## 2.1 Emissions data

## Source details

**Canonical URL:** [2.1 Emissions data](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021112-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021112-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021112-print-pdf.pdf.json)

---

### Emissions data source and construction
- Main variable: carbon dioxide emissions at the firm level.
- Data source: Climate Accountability Institute (CAI) compiled from company regulatory filings and reports to shareholders; raw daily production data converted to annual production.
- Conversion methods: physical quantities of fossil fuels converted to emissions using conversion factors following international standards and conventions in IPCC reports.
- Example (Royal Dutch Shell, 2017):
  - Oil: ‘net equity’ production = 666 Mb (million barrels) → 247 MtCO2 using conversion factor 0.3714 MtCO2/Mb.
  - Gas: ‘gas available for sale’ = 3,894 billion cubic feet (Bcf) → 208 MtCO2 using conversion factor 0.05343 MtCO2/Bcf.
  - Combined oil and gas emissions attributed to Shell in 2017 = 456 MtCO2.

### Emissions scopes and dataset scope
- Emissions scopes used:
  - Scope 1: direct emissions from firms’ activities.
  - Scope 2: indirect emissions from companies’ operations (excluded in this analysis to prevent double-counting).
  - Scope 3: emissions from customers using a company’s products as intended or other non-controlled sources (included).
- Dataset covers Scopes 1 and 3 as defined by the World Resource Institute (WRI); Scope 2 is excluded.
- Coverage rules and adjustments:
  - Includes firms with emissions above 8 million tons of carbon (MtC) per year and reporting on a consolidated basis.
  - Historical mergers and acquisitions are accounted for by back-filling series pro-forma for the combined entity.
- Full dataset composition:
  - Total firms: 96
  - 67 oil and gas producers
  - 25 coal producers
  - 4 cement manufacturers
  - Ownership: 64 investor-owned firms and 32 state-owned enterprises
- ESG availability: ESG data obtainable for 61 of the 64 investor-owned firms.
- Cumulative-emissions coverage:
  - Firms in the dataset account for about 70 percent of cumulative greenhouse gas emissions since 1850.
  - Firms account for 86 percent of the cumulative flow of carbon dioxide emissions since 2002.

### Panel construction and sample used in analysis
- Time period: 2002 to 2017 (ESG scores begin in 2002; emissions available through 2017).
- Exchange rates: IMF International Financial Statistics used to convert firm financial data into U.S. dollars.
- Panel composition:
  - 61 investor-owned firms headquartered in 23 countries form the constructed panel covering 2002–2017; these represented 29.4 percent of global cumulative emissions.
  - Baseline analysis subset: 52 firms with at least 10 years of data, representing 27 percent of global cumulative emissions.
- Note: Data on ownership by institutional investors begins in 2005.

### ESG data description
- ESG source: Refinitiv (Thomson Reuters).
- Scores collected:
  - Overall ESG score and scores for three pillars: Environmental (E), Social (S), Governance (G).
  - Component scores for pillar subcomponents.
  - Score range: 1 to 100.
- Pillar descriptions:
  - Environmental: measures of how companies’ operations impact the environment (e.g., emissions targets, resource efficiency, waste management).
  - Social: company relationships with employees, suppliers, and communities (e.g., flexible work hours, day care services).
  - Governance: effectiveness of management and practices; components include management, corporate social responsibility, and shareholders; governance scores are comparable across industries within firms with the same country of incorporation.
- Caveats on ESG measurement:
  - No industry standard or regulatory guidance; scores can diverge across providers.
  - Disagreements across providers arise from different choices of ESG attributes and assessments of performance.
  - Gibson et al. (2021) find disagreements particularly acute for the governance pillar.
- Despite measurement issues, direct firm-level emissions estimates are available for the large emitters focused on in this study.

### Key methodological points for empirical analysis
- Outcome variable: Yit = log difference in emissions (log differences multiplied by 100 in regressions).
- Main regressors: standardized ESG scores (unit variance); coefficients interpreted as percentage point difference in emissions growth associated with a one standard deviation move in ESG scores.
- Regression specification:
  - Fixed effects model: Yit = αi + αt + β1 Scoreit + γ Xit + εit
  - Fixed effects: firm fixed effects (αi) and year fixed effects (αt); alternative specification uses country instead of firm fixed effects.
  - Controls Xit: lag of log total assets, lagged log revenue, lagged leverage (debt-to-asset ratio), lagged percent change in fixed asset growth, current and lagged real GDP growth, inflation, and unemployment rates.
  - Standard errors: double clustered at firm and year levels (results similar if clustered only at the firm level).
- Sample choices:
  - Baseline focuses on within-firm variation (firm fixed effects) and the subset of firms with at least 10 years of data.
  - Alternative specifications and robustness checks include country fixed effects, different timing (lags/leads of emissions), orthogonalized pillars, scaling emissions by assets or revenue, and broader firm sets.

### Main empirical findings (high-level)
- Raw cross-section and event-period plots:
  - Emissions growth appears largely unrelated to overall ESG score, Environmental pillar score, or Governance pillar score in the cross-section and in four-year windows of largest ESG improvement or largest emissions growth reduction.
- Regressions (overall ESG score):
  - In specification with only year fixed effects, a one standard deviation improvement in the overall ESG score is associated with a 1.1 percentage point reduction in emissions growth.
  - When firm fixed effects or country fixed effects are included (saturated fixed-effects specifications), the relationship is not statistically significant.
  - Interpretation: at best a weak link between ESG and emissions growth; unobservable country and firm characteristics matter.
- Institutional ownership interactions:
  - Institutional ownership is greater for large emitters with faster growing emissions (Panel B, Table 1, year-FE specification).
  - The negative link between ESG scores and emissions growth is concentrated in firms with higher institutional ownership in simple specifications, but the interaction is not statistically significant when firm or country fixed effects are included.
- Environment pillar and components:
  - The Environmental pillar and its components (including Emissions) are essentially unrelated to emissions growth at the firm level in specifications with firm fixed effects; coefficients are small or positive and not statistically significant.
  - Positive coefficients may reflect firms with growing emissions making efforts to improve ratings without slowing emissions growth.
- Governance pillar:
  - A weak negative association: a one standard deviation within-firm improvement in the Governance pillar score or the management component is associated with a 0.7-0.8 percentage point reduction in emissions growth (statistical significance is marginal).
- Size effects:
  - The connection between ESG scores and emissions growth is weaker for larger firms in the panel (Appendix Table A.1).
  - Larger firms may be more effective at managing ESG scores without changing emissions, potentially by emphasizing environmental innovation in disclosures (Appendix Table A.2 and A.3).
- Robustness and additional checks:
  - Results similar when examining emissions in subsequent or prior two years (Appendix Table A.5).
  - No strong link between ESG scores and levels of emissions (Appendix Table A.6).
  - Results remain similar when using the broadest set of firms (not restricting to those with at least 10 years of data) (Appendix Table A.7) and when not winsorizing variables (Appendix Table A.8).
  - Scaling emissions by assets or revenue and examining the Social pillar yields broadly unchanged or somewhat weaker results (Appendix Tables A.9–A.14).
  - No statistically significant connection between overall ESG scores and emissions intensity even with only year fixed effects (Appendix Table A.9).
  - No statistically significant relationship between the Governance Pillar (or management component) and emissions scaled by revenue or assets (Appendix Tables A.10–A.11).

### Key empirical estimates (selected tables and coefficients)
- Main firm-year regression (Panel A, Table 1): Overall ESG Score coefficient on ∆ln(Emissions) = -1.1** (0.39) in the first specification; alternative specs show -0.5 (0.45) and -0.5 (0.55).
- Timing robustness (Table A.5):
  - Emissions two years later: Overall ESG Score = -1.2*** (0.32); other specs -0.8* (0.38) and -0.6 (0.47).
  - Emissions one year later: Overall ESG Score = -1.1** (0.39); other specs -0.4 (0.41) and -0.0 (0.50).
  - Emissions one year ago: Overall ESG Score = -1.1** (0.39); other specs -0.6 (0.45) and -0.4 (0.50).
  - Emissions two years ago: Overall ESG Score = -1.0* (0.55); other specs -0.1 (0.67) and 0.7 (0.98).
- Full-sample replication (Table A.7): Overall ESG Score = -1.2*** (0.36); alternative specs -0.8* (0.43) and -0.5 (0.52).
- Non-winsorized specification (Table A.8): Overall ESG Score = -1.3* (0.59); other specs -0.6 (0.53) and -0.6 (0.90).

### Panel B interaction (institutional ownership) — selected coefficients
- Overall ESG Score = 0.1 (0.51), 0.5 (0.91), -0.3 (0.80) across specs.
- Inst ownership = 0.9*** (0.26), 0.6* (0.34), 0.1 (0.45).
- Overall ESG Score × Inst ownership = -0.2*** (0.04), -0.1 (0.07), -0.0 (0.06).

### Pillar- and component-level coefficients (selected)
- Environment pillar (Table 2 — Panel A):
  - Environment = 0.3 (0.59) in the baseline specification.
  - Emissions (component) = 0.5 (0.54).
  - Resource Use = -0.1 (0.51).
  - Env. Innovation = 0.0 (0.31).
- Governance pillar (Table 2 — Panel B):
  - Governance = -0.8* (0.44).
  - Management = -0.7* (0.39).
  - CSR Strategy = -0.5 (0.37).
  - Shareholders = -0.2 (0.45).
- Social pillar (Appendix Table A.12):
  - Social = -0.3 (0.44).
  - Workforce = 0.4 (0.50).
  - Community = -0.1 (0.36).
  - Human Rights = -0.4 (0.28).
  - Product Responsibility = -0.5* (0.26).
- Appendix Table A.6 — Level of emissions:
  - Overall ESG Score on ln(Emissions) = 0.20** (0.08); alternative specs 0.11 (0.08) and 0.03 (0.04).
  - Pillars on ln(Emissions): Environment = 0.03 (0.04); Governance = 0.04 (0.02); Social = -0.01 (0.02).

### Firm size interactions (selected)
- Table A.1 (Overall ESG × Log Lag Assets):
  - Overall ESG Score = -10.4** (3.77), -9.1* (4.29), -5.0 (4.59) across specs.
  - Log Lag Assets = -0.5 (1.58), -1.7 (1.79), -1.0 (1.66).
  - Overall ESG Score × Log Lag Assets = 0.4** (0.14), 0.3* (0.17), 0.2 (0.18).
- Table A.2 (Environment pillar × firm size components):
  - Environment = -6.6 (4.68); Environment × Log Lag Assets = 0.3 (0.19).
  - Emissions component = -0.9 (3.36); Emissions × Log Lag Assets = 0.1 (0.12).
  - Resource Use = -7.1 (4.63); Resource Use × Log Lag Assets = 0.3 (0.19).
  - Env. Innovation = -7.4* (3.54); Env. Innovation × Log Lag Assets = 0.3* (0.14).

### Mechanisms and inputs to Environment scores (Table A.3)
- Attempt reduction in volatile organic compounds → Overall ESG Score = 0.56*** (0.12); Environment = 0.72*** (0.11).
- Recognize climate risks & opportunities → Overall ESG Score = 0.60*** (0.10); Environment = 0.52*** (0.10).
- Report environmental investments → Overall ESG Score = 0.41*** (0.11); Environment = 0.30** (0.12).
- Report environmental partnerships → Overall ESG Score = 0.34*** (0.11); Environment = 0.37*** (0.10).
- Orthogonalized Environment score (Table A.4): Environment (orth.) = -0.5 (0.49), 0.2 (0.53), 0.7 (0.48) across specs.

### Emissions intensity (scaled by revenue or assets) — selected
- Appendix Table A.9 — Overall ESG Score on emissions scaled by revenue:
  - Overall ESG Score = 0.0 (0.51), 0.0 (0.85), 0.3 (1.19).
- Appendix Table A.9 — Overall ESG Score on emissions scaled by assets:
  - Overall ESG Score = -0.3 (0.48), -0.7 (0.45), -1.4 (1.05).
- Tables A.10 and A.11 repeat pillar analyses for emissions scaled by revenue and assets with generally imprecise coefficients.

### Global carbon budgets and scenario analysis (Table 3)
- Assumptions:
  - Budgets from IPCC (2018a) allocated to 52 firms in sample by 2017 share of global carbon emissions.
  - Business-as-usual growth rate for firms 2002-2017 = 3.85 percent.
  - A one standard-deviation improvement in ESG assumed to reduce emissions growth by 1.1 percent (based on Panel A, Table 1 first spec).
- Years left (business-as-usual) and years gained by ESG improvement:
  - Probability of warming under 1.5°C:
    - 66 percent: Years left = 10.6; Years gained by ESG +1SD = 0.6; ESG +2SD = 1.3.
    - 50 percent: Years left = 13.8; Years gained by ESG +1SD = 0.9; ESG +2SD = 2.1.
    - 33 percent: Years left = 18.2; Years gained by ESG +1SD = 1.6; ESG +2SD = 3.6.
  - Probability of warming under 2°C:
    - 66 percent: Years left = 22.9; Years gained by ESG +1SD = 2.4; ESG +2SD = 5.7.
    - 50 percent: Years left = 26.9; Years gained by ESG +1SD = 3.3; ESG +2SD = 7.7.

### Visual and descriptive evidence
- Figure 1: Stock and flow of global carbon emissions traced to large emitters; Panel A shows stock since 1940; Panel B shows global carbon emissions since 2002 with emitter-type shares by year.
- Figure 2: ESG and emissions growth cross-sectional bubble plots for Overall ESG, Governance, and Environment pillars; columns show full sample average, four-year period with largest increase in emissions, and four-year period with largest improvement in ESG scores; shading indicates economic conditions (darker red = slower economic growth); outliers trimmed.
- Appendix Figures A.1–A.5: comparisons of ESG scores for large emitters vs S&P500; alternative scaling of emissions (revenue, assets); Social pillar plots; level of emissions panels.

### Robustness and sensitivity checks
- Firm-year fixed effects, country controls (GDP growth, unemployment, inflation), and firm controls (lagged log assets, lagged log revenue, lagged percent change in leverage, lagged fixed asset growth) are used across specifications.
- Results presented with standard errors double-clustered at firm and year levels; significance denoted *, **, *** for 10 percent, 5 percent, and 1 percent levels respectively.
- Multiple robustness tables examine timing, winsorization, full sample, orthogonalized pillars, interactions with firm size, and emissions intensity measures.

### Scenario analysis and implications
- Method:
  - Estimate firms' future carbon dioxide emissions based on their average growth rate over the sample period.
  - Calculate cumulative emissions and the difference between when they would run out of their carbon budget under the status quo versus under a scenario with slower emissions growth.
- Main quantitative findings:
  - With business as usual, less than 14 years are left before the odds of containing global warming to under 1.5 degrees Celsius would be worse than even.
  - The reduction in emissions growth associated with a two standard deviation improvement in ESG scores would buy only an additional two years before the 1.5 degrees Celsius objective would be out of reach.
  - The number of years gained is similarly small relative to the baseline across a range of different climate objectives.
- Implications for ESG investing and policy:
  - Weak link between ESG scores and emissions growth: Even reductions in emissions growth corresponding to large improvements in ESG scores would do little to help with climate change.
  - High ESG scores are not necessarily related to companies’ actual carbon footprints.
  - Sustainable investing strategies conditioned solely on ESG indicators have limited scope to meaningfully mitigate climate change by shifting production incentives for large emitters.
  - Environmentally-conscious investors, shareholders, and policymakers should note the limited translation of ESG-score improvements into emissions reductions.
  - Central banks managing the climate footprint of their interventions should be wary of over-reliance on ESG indicators.

### Conclusion
- Although greater attention to ESG considerations could in principle lead firms to improve ESG scores and reduce emissions, an important prerequisite is that firms that reduce their contributions to climate change be rewarded with better ESG scores.
- Using data on a small but important set of firms among the largest emitters worldwide, the analysis finds a weak relationship between ESG scores and growth rate of emissions, if any.
- Even a remarkable rise in sustainable investing incentivizing large emitters to significantly improve their ESG scores would not necessarily translate into meaningful momentum in dealing with climate change.

*Source: 2.1 Emissions data — wpiea2021112-print-pdf*

### 2.1    Emissions data

### 2.1 Emissions data

### Emissions data source and construction
- Main variable: carbon dioxide emissions at the firm level.
- Data source: Climate Accountability Institute (CAI) compiled from company regulatory filings and reports to shareholders; raw daily production data converted to annual production.
- Conversion methods: physical quantities of fossil fuels converted to emissions using conversion factors following international standards and conventions in IPCC reports.
- Example (Royal Dutch Shell, 2017):
  - Oil: ‘net equity’ production = 666 Mb (million barrels) → 247 MtCO2 using conversion factor 0.3714 MtCO2/Mb.
  - Gas: ‘gas available for sale’ = 3,894 billion cubic feet (Bcf) → 208 MtCO2 using conversion factor 0.05343 MtCO2/Bcf.
  - Combined oil and gas emissions attributed to Shell in 2017 = 456 MtCO2.

### Emissions scopes and dataset scope
- Emissions scopes used:
  - Scope 1: direct emissions from firms’ activities.
  - Scope 2: indirect emissions from companies’ operations (excluded in this analysis to prevent double-counting).
  - Scope 3: emissions from customers using a company’s products as intended or other non-controlled sources (included).
- The dataset covers Scopes 1 and 3 as defined by the World Resource Institute (WRI); Scope 2 is excluded.
- Coverage rules and adjustments:
  - Includes firms with emissions above 8 million tons of carbon (MtC) per year and reporting on a consolidated basis.
  - Historical mergers and acquisitions are accounted for by back-filling series pro-forma for the combined entity.
- Full dataset composition:
  - Total firms: 96
  - 67 oil and gas producers
  - 25 coal producers
  - 4 cement manufacturers
  - Ownership: 64 investor-owned firms and 32 state-owned enterprises
- ESG availability: ESG data obtainable for 61 of the 64 investor-owned firms.
- Cumulative-emissions coverage:
  - Firms in the dataset account for about 70 percent of cumulative greenhouse gas emissions since 1850.
  - Firms account for 86 percent of the cumulative flow of carbon dioxide emissions since 2002.

### Panel construction and sample used in analysis
- Time period: 2002 to 2017 (ESG scores begin in 2002; emissions available through 2017).
- Exchange rates: IMF International Financial Statistics used to convert firm financial data into U.S. dollars.
- Panel composition:
  - 61 investor-owned firms headquartered in 23 countries form the constructed panel covering 2002–2017; these represented 29.4 percent of global cumulative emissions.
  - Baseline analysis subset: 52 firms with at least 10 years of data, representing 27 percent of global cumulative emissions.
- Note: Data on ownership by institutional investors begins in 2005.

### ESG data description
- ESG source: Refinitiv (Thomson Reuters).
- Scores collected:
  - Overall ESG score and scores for three pillars: Environmental (E), Social (S), Governance (G).
  - Component scores for pillar subcomponents.
  - Score range: 1 to 100.
- Pillar descriptions:
  - Environmental: measures of how companies’ operations impact the environment (e.g., emissions targets, resource efficiency, waste management).
  - Social: company relationships with employees, suppliers, and communities (e.g., flexible work hours, day care services).
  - Governance: effectiveness of management and practices; components include management, corporate social responsibility, and shareholders; governance scores are comparable across industries within firms with the same country of incorporation.
- Caveats on ESG measurement:
  - No industry standard or regulatory guidance; scores can diverge across providers.
  - Disagreements across providers arise from different choices of ESG attributes and assessments of performance.
  - Gibson et al. (2021) find disagreements particularly acute for the governance pillar.
- Despite measurement issues, direct firm-level emissions estimates are available for the large emitters focused on in this study.

### Key methodological points for empirical analysis
- Outcome variable: Yit = log difference in emissions (log differences multiplied by 100 in regressions).
- Main regressors: standardized ESG scores (unit variance); coefficients interpreted as percentage point difference in emissions growth associated with a one standard deviation move in ESG scores.
- Regression specification:
  - Fixed effects model: Yit = αi + αt + β1 Scoreit + γ Xit + εit
  - Fixed effects: firm fixed effects (αi) and year fixed effects (αt); alternative specification uses country instead of firm fixed effects.
  - Controls Xit: lag of log total assets, lagged log revenue, lagged leverage (debt-to-asset ratio), lagged percent change in fixed asset growth, current and lagged real GDP growth, inflation, and unemployment rates.
  - Standard errors: double clustered at firm and year levels (results similar if clustered only at the firm level).
- Sample choices:
  - Baseline focuses on within-firm variation (firm fixed effects) and the subset of firms with at least 10 years of data.
  - Alternative specifications and robustness checks include country fixed effects, different timing (lags/leads of emissions), orthogonalized pillars, scaling emissions by assets or revenue, and broader firm sets.

### Main empirical findings (high-level)
- Raw cross-section and event-period plots:
  - Emissions growth appears largely unrelated to overall ESG score, Environmental pillar score, or Governance pillar score in the cross-section and in four-year windows of largest ESG improvement or largest emissions growth reduction.
- Regressions (overall ESG score):
  - In specification with only year fixed effects, a one standard deviation improvement in the overall ESG score is associated with a 1.1 percentage point reduction in emissions growth.
  - When firm fixed effects or country fixed effects are included (saturated fixed-effects specifications), the relationship is not statistically significant.
  - Interpretation: at best a weak link between ESG and emissions growth; unobservable country and firm characteristics matter.
- Institutional ownership interactions:
  - Institutional ownership is greater for large emitters with faster growing emissions (Panel B, Table 1, year-FE specification).
  - The negative link between ESG scores and emissions growth is concentrated in firms with higher institutional ownership in simple specifications, but the interaction is not statistically significant when firm or country fixed effects are included.
- Environment pillar and components:
  - The Environmental pillar and its components (including Emissions) are essentially unrelated to emissions growth at the firm level in specifications with firm fixed effects; coefficients are small or positive and not statistically significant.
  - Positive coefficients may reflect firms with growing emissions making efforts to improve ratings without slowing emissions growth.
- Governance pillar:
  - A weak negative association: a one standard deviation within-firm improvement in the Governance pillar score or the management component is associated with a 0.7-0.8 percentage point reduction in emissions growth (statistical significance is marginal).
- Size effects:
  - The connection between ESG scores and emissions growth is weaker for larger firms in the panel (Appendix Table A.1).
  - Larger firms may be more effective at managing ESG scores without changing emissions, potentially by emphasizing environmental innovation in disclosures (Appendix Table A.2 and A.3).
- Robustness and additional checks:
  - Results similar when examining emissions in subsequent or prior two years (Appendix Table A.5).
  - No strong link between ESG scores and levels of emissions (Appendix Table A.6).
  - Results remain similar when using the broadest set of firms (not restricting to those with at least 10 years of data) (Appendix Table A.7) and when not winsorizing variables (Appendix Table A.8).
  - Scaling emissions by assets or revenue and examining the Social pillar yields broadly unchanged or somewhat weaker results (Appendix Tables A.9–A.14).
  - No statistically significant connection between overall ESG scores and emissions intensity even with only year fixed effects (Appendix Table A.9).
  - No statistically significant relationship between the Governance Pillar (or management component) and emissions scaled by revenue or assets (Appendix Tables A.10–A.11).

*Source: 2.1 Emissions data — wpiea2021112-print-pdf*

### 2017.   We  then  estimate  their  future  carbon  dioxide  emissions  based  on  their  average  growth

### wpiea2021112-print-pdf - 2017.   We  then  estimate  their  future  carbon  dioxide  emissions  based  on  their  average  growth

### Scenario analysis and findings
- Method:
  - Estimate firms' future carbon dioxide emissions based on their average growth rate over the sample period.
  - Calculate cumulative emissions and the difference between when they would run out of their carbon budget under the status quo versus under a scenario with slower emissions growth.
- Main quantitative findings:
  - With business as usual, less than 14 years are left before the odds of containing global warming to under 1.5 degrees Celsius would be worse than even.
  - The reduction in emissions growth associated with a two standard deviation improvement in ESG scores would buy only an additional two years before the 1.5 degrees Celsius objective would be out of reach.
  - The number of years gained is similarly small relative to the baseline across a range of different climate objectives.

### Implications for ESG investing and policy
- Weak link between ESG scores and emissions growth:
  - Even reductions in emissions growth corresponding to large improvements in ESG scores would do little to help with climate change.
  - High ESG scores are not necessarily related to companies’ actual carbon footprints.
- Implications for investors and policymakers:
  - Sustainable investing strategies conditioned solely on ESG indicators have limited scope to meaningfully mitigate climate change by shifting production incentives for large emitters.
  - Environmentally-conscious investors, shareholders, and policymakers should note the limited translation of ESG-score improvements into emissions reductions.
  - Central banks managing the climate footprint of their interventions should be wary of over-reliance on ESG indicators.

### Conclusion
- Although greater attention to ESG considerations could in principle lead firms to improve ESG scores and reduce emissions, an important prerequisite is that firms that reduce their contributions to climate change be rewarded with better ESG scores.
- Using data on a small but important set of firms among the largest emitters worldwide, the analysis finds a weak relationship between ESG scores and growth rate of emissions, if any.
- Even a remarkable rise in sustainable investing incentivizing large emitters to significantly improve their ESG scores would not necessarily translate into meaningful momentum in dealing with climate change.

*https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021112-print-pdf.pdf*

### Chapter 6, October 2019, International Monetary Fund.

### Chapter 6, October 2019, International Monetary Fund

### Key empirical finding: ESG scores and emissions growth
- Main firm-year regression (Panel A, Table 1): Overall ESG Score coefficient on ∆ln(Emissions) = -1.1** (0.39) in the first specification; alternative specs show -0.5 (0.45) and -0.5 (0.55).
- Timing robustness (Table A.5):
  - Emissions two years later: Overall ESG Score = -1.2*** (0.32); other specs -0.8* (0.38) and -0.6 (0.47).
  - Emissions one year later: Overall ESG Score = -1.1** (0.39); other specs -0.4 (0.41) and -0.0 (0.50).
  - Emissions one year ago: Overall ESG Score = -1.1** (0.39); other specs -0.6 (0.45) and -0.4 (0.50).
  - Emissions two years ago: Overall ESG Score = -1.0* (0.55); other specs -0.1 (0.67) and 0.7 (0.98).
- Full-sample replication (Table A.7): Overall ESG Score = -1.2*** (0.36); alternative specs -0.8* (0.43) and -0.5 (0.52).
- Non-winsorized specification (Table A.8): Overall ESG Score = -1.3* (0.59); other specs -0.6 (0.53) and -0.6 (0.90).

### Role of institutional ownership (interaction effects)
- Panel B, Table 1:
  - Overall ESG Score = 0.1 (0.51), 0.5 (0.91), -0.3 (0.80) across specs.
  - Inst ownership = 0.9*** (0.26), 0.6* (0.34), 0.1 (0.45).
  - Overall ESG Score × Inst ownership = -0.2*** (0.04), -0.1 (0.07), -0.0 (0.06).
- Interpretation: significant negative interaction in the preferred specification (coefficient -0.2***), implying stronger ESG effects on emission reductions when institutional ownership is higher.

### Pillar-level results: Environment, Governance, Social
- Table 2 — Environment pillar (Panel A):
  - Environment = 0.3 (0.59) in the baseline specification.
  - Emissions (component) = 0.5 (0.54).
  - Resource Use = -0.1 (0.51).
  - Env. Innovation = 0.0 (0.31).
- Table 2 — Governance pillar (Panel B):
  - Governance = -0.8* (0.44).
  - Management = -0.7* (0.39).
  - CSR Strategy = -0.5 (0.37).
  - Shareholders = -0.2 (0.45).
- Appendix Table A.12 — Social pillar and components:
  - Social = -0.3 (0.44).
  - Workforce = 0.4 (0.50).
  - Community = -0.1 (0.36).
  - Human Rights = -0.4 (0.28).
  - Product Responsibility = -0.5* (0.26).
- Appendix Table A.6 — Level of emissions:
  - Overall ESG Score on ln(Emissions) = 0.20** (0.08); alternative specs 0.11 (0.08) and 0.03 (0.04).
  - Pillars on ln(Emissions): Environment = 0.03 (0.04); Governance = 0.04 (0.02); Social = -0.01 (0.02).

### Firm size interactions
- Table A.1 (Overall ESG × Log Lag Assets):
  - Overall ESG Score = -10.4** (3.77), -9.1* (4.29), -5.0 (4.59) across specs.
  - Log Lag Assets = -0.5 (1.58), -1.7 (1.79), -1.0 (1.66).
  - Overall ESG Score × Log Lag Assets = 0.4** (0.14), 0.3* (0.17), 0.2 (0.18).
- Table A.2 (Environment pillar × firm size components):
  - Environment = -6.6 (4.68); Environment × Log Lag Assets = 0.3 (0.19).
  - Emissions component = -0.9 (3.36); Emissions × Log Lag Assets = 0.1 (0.12).
  - Resource Use = -7.1 (4.63); Resource Use × Log Lag Assets = 0.3 (0.19).
  - Env. Innovation = -7.4* (3.54); Env. Innovation × Log Lag Assets = 0.3* (0.14).

### Mechanisms and inputs to Environment scores
- Table A.3 — Inputs associated with ESG and Environment standardized scores:
  - Attempt reduction in volatile organic compounds → Overall ESG Score = 0.56*** (0.12); Environment = 0.72*** (0.11).
  - Recognize climate risks & opportunities → Overall ESG Score = 0.60*** (0.10); Environment = 0.52*** (0.10).
  - Report environmental investments → Overall ESG Score = 0.41*** (0.11); Environment = 0.30** (0.12).
  - Report environmental partnerships → Overall ESG Score = 0.34*** (0.11); Environment = 0.37*** (0.10).
- Orthogonalized Environment score (Table A.4): Environment (orth.) = -0.5 (0.49), 0.2 (0.53), 0.7 (0.48) across specs.

### Emissions intensity (scaled by revenue or assets)
- Appendix Table A.9 — Overall ESG Score on emissions scaled by revenue:
  - Overall ESG Score = 0.0 (0.51), 0.0 (0.85), 0.3 (1.19).
- Appendix Table A.9 — Overall ESG Score on emissions scaled by assets:
  - Overall ESG Score = -0.3 (0.48), -0.7 (0.45), -1.4 (1.05).
- Tables A.10 and A.11 repeat pillar analyses for emissions scaled by revenue and assets with generally imprecise coefficients (see tables for component values).

### Global carbon budgets and implications of ESG improvements (Table 3)
- Assumptions:
  - Budgets from IPCC (2018a) allocated to 52 firms in sample by 2017 share of global carbon emissions.
  - Business-as-usual growth rate for firms 2002-2017 = 3.85 percent.
  - A one standard-deviation improvement in ESG assumed to reduce emissions growth by 1.1 percent (based on Panel A, Table 1 first spec).
- Years left (business-as-usual) and years gained by ESG improvement:
  - Probability of warming under 1.5°C:
    - 66 percent: Years left = 10.6; Years gained by ESG +1SD = 0.6; ESG +2SD = 1.3.
    - 50 percent: Years left = 13.8; Years gained by ESG +1SD = 0.9; ESG +2SD = 2.1.
    - 33 percent: Years left = 18.2; Years gained by ESG +1SD = 1.6; ESG +2SD = 3.6.
  - Probability of warming under 2°C:
    - 66 percent: Years left = 22.9; Years gained by ESG +1SD = 2.4; ESG +2SD = 5.7.
    - 50 percent: Years left = 26.9; Years gained by ESG +1SD = 3.3; ESG +2SD = 7.7.

### Visual and descriptive evidence
- Figure 1: Stock and flow of global carbon emissions traced to large emitters; Panel A shows stock since 1940; Panel B shows global carbon emissions since 2002 with emitter-type shares by year.
- Figure 2: ESG and emissions growth cross-sectional bubble plots for Overall ESG, Governance, and Environment pillars; columns show full sample average, four-year period with largest increase in emissions, and four-year period with largest improvement in ESG scores; shading indicates economic conditions (darker red = slower economic growth); outliers trimmed.
- Appendix Figures A.1–A.5: comparisons of ESG scores for large emitters vs S&P500; alternative scaling of emissions (revenue, assets); Social pillar plots; level of emissions panels.

### Robustness and sensitivity checks
- Firm-year fixed effects, country controls (GDP growth, unemployment, inflation), and firm controls (lagged log assets, lagged log revenue, lagged percent change in leverage, lagged fixed asset growth) are used across specifications.
- Results presented with standard errors double-clustered at firm and year levels; significance denoted *, **, *** for 10 percent, 5 percent, and 1 percent levels respectively.
- Multiple robustness tables examine timing, winsorization, full sample, orthogonalized pillars, interactions with firm size, and emissions intensity measures.

*International Monetary Fund — Chapter 6, October 2019 (wpiea2021112-print-pdf)*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021112-print-pdf.pdf_