## Deconstructing ESG Scores: How to Invest with your own Criteria? — Executive Summary, Data Coverage, and Global Screening (wpiea2023057-print-pdf) ## Source details **Canonical URL:** [Deconstructing ESG Scores: How to Invest with your own Criteria? — Executive Summary, Data Coverage, and Global Screening (wpiea2023057-print-pdf)](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023057-print-pdf.pdf) ## Other formats - [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023057-print-pdf.pdf.md) - [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023057-print-pdf.pdf.json) --- ### Executive Summary — Overview and Key Strategy Messages - ESG scores amalgamate fundamentally different factors; weak scores in one pillar can offset strong scores in another pillar. - The paper demonstrates an investment strategy that deconstructs ESG scores and targets specific ESG categories (e.g., emissions reduction, human rights). - Strategy implementation: exclude firms with the lowest scores in selected ESG categories and implement a best-in-class reinvestment within region-sector cells to preserve benchmark composition. - Best-in-class outcome: - reduces tracking error relative to exclusions-only, - slightly improves risk-adjusted performance, - preserves regional and sectoral composition of the benchmark, - yields a large gain in the targeted ESG category score. ### Executive Summary — Methodology and Data - Data provider: Refinitiv (formerly Thomson Reuters Asset4). - Refinitiv coverage: 450 data points collected, of which 186 comparable measures are used for ESG scoring. - Aggregation: 186 comparable measures → 10 categories → 3 pillars (E, S, and G). Scores range from 0 to 100. - Benchmark and universe: MSCI All Country World Index (ACWI); investable universe = constituents of the benchmark. - Example category counts/themes: - Environmental: Emission reduction — 28 measures; Innovation — 20; Resource use — 20. - Social: Community — 10; Human rights — 8; Product responsibility — 30; Workforce — 14. - Governance (weights shown): CSR strategy (0.13) — 12; Management (0.67) — 9; Shareholders (0.20) — 35. - Refinitiv sources: public sources (annual reports, CSR reports, company websites, news); controversies captured from global media. - Refinitiv methodological detail: for Boolean indicators, a default value is assigned when no relevant data is found (0 when a “yes” is positive from a sustainability point of view; 1 when “yes” is negative). ### Executive Summary — Key Results and Quantitative Findings - ESG investing scale and projection: - ESG investing has reached $35trillion – more than one third of global total assets under management (GSIA, 2020). - ESG funds’ assets under management could exceed $50trillion by 2025 (Bloomberg, 2021). - Rating divergence: average correlation between ESG ratings across major providers is only slightly above 50%; credit ratings correlation usually close to 99%. - Screening effect (33% screening threshold example): - Portfolio category score improves by 18 pp on average across ESG categories (from 62 to 80, maximum possible = 100). - When the overall ESG score is targeted, the increase is 11 pp. - Tracking error and costs: - Tracking error of the intermediate benchmark (constituents with an ESG score) relative to MSCI ACWI = 0.9% per year. - Additional cost due to screening is below 0.7% per year even with a 33% screening threshold. - Financial performance impact: simple exclusions substantially improve headline ESG scores while the portfolio’s financial performance only suffers a marginal impact relative to a broad stock market benchmark. ### Executive Summary — Investment Strategies and Practical Implications - Exclusion (negative screening): - Pros: substantial improvement in targeted category scores; marginal financial performance hit versus broad benchmark. - Cons: introduces regional and sectoral biases relative to the benchmark. - Best-in-class screening: - Exclude lowest scorers and reinvest proceeds in highest scorers within the same region-sector cells to maintain benchmark regional and sectoral composition. - Pros: reduces tracking error, slightly improves risk-adjusted performance, preserves benchmark exposures while improving targeted ESG scores. - Applicability: - Targeted screening strategies work better for categories supported by numeric disclosure (e.g., annual CO2 emissions in tons). - Boolean indicators and missing disclosure limit differentiation; in Refinitiv data missing and zero values can be indistinguishable. ### Executive Summary — Data Issues and Limitations - Low comparability across ESG data providers (measurement, scope, weighting differences). - Treatment of missing data: in Refinitiv, missing and zero values can be indistinguishable for some indicators, biasing screening outcomes. - Boolean indicators provide limited differentiation when many firms have missing or identical logical values. - Implication: screening strategies are best suited for categories with robust numeric disclosure. --- ### 2.2 Data Coverage — Scope, Sample, and Limitations - Data source: Refinitiv database. - Time span used: 2010 to 2019. - Last data download: March 2021 (some 2020 data present but incomplete). - Sample definition: complete set of firms in Refinitiv with a market capitalization available in a given year. - Total number of firms in Refinitiv database as of 2020: 10,142 firms. - Analysis of ESG data ends in 2019 because portfolio performance evaluation uses stock returns in yeart+1. ### 2.2 Data Coverage — Refinitiv revisions and coverage statistics - Refinitiv data are subject to backward changes of up to five years as new information is disclosed. - Cited evidence: median ESG score in rewritten data are 18% lower than in the initial data; deviation of 44% for the E score. - Coverage (firm-level): - Proportion of firms with market capitalization in 2010: 74.8%. - Among 7,590 firms in 2010, 3,911 firms (51.5%) also have a Refinitiv ESG score in the respective year. - ESG score coverage over time: slightly above 50% until 2014; improves starting 2015; maximum in 2019: 82.3% of firms have an ESG score. - Regional coverage in 2019 (by market capitalization): - North America: 39% of firms. - Europe: 20% of firms. - Pacific: 14% of firms. - Emerging countries: 23% of firms. - Disclosure of numeric indicators (aggregate and pillar gaps): - Overall proportion of valid values for numeric indicators: close to 40% on average worldwide. - E pillar: 20%; S pillar: 30%; G pillar: 80%. - Emission reduction indicators: 22% worldwide on average. - Renewable energy use ratio: approximately 10% provide data. - CO2 equivalent Scope 3 indirect emissions: 25% report data. - Innovation and Product responsibility: close to 10% valid values on average worldwide. ### 2.2 Data Coverage — Boolean indicators, zeros, and category sparsity - Refinitiv handling: when information not available, assigns 0 (corresponding to 'no') if 'yes' is positive; assigns 1 if 'yes' is negative. - Implication: proportion of valid values effectively 100% for most Boolean categories but evolution is ambiguous (reporting vs. policy change). - Examples of high proportions of zeros in 2019: - Environmental expenditures reporting: proportion of 0 = 74%. - Total individual compensation reporting: proportion of 0 = 95%. - Category zero-score examples (2019): - Innovation: 56% of firms worldwide have Innovation score equal to 0. - Emission score = 0 for 28% worldwide. - Resource use score = 0 for 29% worldwide. - Human rights score = 0 for 42% worldwide. - Product responsibility score = 0 for 10% worldwide. - CSR strategy (Boolean-only): approximately 34% of scores equal to 0. ### 2.2 Data Coverage — Patterns by region, sector, and firm size - Average scores: E pillar lower than S and G on average. - Regional heterogeneity: - Europe: higher scores, particularly E and S. - North America and Emerging: lower E scores. - Time trends: Pacific and Emerging large increases in ESG score (mainly E and S); North America ESG score does not improve overall due to decrease in E pillar. - Sector examples (2019 E/Emission scores): - Emission scores: Energy 45%; Utilities 47%; Basic materials 44%; Health care E score 17.6%; Financial 30%; Technology 32%; Utilities average E score 44.8%. - Interpretation: energy and utilities report more on emission policies → higher Emission scores despite higher emissions; health care and financials often do not report → low Emission scores despite low carbon intensity. - Firm-size effects: - Numeric indicator valid values: 31% for lowest quartile vs. 45% for highest quartile. - Proportion of firms with category scores equal to 0: ~50% for small firms vs. ~10% for large firms. - Small firms: average ESG score ≈ 30% over the sample. - Large firms: average ESG score increased from 49% in 2010 to 58% in 2019. ### 2.2 Data Coverage — Implications for screening - Two main issues: - Large proportion of scores equal to 0 for 6 out of 10 categories complicates low-threshold screening. - Large heterogeneity across regions and sectors creates significant regional and sectoral biases when screening. - Screening approaches: - Global exclusion and proportional reinvestment (can generate regional/sectoral biases). - Region-sector level screening with reinvestment within same region-sector (“best-in-class”) to preserve market structure. - MSCI ACWI subset coverage: coverage with valid scores above 95% in developed markets from 2010 on; emerging markets coverage above 90% from 2010 and above 95% from 2017 on. --- ### 3.1 Global Screening — Construction and Implementation - Global screening uses all firms in the benchmark with scores available at the end of yeart. - For screening threshold θ (e.g., 25%), identify firms with lowest scores until cumulative market cap ≈ θ of the benchmark market cap. - Excluded-firm set: IEx,t = {1{Si,t ≤ q(S)θ,t}}Nt i=1. - Included-firm set: IIn,t = {1{Si,t > q(S)θ,t}}Nt i=1. - Pure exclusion portfolio weights: - w(p)i,t = 0 for i ∈ IEx,t with ∑i∈IEx,t w(b)i,t ≈ θ. - w(p)i,t = w(b)i,t (1 / (1 − ∑i∈IEx,t w(b)i,t)) for i ∈ IIn,t. - Evaluation: portfolio built at end of yeart evaluated at end of yeart+1 using next-year returns. - Investors may target specific pillars or combine categories. ### 3.1 Global Screening — Empirical composition and constraints - Data sample: 2010–2019; MSCI ACWI used as market index; missing-score impact: world index loses only 2.4% of market cap on average due to lack of Refinitiv scores among MSCI ACWI firms. - Zero-score prevalence and exclusion implications: - Low-score firms tend to have low market cap; exclusion often removes many small firms. - 10% screening: exclude 9.9% of market cap but 26.2% of firms with lowest ESG scores. - 25% screening: exclude 24.7% of market value but 50.4% of firms. - 33% screening: exclude 33% of market value and 60.2% of firms. - Categories with many zero scores set lower bounds on feasible screening: - Innovation: 40.3% of firms have score = 0 (26.1% of market cap). Lowest screening threshold feasible = 26.1% quantile. - Human Rights: cannot exclude less than 21.9% of market cap (40% of firms). - CSR strategy: lower bound = 9.5% of market cap. - Implication: categories with many zero scores make low-threshold screening infeasible; screening often corresponds effectively to ≈25% threshold. ### 3.1 Global Screening — Score gains and category-level effects - Score gains (portfolio score minus benchmark score): - 10% threshold: gain usually between 4 and 7 points; relative gain between 6% and 10%. - Market-cap weighting limits score gain because large firms tend to have higher scores (25% smallest firms ESG = 42.1; 25% largest firms ESG = 58.6). - Category-level: - E category shows highest score gains. - For same 10% excluded firms, Resource use and Emission deliver highest gains, above 7pp. - Aggregate E pillar gain smaller than category gains because category scores summed at firm level first and categories not perfectly correlated across firms. - Examples: - 25% threshold: Product responsibility gain = 14.2; Workforce gain = 11.5. - 33% threshold: gains by category range from 14.1pp (Workforce) to 21.4pp (Human rights); overall ESG increases by 11.2pp. - Note: with 33% threshold, methodological issues related to missing values in numeric and Boolean indicators no longer affect portfolio construction. ### 3.1 Global Screening — Financial performance, tracking error, and exposures - Sharpe ratio (selected): - Benchmark Sharpe ratio = 0.69. - CSR strategy Sharpe ratio = 0.63. - Community score Sharpe ratio = 0.73. - Tracking error versus MSCI ACWI: - Annual tracking error of benchmark (MSCI ACWI constituents with ESG scores) = 0.9% relative to MSCI ACWI. - Screening portfolios tracking errors: - 25% threshold → 1.4% annual tracking error. - 33% threshold → 1.5% annual tracking error. - Regional and sectoral exposure shifts (E score, 33% threshold, sample averages): - Overweight Europe by 6.4pp: from 23.7% to 28.7%. - Underweight Emerging countries by 3.5pp: from 11% to 7.5%. - Overweight financials by 1.9pp: from 11.7% to 13.6%. - Underweight health care by 1.7pp: from 17.8% to 16.1%. - Key takeaways: - Screening can substantially increase portfolio category scores while keeping Sharpe ratios in the same ballpark as the market benchmark. - Screening tends to exclude many small firms. - Categories with many zero scores limit minimal feasible exclusion proportion and can produce larger-than-intended reallocation. - Screening generates modest tracking error increases (~0.5–0.6pp above benchmark’s 0.9%, due mostly to missing-score constituents). ### 3.1 Global Screening — Empirical table and figure highlights (selected numeric points) - Exact exclusion outcomes by threshold: - 10% screening: excludes 9.9% of market cap but 26.2% of firms. - 25% screening: excludes 24.7% of market value but 50.4% of firms. - 33% screening: excludes 33% of market value and 60.2% of firms. - Example category feasibility lower bounds: - Innovation cannot screen below 26.1% market cap because 40.3% of firms have score = 0 (representing 26.1% market cap). - Human Rights cannot exclude less than 21.9% of market cap. - CSR strategy lower bound = 9.5% of market cap. - Benchmark and exclusion portfolio summary (reported in tables): - MSCI ACWI benchmark score: 9.71. - Benchmark example: Ann. return 14.34; Sharpe 0.69; tracking error 0.94. - Exclusion portfolios (25%/33%): tracking error around 1.4%–1.6% annual. --- *Source: wpiea2023057-print-pdf* ### Executive Summary ### Executive Summary ### Overview - ESG scores are increasingly important tools for asset managers to design and implement ESG investment strategies. - Headline ESG scores amalgamate a broad range of fundamentally different factors, creating ambiguity: weak scores in one pillar can offset strong scores in another pillar. - The paper demonstrates an investment strategy that deconstructs ESG scores and focuses on specific underlying ESG categories such as emissions reduction and human rights. - Strategy implementation: exclude firms with the lowest scores in selected ESG categories and implement a best-in-class reinvestment within region-sector cells. ### Methodology and Data - Data provider: Refinitiv (formerly Thomson Reuters Asset4). - Refinitiv coverage: 450 data points collected, of which 186 comparable measures are used for ESG scoring. - Aggregation: 186 comparable measures → 10 categories → 3 pillars (E, S, and G). Scores range from 0 to 100. - Benchmark: MSCI All Country World Index (ACWI) as a broad equity benchmark; investable universe = constituents of the benchmark. - Example pillar/category counts and themes: - Environmental: (1) Emission reduction — 28 comparable measures; (2) Innovation — 20; (3) Resource use — 20. - Social: (1) Community — 10; (2) Human rights — 8; (3) Product responsibility — 30; (4) Workforce — 14. - Governance (weights shown): (1) CSR strategy (0.13) — 12; (2) Management (0.67) — 9; (3) Shareholders (0.20) — 35. - Refinitiv collects information from public sources (annual reports, CSR reports, company websites, news). Data reflect firm disclosure; controversies also captured from global media. - Refinitiv methodological detail: for Boolean indicators, a default value is assigned when no relevant data is found (0 when a “yes” is positive from a sustainability point of view; 1 when “yes” is negative). ### Key Results and Quantitative Findings - ESG investing scale: ESG investing has reached $35trillion – more than one third of global total assets under management (GSIA, 2020). - Projected growth: ESG funds’ assets under management could exceed $50trillion by 2025 (Bloomberg, 2021). - Rating divergence: correlation between credit ratings is usually close to 99%; average correlation between ESG ratings across major providers is only slightly above 50%. - Screening effect (example using a 33% screening threshold): - Portfolio category score improves by 18 pp on average across ESG categories (from 62 to 80, maximum possible = 100). - When the overall ESG score is targeted, the increase is 11 pp. - Tracking error and costs: - Tracking error of the intermediate benchmark (constituents with an ESG score) relative to MSCI ACWI is on average equal to 0.9% per year. - Additional cost due to screening is below 0.7% per year even with a 33% screening threshold. - Financial performance impact: simple exclusions substantially improve headline ESG scores while the portfolio’s financial performance only suffers a marginal impact relative to a broad stock market benchmark. - Best-in-class strategy outcome: excluding firms with the lowest category scores and reinvesting in firms with the highest scores within the same region-sector - reduces tracking error relative to exclusions-only, - slightly improves risk-adjusted performance, - preserves regional and sectoral composition of the benchmark, - still yields a large gain in the targeted ESG category score. ### Investment Strategies and Practical Implications - Exclusion (negative screening): - Exclude firms with the lowest category scores of interest. - Pros: substantial improvement in targeted category scores; marginal financial performance hit versus broad benchmark. - Cons: introduces regional and sectoral biases relative to the benchmark. - Best-in-class screening: - Exclude lowest scorers and reinvest proceeds in highest scorers within the same region-sector cells to maintain benchmark regional and sectoral composition. - Pros: reduces tracking error, slightly improves risk-adjusted performance, preserves benchmark exposures while improving targeted ESG scores. - Applicability: - Targeted screening strategies work better for categories supported by numeric disclosure (e.g., annual CO2 emissions in tons). - Boolean indicators and missing disclosure limit differentiation; in Refinitiv data missing and zero values can be indistinguishable, reducing effectiveness of screening unless degree of exclusion is sufficiently high. ### Data Issues and Limitations - Major challenges: - Low comparability across ESG data providers (measurement, scope, weighting differences). - Treatment of missing data: in Refinitiv, missing and zero values can be indistinguishable for some indicators, which can bias screening outcomes. - Translating qualitative information into numeric values: Boolean questions provide limited differentiation when many firms have missing or identical logical values. - Implication: screening strategies are best suited for categories with robust numeric disclosure; Boolean-heavy categories may yield skewed exclusion results. ### Related Literature and Context - Divergence sources in ESG ratings: scope, measurement, and weights; measurement differences explain most divergence. - Empirical findings in literature: - High disagreement across providers tends to associate with lower subsequent stock returns. - Large and repeated changes in historical ESG scores have been documented. - ESG scores do not necessarily capture differences in firm carbon emissions. - Investor preference framing: - “Value-alignment” investors: aversion to companies not aligned with investor values (dominant institutional preference per authors’ assumption). - “Impact-seeking” investors: value measurable societal impact (characterized as niche for institutional mandates). - Theoretical context: - ESG-efficient frontier concept: highest attainable Sharpe ratio for each ESG level. - In equilibrium, green assets may have negative alphas and brown assets positive alphas, but realized recent returns for green assets have been higher due to demand pressure. ### Conclusion - Deconstructing ESG scores into granular categories enables asset managers to implement targeted ESG objectives with virtually no cost to financial performance relative to a diversified benchmark. - Simple exclusions improve category scores substantially but create regional and sectoral biases; best-in-class reinvestment mitigates these biases while preserving ESG improvements. - The effectiveness of category-based screening depends critically on the nature and completeness of underlying ESG disclosure; numeric indicators enable sharper screens than Boolean indicators. *Source: wpiea2023057-print-pdf - Executive Summary* ### 2.2 Data Coverage ### 2.2 Data Coverage ### Data scope and sample definition - Data source: Refinitiv database. - Time span used: 2010 to 2019. - Last data download: March 2021 (some 2020 data present but incomplete). - Sample definition: the complete set of firms included in the Refinitiv database for which a market capitalization is available in a given year. - Total number of firms in Refinitiv database as of 2020: 10,142 firms. - The analysis of ESG data ends in 2019 because: - substantial time lag for complete database updates (many firms missing 2020 data at last download); - portfolio performance evaluation uses stock returns in yeart+1 (performance of portfolio built with 2019 data uses returns at the end of 2020). ### Refinitiv database limitations and historical revisions - Refinitiv data are subject to backward changes of up to five years as new information is disclosed. - Evidence cited (Berget al.(2020)) indicates: - median ESG score in rewritten data are 18% lower than in the initial data; - deviation of 44% for the E score. - In early 2021, Refinitiv opened a portal for firms to submit ESG data (note: not applicable to sample period ending fiscal year 2019). ### Coverage statistics (firm-level) - Proportion of firms with available market capitalization in 2010: 74.8%. - Among these 7,590 firms in 2010, 3,911 firms (51.5%) also have a Refinitiv ESG score in the respective year. - ESG score coverage over time: - fairly stable at slightly above 50% until 2014; - improves steadily starting in 2015; - maximum in 2019: 82.3% of firms in the sample have an ESG score. - Coverage for other metrics (3 pillars and 10 categories): essentially the same as aggregate ESG score. ### Regional coverage in 2019 (by market capitalization) - North America: 39% of firms. - Europe: 20% of firms. - Pacific: 14% of firms. - Emerging countries: 23% of firms. - Regional differences in ESG score availability: - North America: relatively low coverage at the beginning of the sample, below 50% until 2014. - Emerging countries: proportion with an available ESG score is above 50% since 2011. - Europe and the Pacific: ESG score coverage relatively complete over the full sample. ### Disclosure of ESG information — missing values in numeric indicators (summary) - Numeric indicator scoring: a score based on relative percentile ranking is calculated only if the firm has reported the information; otherwise a missing value is assigned. - Method for proportion of valid (non-missing) values for a numeric indicator in a given year: - calculate number of firms for which indicator is available in that year; - identify industries for which the indicator is material; - calculate number of firms with valid data in these industries and divide by total number of firms in these industries. - Aggregate results (Table 2 Panel A summary): - overall proportion of valid values for numeric indicators: close to 40% on average worldwide over the sample. - gap by pillar: E and S pillars have 20% and 30% respectively, G pillar has 80%. - within pillars, proportion usually homogeneous. - Emission reduction indicators: proportion of valid values = 22% worldwide on average. - Renewable energy use ratio: approximately 10% of firms provide data. - CO2 equivalent Scope 3 indirect emissions: 25% of firms report data. - Innovation (E pillar) and Product responsibility (S pillar): close to 10% valid values on average worldwide. ### Disclosure of ESG information — Boolean indicators and proportion of zeros - Refinitiv handling of Boolean indicators: - When information not available, Refinitiv assigns 0 (corresponding to 'no') if answering 'yes' would be positive; assigns 1 (corresponding to 'yes') when answering 'yes' would be negative. - This penalizing strategy (assigning 0 to non-reporting when 'yes' is positive) is a recent methodological change intended to encourage disclosure. - Implications: - Proportion of valid values is effectively 100% for most Boolean categories because missing is mapped to a value. - Evolution of a Boolean indicator over time can be ambiguous: cannot distinguish better reporting from actual policy improvement. - Examples of high proportions of zeros in 2019: - reporting on firm’s environmental expenditures: proportion of 0 = 74%. - reporting about total individual compensation of all executives and board members: proportion of 0 = 95%. - Category-level impact: - Some categories (Human rights and CSR strategy) are exclusively based on Boolean indicators → substantial proportion of firms report a score equal to 0. - In 2019, proportion of firms with category score equal to 0 (selected examples): - Innovation: 56% of firms worldwide have Innovation score equal to 0. - Emission score = 0 for 28% worldwide in 2019. - Resource use score = 0 for 29% worldwide in 2019. - Human rights score = 0 for 42% of firms in 2019. - Product responsibility score = 0 for 10% of firms in 2019. - CSR strategy (Boolean-only): approximately 34% of scores equal to 0 in 2019. ### Scores at category and pillar level — patterns and heterogeneity - Categories based on Boolean indicators only (Human rights in S; CSR strategy in G) or with few numeric indicators and many missing values (Innovation in E) are associated with low average scores. - On average, scores are lower for the E pillar than for the S and G pillars. - Regional heterogeneity: - European firms have higher scores, particularly for E and S categories. - Firms in North America and Emerging countries have lower E scores. - Time trends: - On average, scores tend to improve over time. - Pacific and Emerging countries have large increases in the ESG score, largely from E and S pillars. - North America: ESG score does not improve overall, mainly because of a decrease in the E pillar. - Sector patterns (selected 2019 values): - Emission scores in 2019: - Energy: 45% - Utilities: 47% - Basic materials: 44% - Health care: 17.6% (average E score for health care firms) - Financial: 30% - Technology: 32% - Utilities average E score reported as 44.8% in 2019. - Interpretation: - Energy and utilities report more on emission policies and targets → higher Emission scores despite higher emissions. - Health care and financials often do not report on emission policy → low Emission scores despite low carbon intensity. - S pillar average ranges across sectors in 2019: between 42.9 and 46.3. - G pillar average ranges across sectors in 2019: between 40.6 and 53.5. ### Firm size effects on disclosure and scores - Large firms (highest market cap quartile) disclose more than small firms: - For numeric indicators, proportion of valid values: 31% for lowest quartile vs. 45% for highest quartile. - Proportion of firms with category scores equal to 0: approximately 50% for small firms vs. approximately 10% for large firms. - Resulting score differences: - Small firms: average ESG score close to 30% over the sample. - Large firms: average ESG score increased from 49% in 2010 to 58% in 2019. - Consistent with evidence that large firms spend considerable resources on ESG reporting. ### Implications for ESG screening and portfolio construction (preview) - Two main issues for category-level ESG screening: - Large proportion of scores equal to 0 for 6 out of 10 categories complicates low-threshold screening (e.g., excluding 1% or 5% of firms may exclude some firms with score 0 while keeping others with score 0). - Large heterogeneity of scores across regions and sectors creates significant regional and sectoral biases relative to market exposures when screening. - Screening strategies discussed: - Global exclusion and proportional reinvestment in remaining firms (can generate regional/sectoral biases). - Region-sector level screening with reinvestment within same region-sector (a “best-in-class” approach) to preserve market structure. - Benchmark used for portfolio experiments: MSCI ACWI (covers developed and emerging markets). - Coverage for MSCI ACWI subset: - Coverage with valid scores above 95% in developed markets from 2010 on. - Coverage in emerging markets above 90% from 2010 on and above 95% from 2017 on. *Source: wpiea2023057-print-pdf - 2.2 Data Coverage* ### 3.1 Global Screening ### 3.1 Global Screening ### Methodology: global screening construction - The global screening is based on all the firms in the benchmark with scores available at the end of yeart. - For a screening threshold of θ (say, 25%), identify firms with the lowest scores until their cumulative market cap represents proportion θ of the market cap of the benchmark portfolio. - Notation: - For score Si,t, q(S)θ,t denotes the threshold corresponding to probability θ. - Excluded-firm set: IEx,t = {1{Si,t ≤ q(S)θ,t}}Nt i=1. - Included-firm set: IIn,t = {1{Si,t > q(S)θ,t}}Nt i=1. - Excluded firms chosen so that ∑Nt i=1 w(b)i,t 1{Si,t ≤ q(S)θ,t} is as close as possible to θ, where w(b)i,t is firm i market weight in the benchmark. - Reinvestment: - Proceeds of exclusion are reinvested in remaining firms IIn,t in proportion to their market weight. - Pure exclusion portfolio weights: - w(p)i,t = 0 for i ∈ IEx,t with ∑i∈IEx,t w(b)i,t ≈ θ. - w(p)i,t = w(b)i,t (1 / (1 − ∑i∈IEx,t w(b)i,t)) for i ∈ IIn,t. - Evaluation: - A portfolio built at end of yeart is evaluated at end of yeart+1 using next-year stock market returns. - Investors may target specific ESG pillars or combine categories. ### Empirical composition and screening constraints - Data sample: 2010–2019; MSCI ACWI used as market index; benchmark contains MSCI ACWI constituents with Refinitiv ESG scores available. - Missing-score impact: - For the world index, we lose only 2.4% of the market cap on average due to lack of Refinitiv scores among MSCI ACWI firms. - Zero-score prevalence and operational implications: - First two columns of Table 6: proportion of firms and proportion of market value with scores equal to zero (used to assess feasibility of low-threshold screening). - Low-score firms tend to have low market cap, so exclusion often removes a large fraction of small firms. - Exact exclusion outcomes (examples from Panels): - 10% screening: exclude 9.9% of the market cap but 26.2% of the firms with the lowest ESG scores. - 25% screening: exclude 24.7% of the market value but 50.4% of the firms. - 33% screening: exclude 33% of the market value and 60.2% of the firms. - Categories with many zero scores (lower bounds on feasible screening): - Innovation: 40.3% of firms in the MSCI index have a score equal to zero (26.1% of market cap). Therefore the lowest screening threshold that can be applied is the 26.1% quantile. - Human Rights: cannot exclude less than 21.9% of the market cap (40% of the firms). - CSR strategy: lower bound for screening is 9.5% of the market cap. - Implication: for categories with a large proportion of zero scores, low-threshold screening cannot be implemented and a screening strategy effectively corresponds to approximately a 25% screening. ### Score gains from global screening - Gain on the score (portfolio score minus benchmark score): - For the 10% threshold, the gain is usually between 4 and 7 points. - In relative terms (gain divided by benchmark score), the gain is between 6% and 10%. - Market-cap weighting limits score gain because large firms tend to have higher scores: on average, the 25% smallest firms in the ACWI index have an overall ESG score equal to 42.1, whereas the 25% largest firms have an overall ESG score equal to 58.6. - Category- and pillar-level observations: - E category shows the highest score gains. - For the same 10% excluded firms, Resource use and Emission category scores deliver the highest score gains, above 7pp. - Aggregate E pillar gain is much smaller than gains in its component categories because category scores are summed at the firm level first and category scores are not perfectly correlated across firms. - Similar results hold for S and G pillars and for the aggregate ESG score. - Examples of category gains (selected values reported in text): - For the 25% threshold, Product responsibility gain = 14.2; Workforce gain = 11.5. - With the 33% threshold, gains by category range from 14.1pp for Workforce to 21.4pp for Human rights, while the overall ESG score increases by 11.2pp. - Implementation note: - With the 33% screening threshold, methodological issues related to missing values in numeric and Boolean indicators no longer affect portfolio construction. ### Financial performance of global screening portfolios - Risk-adjusted performance metric used: Sharpe ratio (ESG risk-factor adjustment beyond scope). - Selected threshold focus: 33% threshold (implementable for all categories). - Sharpe ratios (last columns of Table 6): - Benchmark Sharpe ratio = 0.69. - CSR strategy Sharpe ratio = 0.63. - Community score Sharpe ratio = 0.73. - Tracking error versus MSCI ACWI: - Annual tracking error of the benchmark (including only MSCI ACWI constituents with ESG scores) = 0.9% relative to MSCI ACWI. - Screening portfolios tracking errors: - 25% threshold → 1.4% annual tracking error. - 33% threshold → 1.5% annual tracking error. - Regional and sectoral exposure shifts (example: E score, 33% threshold, sample averages): - Overweight Europe by 6.4pp: from 23.7% to 28.7%. - Underweight Emerging countries by 3.5pp: from 11% to 7.5%. - Overweight financials by 1.9pp: from 11.7% to 13.6%. - Underweight health care by 1.7pp: from 17.8% to 16.1%. - Implication: regional and sectoral biases may be undesirable for investors seeking otherwise passive exposures; addressed in region-sector screening. ### Key empirical takeaways on global screening - Screening can substantially increase portfolio category scores while maintaining Sharpe ratios in the same ballpark as the market benchmark. - Screening tends to exclude many small firms (large fraction of firms excluded relative to market cap excluded). - Categories with many zero scores limit the minimal feasible exclusion proportion and can produce larger-than-intended reallocation. - Screening generates modest tracking error increases (around 0.5–0.6pp above the benchmark's 0.9% due mostly to missing-score constituents). *Source: IMF working paper chapter 3.1 (Global Screening) from the provided PDF content.* ### References ### References (Appendix and Data Tables from wpiea2023057-print-pdf) ### Data coverage and sample - Sample period: 2010 to 2019 for Refinitiv database coverage and scores; financial performance measures computed from 2011 to 2020. - Global coverage table reports numbers and proportions of firms with valid market capitalization and valid ESG scores by region (World, North America, Europe, Pacific, Emerging countries) for 2010–2019. - MSCI coverage: number and proportion of MSCI constituents covered by Refinitiv by region for 2010–2019. ### Missing values and data validity (Technical Appendix A) - Aggregate proportion of valid numeric indicators (World, ESG): 2010 = 37.2, 2015 = 39.1, 2019 = 39.7. - Regional average proportion of valid values (across indicators): - Europe average proportion: 48 percent (noted in prose). - North America average proportion: 35 percent. - Pacific average proportion: 35 percent. - Emerging countries average proportion: 38 percent. - E and S pillars: larger regional differences—Europe: 32 percent (E), 38 percent (S); North America: 10 percent (E), 21 percent (S). - Industry differences: - Financials: low record of numeric indicators (below 8 percent average for E indicators). - Basic materials and utilities: highest reporting standards for E indicators (23 percent and 20 percent, respectively). - Energy, utilities, basic materials: disclosure relatively higher for Emission reduction (~20 percent on average) and Resource use. - Firm-size differences (Table A.3): proportions of valid numeric indicators reported by size quartiles (Small, Medium-small, Medium-large, Large) with numeric examples: - Small firms ESG: 2010 = 30.0, 2015 = 30.6, 2019 = 31.5. - Large firms ESG: 2010 = 40.8, 2015 = 45.1, 2019 = 48.8. ### Zero values and score sparsity (Technical Appendix B) - Many E-category scores exhibit high frequency of zero values across regions and sectors. - World-level proportions of scores with zero value (selected): - Environment: 2010 = 17.9, 2015 = 21.0, 2019 = 21.2. - Emission: 2010 = 26.6, 2015 = 29.8, 2019 = 28.2. - Innovation: 2010 = 56.1, 2015 = 56.6, 2019 = 56.5. - Resource use: 2010 = 27.6, 2015 = 29.1, 2019 = 29.4. - Regional zero-value examples (2019): - Europe Environment zero: 2019 = 6.0. - North America Human rights zero: 2019 = 13.7. - Emerging countries CSR Strategy zero: 2019 = 16.4. - Sector heterogeneity in zero-value prevalence (selected 2019 figures): - Health Care Innovation zero: 2019 = 90.3 (noted in prose). - Utilities Innovation zero: 2019 = 17.6. - Health Care Emission zero: 2019 = 50.9. - Financials show substantial zero frequencies for some categories. - Firm-size zero-value patterns (Table A.6, selected 2019): - Small firms E Score zero: 2019 = 43.0. - Large firms E Score zero: 2019 = 10.7. - Innovation Score zero: Small = 79.6 (2015), Large = 35.6 (2019). - Prose finding: proportions of zero scores tend to decrease over time for most sectors but not universally; health care reports higher zero proportions in 2019 than 2010 for several E and G categories. ### Average scores by region, sector, and firm size (Technical Appendix C and main tables) - World average ESG scores (selected years): - World ESG: 2010 = 40.4, 2015 = 41.4, 2019 = 42.8. - Regional average ESG scores (2010/2015/2019, Table 3): - North America ESG: 2010 = 38.9, 2015 = 37.5, 2019 = 37.1. - Europe ESG: 2010 = 49.9, 2015 = 51.9, 2019 = 52.5. - Pacific ESG: 2010 = 36.1, 2015 = 38.4, 2019 = 45.0. - Emerging countries ESG: 2010 = 36.6, 2015 = 41.4, 2019 = 43.3. - Sector average ESG scores (selected sectors, Table 4, 2010/2015/2019): - Energy ESG: 2010 = 36.7, 2015 = 38.9, 2019 = 42.8. - Financials ESG: 2010 = 41.4, 2015 = 41.6, 2019 = 42.3. - Health Care ESG: 2010 = 39.6, 2015 = 38.5, 2019 = 36.3. - Technology ESG: 2010 = 42.3, 2015 = 42.2, 2019 = 44.7. - Firm-size average scores (Table A.7, selected): - Small firms ESG: 2010 = 30.0, 2015 = 30.6, 2019 = 31.5. - Large firms ESG: 2010 = 35.8, 2015 = 38.6, 2019 = 44.1. - E Score (Large firms): 2019 = 52.8. - G Score (Large firms): 2019 = 57.7. ### Exclusion portfolios: construction and financial performance (Tables 6–8 and Tables 7–8 panels) - Benchmark and MSCI ACWI baseline (reported across tables): - MSCI ACWI benchmark score: 9.71. - Benchmark (when reported): score 9.91; Ann. return 14.34; Sharpe 0.69; tracking error 0.94. - MSCI ACWI Ann. return: 14.07; Sharpe: 0.69 (MSCI ACWI row). - Exclusion strategy summary: exclusion portfolios built by excluding bottom x% of firms by score for pillars or categories and reinvesting (thresholds: 10%, 25%, 33%; also analysis of 5%–50% thresholds in figures). - Key summary statistics for exclusion portfolios (selected highlights): - Panel A: 10% threshold (Table 6): - ESG exclusion portfolio: Port‑folio score = 64.0; Ann. return = 9.8? (table shows fragmented entries—preserve shown numbers) - Reported example row (ESG, Panel A 10%): "ESG 0.00.0 26.29.964.0 68.4 4.49.85 14.41 0.681.16" — table entries indicate proportions, scores, Ann. return, Ann. vola, Sharpe, Ann. track. error (interpretation follows table layout). - Panel B: 25% threshold (Table 6): - ESG: Port‑folio score 64.0; Ann. gain/return entries (table layout): "ESG 0.00.0 50.424.764.0 73.0 9.19.81 14.33 0.681.40" - Panel C: 33% threshold (Table 6): - ESG: Port‑folio score 64.0; "ESG 0.00.0 60.233.064.0 75.2 11.29.70 14.29 0.681.60" - Regional, sectoral, and sector-region exclusion at 25% threshold (Table 7, Panel A–C, selected): - Panel A (Regional exclusion and reinvestment), ESG 25%: "ESG 45.024.464.0 72.2 8.29.72 14.25 0.681.47" (proportion excluded, excluded market cap, portfolio score, Ann. return, Ann. vola, Sharpe, Ann. track. error). - Panel B (Sectoral exclusion and reinvestment), ESG 25%: "ESG 49.224.364.0 72.7 8.79.88 14.33 0.691.71". - Panel C (Sector-region exclusion and reinvestment), ESG 25%: "ESG 42.023.364.0 74.2 10.39.56 14.34 0.671.50". - Higher exclusion thresholds generally: - Increase portfolio ESG/E/S/G scores (figures: Score gain when threshold increased from 5% to 50%). - Affect Sharpe ratios (Figure 9 shows Sharpe ratio when threshold increased from 5% to 50%; axis range ~0.60–0.72). - Increase annual tracking error (Figure 10 axis range 1–2 for tracking error when threshold increased from 5% to 50%). - Sector- and category-specific exclusion results (33% threshold examples, Table 8): - Panel A (Regional exclusion) ESG 33%: "ESG 54.532.464.0 74.1 10.29.51 14.19 0.671.72". - Panel B (Sectoral exclusion) ESG 33%: "ESG 58.732.364.0 74.7 10.89.72 14.37 0.681.83". - Panel C (Sector-region) ESG 33%: "ESG 51.631.064.0 75.8 11.99.50 14.30 0.661.54". ### Temporal and distributional visual findings (figures) - Figures show distributions (2019) for E, S, G pillar scores and categories; cross-section distributions indicate wide dispersion and many low/zero observations for several categories. - Time-series figures display evolution (2010–2020) of benchmark and exclusion portfolio scores across pillars and categories; exclusion portfolios based on 10%, 25%, 33% thresholds consistently yield higher pillar/category scores than benchmark. - Figures 8–10: illustrate score gain, Sharpe ratio, and tracking error as exclusion threshold varies from 5% to 50% (quantitative axes preserved in figures: Score gain axis 0–0.5; Sharpe range 0.60–0.72; Tracking error range 1–2). ### Notable methodological and interpretative notes (from table notes and technical appendix) - Table note: The Refinitiv coverage and valid-score proportions are reported relative to the total number of firms listed in the database in the given region (as of 2020). - Category scores are computed from underlying indicators that are preselected for sector-level materiality; sectoral heterogeneity in zero scores is not solely due to indicator immateriality. - Exclusion portfolios constructed relative to MSCI ACWI; first two columns in exclusion tables report proportion of firms and fraction of market value with zero scores; next two columns report proportion of excluded firms and excluded market value relative to MSCI ACWI. - Financial performance measures for exclusion portfolios computed from 2011 to 2020. *Deconstructing ESG Scores: How to Invest with your own Criteria? — Working Paper No. WP/2023/057 (References, Tables, Figures, and Technical Appendix extracted from wpiea2023057-print-pdf)* --- _Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023057-print-pdf.pdf_