## online-annex-ch3

## Source details

**Canonical URL:** [online-annex-ch3](https://www.imf.org/-/media/files/publications/gfsr/2021/october/english/online-annex-ch3.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/gfsr/2021/october/english/online-annex-ch3.pdf.md)
- [Structured JSON version](/-/media/files/publications/gfsr/2021/october/english/online-annex-ch3.pdf.json)

---

### Distributional Figures and Decomposition (2020:Q4 and yearly changes)
- Figures reported (descriptive distributions, 2020:Q4):
  - Transition-Opportunity Score Distribution, Sustainable versus Conventional Funds (x-axis: score 0-100; y-axis: percent).
  - Carbon-Intensity Score Distribution, Sustainable versus Conventional Funds (x-axis: tons of CO2 equivalent per million dollars of revenue; y-axis: percent).
  - Morningstar Carbon Solutions Involvement Distribution, Climate versus Conventional Funds (x-axis: score 0-100; y-axis: percent).
  - Morningstar Carbon Risk Score Distribution, Climate versus Conventional Funds (x-axis: score 0-100; y-axis: percent).
  - Alternative Transition-Opportunity Score Distribution, Climate versus Conventional Funds (x-axis: score 0-100; y-axis: percent).
  - Ownership of Firms by Type of Fund, Various Industries (Percent of total market capitalization).
- Decomposition (Online Annex Figure 3.2.2):
  - Yearly Change in Transition Opportunity Score (Change in index; index takes values between 0 and 100).
  - Yearly Change in Carbon Intensity (Tons of CO2 equivalent emissions per million dollars of revenue).
- Data sources: Morningstar; Bloomberg Finance L.P.; Lipper; FactSet; Refinitiv; and IMF staff.

### Fund Flow Analysis: Model, Data, and Key Parameters
- Estimated model (Equation (1)) examines relation between fund sustainability labels, Morningstar scores, and net fund flows:
  - Dependent variable: net flow into fund 푖 (푓푓퐹퐹퐹퐹퐹퐹퐹퐹 푖,푐,푟,푡).
  - Key regressors:
    - Vector of dummies capturing fund sustainability label, environment label, and climate label (퐿퐿퐿퐿퐿퐿퐿퐿퐹퐹).
    - Dummy for top decile of funds by broad sustainability rating.
    - Transition opportunity and carbon intensity scores.
    - Fund-level controls: lagged flows, lagged returns, logarithm of fund size, expense ratio, fund age.
  - Fixed effects: category-year (휈휈 푐,푦) and region-year (훿훿 푟,푦).
  - Standard errors clustered at the fund level.
- Sample:
  - 6,454 funds from 33 countries.
  - Sample period: 2010:Q1 to 2020:Q4.

### Proxy Voting Analysis: Model, Data, and Scope
- Estimated model (Equation (2)) examines relation between labels/scores and proxy voting:
  - Dependent variable: percentage of votes on climate resolutions by the fund that were cast in favor.
  - Key regressors similar to Equation (1); climate label not included separately due to very small number of observations.
  - Additional controls include a passive-fund dummy, logarithm of fund size, expense ratio, fund age.
- Sample:
  - 1,521 funds from the United States.
  - Sample period: 2015 to 2020.

### Issuance Analysis (Firm-Level Securities Issuance)
- Coverage and data:
  - Firms in transition-sensitive sectors issuing bonds or SEOs at least once during 2010:Q1-2021:Q1.
  - Total number of firms: 6,449, of which 5,446 issued equity and 3,722 issued bonds.
- Regression specification (fixed-effect panel):
  - Dependent variable: issuance measures on the intensive margin (issuance amount conditional on issuance) and extensive margin (likelihood of issuance).
  - Key regressors and interactions:
    - Firm “Green” indicators: ESG score, E score, transition opportunity score, carbon intensity (higher value = greener).
    - Flow exposure variables:
      - 퐹퐹퐹퐹퐹퐹퐹퐹_푆푆푆푆푆푆푆푆 and 퐹퐹퐹퐹퐹퐹퐹퐹_푆푆푭퐹푆푆 (exposure to net inflows to sustainable and conventional funds respectively), defined as weighted averages of net flows into funds holding a firm’s securities.
  - Controls: sector fixed effects (훼훼 푠), quarter fixed effects (휌휌 푡), market capitalization, leverage ratio, market-to-book ratio, ROA, tangibility, short-term-debt-to-asset ratio, stock returns (measured at the end of the first lag of quarter t’s calendar year).
- Notes:
  - Quarterly values of firm-level green indicators obtained via linear interpolation of yearly values.
  - Identification of transition-sensitive sectors: SIG codes in Dealogic hand-matched with 4-digit NACE Rev2 sector codes; housing sector excluded.

### Stock Return Analysis: Price Pressure from Flows
- Price-pressure measure constructed as in Khan et al. (2012):
  - Difference between buying pressure from high-net-inflow funds and selling pressure from high-net-outflow funds, normalized by shares outstanding (Equation (4)).
  - Measures constructed separately for sustainable-fund flows and conventional-fund flows.
- Econometric approach:
  - Regress firm i’s h-quarter-ahead CAPM-based abnormal returns on current-quarter price pressure from sustainable funds, the firm’s lagged green score, and their interaction.
  - Controls include same variables as issuance regressions plus price pressure from conventional funds and its interaction with the firm green score.
- Key empirical findings:
  - A one percentage point increase in price pressure leads to a significantly larger contemporaneous increase in abnormal returns for firms with a higher ESG score or E score (Online Figure 3.4.1).
  - There is almost no difference in price impact for firms with higher transition opportunity score or lower carbon intensity.

### Climate Change News, Fund Performance, and Scores
- Climate news events:
  - Identified from four indices based on New York Times, Wall Street Journal, and Google News; aggregated to quarterly frequency.
  - Dummies constructed for top decile of quarterly distribution of each index; union of dummies yields nine quarters of heightened climate news attention.
  - Examples: Paris Agreement in 2015:Q4 and announcement of the US withdrawal from the Paris Agreement in 2017:Q2.
- Estimated model (Equation (1) in this annex) for news impact:
  - Dependent variable Y: fund return, net inflow, change in carbon intensity, or change in transition opportunity score.
  - Key regressors:
    - Climate news dummy (푁푁퐿퐿퐹퐹푆푆 푡).
    - Carbon intensity and transition opportunity score (fund-level).
    - Lagged flows, lagged returns, ln(fund size), fund age, expense ratio, category-year, region-year fixed effects.
  - Panel estimated for each climate news event separately in baseline.
- Robustness checks:
  - Panel regression with all nine news events as shock source.
  - Alternative specification with fund-year fixed effects.
  - Separating carbon intensity and transition opportunity into two equations.
  - Using individual news index series as source of events.
- Key empirical findings (Online Annex Figure 3.5.2):
  - Climate news events have had limited impact on fund flows regardless of funds’ carbon intensity and transition opportunity score.
  - Following a climate news event:
    - Funds with a higher transition opportunity score have enjoyed higher returns on average.
    - Funds with higher carbon intensity experienced lower performance.
    - Funds with higher carbon intensity appear to have reduced their carbon exposure.
    - Funds with higher transition opportunity score further increased this score.
  - Results robust to pooled panel model and alternative sustainability identifier (highest 32.5 or 10 percentiles), but not robust to replacement by fund and year fixed effects in some specifications.
- Note on statistical reporting:
  - Solid bars in figures denote statistical significance at the 10 percent level or less.

### Funds’ Cash Balances and Transition-Related Scores
- Estimated model (Equation (2) in this annex) assesses impact of transition opportunity score and carbon intensity on fund cash balances:
  - Dependent variable: Cash (fund’s net liquidity buffers held as cash or cash equivalents in quarter t).
  - Key regressors and interactions:
    - Label (dummy for sustainable fund), Carbon (carbon intensity), Opportunity (transition opportunity score).
    - Interactions of label with Carbon and Opportunity.
  - Fund-specific controls: lagged flows, log size, annual expense ratio, ETF dummy.
  - Macro-financial controls: VIX, term spread (10-year minus 3-month US treasuries), credit risk spread (CSI BBB/Baa corporate debt index minus 10-year treasury yield), 3-month US treasury yield, Bloomberg US dollar index versus ten leading currencies.
  - Fixed effects: domicile region × year and fund category × year.
  - Estimation methods: OLS and unconditional quantile regressions to test heterogeneous effects across distribution of cash buffers.
- Robustness analysis:
  - Results remain qualitatively similar in pooled panel model.
  - Results partially robust when replacing sustainability label with identifier based on high sustainability scores (top 32.5 or 10 percentiles); in that replacement, interaction coefficients with sustainability variable become positive and significant, offsetting coefficients for transition opportunity and carbon intensity for some deciles.
  - Inclusion of news events and their interactions does not change main results.
  - Results not robust when replacing fixed-effects specification with fund and year fixed effects.

### Sensitivities to One Standard Variation Shocks — Cash Buffer Deciles and Flow-Performance Model
- Flow-Performance Model Specification:
  - Model evaluated:
    Y_{i,c,r,t} = β1 (fFUND_{i,c,r,t-1} / SLSCORE_{i,c,r,t-1}) + β2 LAGFUND_{i,c,r} + β3 LAGFUND_{i,c,r} x (fFUND_{i,c,r,t-1} / SLSCORE_{i,c,r,t-1}) + β4 CFFSCORE_{i,c,r,t} + ν_{c,y} + δ_{r,y} + ε_{i,c,r,t}
  - Dependent variable: contemporaneous fund flows or returns.
  - Right-hand side includes:
    - Lagged flows and returns.
    - Dummy indicating whether fund has a sustainable label.
    - Interactions of the sustainability dummy with lagged returns and flows.
    - Fund-specific controls: fund log size, age, and annual expense ratio (all in quarter t).
    - Fixed effects: combined domicile region r and year y, and combined fund category c and year y.
  - Sustainability labels: union of labels from Bloomberg, Lipper, Morningstar, and a textual analysis of fund names.
  - Estimation approaches: OLS for mean coefficients and unconditional quantile models (recentered influence function regression methodology) for quantile coefficients to assess heterogeneity across the distribution.
- Key Notes on Figures and Significance:
  - Solid bars denote statistical significance at the 10 percent level or less.
  - In panel 2, the high sustainability score is based on the top 32.5 percentiles.
  - The 5th globe sustainability dummy is the sustainability identifier based on the highest decile of the cross-sectional distribution of funds’ sustainability scores as described in the main text.

### Robustness Analysis — Main Checks and Outcomes
- Pooled panel estimation with no fixed effects:
  - Mean regressions: results remain robust for flow persistence, but not for the lower sensitivity of sustainable fund flows to past returns.
  - Quantile regressions: results remain robust (Annex Figure 3.6.1., panels 1-2).
- Alternative fixed effects specification (replace joint fixed effects with fund-year fixed effects):
  - Results for both mean and quantile regressions are robust.
  - Magnitude of persistence is reduced in the mean regression and for higher flow deciles.
- Replacing sustainability label with identifier based on highest quantiles of funds’ portfolio sustainability scores (top 10 percentiles):
  - Flows are not significantly less sensitive to past performance for sustainable funds in this specification.
  - Reinforcement of flow persistence for sustainable funds flips to a marginal moderation (Online Annex Figure 3.6.1, panels 3 and 4).
- Controlling for strong vs weak past performance and for high vs low volatility in financial markets (mean regression):
  - Equation (1) augmented by triple interaction: dummy for whether previous-quarter return was negative, past returns, and sustainability indicator.
  - Raw return and FF3 alpha used separately.
  - Results using FF3 alpha point to lower sensitivity of sustainable funds to past poor performances; these results do not hold for raw returns.
  - High-volatility dummy constructed as the top one-third quantile of the VIX distribution.
  - Results suggest sensitivity of sustainable funds has not differed from that of conventional funds during high volatility episodes.
- Restricting sustainability labels to each of the four sources within mean regressions:
  - Lower sensitivity of flow to past performance, and persistence of flows for funds whose name include a sustainability term are confirmed.
  - Same conclusion cannot be drawn for funds with climate-related terms in their names.

### Flow Sensitivity and Flow Persistence Results (Descriptive)
- Flow sensitivity to lagged returns reported as:
  - Percent, for 1 percentage point shock to lagged returns; flows normalized by lagged total net assets.
- Flow persistence reported as:
  - Percent, for 1 percentage point shock to lagged flows; flows normalized by lagged total net assets.
- Baseline results qualitatively hold if fixed effect model replaced by pooled panel model for both flow-return sensitivities and flow persistence.
- Using a high sustainability score dummy instead of sustainability label:
  - Return sensitivities: results qualitatively unchanged.
  - Flow persistence: for funds with a high sustainability score, flow persistence is lower than for other funds.

### Online Annex 3.7 — Box on Survey of Asset Managers
- Survey timing: July 21-August 16, 2021.
- Respondents: 26 respondents representing 1 1 large and very large asset managers, and 1 asset owner.
- Respondent geographies: United States, Advanced Europe, Australia, and Japan.
- Fund types managed by surveyed portfolio managers: equity funds, fixed income funds, private debt funds, or infrastructure funds.
- Multiple-choice questions covered:
  1. Integration of climate change mitigation in investment strategy for funds with a sustainability/environment/climate label (options include Negative/exclusionary screening; ESG/Climate change risk integration; Positive/best-in-class screening; Sustainability/climate-themed investment; Impact investing; Other).
  2. Tools/heuristics used to incorporate transition risks and opportunities (options include Sector/industry classification; ESG-type score; Carbon footprint related to securities in portfolio; Expected carbon reduction related to securities in portfolio; Valuation model / scenario analysis; Other).
  3. Obstacles in integrating transition risks and opportunities (ranked 1=most severe to 4=less severe): Lack of current data; Lack of forward-looking data; Lack of a commonly accepted taxonomy; Multiplicity of disclosure standards.
  4. Rank the three most important factors for portfolio decisions over the next 3 years (1 = most important to 3 = least important) for risks (Increase in intensity/severity of climate-change-related physical events; Increase in carbon taxation and emissions-related regulation; Technological change; Changes to consumer preferences towards green products and services; Litigation) and opportunities (Climate change mitigation; Climate change adaptation; Technological change; Changes to consumer preferences towards green products and services).
- Box Figure 3.1.1. shows survey results by institution for the first two questions, and for every individual respondent for the third and fourth questions.

*Sources: FactSet; Morningstar; Refinitiv; Bloomberg Finance L.P.; Lipper; and IMF staff calculations.*

### 1. Transition-Opportunity Score Distribution, Sustainable versus

### online-annex-ch3 - 1. Transition-Opportunity Score Distribution, Sustainable versus

### Distributional Figures and Decomposition (2020:Q4 and yearly changes)
- Figures reported (descriptive distributions, 2020:Q4):
  - Transition-Opportunity Score Distribution, Sustainable versus Conventional Funds (x-axis: score 0-100; y-axis: percent).
  - Carbon-Intensity Score Distribution, Sustainable versus Conventional Funds (x-axis: tons of CO2 equivalent per million dollars of revenue; y-axis: percent).
  - Morningstar Carbon Solutions Involvement Distribution, Climate versus Conventional Funds (x-axis: score 0-100; y-axis: percent).
  - Morningstar Carbon Risk Score Distribution, Climate versus Conventional Funds (x-axis: score 0-100; y-axis: percent).
  - Alternative Transition-Opportunity Score Distribution, Climate versus Conventional Funds (x-axis: score 0-100; y-axis: percent).
  - Ownership of Firms by Type of Fund, Various Industries (Percent of total market capitalization).
- Decomposition (Online Annex Figure 3.2.2):
  - Yearly Change in Transition Opportunity Score (Change in index; index takes values between 0 and 100).
  - Yearly Change in Carbon Intensity (Tons of CO2 equivalent emissions per million dollars of revenue).
- Data sources: Morningstar; Bloomberg Finance L.P.; Lipper; FactSet; Refinitiv; and IMF staff.

### Fund Flow Analysis: Model and Data
- Estimated model (Equation (1)) examines relation between fund sustainability labels, Morningstar scores, and net fund flows:
  - Dependent variable: net flow into fund 푖 (푓푓퐹퐹퐹퐹퐹퐹퐹퐹
    푖,푐,푟,푡).
  - Key regressors:
    - Vector of dummies capturing fund sustainability label, environment label, and climate label (퐿퐿퐿퐿퐿퐿퐿퐿퐹퐹).
    - Dummy for top decile of funds by broad sustainability rating.
    - Transition opportunity and carbon intensity scores.
    - Fund-level controls: lagged flows, lagged returns, logarithm of fund size, expense ratio, fund age.
  - Fixed effects: category-year (휈휈
    푐,푦) and region-year (훿훿
    푟,푦).
  - Standard errors clustered at the fund level.
- Sample:
  - 6,454 funds from 33 countries.
  - Sample period: 2010:Q1 to 2020:Q4.

### Proxy Voting Analysis: Model and Data
- Estimated model (Equation (2)) examines relation between labels/scores and proxy voting:
  - Dependent variable: percentage of votes on climate resolutions by the fund that were cast in favor.
  - Key regressors similar to Equation (1); climate label not included separately due to very small number of observations.
  - Additional controls include a passive-fund dummy, logarithm of fund size, expense ratio, fund age.
- Sample:
  - 1,521 funds from the United States.
  - Sample period: 2015 to 2020.

### Issuance Analysis (Firm-Level Securities Issuance)
- Coverage and data:
  - Firms in transition-sensitive sectors issuing bonds or SEOs at least once during 2010:Q1-2021:Q1.
  - Total number of firms: 6,449, of which 5,446 issued equity and 3,722 issued bonds.
- Regression specification (fixed-effect panel):
  - Dependent variable: issuance measures on the intensive margin (issuance amount conditional on issuance) and extensive margin (likelihood of issuance).
  - Key regressors and interactions:
    - Firm “Green” indicators: ESG score, E score, transition opportunity score, carbon intensity (higher value = greener).
    - Flow exposure variables:
      - 퐹퐹퐹퐹퐹퐹퐹퐹_푆푆푆푆푆푆푆푆 and 퐹퐹퐹퐹퐹퐹퐹퐹_푆푆퐹퐹푆푆 (exposure to net inflows to sustainable and conventional funds respectively), defined as weighted averages of net flows into funds holding a firm’s securities (see Equation (2) and (3)).
  - Controls: sector fixed effects (훼훼
    푠), quarter fixed effects (휌휌
    푡), market capitalization, leverage ratio, market-to-book ratio, ROA, tangibility, short-term-debt-to-asset ratio, stock returns (measured at the end of the first lag of quarter t’s calendar year).
- Notes:
  - Quarterly values of firm-level green indicators obtained via linear interpolation of yearly values.
  - Identification of transition-sensitive sectors: SIG codes in Dealogic hand-matched with 4-digit NACE Rev2 sector codes; housing sector excluded.

### Stock Return Analysis: Price Pressure from Flows
- Price-pressure measure constructed as in Khan et al. (2012):
  - Difference between buying pressure from high-net-inflow funds and selling pressure from high-net-outflow funds, normalized by shares outstanding (Equation (4)).
  - Measures constructed separately for sustainable-fund flows and conventional-fund flows.
- Econometric approach:
  - Regress firm i’s h-quarter-ahead CAPM-based abnormal returns on current-quarter price pressure from sustainable funds, the firm’s lagged green score, and their interaction.
  - Controls include same variables as issuance regressions plus price pressure from conventional funds and its interaction with the firm green score.
- Key empirical findings:
  - A one percentage point increase in price pressure leads to a significantly larger contemporaneous increase in abnormal returns for firms with a higher ESG score or E score (Online Figure 3.4.1).
  - There is almost no difference in price impact for firms with higher transition opportunity score or lower carbon intensity.

### Climate Change News, Fund Performance, and Scores
- Climate news events:
  - Identified from four indices based on New York Times, Wall Street Journal, and Google News; aggregated to quarterly frequency.
  - Dummies constructed for top decile of quarterly distribution of each index; union of dummies yields nine quarters of heightened climate news attention.
  - Examples: Paris Agreement in 2015:Q4 and announcement of the US withdrawal from the Paris Agreement in 2017:Q2.
- Estimated model (Equation (1) in this annex) for news impact:
  - Dependent variable Y: fund return, net inflow, change in carbon intensity, or change in transition opportunity score.
  - Key regressors:
    - Climate news dummy (푁푁퐿퐿퐹퐹푆푆
      푡).
    - Carbon intensity and transition opportunity score (fund-level).
    - Lagged flows, lagged returns, ln(fund size), fund age, expense ratio, category-year, region-year fixed effects.
  - Panel estimated for each climate news event separately in baseline.
- Robustness checks:
  - Panel regression with all nine news events as shock source.
  - Alternative specification with fund-year fixed effects.
  - Separating carbon intensity and transition opportunity into two equations.
  - Using individual news index series as source of events.
- Key empirical findings (Online Annex Figure 3.5.2):
  - Climate news events have had limited impact on fund flows regardless of funds’ carbon intensity and transition opportunity score.
  - Following a climate news event:
    - Funds with a higher transition opportunity score have enjoyed higher returns on average.
    - Funds with higher carbon intensity experienced lower performance.
    - Funds with higher carbon intensity appear to have reduced their carbon exposure.
    - Funds with higher transition opportunity score further increased this score.
  - Results robust to pooled panel model and alternative sustainability identifier (highest 32.5 or 10 percentiles), but not robust to replacement by fund and year fixed effects in some specifications.
- Note on statistical reporting:
  - Solid bars in figures denote statistical significance at the 10 percent level or less.

### Funds’ Cash Balances and Transition-Related Scores
- Estimated model (Equation (2) in this annex) assesses impact of transition opportunity score and carbon intensity on fund cash balances:
  - Dependent variable: Cash (fund’s net liquidity buffers held as cash or cash equivalents in quarter t).
  - Key regressors and interactions:
    - Label (dummy for sustainable fund), Carbon (carbon intensity), Opportunity (transition opportunity score).
    - Interactions of label with Carbon and Opportunity.
  - Fund-specific controls: lagged flows, log size, annual expense ratio, ETF dummy.
  - Macro-financial controls: VIX, term spread (10-year minus 3-month US treasuries), credit risk spread (CSI BBB/Baa corporate debt index minus 10-year treasury yield), 3-month US treasury yield, Bloomberg US dollar index versus ten leading currencies.
  - Fixed effects: domicile region × year and fund category × year.
  - Estimation methods: OLS and unconditional quantile regressions to test heterogeneous effects across distribution of cash buffers.
- Robustness analysis:
  - Results remain qualitatively similar in pooled panel model.
  - Results partially robust when replacing sustainability label with identifier based on high sustainability scores (top 32.5 or 10 percentiles); in that replacement, interaction coefficients with sustainability variable become positive and significant, offsetting coefficients for transition opportunity and carbon intensity for some deciles.
  - Inclusion of news events and their interactions does not change main results.
  - Results not robust when replacing fixed-effects specification with fund and year fixed effects.

Sources: FactSet; Morningstar; Refinitiv; Bloomberg Finance L.P.; Lipper; and IMF staff calculations.

*International Monetary Fund | October 2021 — Online annex to Chapter 3 of the Global Financial Stability Report — Investment Funds: Fostering the Transition to A Green Economy*

### 1. Sensitivities to One Standard Variation Shocks, Various Deciles of

### 1. Sensitivities to One Standard Variation Shocks, Various Deciles of the Cash Buffer Distribution

### Flow-Performance Model Specification
- The flow-performance relationship is evaluated using the model:
  Y_{i,c,r,t} = β1 (fFUND_{i,c,r,t-1} / SLSCORE_{i,c,r,t-1}) + β2 LAGFUND_{i,c,r} + β3 LAGFUND_{i,c,r} x (fFUND_{i,c,r,t-1} / SLSCORE_{i,c,r,t-1}) + β4 CFFSCORE_{i,c,r,t} + ν_{c,y} + δ_{r,y} + ε_{i,c,r,t}
- Dependent variable: contemporaneous fund flows or returns.
- Right-hand side includes:
  - Lagged flows and returns.
  - Dummy indicating whether fund has a sustainable label.
  - Interactions of the sustainability dummy with lagged returns and flows.
  - Fund-specific controls: fund log size, age, and annual expense ratio (all in quarter t).
  - Fixed effects: combined domicile region r and year y, and combined fund category c and year y.
- Sustainability labels: union of labels from Bloomberg, Lipper, Morningstar, and a textual analysis of fund names (see Online Annex 3.1).
- Estimation approaches: OLS for mean coefficients and unconditional quantile models (recentered influence function regression methodology) for quantile coefficients to assess heterogeneity across the distribution.

### Key Notes on Figures and Significance
- Solid bars denote statistical significance at the 10 percent level or less.
- In panel 2, the high sustainability score is based on the top 32.5 percentiles.
- The 5th globe sustainability dummy is the sustainability identifier based on the highest decile of the cross-sectional distribution of funds’ sustainability scores as described in the main text.

### Robustness Analysis — Main Checks and Outcomes
- Pooled panel estimation with no fixed effects:
  - Mean regressions: results remain robust for flow persistence, but not for the lower sensitivity of sustainable fund flows to past returns.
  - Quantile regressions: results remain robust (Annex Figure 3.6.1., panels 1-2).
- Alternative fixed effects specification (replace joint fixed effects with fund-year fixed effects):
  - Results for both mean and quantile regressions are robust.
  - Magnitude of persistence is reduced in the mean regression and for higher flow deciles.
- Replacing sustainability label with identifier based on highest quantiles of funds’ portfolio sustainability scores (top 10 percentiles):
  - Flows are not significantly less sensitive to past performance for sustainable funds in this specification.
  - Reinforcement of flow persistence for sustainable funds flips to a marginal moderation (Online Annex Figure 3.6.1, panels 3 and 4).
- Controlling for strong vs weak past performance and for high vs low volatility in financial markets (mean regression):
  - Equation (1) augmented by triple interaction: dummy for whether previous-quarter return was negative, past returns, and sustainability indicator.
  - Raw return and FF3 alpha used separately.
  - Results using FF3 alpha point to lower sensitivity of sustainable funds to past poor performances; these results do not hold for raw returns.
  - High-volatility dummy constructed as the top one-third quantile of the VIX distribution.
  - Results suggest sensitivity of sustainable funds has not differed from that of conventional funds during high volatility episodes.
- Restricting sustainability labels to each of the four sources within mean regressions:
  - Lower sensitivity of flow to past performance, and persistence of flows for funds whose name include a sustainability term are confirmed.
  - Same conclusion cannot be drawn for funds with climate-related terms in their names.

### Flow Sensitivity and Flow Persistence Results (Descriptive)
- Flow sensitivity to lagged returns reported as:
  - Percent, for 1 percentage point shock to lagged returns; flows normalized by lagged total net assets.
- Flow persistence reported as:
  - Percent, for 1 percentage point shock to lagged flows; flows normalized by lagged total net assets.
- Baseline results qualitatively hold if fixed effect model replaced by pooled panel model for both flow-return sensitivities and flow persistence.
- Using a high sustainability score dummy instead of sustainability label:
  - Return sensitivities: results qualitatively unchanged.
  - Flow persistence: for funds with a high sustainability score, flow persistence is lower than for other funds.

### Online Annex 3.7 — Box on Survey of Asset Managers
- Survey timing: July 21-August 16, 2021.
- Respondents: 26 respondents representing 1 1 large and very large asset managers, and 1 asset owner.
- Respondent geographies: United States, Advanced Europe, Australia, and Japan.
- Fund types managed by surveyed portfolio managers: equity funds, fixed income funds, private debt funds, or infrastructure funds.
- Multiple-choice questions covered:
  1. Integration of climate change mitigation in investment strategy for funds with a sustainability/environment/climate label (options include Negative/exclusionary screening; ESG/Climate change risk integration; Positive/best-in-class screening; Sustainability/climate-themed investment; Impact investing; Other).
  2. Tools/heuristics used to incorporate transition risks and opportunities (options include Sector/industry classification; ESG-type score; Carbon footprint related to securities in portfolio; Expected carbon reduction related to securities in portfolio; Valuation model / scenario analysis; Other).
  3. Obstacles in integrating transition risks and opportunities (ranked 1=most severe to 4=less severe): Lack of current data; Lack of forward-looking data; Lack of a commonly accepted taxonomy; Multiplicity of disclosure standards.
  4. Rank the three most important factors for portfolio decisions over the next 3 years (1 = most important to 3 = least important) for risks (Increase in intensity/severity of climate-change-related physical events; Increase in carbon taxation and emissions-related regulation; Technological change; Changes to consumer preferences towards green products and services; Litigation) and opportunities (Climate change mitigation; Climate change adaptation; Technological change; Changes to consumer preferences towards green products and services).
- Box Figure 3.1.1. shows survey results by institution for the first two questions, and for every individual respondent for the third and fourth questions.

*Sources: Bloomberg Finance L.P.; FactSet; Morningstar; Refinitiv; and IMF staff calculations.*

---


_Source: https://www.imf.org/-/media/files/publications/gfsr/2021/october/english/online-annex-ch3.pdf_