USDT gross inflows scaled by GDP.
  - Regressors: country-specific factors (inflation; exchange-rate volatility vs USD; institutional and policy stability; MMF access indicator), economic flows (exports-to-GDP, financial-flows-to-GDP, remittances-to-GDP, unbacked-crypto inflows-to-GDP), global factors (effective federal funds rate, VIX, DXY), crypto-market factors (Bitcoin supply growth, BTC volatility, BTC returns).
  - Baseline estimated by pooled OLS; standard errors clustered at country level; all continuous variables standardized; time-varying domestic factors lagged.
- Regression results and robustness:
  - Baseline results robust to time fixed effects; country-specific coefficients weaker once country fixed effects included.
  - Inflation remains statistically significant even with country fixed effects.
  - Stablecoin flows associated with inflation, exchange-rate volatility, institutional/political stability, and limited access to on-shore USD cash-like instruments (MMF access), supporting role as “digital dollars.”
  - Unbacked crypto (BTC+ETH) behaves like a risk asset driven by crypto-market dynamics; domestic variables less significant.
  - Results robust to correcting for anomalous USDT spikes in April 2024.

*Source: Global Financial Stability Report Chapter 2 Online Annex, IMF | April 2026*

### 1. External Portfolio Debt and Equity Liabilities, by Investor

### 1. External Portfolio Debt and Equity Liabilities, by Investor

### Overview and purpose
- Quantifies how portfolio debt and equity liability flows to emerging markets respond to global risk factors.
- Compares sensitivities before and after the global financial crisis (GFC).
- Contrasts behavior of NBFI (non-bank financial intermediation) flows with banking flows and flows to advanced economies.

### Data, sample, and measurement
- Data sources: IMF Balance of Payments (BOP) statistics (Portfolio Investment Positions by Counterpart Economy (PIP)), BIS Locational Banking Statistics (LBS), EPFR, Bank for International Settlements, International Debt Statistics, and IMF staff calculations.
- Baseline sample: 20 major emerging markets (ARG, BGR, BRA, CHL, CHN, COL, HUN, IDN, IND, MEX, MYS, PER, PHL, POL, ROU, RUS, THA, TUR, UKR, and ZAF).
- Expanded emerging market sample: 41 emerging markets (adds AGO, ALB, BLR, BOL, CRI, DOM, ECU, EGY, GHA, GTM, JAM, JOR, KAZ, LBN, LKA, MAR, MKD, PAK, TUN, URY, VNM).
- Sample period: 2003Q2-2025Q1 (with specific estimations using 2009Q2 onward for post-GFC comparisons).
- Flows normalized by country c’s (seasonally-adjusted) quarterly GDP at t-1 and standardized by flow type to compare sensitivities.
- Global risk factors used: VIX (baseline) and Risk-on/Risk-off (RORO) index (robustness); also evaluated World Uncertainty index, MOVE index, and US dollar index (DXY).
- Banking flows measured using FX- and break-adjusted cross-border claims in BIS LBS.
- Investment-fund flows proxied by EPFR country-level fund flows (mutual funds and ETFs).

### Summary statistics (Overall sample 2003Q2–2025Q1)
- Banking inflows: count 1740; mean 0.614; sd 3.982; min -41.583; p25 -1.140; p50 0.352; p75 2.009; max 29.838 (percent of lagged GDP).
- Overall portfolio debt liability flows: count 1740; mean 0.992; sd 3.048; min -19.212; p25 -0.499; p50 0.464; p75 2.152; max 19.792 (percent of lagged GDP).
- Overall portfolio equity liability flows: count 1740; mean 0.231; sd 1.351; min -21.728; p25 -0.202; p50 0.064; p75 0.558; max 10.323 (percent of lagged GDP).
- Investment fund debt inflows: count 1664; mean 0.233; sd 1.385; min -7.151; p25 -0.283; p50 0.084; p75 0.724; max 8.840 (percent of lagged GDP).
- Investment fund equity inflows: count 1740; mean 0.082; sd 0.375; min -1.635; p25 -0.080; p50 0.010; p75 0.197; max 2.552 (percent of lagged GDP).
- Distributional notes: portfolio debt liability flows average 1 percent of GDP (upper quartile > 2 percent; max up to 20 percent). Portfolio equity flows average 0.2 percent of GDP (upper quartile ~0.5 percent; max up to 10 percent). Banking flows average 0.6 percent of GDP (can reach up to 30 percent of GDP).
- Cross-flow standard deviations: banking sd 4.0; portfolio debt sd 3.0; portfolio equity sd 1.4.

### Baseline empirical specification and identification
- Baseline regression (Equation 2.3.1) estimates sensitivity of flow Yc,t^k (overall portfolio debt or equity flows, or banking inflows, standardized) to global risk factor (VIX) and a post-2009Q2 dummy D_ppp_t, including country-year fixed effects η_c,t(y).
- Identification of sensitivity to global risk shocks comes from quarterly variation in the global risk factor.
- Driscoll-Kraay standard errors used to account for cross-sectional dependence and serial correlation.

### Main empirical findings
- Post-GFC change in relative sensitivity:
  - The sensitivity of portfolio debt inflows to emerging markets increased relative to that of banking flows in the post-2009Q2 period.
  - Portfolio equity liability flows show a similar pattern post-GFC, but the difference is not always statistically significant.
- Magnitudes (selected comparisons reported in text):
  - Using RORO in the post-GFC period, a one-standard-deviation increase in the RORO is associated with a decline of about 0.3 standard deviations in portfolio debt flows (comparable to VIX effect).
  - Corresponding response of portfolio equity inflows to RORO is about 0.3 standard deviations, exceeding sensitivity to the VIX.
  - For investment funds, a one-standard-deviation increase in the RORO is associated with declines of about 0.9 (portfolio debt) and 0.5 (portfolio equity) standard deviations in the post-GFC period.
- Comparison with banking flows:
  - Prior to the GFC, portfolio debt liability flows appeared less sensitive to the VIX than banking inflows; post-GFC, portfolio debt sensitivity increased relative to banking inflows, becoming statistically similar.
  - Using RORO, both portfolio debt and equity liability flows appear significantly more sensitive to global risk than banking flows.
  - The sensitivity of banking flows to global risk declines significantly in the post-GFC period and becomes statistically indistinguishable from zero in several specifications (consistent with Avdjiev and others (2025)).
- Advanced economies vs emerging markets:
  - Portfolio debt liability flows to emerging markets are more sensitive to the RORO than those to advanced economies, although the difference is not statistically significant in some specifications.
  - Results are robust across alternative global risk measures (MOVE, World Uncertainty index, dollar index) and across expanded samples.

### Robustness checks and alternative specifications
- Alternative risk factors: RORO, MOVE, World Uncertainty index, and US dollar index produce qualitatively similar results; RORO often shows larger sensitivities for portfolio and investment-fund flows than the VIX.
- Expanded sample: In the expanded sample of 41 emerging markets, sensitivity of portfolio debt flows to a one-standard-deviation increase in the VIX is smaller (~1¼ standard deviations) than baseline, while portfolio equity and banking flow sensitivities remain broadly similar.
- Global and country controls (Equation 2.3.2): include change in U.S. monetary policy (actual/shadow federal funds rate), world GDP growth, lagged domestic real GDP growth, private credit-to-GDP ratio, ex-ante real policy rate, capital account openness, institutional quality, official international reserves, real effective exchange rate gap, and linear annual trend. Under this specification:
  - Portfolio debt flows remain sensitive to increases in the VIX, though attenuated versus baseline.
  - Banking flows’ sensitivity to the VIX declines in the post-GFC period and becomes statistically insignificant.
- Specification comparing flow types jointly (Equation 2.3.3): estimates within-country-quarter differences across bank vs portfolio debt vs portfolio equity flows; findings confirm greater post-GFC sensitivity of portfolio debt relative to banking flows.
- Robustness to exclusion tests: main results are robust when excluding China, India, and groups of large emerging markets (Brazil, China, India, Mexico), and across regional subsamples (Asia Pacific, Emerging Europe, Latin America).

### Investment-fund (NBFI) flows findings
- Investment-fund (EPFR) flows:
  - More sensitive to global risk than overall portfolio debt liability flows in the post-2009Q2 period.
  - Under baseline specification, a one-standard-deviation increase in the RORO is associated with declines of about 0.9 (debt) and 0.5 (equity) standard deviations in investment-fund flows in the post-GFC period.
- Caveats on EPFR data:
  - EPFR primarily captures mutual funds and ETFs and excludes sovereign wealth funds, pension funds outside mutual vehicles, insurance companies, and banks’ trading books.
  - Fund domicile does not necessarily correspond to the residence of the ultimate investor; some flows may reflect domestic investors using foreign-domiciled fund vehicles.

### Key technical notes
- “Off-shore debt” defined as the difference between nationality-based and residence-based international debt security issuances (based on BIS International Debt Statistics).
- Panel decompositions follow Avdjiev and others (2022).
- Seasonally adjusted quarterly nominal GDP constructed using seasonally adjusted quarterly data when available; otherwise proxied as a quarter of the corresponding annual nominal GDP.
- Actual/shadow federal funds rate used: effective federal funds rate when non-zero and the Wu-Xia shadow rate during the zero lower bound period (2009Q1-2015Q4).

*Source: Global Financial Stability Report Chapter 2 Online Annex, IMF | April 2026*

### 1. Portfolio Liability Flows to Emerging

### ch2annex - 1. Portfolio Liability Flows to Emerging

### Sensitivity of portfolio liability flows to global risk (VIX and RORO)
- Charts estimate sensitivity of portfolio liability flows (relative to GDP, expressed in standard deviations) to a one-standard-deviation increase in the VIX or RORO, with 90 percent confidence intervals.
- Baseline and robustness checks:
  - Baseline specification: (Equation 2.3.1) estimated on baseline sample of 20 emerging markets.
  - Expanded sample: same specification estimated on 41 emerging markets.
  - Alternative specification: (Equation 2.3.2) uses a large set of global and domestic controls, based on baseline sample of 20 emerging markets.
- Key findings:
  - Using the expanded sample of 41 emerging market economies and (Equation 2.3.1), the sensitivity of investment-fund flows to a one-standard-deviation increase in the VIX is broadly the same as results in the main text (Figure 2.7, Panel 3).
  - Under the alternative specification (Equation 2.3.2), a one-standard-deviation increase in the VIX is associated with a decline of about 0.3 standard deviations in portfolio debt inflows.
  - The response of investment-fund portfolio equity inflows is described as mild and not statistically significant at conventional values.

### Comparison: overall portfolio liability flows vs investment-fund flows
- Estimation uses (Equation 2.3.3); panels compare overall portfolio liability flows with investment-fund inflows using VIX or RORO as global risk factors.
- Sample coverage in presented panels:
  - Baseline sample: 20 emerging markets (Panels 1 and 2).
  - Expanded sample: 40 emerging markets for debt liability flows and 38 for equity liability flows (Panels 3 and 4).
- Key findings:
  - In the post-GFC period, investment funds portfolio debt inflows are more sensitive to the VIX than overall portfolio debt inflows (panel 1, fourth bar).
  - The difference in sensitivity for equity flows is statistically insignificant.
  - Using RORO yields similar results: both investment funds debt and equity inflows exhibit greater sensitivity to global risk than overall portfolio debt and equity liability flows.
  - Results are broadly similar using the expanded sample.

### Nonlinear effects of global risk on NBFI flows
- Extended baseline specification with quadratic term: (2.3.5). Marginal effect of the VIX given by 훽k + 2휃k VIX and evaluated across VIX distribution.
- Estimation period and sample: baseline sample of 20 emerging markets over 2009Q2-2025Q1.
- Key findings:
  - Effect of global risk shocks on NBFI behavior depends on shock size (Online Annex Figure 2.3.6).
  - Portfolio debt liability flows:
    - Remain relatively unresponsive at low levels of global risk—when the VIX is below its historical mean of about 20.
    - Become increasingly sensitive as the VIX rises, reacting strongly at higher levels.
  - Portfolio debt liability flows also exhibit a non-linear response to the RORO index.
- Quantile (“capital-flows-at-risk”) approach: (2.3.6) uses quantile regressions to estimate VIX impact across distribution of capital flows.
  - Results suggest in the post-GFC period, portfolio debt liability flows intermediated by investment funds exhibit the most pronounced non-linear effects.
  - Increases in global risk have stronger effects during large outflows and the strongest effects during periods of large inflows.
- Confidence/statistical notation in figures:
  - Panels 3 and 4 show effects of a one-standard-deviation increase in VIX or RORO at different percentiles of gross portfolio inflows.
  - Solid markers indicate statistical significance at 10 percent or lower; hollow markers indicate not statistically significant.
  - Ranges correspond to 90 percent confidence intervals.

### Role of country policies and vulnerabilities
- Interaction specification: (2.3.7). Country characteristics V i,t(y) f include institutional quality, reserve-to-import ratio, debt-to-GDP ratio, capital account openness, and exchange rate regime; each turned into binary indicator above/below cross-country median. Interactions use four-quarter lags. Estimation sample: 20 emerging market economies, 2009Q2-2025Q1.
- Main results:
  - Additional sensitivity of portfolio debt inflows (overall or investment funds) to a one-standard-deviation increase in the VIX is larger in countries with weaker fundamentals relative to those with stronger fundamentals.
  - Among seven components of institutional quality, government stability is the primary driver of the differential sensitivity.
  - Even in some cases for countries with stronger fundamentals, an increase in the VIX may be associated with lower inflows.
  - Specifically: a one-standard-deviation increase in the VIX is associated with a one-standard-deviation decline in investment-fund portfolio debt flows.
  - Effects on overall debt flows and on portfolio equity flows (both overall and investment-fund flows) are not statistically significant at conventional levels.
- Using RORO instead of VIX:
  - Replacing the VIX with the RORO index yields similar qualitative patterns.
  - Interaction terms suggest stronger institutions and external buffers are associated with a more muted response of portfolio debt inflows to global risk.
  - Higher public debt is associated with greater sensitivity.
- Note on exchange rate regime classification: based on Ilzetzki, Reinhart and Rogoff (2021), Rethinking Exchange Rate Regimes.

### Which types of NBFI investors are most sensitive? — Institutional investors (sample coverage)
- Data source: FactSet Ownership security-level holdings data.
- Coverage notes:
  - Coverage of securities held by individual investors is relatively high for mutual fund managers: over 70 percent of assets under management.
  - Coverage is lower for pension fund managers and insurance companies: around 20 percent of assets under management.
- Stylized observation from figures:
  - Reported nonbank holdings account for a larger share of the market value of equity than that of debt securities.
  - Emerging markets securities holdings of nonbanks have increased substantially over the sample period.

*Sources: IMF BOP, BIS LBS, EPFR, FactSet Ownership, and IMF staff calculations.*

### 2. Emerging Market Security Holdings of the Nonbanks

### 2. Emerging Market Security Holdings of the Nonbanks

### Overview and coverage
- FactSet Ownership data and IMF staff calculations assemble holdings of a range of nonbank institutional investors (NBFIs), covering hedge funds, insurance companies, mutual funds, pension funds, sovereign wealth funds, family offices, foundations, and endowments.
- FactSet captures roughly 65 percent of the total portfolio equity positions intermediated by non-bank financial institutions in the PIP (formerly CPIS).
- For debt securities, coverage has gradually increased over time, reaching close to 25 percent in 2024.
- For investment funds, coverage appears higher: up to 70 percent for equity holdings and more than 30 percent for debt holdings.
- The emerging market securities holdings of the NBFIs satisfying the sample restrictions account for 74 percent of the total in the data.

### Data composition and magnitudes (selected 2024:Q4 numbers from the sample)
- Mutual fund (Advanced economies): Number nonresident 7,703; Number resident 7,482; Holdings nonresident 17.11; Holdings resident 37.57.
- Mutual fund (Emerging markets): Number nonresident 4,357; Number resident 1,258; Holdings nonresident 2.02; Holdings resident 1.91.
- Hedge fund (Advanced economies): Number nonresident 2,893; Number resident 2,889; Holdings nonresident 0.84; Holdings resident 1.79.
- Hedge fund (Emerging markets): Number nonresident 1,523; Number resident 74; Holdings nonresident 0.06; Holdings resident 0.01.
- Pension fund (Advanced economies): Number nonresident 208; Number resident 162; Holdings nonresident 0.90; Holdings resident 1.12.
- Pension fund (Emerging markets): Number nonresident 135; Number resident 49; Holdings nonresident 0.08; Holdings resident 0.11.
- Insurance company (Advanced economies): Number nonresident 167; Number resident 158; Holdings nonresident 0.05; Holdings resident 0.10.
- Insurance company (Emerging markets): Number nonresident 96; Number resident 78; Holdings nonresident 0.01; Holdings resident 0.24.
- Other nonbanks (Advanced economies): Number nonresident 140; Number resident 153; Holdings nonresident 1.38; Holdings resident 0.08.
- Other nonbanks (Emerging markets): Number nonresident 70; Number resident 21; Holdings nonresident 0.14; Holdings resident 0.05.

(Note: holdings are in trillions of US dollars as reported.)

### Methodology (key definitions and estimands)
- Main outcome: change in the share of a security held by an investor, defined as the change in holdings not attributable to valuation changes (equation (2.4.1)).
  - For security i, investor j, quarter t: H... denotes market value of holdings and S... denotes market value of the security.
  - Change is defined only when the investor held the security in the previous period (H...t−1 > 0) and the security has not been redeemed (S...t > 0), thereby excluding issuance and redemptions.
- Portfolio weight of a security in an investor’s portfolio is given by equation (2.4.2).
- Baseline empirical specification (equation (2.4.3)) estimates sensitivity of nonresident relative to resident holdings to a global risk factor (GRF):
  - Fixed effects: issuer-security-type-time (γf,s,t) and issuer-investor-security-type (δf,j,s) absorb common and time-invariant issuer/security characteristics.
  - Coefficient β measures sensitivity of nonresident investors relative to resident investors to the global risk factor.
- Individual-investor sensitivity estimated with equation (2.4.4); only coefficients statistically significant at the 5 percent level are retained for heterogeneity analysis.
- Flightiness measure (equation (2.4.5)): share of holdings by investors whose statistically significant βj is below threshold β̄ (β̄ = 0 for individual securities; β̄ = 20th percentile of the βj distribution at country level).
- Portfolio rebalancing specification (equation (2.4.6)) examines rotations across asset classes (corporate/sovereign bond/equity) using asset-class controls and fixed effects to capture within-issuer or within-country reallocations.
- Sample restrictions:
  - Each issuer-investor pair must have at least 25 quarters of holdings to be included.
  - Only investors whose emerging market security holdings exceed 10 million US dollars are included.

### Key empirical findings
- Nonresident NBFIs reduce their emerging market investments more than resident NBFIs when global risk aversion rises:
  - Differential response is significant for emerging market economies but not for advanced-economy investments (advanced economy sample excludes Japan, Switzerland, and the United States).
  - When the VIX is elevated (top decile), nonresident reductions in emerging market holdings are more pronounced.
- Sensitivity to the US dollar:
  - A one-standard-deviation appreciation of the US dollar against other major currencies is followed by a decrease of 1.4 percentage points in nonresident emerging market investments—substantially larger than the effect from an analogous increase in the VIX.
- Flighty investors:
  - There are 167 nonresident NBFIs in the sample with statistically significant negative βj at the 5 percent level: 120 mutual funds and 32 hedge funds.
  - These flighty nonresident investors account for 23 percent of the emerging market bond holdings and 12 percent of the equity holdings of the NBFIs in the sample.
  - Substantial heterogeneity in the flighty share across countries and over time.
- State dependence and market depth:
  - Emerging market investments of nonresident NBFIs are more sensitive to the VIX when it increases from a low level than when it decreases from a high level, though the two responses are not statistically significantly different.
  - Nonresident reductions are larger in countries with shallower financial markets, proxied by the ratio of private debt securities to GDP.
- Recipient vulnerabilities amplify responses of flighty investors:
  - Flightier nonresident investors cut emerging market investments more following VIX increases in countries with lower international reserves, weaker institutions, and higher sovereign credit risk.
  - Higher foreign exchange reserves substantially moderate the fall in nonresident investments, particularly in countries with managed exchange rates.
  - Differences between high and low vulnerability groups are statistically significant for institutional quality, and for reserve adequacy in countries with managed exchange rates, but not for all other vulnerability variables.

### Robustness checks and additional results
- Alternative fixed effects specification:
  - Including time–security fixed effects (a more saturated specification) yields similar β estimates to the baseline (2.4.3); the advanced-economy coefficient becomes slightly larger in absolute value and more statistically significant.
- Leave-one-country-out exercise:
  - Excluding each emerging market country one at a time yields coefficient estimates tightly concentrated around the baseline; excluding China and India produces the largest and smallest absolute estimates, respectively, but neither is economically significantly different from baseline.
- Sample design choices:
  - Aggregation to investor parent level is used to focus heterogeneity across NBFI types and to avoid noise from smaller legal-entity level holdings.

*Source: FactSet; and IMF staff calculations.*

### 1. Change in the Growth of Nonresident Securities

### 1. Change in the Growth of Nonresident Securities

### Key findings on sensitivity to global risk (VIX)
- Regression estimates measure changes in the growth of valuation-adjusted emerging market securities holdings of nonresident NBFIs, relative to resident NBFIs, for a one-standard-deviation increase in the VIX.
- Panel 1: Sensitivity varies with the prior-quarter value of the VIX; low (high) VIX denotes values below (above) its mean over the sample period.
- Panel 2: Sensitivity varies by financial depth; low (high) financial depth when the ratio of private debt securities to GDP is below (above) the sample median in the previous year.
- Panels 3 and 4: "Flighty" nonresident investors are those nonresident NBFIs for which the sensitivity of their emerging market investments to the VIX is statistically significantly different from 0 at the 5 percent level and negative.
- Empirical regularities:
  - Flighty nonresident NBFIs reduce their emerging market investments more in jurisdictions with elevated vulnerabilities.
  - Higher international reserves moderate nonresident outflows in jurisdictions with shallower financial markets.

### Portfolio rebalancing patterns when volatility rises
- Nonresident NBFIs holding both equity and debt of a given emerging market issuer tend to rebalance from debt to equity (Online Annex Figure 2.4.8, panel 1).
- Across emerging markets, nonresident investors rotate from corporate to sovereign bonds.
- Issuers with both local and foreign currency bonds outstanding see nonresident investors decrease the portfolio share of foreign currency bonds more than local currency bonds (Online Annex Figure 2.4.8, panel 2).
- Interpretation: Among emerging market securities, corporate bonds are the most prone to being sold by nonresident NBFIs in periods of elevated volatility; the currency denomination choice (local vs foreign) can influence the investor base and market pricing, especially during market stress episodes.

### Methodology and definitions (as in source)
- Data sources: BIS; Chicago Board Options Exchange; CMA; FactSet; PRS Group; and IMF staff calculations.
- Definitions used in regressions:
  - Low/high VIX: below/above mean over the sample period (previous quarter).
  - Financial depth: low/high when the ratio of private debt securities to GDP is below/above the sample median in the previous year.
  - Flighty nonresident investors: nonresident NBFIs with sensitivity to the VIX statistically significantly different from 0 at the 5 percent level and negative.
  - Foreign exchange adequacy (FX reserves relative to imports): low/high when below/above the sample median for the previous year.
  - Institutional quality: low/high when the sum of relevant ICRG political risk components (bureaucracy quality, corruption, democratic accountability, government stability, investment profile, law and order, socioeconomic conditions) is below/above the sample median for the previous year.
  - Sovereign risk: low/high when the average of the daily values of the sovereign credit default swap spread in the previous quarter is below/above the sample median.
- Specifications include time–issuer–security type (bond/equity) and investor–issuer–security type fixed effects.
- Error bars: 90 percent confidence intervals obtained using Driscoll-Kraay standard errors, with the number of lags equal to 4.

### Numeric labels and axis values shown in figures (as presented)
- Axis/marker values appearing in the figures: -2.4, -2.0, -1.6, -1.2, -0.8, -0.4, 0.0, 0.4, 0.8
- Additional axis values appearing in the figures: -3.0, -2.5, -2.0, -1.5, -1.0, -0.5, 0.0, 0.5
- Extended axis values appearing in the figures: -4.5, -4.0, -3.5, -3.0, -2.5, -2.0, -1.5, -1.0, -0.5, 0.0
- Figure labels/conditions shown: Low VIX, High VIX; Low financial depth, High financial depth; Low, High (for Reserve adequacy, Institutional quality, Sovereign credit risk); Low reserve adequacy, High reserve adequacy; Managed exchange rate, Floating exchange rate.

*Source: IMF staff calculations (data: BIS; Chicago Board Options Exchange; CMA; FactSet; PRS Group) — Online Annex Figure 2.4.8, Global Financial Stability Report Chapter 2, Online Annex.*

### 2. Sensitivity of the changes in relative portfolio shares of

### 2. Sensitivity of the changes in relative portfolio shares of 

### B. Mutual Funds/ETFs — Data, sample, and variable construction
- Data coverage:
  - Quarterly panel of fund–security observations spanning 2013q2-2025q2.
  - Around 4k funds, 12k securities from 52 emerging countries, and 5 million fund–security–time observations.
  - Securities roughly equally split between equities, corporate bonds, and sovereign bonds.
  - Global risk measured using the CBOE VIX index, aggregated to quarterly frequency by taking the within-quarter average and standardized by its sample standard deviation.
  - Note: One standard deviation of VIX in our quarterly sample is around 5.22.
- Sample construction and filters:
  - Restricted to funds that invest in at least one security over the sample period issued in countries belonging to emerging markets included in the IMF’s Vulnerability Exercise.
  - Only fund–issuer pairs observed for at least 25 quarters are retained (25-quarter requirement is imposed at the fund–issuer level).
  - Analysis focuses on intensive-margin portfolio adjustments (continuing positions only).
  - Panel is unbalanced with an interquartile range of 4–27 quarters for fund–security pairs; fund–issuer pairs typically longer.
- Construction of flow variables:
  - Security-level flows computed by adjusting changes in holdings for contemporaneous price movements to isolate net trading; flows defined on the intensive margin (continuing positions only).
  - Fund j’s flow into security r at quarter f is defined as:
    - F... = 100 * [ (Δ_it H..._ijt − H..._ijt−1) / H..._ijt−1 ] with Δ_it = P_it−1 / P_it and P_it being the average price of the security at quarter f (formula text preserved in source).
  - Investor flows: Fund j’s investor flow at quarter f is defined as:
    - V..._jt = 100 * [ (Fund_t − Fund_t−1) / (1 + G_j,t) Fund_t−1 ] with G_j,t the return of the fund for the period of quarter f−1 to f (formula text preserved in source).

### Synthetic leverage / hedging (measurement and classification)
- Methodology follows Fricke (2025): compare realized fund returns with returns implied by reported security holdings.
- Definitions:
  - G_jt^H = O_j,t−1 b' r_t denotes the buy-and-hold portfolio return using beginning-of-period holdings weights.
  - Return gap Δ_jt = G_jt − G_jt^H.
  - Decomposition into three components:
    - Passive component: return from mechanical price movements (value-adjusted lagged holdings w_j,t−1).
    - Active component: effect of changes in portfolio weights between w_j,t−1 and observed current holdings w_j,t (discretionary trading/rebalancing).
    - Contemporaneous residual: difference between realized returns and return implied by current holdings, capturing derivatives, borrowing, off-balance-sheet instruments.
- Regression to isolate synthetic exposures:
  - Residual return gap (after removing active component) regressed on holdings-implied return and fund controls: (Δ_jt − Δ_jt^Active) = α_j + β̄_j ⋅ G_jt^H + controls + error.
  - Interpretation: significantly positive β̄_j → synthetic leverage; significantly negative β̄_j → synthetic hedging (significance at the 10 percent level).
- Control variables include: end-investor flows (scaled by lagged AUM), flows squared, turnover, lagged portfolio concentration (HHI), lagged fund size, VIX and its lag, and lagged return gap Δ_j,t−1.
- Empirical findings on portfolio composition:
  - Synthetically hedged funds allocate the largest share of their portfolios to emerging market sovereign bonds, while still holding a sizable equity component (Online Annex Figure 2.4.11).
  - Synthetically leveraged funds invest mostly in emerging market equities.

### Descriptive statistics and sample composition (key figures preserved)
- Quarterly fund–security flows centered around median change of 0.2% of holdings but with large dispersion and fat tails (interdecile range roughly −19% to +27%).
- Position sizes:
  - Median holdings USD 0.7 million; pronounced right tail exceeding USD 8 million.
- Panel structure:
  - Median security-level tenure: 12 quarters; issuer-level tenure: 33 quarters.
- Monthly fund size and end-investor flows (monthly frequency):
  - Fund size (Million USD): p10 30.7; p25 87.9; p50 312.65; p75 1079.37; p90 3339.37.
  - End-investor flow (% of lagged AUM): p10 −3.61; p25 −1.34; p50 −0.08; p75 1.13; p90 4.09.
- Online Annex Table 2.4.2 (selected distributional statistics, p10–p90):
  - Flow (thousand USD, thousands): p10 −346.6; p25 −43.55; p50 56.09; p75 53.17; p90 357.51.
  - Flow (% of previous holding): p10 −18.59; p25 −6.4; p50 0.22; p75 8.52; p90 26.84.
  - Holding value (USD, thousands): p10 72; p25 215.43; p50 730; p75 2611.6; p90 8356.
  - Holding's portfolio share (%): p10 0.01; p25 0.02; p50 0.09; p75 0.34; p90 1.
  - Number of observations for fund-security pairs: p10 2; p25 4; p50 12; p75 27; p90 39.
  - Number of observations for fund-issuer pairs: p10 26; p25 28; p50 33; p75 42; p90 48.
  - Number of securities a given fund holds: p10 1; p25 2; p50 7; p75 20; p90 74.
  - Number of funds that hold a given security: p10 1; p25 3; p50 7; p75 21; p90 57.
- Summary statement: Fund–security flows typically small but highly dispersed with fat tails; portfolios heterogeneous, broadly diversified, episodic rebalancing and potential for correlated trading.
- Fund-type composition:
  - Actively managed mutual funds are the dominant investors throughout the sample period by assets and fund counts; passive vehicles (including ETFs and passive mutual funds) have steadily gained share over time.
  - Increase more pronounced in holdings than in counts, particularly for ETFs, suggesting larger average fund sizes.

### Empirical specification (holdings response to global risk)
- Baseline holdings regression:
  - FLOW_irt = τ1 ⋅ GGRF_t × Type_d_j + μ_f,p,t + ξ_f,p,j + ε_ijt
  - Dependent variable: valuation-adjusted flows into security r by fund j at time f.
  - GGRF_t denotes standardized quarterly VIX.
  - Type_d_j indicates fund characteristics (foreign vs domestic domicile, passive vs active, ETF vs mutual fund, institutional vs retail investor base).
  - Fixed effects μ_f,p,t absorb issuer–security type–time shocks common across funds; ξ_f,p,j absorb time-invariant fund–issuer–security type relationships.
  - Identification from within-fund–issuer variation in responses to changes in global risk.
  - Inference: two-way cluster-robust standard errors allowing arbitrary dependence within funds over time and within securities across funds.
- Clustering rationale:
  - Residual-dependence diagnostics show economically meaningful dependence within funds and within securities; within-time covariances several orders of magnitude smaller.
  - Therefore standard errors clustered along cross-sectional dimensions (fund and security) capture dominant sources of residual correlation.

### Additional results and decomposition of fund sensitivity to global risk
- Decomposition strategy isolates three channels:
  1. End‑investor behavior (end-investor redemptions).
  2. Fund structure and liquidity management (degree to which redemptions are passed through into asset sales).
  3. Portfolio rebalancing independent of end‑investor flows (stress-induced discretionary or rule‑based adjustments).
- Estimation steps:
  - Estimate heterogeneity in end-investor flow sensitivity to global risk with:
    - V..._jt^m = γ1 ⋅ GGRF_t^m × Type_d_j + Controls + λ_t + ν_j + error, with two-way clustered standard errors by funds and time.
  - Incorporate realized investor flows directly into security-level holdings regression:
    - FLOW_irt = β1 ⋅ GGRF_t × Type_d_j + β2 ⋅ V..._jt + β3 ⋅ V..._jt × Type_d_j + μ_f,p,t + ξ_f,p,j + ω_ijt.
    - β2 measures pass-through of end-investor flows into asset sales.
    - β3 captures incremental pass-through for funds of type j.
    - β1 identifies stress-induced portfolio rebalancing independent of contemporaneous investor flows.
  - Aggregate mapping:
    - ∂FLOW/∂GGRF = β1 Type_d_j + β2 ∂V/∂GGRF + β3 ∂V/∂GGRF Type_d_j.
    - Relation to τ1: τ1 = β1 + β2 γ1 + β3 γ0 + β3 γ1 (definitions preserved).
- Key decomposition findings:
  - Passive funds retrench more sharply from emerging market assets than active funds because:
    - Investor outflows translate more directly into asset sales (limited use of internal liquidity buffers).
    - Passive funds engage in more pronounced stress-induced portfolio rebalancing (liquidity management, not benchmark composition change).
  - ETFs (mostly passive) demonstrate:
    - Strong passthrough from investor flows to asset sales.
    - Somewhat more stress‑sensitive investor base than mutual funds.
  - Institutional funds experience larger investor redemptions during global stress (more procyclical adjustments by professional investors).
  - Retail-oriented funds, on aggregate, exhibit higher passthrough from investor flows to asset sales.
  - Among passive funds:
    - Institutional mandates seem able to smooth adjustments.
    - Active institutional funds serve performance-sensitive investors whose withdrawals are larger, leading to greater passthrough and higher sensitivity.
- Overall conclusion from decomposition:
  - Observed heterogeneity in fund sensitivity to global risk arises from specific combinations of vehicle design, investor base, and portfolio management constraints rather than a single attribute.

*Source: Chicago Board Options Exchange; FactSet; and IMF staff calculations.*

### 1. Decomposition of Non-Residents’ Emerging Market

### 1. Decomposition of Non-Residents’ Emerging Market

### Flow sensitivity decomposition
- Decomposition approach:
  - Decomposes funds’ emerging market holdings sensitivity to a one–standard deviation increase in the CBOE Volatility Index (VIX) into:
    - sensitivity of end-investor flows to VIX;
    - passthrough of those flows to emerging market asset sales;
    - portfolio rebalancing independent of flows;
    - a residual.
  - Decomposition obtained using two auxiliary regressions plus the baseline regression for total sensitivity.
  - Regressions include fund × issuer × security-type and issuer × security-type × time fixed effects (baseline).
  - Dependent variables are reported as percentage points, relative to initial holdings, valuation-adjusted.

### Fund-type differences and key patterns
- Emerging Market-Targeted vs Emerging Market-Invested funds:
  - Emerging market-targeted funds are significantly more sensitive to the VIX than emerging market-invested funds.
  - Drivers of heightened sensitivity for targeted funds:
    - stronger stress-induced portfolio rebalancing (active precautionary liquidity management with higher liquidity mismatch between redeemable liabilities and relatively illiquid emerging market assets);
    - a more risk-responsive end-investor base (larger ETF presence);
    - greater concentration in equities, increasing transmission to emerging market equity markets.
- Granular fund breakdowns:
  - Sensitivities of ETFs and passive retail funds are driven by high passthrough and portfolio rebalancing.
  - Sensitivities of institutional active mutual funds reflect more procyclical redemptions and limited ability to smooth adjustments.
  - Salganik‑Shoshan (2016) evidence: institutional mutual fund investors exhibit more performance‑sensitive and correlated flow behavior, implying larger and more synchronized withdrawals and higher passthrough to portfolio adjustments in institutional active funds.

### Alternative fixed-effect structures (robustness)
- Alternative specification:
  - Replace fund × issuer × security-type and issuer × security-type × time fixed effects with fund × security and security × time fixed effects to absorb all time-varying heterogeneity at the individual security level (ratings, currency denomination, maturity, liquidity).
- Result:
  - Estimates remain robust under the more stringent control structure (Online Annex Figure 2.4.15), indicating differential behavior across fund types holding the same securities—not driven by security composition.

### Macro-financial implications — overview
- General empirical design elements appearing across analyses:
  - Share of flighty investors (퐹퐹퐺퐺푟푟푏푏ℎ푓푓푒푒
    푐푐,푡푡
    푞푞
    −1) is measured as the average over the preceding four quarters and is lagged by one quarter.
  - VIX indicators:
    - For international bond issuance regressions, a dummy 퐺퐺퐺퐺퐹퐹
      푡푡
      푤푤
      −1
      퐷퐷 equals one if the average VIX over the preceding week (푓푓
      퐻퐻
      −7 to 푓푓
      퐻퐻
      −1) exceeds the 90th percentile, zero otherwise.
    - For firm-quarter analyses, 퐺퐺퐺퐺 퐹퐹
      푡푡
      푞푞
      −1
      퐷퐷 equals one if the VIX in quarter 푓푓
      푞푞
      −1 exceeds the 80th percentile.
  - Control variables and fixed effects vary by equation but include rating, duration, subordination, callability, floating-rate/fixed-rate indicators, coupon, logarithm of issuance volume (for spreads), firm-level controls (log total assets, log total cash holdings, Tobin’s q, leverage ratio, ROA), and country/currency/time/sector fixed effects as appropriate.
  - Dependent variables are winsorized at the one-percent level on both tails.
  - Analysis in this section uses data up to 2024.

### Impacts on international bond issuance
- Estimation equation (summary):
  - Dependent variable 푌푌
    푖푖,푡푡
    푑푑 denotes either logarithm of the volume issued of an international bond on day 푓푓
    퐻퐻 or the bond’s spread (basis points).
  - The effect of a one-percentage-point increase in the share of flighty investors is captured by 훽̂1 in normal times and by 훽̂1+훽̂3 during stress periods.
- Flighty investors definition and prevalence:
  - Flighty investors are defined, separately for sovereign and corporate bond markets, as the top quintile of nonresident investors with the highest sensitivity to the VIX.
  - The share of flighty investors is around 1 percent on average across emerging market sovereign and corporate bond markets, and up to 14 percent in some countries.
- Additional notes:
  - While the analysis examines the effect of the VIX on issuance, issuance activity may influence the VIX. The VIX is measured using information lagged by one day to mitigate simultaneity.

### Impacts on domestic bond issuance and syndicated loans
- Estimation equation (summary):
  - Dependent variable 푉푉
    푖푖,푡푡
    푑푑 is either logarithm of the volume of domestic bond issuance on day 푓푓
    퐻퐻 or logarithm of the volume of syndicated loan on day 푓푓
    퐻퐻.
  - For syndicated loans, firm-level controls are included (log total assets, log cash holdings, Tobin’s q, leverage, ROA).
  - The impact of a one-percentage-point increase in the share of flighty investors during stress periods is given by 훽̂1+훽̂3.
- Empirical findings (summary statements shown in figures and text):
  - No substitution toward domestic bond markets or syndicated loan markets is observed when access to international bond markets becomes constrained for firms with greater reliance on flighty investors.

### Impacts on firms’ deleveraging and investment
- Estimation equation (summary):
  - Dependent variable 푍푍
    푗푗,푡푡
    푞푞 denotes either firm j’s total debt or capital expenditure in quarter 푓푓
    푞푞 scaled by total assets in the previous year.
  - Control variables include log total assets, log cash holdings, Tobin’s q, leverage ratio, ROA at 푓푓
    푞푞
    −1. Firm and time fixed effects are included. Variables are winsorized at the one-percent level on both tails.
  - The effect during stress periods is given by 훽̂1+훽̂3.
- Empirical findings (summary statements shown in figures and text):
  - For firms with greater reliance on flighty investors, access to international bond markets becomes constrained during stress periods.
  - Because there is no substitution toward domestic bond markets or syndicated loan markets, constrained access leads to deleveraging and reduced investment (results presented in panel 2 of Figure 2.11).

### Additional empirical notes and sample composition
- Chinese issuers composition:
  - Chinese issuers account for about 10 percent of the observations in the sovereign bond sample and roughly 60 percent in the corporate bond sample.
  - Excluding China does not materially affect the baseline results.
- Maturity restriction:
  - Sample restricted to bonds with a duration of up to ten years; including bonds with maturity above ten years yields qualitatively similar but statistically insignificant results.

*Sources: FactSet; Lipper; MSCI; IMF staff calculations (Global Financial Stability Report Chapter 2 Online Annex, IMF | April 2026).*

### 3. Volume and Spreads of Sovereign

### 3. Volume and Spreads of Sovereign

### Amplified effects for emerging market issuers with preexisting vulnerabilities
- The analysis examines how reliance on "flighty investors" affects issuance volumes and bond spreads, allowing effects to vary with issuers’ preexisting vulnerabilities proxied by debt levels.
- Vulnerability definitions:
  - Sovereign issuer vulnerable if public debt-to-GDP ratio exceeds the sample median.
  - Corporate issuer vulnerable if leverage ratio exceeds the industry-specific median.
- Interaction specification (summary):
  - Includes a dummy 푂푗퐷 equal to one for vulnerable issuers.
  - Allows for persistent responses to stress by constructing 퐺퐺퐺퐺퐹퐹푡푡
    푤푤
    −ℎ
    퐷퐷 using lags h = 4 (approximately one month).
  - All other specifications follow equation (2.5.1).
- Key empirical quantities and interpretation:
  - The effect of a one-percentage-point increase in the share of flighty investors for vulnerable issuers during stress periods is given by 훾̂1 ∙ (β̂1 + β̂3).
  - Volume results are reported in panel 3 of Figure 2.11.
  - Spread results are presented in panel 2 of Online Annex Figure 2.5.1.
- Robustness check:
  - Sovereign CDS spreads are used instead of public debt-to-GDP ratio; low (high) CDS issuers defined as those below (above) the 45th percentile.
  - Key empirical results are largely unchanged (Online Annex Figure 2.5.1, panel 3).

### Growth-at-Risk (GaR) specification and inclusion of flighty investor measure
- GaR definition and model components:
  - 푒푖,푡+ℎ(휏) denotes h-quarter ahead GDP growth for country r realized at f+ h (annualized).
  - Model includes country-specific and time-specific constants 훽ℎ,푖(휏) and 훽ℎ,푡(휏).
  - Regressors include realized GDP growth 푒푖,푡, country’s financial conditions index 퐹𝐶푉푖,푡, and other terms across quantile levels 휏 = 0.05,0.10, ... ,0.95 and horizons h = 1,...,12.
- Extension to include flighty investor base:
  - Adds a measure 퐹퐹푉푉퐺퐺퐻퐻푇푇푌푌𝑖,𝑡 and its interaction with the VIX.
  - Uses 퐹𝐶푉̃𝑖,𝑡 as residuals from regressing original financial conditions index on the VIX, country fixed effects and time fixed effects (orthogonal to VIX).
- Estimation details:
  - Panel of 18 emerging markets with available data from 2013q4 to 2024q1.
  - Standard errors are bootstrapped.
  - Figure 13 in the main text shows the estimated results.

### Online Annex 2.6 — Cross-Border Stablecoin Flows to Emerging Markets: data and methodology
- Data source and coverage:
  - Uses Chainalysis on-chain transfers between crypto wallets, allocates wallets to exchanges via a proprietary algorithm, and assigns exchanges to countries using web-traffic patterns.
  - Dataset covers BTC, ETH, Tether (USDT), USD Coin (USDC), and other assets from January 2019 to February 2025 at daily frequency for bilateral gross and net flows.
- Data preparation steps:
  - Chainalysis classifies exchange-to-exchange transfers as “direct” and transfers passing through non-exchange wallets or extra transactions as “indirect”; these are summed to obtain gross and net flows for each asset.
  - Bilateral flows aggregated by destination country to derive country-level gross and net inflows.
  - Daily flows aggregated to monthly frequency for regression analysis.
  - Exclude residual observations not allocated to a specific country (classified as “defi”) and domestic (within-country) flows.

### Stylized facts on cross-border stablecoin activity
- Three additional observations:
  - Large share of emerging market-related cross-border stablecoin activity reflects both within-group and across-group flows (latter primarily with advanced economies); low-income countries (LICs) account for only a very small fraction of overall gross volume (Online Annex Figure 2.6.1, panel a).
  - Asia accounts for the largest share of cross-border stablecoin activity, followed by Europe and the Americas; activity in Oceania and Africa remains limited (Online Annex Figure 2.6, panel b).
  - In nominal terms, several large emerging markets—Russia, India, Türkiye, and Brazil—rank among the largest recipients of gross stablecoin inflows in 2024, with volumes comparable to the largest advanced economy recipients (Online Annex Figure 2.6, panel c).
- Net recipients in 2024:
  - Russia, Türkiye, Vietnam, Korea, and Ukraine are the largest recipients in net terms.

### Panel regression for drivers of gross stablecoin inflows
- Dependent variable:
  - 푆푆𝐶𝑏𝑟𝐺𝑟𝑟𝑏𝑓𝐺𝐺𝑂𝐺𝐷𝑃𝑃𝑖,𝑡 denotes the sum of USDC and USDT gross inflows into country r in month f, scaled by GDP.
- Regressors and controls:
  - Vectors of country-specific, global, and crypto-market-specific factors: country-specific factors include inflation, volatility of bilateral exchange rate vs the U.S. dollar, institutional and policy stability, and indicator of access to short-term dollar assets proxied by presence of U.S.-dollar-asset money market funds domiciled in the country.
  - Economic flows included: exports-to-GDP, financial-flows-to-GDP, remittance-flows-to-GDP, and unbacked crypto-flows-to-GDP (sum of BTC and ETH).
  - Global factors: effective federal funds rate, VIX, dollar index (DXY).
  - Crypto-specific factors: growth rate of Bitcoin supply, volatility of Bitcoin price, BTC returns in U.S. dollar terms.
- Sample and estimation:
  - Monthly dataset comprises 18 emerging markets between January 2019 to February 2025.
  - Baseline regression estimated through pooled ordinary least squares, with standard errors clustered at the country level.
  - Fixed effects included in robustness exercises.
  - All continuous variables are standardized.
  - Time-varying domestic factors are lagged to reduce endogeneity.
  - Note: The inclusion of country fixed effects precludes the MMF-access variable as it does not vary through time.

### Results from baseline regression and robustness exercises
- Table and column summary:
  - Online Annex Table 2.6.1 reports baseline coefficients in Column 1 (as shown in Box Figure 2.2.1.d). Columns 2 and 3 augment the baseline with time- and country-fixed effects, respectively.
  - Column 4 uses unbacked-crypto gross inflows (sum of BTC+ETH) as dependent variable and controls for stablecoin flows as percent of GDP (a falsification check).
- Robustness and interpretation:
  - Baseline results robust to inclusion of time fixed effects.
  - Country-specific factor coefficients are weaker once country fixed effects are included — much of the variation reflects cross-country differences given short sample and persistence of country fundamentals.
  - Inflation remains statistically significant even after controlling for country fixed effects.
  - Column 4 comparison (crypto flow dependent variable) indicates:
    - Stablecoin flows are associated with inflation, exchange-rate volatility, institutional and political stability, and limited access to on-shore cash-like U.S.-dollar instruments (proxied by MMF access), supporting stablecoins functioning as “digital dollars” to manage macroeconomic risk and domestic financial frictions.
    - Unbacked crypto (BTC+ETH) behaves more like a risk asset driven by crypto-market dynamics and global risk factors; domestic variables become statistically insignificant.
    - Differential responses to global risk aversion (negative versus positive coefficients) are consistent with their roles as speculative versus safe assets.
- Data anomaly note:
  - Regression results remain robust to correcting for unusual spikes in USDT flows in April 2024, which appear to reflect a data anomaly.

*Source: Global Financial Stability Report Chapter 2 Online Annex (IMF | April 2026).*

---


_Source: https://www.imf.org/-/media/files/publications/gfsr/2026/april/english/ch2annex.pdf_