Investment funds are responsible for substantial environmental and social impacts

Abstract

Sustainable investing is under risk of being watered down by greenwashing given the lack of standardized and reliable indicators for measuring impact at financial product level. Here we propose 13 environmental and 13 social life-cycle-based, ready-to-use, and policy-relevant impact indicators that can be adapted for sustainability assessment of financial products. These indicators are aligned primarily with the EU Sustainable Finance Disclosure Regulation (SFDR) for investment funds. As practical application, we estimate the impacts of a sample of 230 self-labeled sustainable investment funds for all indicators. Their total estimated environmental impacts and social impacts are large and vary between 2.1 and 28.4 times the impacts associated with the consumption of a one million EU citizens, depending on the indicator. Moreover, we have found similar impact ranges within a sample of conventional funds, given the heterogeneity of funds. However, when comparing two funds that are equivalent sustainable/non-sustainable pairs, we find that the sustainable fund is better on specific impact categories, but not all of these, therefore hinting towards trade-offs in terms of impact categories also for sustainable funds.

Introduction

Monetary flows towards so-called sustainable investment funds are projected to grow to one third of the global market by 2025 (53 trillion USD)¹. However, the effectiveness of sustainable financial products in driving changes in the real economy is heavily scrutinized^2,3 as there is no top-down, thorough assessment of their sustainability claims^4,5, leading to an increased risk of greenwashing⁶. This risk is being progressively compensated by investors from a bottom-up perspective by the growing focus on measuring the sustainability of financial products such as investment funds^7,8. Investors’ awareness is, among others, driven by the imminence of climate (but also other environmental and social) risks that can jeopardize the long-term profitability of investments^9,10. Next to policy makers, investors are an additional lever to push for a sustainable economy, by exerting influence on the companies they invest in, which can themselves influence reduction of impacts in their supply networks¹¹. Having access to reliable and complete sustainability information is vital to the integration of sustainability in the investment decision making process^12,13. Existing environmental, social, and governance (ESG) ratings, widely used as proxy for sustainable performance, are mostly unreliable and do not offer sound quantitative information to investors^8,14. To tackle this issue, initiatives to standardize sustainability assessment at financial product level are emerging globally¹⁵, with the regulations under the EU Sustainable Finance Action Plan being regarded as the most ambitious^16,17. The Sustainable Finance Disclosure Regulation (SFDR)¹⁸ is the piece of legislation addressed to financial institutions and is the driver of our research. However, indicators under the SFDR, so-called principal adverse impacts (PAIs), are not comprehensive enough, while at the same time reliable company data to aid in sustainability reporting is unavailable.

As fostered to a considerable degree in regulations^19,20, life cycle assessment (LCA) can provide consistent estimation of sustainability impact at product and organization level^8,21,22,23. In this paper, we demonstrate that LCA-based, scientific and ready-to-use environmental and social indicators aligned with regulations can already be operationalized in a coherent framework for sustainability assessment of investment funds. Adopting a life cycle perspective to impact assessment is crucial in order to avoid shifting of impact to indirect stages of production and consumption. Moreover, environmental LCA methodology includes a comprehensive set of impact indicators, to avoid trade-offs between these^8,24. Importantly, social impact indicators can be retrieved using Social LCA²⁵, which has not yet been broadly applied in this context. Input-output LCA (IOLCA) is a good fit in the case of company and fund-level assessments as it ensures a comprehensive coverage of all indirect impacts connected to an economic sector, due its completeness in covering economic transactions globally.

As a first contribution, we designate 13 environmental and 13 social life-cycle-based impact indicators that can be estimated at company and investment fund-level using IOLCA²⁶ and show their match to sustainable finance reporting regulations (Fig. 1). In our approach, we first derive country and sector specific impact factors^27,28 adapted from two different input-output databases, EXIOBASE for environmental impacts and PSILCA for social impacts. We then estimate firm-level life-cycle-based impacts by using detailed country and sector revenue breakdown at company level to assign revenue-based weights for the different impact factors. Firm-level estimates are then aggregated at investment fund level allocating responsibility of impact based on share invested. This approach has been applied by ourselves and others in past work^29,30. As a second contribution, we apply this framework to a sample of 230 self-labeled sustainable equity investment funds listed on the Luxembourg Green Exchange. This is the biggest exchange for green financial products, which represent more than 13% of the assets under management (AuM) of all funds self-labeled sustainable under the EU SFDR³¹: the so-called article 8 funds, which only declare to use ESG criteria in the investment decision and article 9 funds, which promise an environmental or social contribution, thus having a stricter mandate. We complement the analysis with an equivalent sample of conventional, non-ESG funds, to benchmark the results and test whether there are differences between fund samples, when looking at the extended set of indicators. We analyse the magnitude of impact per impact category, attributed to the sustainable funds’ sample and we discuss the differences between direct and indirect share of impact. We then compare the sustainable funds sample with conventional funds. We showcase how our data can be used to compare two funds by an investor. To better grasp why there are differences between funds, we group companies by industry and show main drivers of impact, by category. We study correlations between impact categories and within the group of social and the group of environmental indicators, to identify synergies, and trade-offs. As a final analysis, we look at the top funds holdings that drive the lion’s share of fund-level impact and discuss implications.

Fig. 1: Environmental and social impact indicators, under the EU sustainable finance regulations linked to proposed life-cycle-based ready-to-use indicators.

a environmental indicators, b social indicators. The first column refers to environmental objectives and minimum social safeguards from the EU Taxonomy. The second column refers to indicators proposed under the CSRD. The third column refers to indicators under the SFDR. The last column shows our proposed life-cycle based indicators. One indicator can be connected to more indicators from another column. The indicators and objectives that are in orange boxes are not directly matched between the regulations and the life-cycle-based indicators. The darker-colored SFDR PAI indicators refer to the mandatory indicators.

Results

A set of consistent and ready-to-use environmental and social impact indicators

The three main pieces of regulations that mandate sustainability-level disclosures are the EU Sustainable Finance Disclosure Regulation¹⁸ (SFDR) for financial product providers, the EU Corporate Sustainability Reporting Directive (CSRD)³², for firm-level reporting and the EU Taxonomy³³, for designating green economic activities. As a first contribution of this paper (Fig. 1), we propose a set of curated life-cycle-based indicators that are science-based and ready-to-use at investment fund level, to estimate sustainability on a more complete set of indicators that are policy-relevant, via links to the SFDR PAI indicators for investments in investee companies (Fig. 1). The proposed indicators can also serve as a bridge between the EU SFDR and the two other regulations, by following a life-cycle perspective as mandated in the EU Taxonomy and being specific in terms of indicator methodology, as attempted in the CSRD. We point out the (mis-)matchings between the sustainability indicators proposed by SFDR and CSRD indicators and the EU Taxonomy environmental objectives.

The comprehensiveness and rigorousness of SFDR-proposed indicators is unsatisfactory, when compared to the state-of-the-art indicators in the sustainability assessment field. First, there is an inconsistent coverage of sustainability issues, as compared to widely accepted frameworks for sustainability assessment: only GHG emissions, water pollution, biodiversity, and waste are mandatory PAIs, with other issues related to land use or different polluting substances being voluntary³⁴. Second, disclosure over the life cycle is not mandated for other indicators beyond climate change, which gives way to impact shifting from direct to indirect stages of production and ultimately greenwashing. Finally, there is no clear methodology to underpin the indicators proposed, which may lead to reported data not being comparable between financial institutions (argumentation is further detailed in the SI). By proposing indicators for which current methods exist to estimate impact over the life cycle we ensure better coverage of issues that funds may be exposed to via the supply chains of companies that they invest in. The usage of life-cycle-based indicators to estimate impact across investment funds would ensure that the same methodology is used by financial product providers, which would in turn allow comparisons between different funds. We further discuss the matching process in the Methods section and provide detailed background on the legislation in the Supplementary Notes 1 (Supplementary Table 1).

Magnitude of direct and indirect impacts at fund level

We first estimate the total life cycle impact of the 230 sustainable investment funds-sample, based on the amounts invested, and plot the proportion of direct and indirect impacts (covering the upstream life cycle chain) alongside absolute direct and indirect impacts (Fig. 2) or all impact indicators selected in the previous section.

Fig. 2: Estimation of the total direct and indirect environmental and social impacts of a sample of sustainability-labeled funds and equivalence in impact of millions of EU citizens.

The stacked bars show the proportion between direct and indirect impacts, for (a). 13 environmental indicators (with specific unit) and (b). 13 social indicators (measured in medium worker hours across all indicators). The labels of the bars represent the actual values for the direct and indirect estimated total impacts for the sample of fund, for each indicator. The sample contains 230 sustainability-labeled funds. The impact at fund-holding level is computed for the year 2019. The red line shows the equivalent of the total estimated life cycle impacts in million EU citizens.

For all indicators, indirect impacts are considerably larger than direct impacts. The results are not corrected for double counting and thus impacts could be overestimated. However, from a risk perspective, this is suitable as it shows the aggregated risk exposure³⁵.

Results vary by impact indicator. Indirect proportion is lower for particulate matter, photochemical ozone formation, and GHG emissions, where the contribution of direct impacts to the total life cycle impact is higher than 40%. This is mainly due to funds investing considerable amounts in Energy and Utilities companies, which have most of climate change impacts coming from the Production of energy as direct impacts. A similar contribution is observed among the social indicators, namely for lack of rights of association and rights to strike. These results are driven by the portfolio allocations of funds, and implicitly by the market valuation of funds. They are thus complementary to traditional IO analysis performed to determine the impacts of industries and countries in the real economy, as the results at fund level are influenced by revenue allocation and by market capitalization of companies. The sectors that have the biggest output in terms of revenue generated in the real economy may not match the sectors with the biggest stock market value in the capital markets, due to the different valuations of companies, as valuations also take into account intangible assets, in addition to tangible assets that are considered in the real economy.

To further contextualize the estimated impacts of the funds sample, we express this in the population equivalents of the impact of EU citizens (assuming an EU population of 447 million citizens in 2019), based on the categories of the EU final demand in EXIOBASE (household and government consumption, inter alia). We show these impacts in the Supplementary Table 2. The funds’ investments are equivalent to the impacts of between 2.1 and 28.2 million EU citizens, depending on the impact indicator chosen. For example, for climate change mitigation, the total sample of investment funds is responsible for 62.2 million tons of CO₂-equivalents (MtCO₂-eq) direct emissions and 70.9 MtCO₂-eq indirect emissions. This is equivalent to the life cycle climate change impact attributable to the final consumption of Belgium in 2019 (11.5 million inhabitants), corresponding to 146.4 MtCO₂-eq, based on input-output calculations.

The variation in the million EU citizen equivalents is explained by the different drivers of impact for final demand versus funds’ holdings. First, the investment pool of a fund investing in global public companies tends to be skewed towards companies from Finance, Services, and Tech industries, as these are the companies with largest market valuation and largest share in the capital markets²⁹. Companies in these industries generally have low direct environmental burdens (for finance companies, second-order impacts, via their investments, are not conventionally counted via the life-cycle-based method), compared to consumption goods, which may play a larger role in the final demand attributable to EU citizens, hence the lower citizen amount equivalent in terms of environmental impacts. Second, for some environmental and social categories, we expect European consumption to be more intensive than an average sample of global public companies (as one could describe the funds’ holdings). For example, for the social indicators of trafficking in persons, restricted right to strike and restricted rights of association, the sample of funds has 20 times higher impacts than the total footprint of all EU citizens.

Comparing sustainability-labeled funds to conventional peers

To further explore insights of the fund-level results in our study framework, a one-to-one comparison of matching counterparts would be revelatory. For example, when comparing a Climate Paris-Aligned Fund (i.e., investing in companies that are on track with 1.5 degrees warming by the end of the century) with an Oil&Gas-focused fund, there are important differences that appear (Fig. 3). First, the Oil&Gas fund has the largest materials footprint of the whole sample. Then, the sustainable fund has higher values for some impact categories compared to the Oil&Gas-focused fund, while still doing better for the climate change indicator, which is its environmental objective. Another balanced comparison would be between two funds with Energy thematic focus, one conventional and one clean energy. Interestingly, the clean energy fund does show a lower life cycle GHG emissions intensity, but a higher intensity for human toxicity cancer effects – which could be due to exposure to companies involved in metals processing.

Fig. 3: Comparison of two funds on all proposed environmental indicators.

In (a) we compare two funds, a Climate Paris Aligned PAB ETF an Oil & Gas ETF. In (b) we compare a traditional energy ETF – with a clean energy-focused fund. The values across all impact categories are normalized, from 0 being lowest impact to 1 being highest impact, on the specific impact category. The graph’s highest point is determined by the maximum of each graph. The values represented are the life-cycle impact intensities, measured in unit of impact per million EUR revenue.

Widely spread impact distribution for the sustainability-labeled funds sample and comparison to conventional funds

We plot the distribution of the three samples by impact category in the Supplementary Fig. 3. Comparing the samples in aggregate may not be as useful as comparing two different funds, given that both conventional and sustainable funds may adopt different investment strategies in terms of industries and countries to invest in, which makes it impossible to find an exact match between funds. Moreover, there is a large overlap in terms of companies invested in, when looking at the average sustainable and conventional funds. For example, both conventional and sustainable funds hold large amounts of stocks from “tech giants” (e.g., Google, Microsoft, Alphabet)^30,36,37. This leads to having similar on-average results in terms of impact. The one-on-one comparison is more useful in identifying whether a specific sustainable fund would be more attractive than a conventional peer.

In the comparison of the conventional vs. self-labeled sustainable funds we find large variability of the life cycle impact intensity, which is to be expected given the diverse portfolios of funds. Statistical tests (detailed in document SD3) shown that both sustainable funds sub-samples (sustainable art. 8 and art. 9) and the conventional sample appear to be drawn from the same distributions. All summary statistics and t-test results are displayed in Supplementary Data file SD3.

For climate change indicator, sustainable funds art. 8 (art. 9) sample has a mean of 749 (699) and a standard deviation of 395 (315) tCO₂-eq per MEUR revenue, while the conventional funds sample has a mean of 798 and a standard deviation of 562 tCO₂-eq per MEUR revenue. The difference in the spread is caused by the very distribution of impact factors that were used to compute company and fund-level impacts and could be that for some indicators spread is higher. For the set of social impact indicators, the impact range is more concentrated, especially for the Article 9 funds. The reason for the smaller spread is using a social impact factors database with a lower resolution in terms of industries, i.e., 26 sectors vs. 163 sectors in the case of the environmental database. This leads to less variability at the level of company and fund-level estimates. Yet, outliers appear across all impacts. These can be driven by investments in companies which have activities in countries with large relative impacts, or be skewed towards specific industries, thus leading to a much larger result than the sample mean.

Contribution to impact per category

Taking a more in-depth look at the drivers of impact for funds, interesting differences between funds are revealed. The more sustainable funds (with Impact art. 9 classification) have a notably different allocation between economic sectors, via the companies that they invest in (Fig. 4). As Article 9 funds seek to prove environmental or social impact, it is expected to see these funds more invested in Primary sectors, such as Manufacturing and Electricity production³⁸. There is 16% allocation towards companies in the Transportation and Utilities sectors (compared to 3-5% for the ESG art. 8 and non-ESG samples) and a larger share for Manufacturing metals industries and Telecom, which can be explained by investing in companies that are key for a decarbonized and thus electrified economy. To our surprise, it can be observed that the sample of ESG art. 8 funds and non-ESG funds tend to have similar trends across all impact categories, as driven by the similar allocation of investment amounts between industries. The Finance and Services category of companies invested in by the funds is up to 40% for the non-ESG sample, which explains why the sample of funds has a smaller average impact intensity. For the Impact funds, the share allocated to Finance companies is at half (20%). This is proof that ESG funds may sometimes simply decrease their exposure to polluting industries by increasing invested amounts in already decarbonized sectors, like Finance and Services, a finding highlighted before in the literature³⁹.

Fig. 4: Scatter plot comparing the composition of funds’ samples by main industry class of companies invested in and the proportion of held amount vs. the proportion of owned impact.

The axes of the graphic are not equal across all sub-graphs, and they vary depending on the location of the maximum point on the respective axes. The different colors depict the industry groups used to classify companies and the different shapes depict the fund samples. The table in the bottom right part of the figure shows the total percentage invested by the funds sample in each industry.

Analyzing the scatter plots of Fig. 4, we observe that some industries are more important than others in terms of impact generation. For the environmental impact categories related to emissions of greenhouse gases or particulate matter: climate change, acidification, particulate matter, and photochemical ozone formation, the Article 9 funds drive most impacts from their exposure to Transportation and Utilities companies. The large impacts, even if the funds should be sustainable, can be driven, on the one hand, by investment in companies that still have parts of revenue from polluting economic activities, or, on the other hand, by a too general classification of revenue in the FactSet database, which would lead to our model allocating a more polluting impact factor to an otherwise low-carbon electricity generation source, for example. Fortunately, while for ESG and non-ESG funds, Oil & Gas industries do play a large role in the associated impact, the Impact art. 9 funds have a minimum exposure to companies from this industry. For human toxicity impact category, exposure to companies in the Manufacturing metals sector is driving a large proportion of the impact, especially for non-ESG funds.

It is interesting that for Impact art. 9 funds, companies from the Telecom industry play a large role, mainly companies involved in Electronics, which would be using metals in their indirect upstream operations. For material footprint, but also photochemical ozone formation indicators, non-ESG and ESG art. 8 funds are responsible for large impacts via their investments in the Oil & Gas industry. For biodiversity loss-related indicator, ESG art. 8 funds seem to derive large impacts from holdings in the Agriculture sector, which is not so present for the other impact categories. For this indicator, the impact is spread less equal between industries, with Companies under Other Manufacturing driving up to 70% of the impacts in the sample of funds, a trend similar for the water stress category. This discrepancy compared to other impact categories can be due to employing characterization factors that overweight impact factors attributable to specific sectors in specific regions and may thus be more accurate in the prediction of risk derived from exposure to impact.

For the social impacts, given that the linking between the financial revenue data and social indicators was done on a much more aggregated set of industries, the impacts at individual fund and company level also see less variation. Namely, all sectors included in Other Manufacturing, but also Retail and Trade and Finance and Services are driving the largest share of impacts. This is expected as social impacts are more pronounced in industries that are more human-centered rather than process-centered.

Implications for investors are two-fold. First, for environmental impacts manufacturing-intensive industries are driving the largest share of impact for most funds, due to their large proportion the capital markets and high impacts due to consumption of resources and release of emissions in their processes. Large parts of these impacts could be avoided when companies switch to low-carbon generation of electricity and energy. Impacts that cannot be abated are the ones driven by investments in Oil & Gas companies, for example. A second implication is that there will always be trade-offs within indicators either within or outside the same realm (environmental or social).

Synergies and trade-offs between social and environmental impacts at fund-level

Environmental and social impacts at fund level are not always strongly and positively correlated, as shown in the matrix of Spearman’s rank correlation coefficients in Supplementary Fig. 4, implying that an investment fund or a company can rank highest for one indicator, while scoring lower for other indicators. In general, we observe high correlations within each distinct subset of social and environmental indicators, and low correlations in between the two, meaning that there are larger trade-offs in between environmental and social indicators than within the environmental or social indicators’ groups. Within the set of environmental indicators, some strong synergies are visible, mostly when impact indicators are derived from common environmental flows. For example, acidification is highly correlated (coefficient larger than 0.9) with terrestrial eutrophication and photochemical ozone formation: expectedly so as ammonia emissions are contributing to all these three impact indicators. Human toxicity is correlated to ecotoxicity as heavy metals emissions are an important contributor to these impact indicators. At the opposite, for human toxicity we observe trade-off with water stress, given the low degree of similarity between the two impact indicators. Another trade-off observed at fund-level is for indicator climate change with land-use related biodiversity loss. Biodiversity loss would thus be a very important indicator to measure alongside climate change, in order to avoid causing more harm for biodiversity when investing with reduction of GHG emissions as main goal. Within the set of social indicators, there is a predominance of very high correlations.

Given the fact that funds tend to have high similitude in portfolio allocation – if, for example, more funds follow the market index – correlation is higher when performed at fund level. At fund level, both the portfolio allocation and the sector-country distribution of the held companies’ revenue drive the correlation coefficients. If we do the exercise at company-level, only the inherent economic activities of the company drive the correlations. In the Supplementary Fig. 3, we show correlation matrixes between all impact indicators at company level, separately by main sector group. For Retail and Wholesale Trade sectors, there are strong negative correlation coefficients between most of social and environmental categories, signaling a high trade-off when investing in these industries. Indeed, the trade sector can be described as having low environmental pressures, while having a high impact on workforce. On the contrary, for Transportation and Utilities companies, we see a weaker negative relation between social and environmental indicators, while we observe more pronounced negative correlations between environmental indicators – especially between particulate matter and the other indicators, as companies with very high particulate matter impacts rank lower on other environmental indicators. Companies in the Mining industry are perhaps the most interesting, as they show strong trade-offs between social and environmental indicators, but also within environmental indicators (namely material footprint vs. the rest of environmental indicators).

Concentration of impact in key industries

The impact factors at industry and country level are the key drivers of impact at fund level, and thus understanding these is the key to making better investment decisions. While the differences between impact factors for different indicators are already explored in the literature at the resolution of countries⁴⁰ and sectors²⁷, an extensive analysis for the distribution of company-level impact intensities is not existing. This is an interesting exercise, as companies usually have heterogenous economic activities in their portfolios, that are spread over multiple geographies. In addition, services and other administrative activities can contribute in part to the revenue generation of company, reducing the total impacts that would be associated only with the main activity of a production-oriented company.

Impact categories that relate more to use of resources have high impact factors for industries like Agriculture, but also Manufacturing, as we show the results for the life-cycle impacts (Fig. 5, with raw data reported in SD4). Especially, land use-related biodiversity loss and water stress show much higher impacts for Agriculture companies. Similarly, impacts that relate to emissions release, are rather linked to fossil-fuel industry and companies with energy-intensive processes. Mining appears the sector in which companies have highest impact intensity across most of the social impact categories. For categories gender wage gap and violations of labor requirements, companies in the Trade sector also have high intensities. Agriculture-related companies have large associated impacts, which can be explained by companies hiring more unqualified workers and not offering decent working conditions.

Fig. 5: Boxplot chart of the distribution of life cycle impact intensity at company level across main industry groups, in indicator specific unit per million EUR.

Boxplot chart of the distribution of life cycle environmental (a) and social (b) impact intensity at company level across main industry groups, in indicator specific unit per million EUR. Outliers are removed from the graphs to increase readability of the graphs. We apply normalization by mean to have consistency in the x-axis across the graphs. The same classification and order of industries is used for each graphic. Companies are grouped by one main SIC code, even if one company may be present across multiple industries. The bars show the 25th and the 75th percentiles of values.

Largest impacts traceable to a small number of companies

Finally, we identify companies (grouped by main industry) that drive the lion’s share of impact, by impact category. Surprisingly, less than 1% of the companies invested in drive in aggregate more than 50% of the total life cycle impacts estimated for the funds sample. In Fig. 6, we show results only for two impact indicators, with all the other results in the SI (including listing of top companies in SD5). Fund-level allocation seems concentrated around a very small number of large corporations, meaning that the studied funds tend to hold similar large companies in their portfolios, with different holding amounts. The list of companies at the top varies depending on the impact category analyzed.

Fig. 6: Identifying the industry groups that drive 50% of the funds’ impact for two impact categories.

Holdings are grouped by main industry classification. The first column shows the proportion of market value held, while the second column shows the proportion of impact responsibility. Additional impact categories are available in Supplementary Fig. 4.

For water stress indicator, 50% of the life cycle impact can be traced back to only 27 large corporations (out of almost 5000 different companies). Funds hold aggregated positions amounting to 28 billion USD in these 27 companies (representing only 8.3% of all positions held). The main positions driving impact being investments in Nestlé, Unilever, and Danone—all three large companies from the fast-moving consumer goods (FMCG) sector. It is expected to see FMCG companies bearing the largest share, as these depend on manufacturing of diverse products, but also cultivation and processing of raw materials, in the case of the food processing sub-sector. Similar importance of FMCG companies is seen for land-use related biodiversity loss indicator, while companies in the Paper sector also play a large role here (due to deforestation impacts). For human toxicity impact indicator, companies in the Information Technology (IT) sector have, in aggregate, the highest contribution (largest companies being Schneider Electric SE, Samsung Electronics, and Siemens AG). IT companies, including semiconductor manufacturers, have large market values, hence the large exposure of funds. The high values for human toxicity are driven by the need of metals and other chemical compounds in the manufacturing phase. For climate change impacts, companies from the Utilities and Oil&Gas sectors drive the largest share of impact (biggest contributors Enel SpA, China Petroleum, and Iberdrola SA). For social impacts, there is a more even distribution of impacts between industries invested in. Surprisingly different to the environmental impacts is the prevalence of Finance and Services sector companies as high contributors to negative social impacts. This is because social issues tend to be more prevalent in finance and services-related sectors. Moreover, for indicators anti-competitive behavior and children in employment, companies from the Industrials sector also have a very high contribution.

Discussion

It has become clearer from our results that trade-offs between and within social and environmental impact categories occur. Focusing on one or a few impact indicators in the detriment of others, could lead to doing more harm than good. As stipulated by the EU Taxonomy, impact assessment should include all relevant environmental impact categories to avoid impact shifting within categories (for example, green electricity can lead to reduction in GHG emissions but may cause negative impact on biodiversity). Additionally, it should include social considerations to account for categories of stakeholders not considered when looking solely at environmental issues (for example violating the rights of indigenous people by approving construction and deforestation on native lands). Company-level correlations in terms of impacts, uncover more specific trade-offs. For example, investments in Utilities companies have better scores on social issues and worse score on environmental issues. Our results strengthen previous findings that climate change cannot always be used as a proxy for all environmental impacts^41,42,43.

Compiling a full set of ready-to-use indicators is delimited by the availability of IO databases with associated impact assessment methods of adequate quality. The continuous improvement in IO databases like EXIOBASE and PSILCA will likely lead to more accurate and complete estimates resulting from our proposed framework. Currently, IO databases remain limited at the level of sector and country resolution, despite the fact that for some countries IO tables with detailed sectorial resolution are available—for example the USEEIO tables have data spread across 389 sectors. However, to allow a global analysis, a higher sectorial resolution harmonized for all countries would be needed. For most of the indicators we use, impact factors are available only at country, or even only at regional resolution. For the few indicators where we use regionalized impacts, these are at country level, whereas sometimes differences could be even more localized, in order to obtain more realistic estimates⁴⁴. Using generalized impact factors makes it harder for a portfolio manager to distinguish between two companies operating in the same industry and thus choose the better company. At the same time, given the lack of granularity in the impact factors data, companies actually having more sustainable practices may be penalized by getting the sector average, and vice versa for more polluting companies. Another limitation of IO databases is the degree of availability of background data. For example, EXIOBASE represents very well information about Agricultural sectors – with detailed sub-sectors and detailed information about agriculture-related pollutants, while it has less detailed information about companies in the Chemicals sector and about emissions of toxic substances. In PSILCA the availability of social impacts data from international statistical bodies influences the accuracy of its social extensions—for example, some industries may have more transparent and accessible information than others. These limitations can lead to under- and over-estimation of impacts at company level.

Our modeled results are also susceptible to uncertainty coming from the limited level of detail in revenue reporting – the coarser the level of reporting, the higher the risk to have under- or over-estimated impacts. This is the case for the company Iberdrola SA, where information on the type of electricity produced is not available in FactSet, leading to overestimation of impact for climate change indicator, as average electricity generation impact factors are used instead of factors for renewable energy. Therefore, better reporting from the company side at the level of economic activities undertaken is necessary.

Our results provide evidence that impacts attributable to funds are substantial. The large share of indirect impacts, previously highlighted at industry level²⁷, is also to be observed at fund level. If we regard investment funds as entities carrying the responsibility for their investment, their environmental and social footprints are comparable to that of EU consumers, albeit much larger given the high value of capital markets. Hence, we call upon a stricter approach to disclosure requirements in terms of indirect impacts. For example, measuring and setting indirect impact targets allows investment managers to exert influence over the companies in the supply chain, thus increasing the potential engagement opportunities¹¹. In addition, we have observed a strong concentration of large publicly listed companies in the portfolios of analyzed funds. Depending on the impact category, the industry and companies contributing most impact are shifting. The large exposure to certain companies and impact hotspots can be a driver of engagement with companies, demanding improvements in environmental and social practices¹¹.

Our analysis can serve as a baseline for harmonizing sustainable finance regulations and science-based sustainability assessment and its impact indicators. Standardization would facilitate comparability and reliability of indicators. Main strength of our approach is that impacts are estimated using the same background methodology data for all environmental indicators; for social indicators, similar methodology but a different underlying input-output database are used. Irrespective of the type of reporting requirements or their location, our proposed set of indicators is embedded in international practice related to sustainability assessment and can thus serve as a general framework for sustainability assessment at financial product level.

Methods

Methodological framework

Our method is based on input-output life cycle assessment (IOLCA) and is using life-cycle-based impact factors derived from environmentally and socially extended multi-regional input-output (MRIO) tables to estimate life cycle sustainability impacts associated with investment funds, via the companies that they invest in, to support reporting mandated via sustainable finance regulations.

MRIO databases contain structured data of all transactions taking place in the world in a specific year, with the information being detailed at the level of country and sector (or product). Data is available in a matrix format and is measured in monetary units. In essence, MRIO databases give information about the production recipe of an aggregated sector, based on the monetary transactions that it depends on to produce the respective sector’s output²⁶. Environmental and social information linked with the specific countries and sectors is then added to the MRIO tables, in order to conduct environmentally or socially extended MRIO analysis (EEMRIO). The environmental and social accounts are based on data collected from various trusted sources, such as FAOSTAT for emissions data in Agriculture²⁸, or the International Labor Organization (ILO) for social data⁴⁵. Direct and indirect environmental and social impact factors expressed in impact category unit per monetary output (e.g., in tCO₂-eq per MEUR) are obtained from the direct and life-cycle matrix of transactions of EEMRIO tables, by dividing the total impacts associated with a country-sector by that country-sector output. The obtained impact factors can then be used to derive direct and life-cycle impact for a company, by multiplying the country-sector impact factors with the revenue of a company for the specific country-sector combinations, obtaining an estimate of the absolute impact of a company. Revenue information is used as proxy for the economic activity of the company as it is the only information that is available in standardized form for the largest sample of companies.

Input-output life cycle analysis (IOLCA) allows to estimate regional and sectorial impacts per monetary unit, making it suitable for assessments at organization or company level. Moreover, this type of analysis is useful for the assessment of supply chain impacts, as companies do not have visibility over their indirect suppliers, even if it is usually where the majority of impacts take place²⁷. EEIO models are also recommended by the GHG Protocol for corporate accounting on GHG emissions⁴⁶ in estimating supply chain emissions where company data is lacking and is equally used by companies reporting to CDP for example⁴⁷.

IOLCA has been previously adapted to estimate impacts of financial portfolios, both in academia^29,30, and in the development of proprietary models by different data providers^48,49. While academic studies do use IOLCA to estimate impacts of companies⁵⁰ and of funds^29,30, it has been done either at the level of one company only⁵¹, using a single country and single sector classification of companies^30,52, or only for a limited set of indicators, like GHG emissions only²⁹. Trucost, a proprietary database used in the corporate sector, which offers estimates of company level impacts, however using only a single-region US-only EEIO table⁴⁸. The method developed in Popescu et al.³⁹ proposes the use of a very detailed by-country and by-sector profile of a company’s revenue, in order to obtain more precise estimates of corporate carbon footprints and investment fund carbon footprints. Our method extends the method of ref. ³⁹ to other 12 environmental indicators and proposes the use of a socially extended MRIO database for the estimation of an additional set of 13 social indicators at company and investment fund level. The detailed model framework is presented in step-by-step schema in Fig. 7 and detailed in the sub-sections below.

Fig. 7: Schematic representation of the procedure to ready-to-use life-cycle-based impact assessment for financial products.

The first module is related to extracting the environmental and social impact factors. The second module is the dictionary linking EEMRIO data to financial revenue data. Module 3 is about extracting detailed revenue breakdown for all companies in the dataset. Module 4 is the estimation of impact at company level. Module 5 is the extension of the model to estimate impact at fund level, using company-level invested amounts as weights.

Input-output calculations

Technically, the calculation of impact factors is based on the conventional MRIO modeling framework using matrix calculation. Following the input-output nomenclature, the direct impact factors vector for each impact indicator ($S$) contains impact factors for each country-sector combination and is obtained by dividing matrix $F$ (total impacts by country-sector) by total output $x$:

$$S=\frac{F}{x}$$

(1)

Based on the original input-output table $\left(A\right)$, known as the direct requirements matrix, the total (life cycle) requirements matrix $(L)$ – the Leontief inverse – is computed:

$$L={(I-A)}^{-1}$$

(2)

The Leontief inverse allows us to compute the life cycle impact factors $M$ (or multipliers), by multiplying matrices $S$ and $L$. $M$ is the vector of life-cycle impact factors, and it allows estimating complete supply chain impacts based on tracing all monetary transactions between sectors:

$$M=S\,\times \,L$$

(3)

With the life cycle impact factors vector $M$ we are only capturing direct and supply chain impacts (or upstream impacts). Given the use of an IOLCA framework, the computation of downstream impacts (impacts from use phase onwards) it not straightforward and cannot be derived directly from the IOLCA tables. However, it could be estimated using traditional LCA data, but it is not covered in this paper. The $S$ and $M$ vectors, explained above, are used to derive a database of country-sector direct, indirect and life cycle impact factors $({IF})$, for each indicator and country-sector combination (impact factor by country $j$ and sector $k$)⁸. Depending on the impact category, the indicator could be grouping together more environmental or social flows and their specific weighting factors, based on the importance of a specific indicator flow. The IFs are then linked with the revenue breakdown for a company $i$(${R}_{{ijk}}$), as per Eq. 4 below. The quality and granularity of revenue-level data is a main driver of final reliability of company-level impact estimates.

$${I}_{{ci}}=\mathop{\sum}_{j} \mathop{\sum}_{k}\,{{IF}}_{{cjk}}\,{R}_{{ijk}}$$

(4)

Selecting life-cycle-based indicators and deriving country-sector impact factors

The next step is selecting the indicators for which we want to compute the impact and deriving the impact factors based on selected MRIO databases (Fig. 7 Module 1). As mentioned, the main goal of our method and analysis was to showcase the usability of life-cycle-based impact indicators in assessing the sustainability of investment funds. This exercise is relevant in the context of recent regulations that mandate sustainability reporting for funds on environmental and social issues, as a clear and science-based approach to indicators to be used to this aim is lacking. Therefore, we propose a set of indicators that match the EU Sustainable Finance Disclosure Regulation¹⁸ (SFDR), which targets investment product providers specifically. Since the SFDR is part of the larger EU Sustainable Action Plan, together with the EU Taxonomy³³ and EU Corporate Sustainability Reporting Directive (CSRD)³², we have also analyzed the degree to which the three regulations are aligned (Fig. 1). Under the SFDR, a set of mandatory and voluntary environmental and social indicators are proposed – the Principal Adverse Impact (PAI) indicators for investments in investee companies. We draw the (mis-)matchings between the PAI indicators proposed and the EU Taxonomy and the CSRD. In the Supplementary Note 1, we further discuss the regulations. Given that there are differences at the level of indicator construction, we dive separately into environmental and social indicators. A detailed critical analysis of the indicators proposed under the SFDR, and their type can be consulted in Supplementary Note 2 and Supplementary Fig. 2.

Environmental impact indicators

Especially for the environmental dimension, the difficulty in choosing a sufficient and comprehensive set of indicators lies in the existence of multiple methods to assess the same impact category, even for established impact categories, like climate change. Moreover, environmental impact categories group together more substances (or environmental flows in LCA nomenclature) – for example, all types of toxic emissions. Given the large number of pollutants, these are grouped into a category that measures a specific impact – for example, toxicity. To aggregate more environmental flows, specific weighting factors are used to quantify the contribution of the flow to the impact category. These weighting factors are named characterization factors (CFs) and are based on scientific literature. For certain categories, like human toxicity, the CFs are harder to define and thus methods are continuously developing. The process is detailed in the Supplementary Note 2 and associated Supplementary Fig. 1.

The Environmental Footprint (EF) is a best-practice environmental impact assessment framework, composed of a set of 16 impact indicators and their underlying methods and characterization factors. We choose the EF method as it is based on latest literature developments, can be linked to ready-to-use indicators built using EXIOBASE^53,54 (the selected EMMRIO database) and it is brought forward by EU policy. We are using the latest version, 3.1, updated in 2022. From the 16 impact categories we only selected 13 (Table 1) excluding two EF indicators that are not covered in the EEMRIO data (ozone depletion and ionizing radiation), and additionally grouping its two resource use indicators into the material footprint indicator. The selection of ready-to-use, life-cycle-based indicators is restricted, in a first step, by the availability of raw environmental indicators in the EEMRIO database EXIOBASE, a database developed and maintained under a European research project²⁸, which has been widely used in academia and in practical case studies for environmental assessments^28,55, employed for similar purposes before^30,56. Compared to other input-output databases, it has a greater coverage of environmental accounts and a detailed coverage of European Union economies^57,58 (44 countries and five rest-of-the-world regions, 163 sectors, and 1,114 environmental flows). To derive impact factors by country-sector combination we use the CFs provided in the EF 3.1 methodology and the matching environmental flows that can be retrieved from EXIOBASE. EXIOBASE does not cover all environmental flows that are described under the EF and thus some of the impact factors will be underestimating impacts.

Table 1 13 environmental impact indicators for sustainable finance reporting, based on Environmental Footprint version 3.1. and EXIOBASE

The missing coverage has been previously discussed in literature and input-output analyses would benefit in the future from the inclusion of other key environmental flows⁵³, such as more toxic substances⁵³, or data on use of pesticides⁵⁹. Missing environmental flows also influence the final results, meaning that impact factors used may be different that the values of process based LCA impact factors, where all environmental flows would be considered. The difference in impact magnitude caused by the location where the impact takes place is highly relevant, especially for water stress⁶⁰, pollution⁶¹, toxicity or biodiversity. To improve thus the precision of the estimates, we replace EF indicators water use, particulate matter, and land use with impact indicators that account for differences in impact factors based on the location where the impact takes place – i.e., the country. The updated indicators are water stress, particulate matter health impacts, and land-use related biodiversity loss. These are obtained in final form as country-sector impact factors from the work of ref. ⁶². These indicators have a higher precision that EF indicators, since they account for the location-specific situations for each of the impact categories.

For example, we used country and region specific characterization factors based on the AWARE method⁶³ to move from water use to water stress and thus differentiate based on country and sector, not only on quantity of water used but also the criticality of water for the specific country or sector in a country. As the regionalized impact factors are obtained only for the year 2015, we use impact factors for all categories for year 2015.

Concerning well-matching environmental impact topics, climate change mitigation is the only objective with a one-to-one relationship between all standards, validating the maturity level and consensus on this main environmental issue. Other SFDR indicators are a partial match and could be grouped to link to one life-cycle-based indicator. Conversely, the SFDR PAIs of land degradation and activities negatively affecting biodiversity-sensitive areas can be grouped under the life cycle impact indicator of land-use related biodiversity loss⁶². For water use and exposure to areas of high water stress, we propose the alternative indicator of water stress, which weights water usage based on the characteristics of the region where it takes place – whether the area is more at risk of water stress or not⁶². For SFDR PAI indicator emissions of air pollutants (like ammonia – NH₃), the EF method provides several impact indicators to assess their effects – namely terrestrial acidification and eutrophication⁶⁴. Similarly, emissions to water and emissions of inorganic pollutants are represented by the corresponding life cycle impact indicators of toxicity and eutrophication.

As an exception, the EU taxonomy objective of climate change adaptation has no standardized equivalent indicator in LCA. In the extended literature, there are however many examples of indicators specifically developed to measure adaptation, e.g., flood safety levels for a stormwater management system⁶⁵ measuring the impact compared to a reference scenario. In some cases (e.g., electricity generation) the same indicators used for climate change mitigation can apply¹⁹. Similarly, for the circular economy (CE) objective, there is no direct correspondent to a PEF indicator. While multi-dimensional scoring tools that serve as CE indicators have been previously developed and tested, such as the Circularity Potential Indicator^66,67, these are not yet widely used in LCA. Life-cycle-based indicators assessing resource use and scarcity⁶⁸ may be used to this aim, until better indicators based on reliable collected data will be developed^67,69.

Social indicators

For social issues, the SFDR proposes rather qualitative indicators, some even only concerning due diligence & compliance. For example, the scope of the SFDR-proposed indicator of “violations of UN Global Compact principles and OECD Guidelines for Multinational Enterprises” is too broad and would not give stakeholders a sense of the social impacts that underline a funds’ portfolio. We provide a set of rather quantitative impact indicators matching the SFDR’s PAIs, that are also linked to the social issues identified by the EU Taxonomy and the CSRD. To ensure that one can estimate impact at financial product level, we analyzed indicator availability from PSILCA^29,45, a social IO database built using the input-output tables of EORA database, a widely used EEMRIO, on which social extensions are added. Social input-output databases have only been recently developed, to aid in accounting for the social impacts embodied in the global economy. Therefore, their reliability and use are lower. PSILCA has a large coverage of social indicators – over 90 – and detailed country and sector-level coverage for EU countries. We selected a set of 13 social indicators (Table 2) related with the social indicators categories proposed by UNEP Social LCA Guidelines²⁵ and extracted the country-sector impact factors from PSILCA. Examples of PSILCA indicators are: rate of accidents, children in employment, right to collective bargaining. A winsorization of 5% was applied to correct outliers that appear if, for example, withing sectors in certain countries with very low outputs.

Table 2 13 social impact indicators for sustainable finance reporting, based on PSILCA

While social impacts are by default driven by more abstract characteristics of a company – such as employee policy – more quantitative indicators can be developed, that allow for a clearer assessment of a company or investment. Hereto, quantitative indicators by social impact category are being developed in social LCA^25,70. We propose semi-quantitative risk-based indicators available in the social IOLCA database PSILCA, that can be estimated at sector, company, and financial product level.

Yet, we did not consider the indicator values expressed in raw units, but their translation in medium risk hours equivalents (mrh), as proposed in the PSILCA documentation⁴⁵. The raw indicator unit, while easier to interpret, cannot be extended easily to estimate life cycle impacts. For example, in the case of a raw unit in percentages, one cannot extend that to the estimation of life cycle impacts, as the percentage unit does not function like a physical unit, when the direct percentage is known. Medium risk hours equivalents unit is the multiplication of the hours worked in the sector with a factor that represents the extent of risk based on predefined criteria, where a medium risk has a factor 1. The mrh unit can be used to derive the life cycle impacts and allows for comparison between indicators.

Impact at company level

A key step before estimating impact at company level is defining the sector and country level correspondence (concordance matrix) between the financial revenue database and the input-output databases used to retrieve the impact factors (Fig. 7 Module 2). It is a step that should not be overlooked, as the quality of the matching and also the level of detail in revenue breakdown of a company will define the accuracy of the estimate. For company-level revenue information the FactSet database was used. Specifically, the FactSet RBICS database contains detailed breakdown of revenue for over 45,000 companies in the world. The most detailed classification available has 1700 different sub-sectors, allowing to achieve a high level of detail in the company revenue distribution. For geographical distribution, FactSet GeoRev shows the breakdown of company revenue by country, without any aggregation at region level. As per Fig. 7 Module 3, the company-level breakdown of revenue by sub-country and sub-sector is extracted from the FactSet database. Specifically, we have information on the percentage of revenue a company generates from different geographies and sectors at the same time. A company will be assigned the average impact factors of the general country-sectors combinations that constitute its revenue generation streams. This allows to obtain a life cycle impact per company, expressed in euros of revenue (e.g., kg CO₂ eq. per million EUR of revenue output for GHG emissions indicator), that is the weighted average of the underlying economic activities of the company (Fig. 7 Module 4).

It is important to mention that the company-level estimates represent only an approximation, that overlooks the unique profile of a company. An example of the linking process for an example company is shown in Fig. 8 – the high detail from FactSet database is lost when translating to EXIOBASE sectorial-level classification. However, EXIOBASE aggregation, in this case, is reasonable, as, from an environmental point of view for the impact calculation one can still differentiate between manufacturing different types of products (e.g., diary vs beverages). However, for other sectors, such as Chemicals, there is no distinction in EXIOBASE between industrial chemicals production and pharmaceuticals, whereas the differences in impact factors can be high. Thus, the reduced level of granularity in the IO databases leads to lower accuracy of estimates. The implications of using industry-averages for estimating impact are discussed in more detailed for GHG emissions in ref. ²⁹ and exemplified by running a validation exercise against self-reported company data.

Fig. 8: Correspondence between FactSet classification and EXIOBASE classification for an example company.

a the revenue split by RBICS l6 classification, as extracted from FactSet. b the revenue split by EXIOBASE sector, after applying linking between FactSet RBICS l6 and EXIOBASE 163-sector classification. c the revenue split by country, using FactSet GeoRev. d the revenue split by country, using EXIOBASE 44-country classification, as the other countries appearing with gray in the figure are grouped into one of the five rest of the world (RoW) regions.

The PSILCA IO database is used to derive social impact factors for firm-level estimation, with year 2015 as reference for the input-output data, and 2017 for indicator-level data⁴⁵. However, there is not a common sector classification between countries in PSILCA. To have a harmonized sectorial breakdown for all countries, we have aggregated all the impacts at the level of the common 26 sectors classification which is also the common classification of the EORA26 database, by computing the mean of the impact factors of all sectors linked to one EORA26 sector. This leads to very coarse estimates at company level, as the detailed FactSet RBICS sector classification is reduced to the 26-sectors classification of harmonized PSILCA.

To derive impact estimates at company level, we run a function for each company and impact indicator, which uses the breakdown of company revenue by sector and country expressed in euro currency and matches it with the country-sector average impact factors per monetary unit for the specific impact indicator (Fig. 7 Module 4). For matching environmental impact indicators, we use the concordance matrix developed manually in ref. ²⁹, based on finding the best match between the FactSet RBICS database used and the EXIOBASE nomenclature. The concordance matrix bridges the EXIOBASE sector and country dictionary for impact factors with the FactSet company-level country and sector dictionary for revenue breakdown. For the social concordance matrix, we linked the EORA26 classification to the sectorial classification from the database of financial revenue breakdown, in a 1-to-n linking – meaning that more country-sectors from the revenue database will receive the same impact factor, as they are part of the same aggregated EORA26 sector.

Impact at fund level

Finally, impacts at fund level can be expressed as absolute values in terms of owned impacts, for each impact indicator (category) c, by computing the share of a company’s impact that an investment fund $f$ is responsible for (Fig. 7 Module 5)⁸. The impact is derived based on investment fund-level information: the list of its public equity investments and the amount invested. Practically, for one impact indicator, the total impact of a company $({I}_{{ci}})$ is divided by the market value (${M}_{i}$) – total shares multiplied by price per share. Each shareholder is attributed its share of the holding, per monetary unit of investment. We use the direct and indirect impact amounts at company level to derive the direct and indirect impacts attributable to the fund. We do not adjust the impact factors for double counting, which results in a fund being allocated two times the same quantity of impact if it is part of one company’s direct operations and another company’s indirect operations. However, as this is an exercise to show the exposure of the fund, it is not critical to account for double counting, as the risk comes from both companies, and it is important to acknowledge it. For each company, the weight held by the fund in the company is accounted for (${w}_{{if}}$), which is the invested amount by each fund in each of its company holdings. This measure accounts for the market valuation of a company, dividing the responsibility of impact between all its shareholders:

$${{Fund\; impact}}_{{cf}}=\mathop{\sum}_{i}{w}_{{if}}\frac{{I}_{{ci}}}{{M}_{i}}$$

(5)

Corporate revenue data and investment funds’ holdings data

As source for financial information, such as holding amount at fund level and company-level revenue data, we use the proprietary dataset of FactSet⁷¹, that can be accessed by purchasing a license. The datasets are FactSet Ownership, for investment fund-level data, and GeoRev and RBICS, for information on the distribution of company revenue at country and sector level. Having data in monetary amounts about the revenue distribution of each company is the best available option. Ideally, companies would share information in physical units about the produced amounts and purchased products. However, companies seldom disclose this type of information, and the most reliable and complete data available as proxy for produced amounts is revenue-level data, in monetary units.

A suite of environmental and social input-output databases is available, each with distinct characteristics, but building on the same principles. Widely used databases for environmental assessments are EXIOBASE²⁸, EORA⁷², GTAP, and OECD⁵⁸. The two main databases for social assessment are PSILCA⁴⁵ and the Social Hotspot Database (SHDB). While the environmental databases are fully free, or free for academic use, neither of the two social databases are freely available. Differences between databases are in the level of disaggregation available at country and sector level and in the impact extensions available. Deviations in data and results between input-output databases have been studied in previous work and the main drivers of variation are the structure of the economic flows and the environmental and social accounts data⁵⁷. As such, our results and coverage of proposed indicators and underlying environmental flows are influenced by our choice of primary IO database.

Selection of funds’ sample

According to market research by Morningstar³¹, SFDR-labeled article 8 and article 9 funds amounted to 10,608 in December 2022, representing 37.8% of all the funds available for sale in Europe. Our initial sample, of article 8 and 9 funds listed on the Luxembourg Green Exchange, is of 1389 funds. The assets under management of these funds represent 13.7% of the total AuM of SFDR -labeled funds (i.e., 630.23 billion USD out of 5.01 trillion USD). The total global pool of sustainability-labeled funds, which comprises all types of funds, not only equity funds, is estimated at around 5 trillion, 12 times higher than our sample³¹. Our final sample is reduced to 230 funds, Article 8 and Article 9 funds, after removing funds with more asset classes (as it leads to double counting for impact intensity metrics) and removing non-equity funds (i.e., funds investing in fixed income or money market funds), as for these we cannot directly apply our proposed assessment model. The sample of equity funds is heterogenous in terms of investment theme and size, ranging from 4 million USD to 16.5 billion USD in Assets under Management (AuM). The funds in the selected sample hold together 401 billion USD of investments, which, if compared to the size of an economy, is approximately as large as the gross domestic product (GDP) of Denmark over a (398.3 USD billion in 2021, according to the World Bank data⁷³). To benchmark our results, we perform the analysis on a sample of conventional funds. The year for which we select the funds, and the holdings matches that of the SFDR-labeled sample. The funds selected are also registered in Luxembourg, just as the sustainable funds sample. There is a total of 288 funds, selected based on the sample of funds used in the paper of Popescu et al. (2022), after removing funds not registered in Luxembourg and also adding industry-specific funds to enable a one-to-one comparison, like Energy and Fossil Fuels. Given that funds are heterogenous, there is no straightforward way to select a perfectly matching sample that would have same financial characteristics as the sustainable funds sample. We ensure that both samples represent funds that are available to investors and invest in public equity.

Spearman rank correlation

We apply the Spearman rank correlation, as this is not sensitive to outliers and leads to more reliable results that the traditional, default correlation method used—the Pearson correlation. The Spearman rank correlation uses the rank of the observations on each variable, thus being described using a monotonic function⁷⁴. The correlation is conducted only on funds in the initial sample of SFDR-labeled funds.

T-test for the difference between two samples

We used the non-parametric Mann–Whitney U test to study the statistical difference between the three samples of funds. We apply the test in python.