Master Thesis Quantitative Finance
Are Managers of ESG Mutual Funds Skilled?
Name: Anouar Bousraou
Supervisor: dr. E. Eiling
Coordinator: dr. T. Ladika
Study programme: Quantitative Finance, MSc Finance
Date: 01/07/2022
Abstract
The growth of the ESG market has attracted significant academic attention and a certain number of researchers have found that ESG screening improves returns. However, these results are potentially biased because they can be based on simulated performance, which can be subject to data mining. In this research, the effect of ESG investing is analyzed by examining whether active ESG mutual fund managers are skilled. This research proposes several methods to assess managerial skill, namely by computing net alphas, controlled for false rejected hypotheses, and the value-added, controlled for luck by bootstrapping. The results show that only 0.14% of 6,602 ESG mutual fund managers obtained a positive net alpha when controlled for false rejected hypotheses. Furthermore, the total sample consists of 12,461 ESG and non- ESG mutual funds, of which 0.097% delivered a positive net alpha when controlled for false rejected hypotheses. At last, there is enough statistical evidence to conclude that ESG funds add, on average, a negative value to the market when controlled for luck. All in all, these results indicate that the average ESG fund manager is not skilled.
Statement of Originality
This document is written by Student, Anouar Bousraou, who declares responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion not for the contents.
Table of content
Chapter 1. Introduction……….4
Chapter 2. Literature Review………6
▪ 2.1. Intro to ESG………...6
▪ 2.2. The effect of ESG investing……….………...8
▪ 2.3. Active ESG funds……….………...9
▪ 2.4. Performance measurements and controlling for luck………...11
Chapter 3. Data description……….……….14
Chapter 4. Methodology……….………...17
▪ 4.1. Net alphas……….……….17
▪ 4.2. Value-added…………...……….………...…………..19
Chapter 5. Results……….………...21
▪ 5.1. Net alphas...21
▪ 5.2. Value-added...27
▪ 5.3. Robustness checks...30
▪ 5.3A. The evolvement of net alphas ...30
▪ 5.3B. Bootstrap analysis of value-added...35
Chapter 6. Conclusion...38
Chapter 7. Discussion...39
Chapter 8. Future work...41
Appendix...42
References...46
1. Introduction
In September 2021, the ECB published the results of its economy-wide climate stress test which indicates that firms and banks benefit from adopting green policies early on. Luis de Guindos, vice President of the ECB, announced that policies to transition to a greener economy should be taken swiftly, otherwise physical risks will increase exponentially over time. Furthermore, losses on corporate loan portfolios have risen significantly, driven by physical risk. When comparing different portfolios, the results of the ECB show that portfolios most vulnerable to climate risk are 30% more likely to default in 2050 compared to 2020 (ECB, 2021).
The exposure of portfolios to environmental risk can possibly be measured by ESG scores. The uncertainty regarding this is argued by Berg et al. (2021) who have shown that ESG scores are likely to be affected by data rewriting issues due to the competition amongst ESG rating providers. These providers may beautify ESG scores to attract more customers.
Therefore, the tests that relate returns to ESG scores may be biased. Additionally, this suggests that an ESG portfolio may not be as sustainable or socially responsible as it declares because the portfolio composition is typically based on these ESG scores. Empirical evidence exists that some fund managers characterize as socially responsible, yet make intangible investment decisions (Candelon et al., 2021). This leads to uncertainty for socially responsible investors as they cannot be certain whether a certified socially responsible fund lives up to its name.
Nevertheless, ESG investing has become increasingly popular for both the investment and corporate community. As a result, investment based on ESG screening reached 30.7 trillion USD at the end of 2018 (Dai, 2021) and AUM under ESG principles grew from 6.5 trillion USD in 2006 to 68.4 trillion USD in 2017 (Daugaard, 2020). According to Hale (2020), social motives, like the COVID-19 pandemic, alongside climate change, and the movement for racial justice are found to be catalysts for investors wanting to align their investments with socially responsible funds.
There is a broad spectrum of literature that studies the effect of ESG investing. Berk &
van Binsbergen (2021) evaluated the quantitative impact of socially conscious investing, the results show that impact investing does not have a significant effect on the long-term cost of capital of targeted firms. Moreover, Pastor et al. (2021) show that high returns of green assets are due to the preferences of investors and arising climate concerns. At last, Dai (2021) found that ESG investing improves risk-adjusted returns and can beat the market in China. Despite the few researchers who show that ESG investing can achieve additional returns, the support for this is potentially based on simulated and not actual performance. Future research should clarify whether the simulated performance was due to skill or data mining (Daugaard, 2020).
Managerial skills can be measured in the active asset management field. Pástor et al.
(2020) studied the performance of active sustainable funds during the COVID-19 crisis by using a sample of 3,626 funds. The researchers found that sustainable funds perform better during periods of crisis. This is also acknowledged by Nofsinger & Varma (2014), who studied the performance of socially responsible funds, within a sample of 240 equity mutual funds, during the period 2000-2011. In this research, the analysis starts in 2000 until 2022 and consists of 12,461 mutual funds. This is a larger timeframe and sample than previous research.
Furthermore, the performance measurements conducted in this research, which will be discussed in the next section, differ from previous literature on active sustainable funds and are more in line with relevant findings in the managerial skill literature from Berk & van Binsbergen (2015).
Over the last decades, there has been a significant shift from active to passive fund management due to the rise of index funds and a general tendency of funds to mimic the holdings of benchmark indices more closely (Cremers & Petajisto, 2009). Furthermore, Fama
& French (2010) show that fund managers fail to deliver a net return to their investors. The combination of on one hand ESG investing, which possibly can improve risk-adjusted returns and has increased in popularity, and the shift from active to passive fund management, leads to the query of whether ESG funds deliver abnormal returns to their investors. If actively managed ESG funds can deliver a positive return adjusted for risk, this likely indicates that either the securities in these portfolios are mispriced, which makes the fund managers who can find these securities skilled, or the models that attempt to evaluate the performance fail to capture ESG related risk factors. Jensen (1968) has shown in an early paper that active fund managers do not possess this superior stock-picking ability, as their performance does not differ from what we would expect from random chance.
All in all, the research question is whether ESG fund managers are skilled. To observe this, there must be an appropriate performance measurement. According to some researchers, such as Fama & French (2010), the net alpha measures skills amongst fund managers.
Therefore, in this research, net alphas will be computed concerning benchmarks of four random ETFs following Moraes et al. (2021), the four-factor model, and the six-factor model following Fama & French (2010). By conducting the False Discovery Rate following Benjamini &
Hochberg (1995), similarly used in Moraes et al. (2021), the (un)skilled funds are controlled for (un)luck. Controlling for luck is necessary because of random fluctuations that can take place in the underlying securities of the portfolio and Jensen (1968) has shown that for a fund manager to possess predictive stock-picking abilities, their performance should be better than
what we expect from random fluctuations. On the other hand, Berk et al. (2020) argued that the net alpha is not a relevant indicator of skills among fund managers, this is because investors typically bid up the price of funds that deliver positive alphas, driving the net alpha away.
According to the researchers, size indicators, such as the value-added, are more precise measurements. Therefore, in this research, the value-added following Berk & van Binsbergen (2015), controlled for luck using a bootstrap simulation following Elyasiani et al. (2018), will be conducted.
At first, the results of the total sample show that both ESG and non-ESG fund managers are not skilled, because 0.097% of all funds in the sample deliver a positive net alpha when controlled for luck. Furthermore, out of the 6,602 ESG fund managers in the sample, only 0.14% deliver a positive net alpha when controlled for luck. This indicates that out of 6,602 ESG fund managers, only 9 of them are considered to be skilled. However, when observing the net alphas over time, there is evidence of skills among ESG fund managers, because, during the period 2000-2003, one out of every three ESG managers was skilled when controlled for luck. However, these results are based on 147 ESG funds, which questions the econometric validity due to a lack of observations. Furthermore, the bootstrap simulation shows that there is enough statistical evidence to conclude that the average ESG fund adds a negative value to the market of -13 million gross and -38 million net per month, which makes the average ESG fund manager unskilled. Nevertheless, there is evidence that the top-performing ESG managers add positive value to the market. Lastly, this research consists of the literature review, followed by the data description, empirical framework, results, conclusion, discussion, and future work.
2. Literature Review
This section covers the theoretical framework. At first, an introduction of the topic will be given with certain key relevant findings. Second, the relation between ESG screening and financial performance is discussed by investigating what prior research has found. Third, the performance of active ESG funds will be outlined. At last, the optimal performance measurement and the difference between skill and luck is discussed.
2.1. Intro to ESG investing
Money keeps flowing into ESG while the S&P 500 falls, as beyond $1.2 billion went into ESG- focused ETFs in April 2022 while the S&P 500 dropped 3.8% following its worst decline since 1970 (Bloomberg, 2022).
ESG assets under management have grown consistently over the last decades by its increase in market share and introduction of new asset classes such as responsible bonds,
‘green’ real estate investments, and venture capital (Vandekerckhove et al., 2011). Global sustainable investment reached 30.7 trillion USD in 2018 in Europe, the United States, Japan, Canada, and Australia (Dai, 2021). According to Andonov et al. (2021), ESG preferences and mandates explain 25 to 40 percent of the higher number of infrastructure investments made by public investors. In addition, the number of ESG-related indices has grown rapidly over time (Dai, 2021). Nowadays, one in three US dollars invested is being managed according to ESG principles (Berg et al., 2021). Przychodzen et al. (2016) investigated the motives behind the incorporation of ESG investing strategies by conducting a questionnaire among bank-affiliated mutual fund managers from the US, Canada, and Europe and found that fund managers are driven by herding behavior and risk mitigation rather than value creation. Jiang & Verardo (2018) found a negative relation between herding behavior and skills in the mutual fund industry, antiherding funds have superior investment strategies and are more skilled than herding funds. This indicates that if the average ESG fund manager is driven by herding behavior, they cannot be considered skilled. Nevertheless, investment based on ESG scores has grown in international capital markets over the past decades. A survey by the Forum for Sustainable and Responsible Investment showed that 80% of fund managers were pressured by clients to address ESG issues. In 2020, 38% of advisers recommended ESG funds to clients.
Furthermore, 40% of advisers said that clients have asked them about ESG funds in the last 6 months of 2020 (Financial Planning Association, 2020).
However, there is an ongoing concern in the literature regarding data rewriting of ESG scores as several rating providers, such as, MSCI, Sustainanalytics, and Refinitiv compete so that their scores are useful for data users developing ESG-related investing strategies. Data rewriting changes the results of predictive regressions relating ESG ratings to future stock returns and leads to biased ESG performance of firms (Berg et al., 2021). Furthermore, some asset managers portray their mutual funds as socially responsible for commercial reasons which lead to asymmetry in the socially responsible investment market. This has been found by Candelon et al. (2021), who investigated information asymmetry in the SRI market by looking for mismatches between a manager’s ESG signal and a fund’s ESG rating using a panel of equity funds. All in all, evidence of ESG-washing has been found because the certification of a fund is not necessarily linked to the investment strategy of the manager. The aforementioned leads to uncertainty since investors cannot be sure whether a certified socially responsible fund also incorporates a socially responsible investment strategy in practice.
The growth of ESG investing has also drawn the attention of policymakers leading to more influence on a regulatory scale, for instance, the Pension Disclosure Regulation 2000 in the UK, the discussion in Belgium on making responsible investment a protected label (Vandekerckhove et al., 2011), and the Energy and Environmental Transition law in France, which requires all institutional investors to publicly report how they address socially responsible investing issues. Similar initiatives are currently also discussed at the European level (Alessandrini & Jondeau, 2019), the Sustainable Finance Disclosure Regulation is applicable since 10 March 2021 on a European level to improve transparency in the market for sustainable investment products and covers a broad range of ESG metrics (SFDR, 2022).
Nowadays, many politicians and investors require financial institutions to behave responsibly and address how they handle social issues. Besides, pension funds are more encouraged to apply a responsible investment strategy ever since the financial crisis of 2008 (Vandekerckhove et al., 2011). Whether a socially responsible fund manager incorporates a socially responsible investing strategy is what an investor must take for granted. At last, the growth of ESG investing leads to the question of whether this is due to its ability to improve financial performance, the next section will give an overview of what relation researchers have found between ESG screening and financial performance.
2.2 The effect of ESG investing
There exists a wide range of literature on the effect of ESG screening on financial performance.
La Torre et al. (2020) investigated the effect of ESG investing on stock returns by conducting panel regressions consisting of companies in the Eurostoxx50 index. The analysis showed that the linear correlation between ESG indices and stock returns is very weak and the volatility of returns is still to be found in other factors. The researchers only found a positive significant effect between ESG investing and stock returns in 7 out of 46 companies, which are mainly from the energy and utility sector. Therefore, ESG strategies do not positively impact stock returns in all sectors. Berk & van Binsbergen (2021) evaluated the quantitative impact of socially conscious investing and found that divestment does not have a large effect on the long- term cost of capital of targeted firms. Furthermore, socially conscious investors can have a greater impact by purchasing targetted firms rather than short-selling asocial stocks.
On the other hand, Pastor et al. (2021) show that green assets’ high recent returns are reflecting preferences that drive up green asset prices. By using environmental ratings from MSCI and conducting Sharpe ratios, the researchers show that green stocks strongly
outperform brown stocks in recent years. However, the outperformance likely reflects an unanticipated increase in environmental concerns. By running a panel regression and interacting a stock’s greenness with climate concern shocks, the researchers show that green stocks typically outperform brown stocks when environmental concerns arise. Dai (2021) studied whether ESG investing beats the market in China. More precisely, the researcher conducted Monte Carlo simulations and showed that ESG investing beats the market and has diversification benefits. The risk metrics used are the Sharpe, Sortino, and Omega ratios. All three ratios indicate that ESG equity indices have the highest risk-adjusted returns relative to their benchmark. Furthermore, the researcher conducted spanning tests and found evidence that ESG indices cannot be replicated by their conventional benchmarks. Therefore, investing in ESG indices in addition to conventional benchmarks can improve portfolio diversification.
Eccles et al. (2014) investigated a sample of 180 companies of which 90 were highly sustainable. They found that these high sustainability companies tend to outperform the other 90 low sustainable companies in the sample in the stock market. This result was more likely to be found in sectors where companies compete based on brands and human capital and where products of firms depend on extracting natural resources. Dimson et al. (2015) analyze a database of corporate social responsibility engagements with U.S. public companies from 1999 until 2009 and find that positive engagement with ESG principles leads to abnormal returns and increases the market value of engaged companies. On average, ESG activities increase positive size-adjusted abnormal return by 7.1 percent over the year after a successful engagement. Themes on corporate governance and climate change are most pronounced for positive abnormal returns.
In the literature, there is a persistent stream that implies that ESG investing can achieve additional returns, however, the empirical support for this idea is erratically based on simulated performance and not on actual performance. Future research should challenge this contradictory claim and clarify whether the simulated performance is due to actual skill or data mining (Daugaard, 2020). The actual skill of portfolio managers can be evaluated in the active asset management field.
2.3 Active ESG funds
In recent times, there has been a significant shift from active to passive fund management due to the rise of index funds and a general tendency of active funds to mimic the holdings of benchmark indices more closely (Cremers & Petajisto, 2009). Even in the early research of
Jensen (1968), active fund managers have shown to be unable to perform better than what we would expect from mere random chance. If an active fund manager is incapable of outperforming the benchmark it may lose the interest of investors as those investors can align their position with passive funds where the costs are significantly less. Thus, one of the goals of an active fund manager is to outperform their benchmark and it can do so by taking positions that are different from the benchmark. This can be done in two general ways, either with stock selection or factor timing (Cremers & Petajisto, 2009). This research will mainly focus on stock selection. If an ESG fund manager can find mispriced securities, they can be defined as skilled as they possess superior stock picking abilities which improve their fund returns. In the research of Jensen (1968) this phenomenon is defined as a manager’s predictive ability. An example of a stock selection strategy is the screening strategy. First is negative screening in which fund managers exclude companies that have a high-risk factor to ESG, such as companies that operate in the oil and tobacco industry (Madhavan et al., 2021). Funds that adapted negative screening in their portfolio selection received net inflows during the COVID-19 crisis, whereas funds that did not apply exclusions experienced outflows (Pástor & Vorsatz, 2020). Second, positive screening involves constructing portfolios based on companies that meet certain ESG criteria. At last, the integrated strategy where ESG factors are included alongside more traditional investing strategies, such as value investing. Vanguard describes the strategy of its Global ESG Select fund as an actively managed fund that seeks to invest in global mid- and large-capitalization companies with high financial productivity and leading ESG practices (Vanguard, 2022).
ESG investing has become mainstream and the largest asset managers have taken steps toward integrating ESG factors in their investment processes. Madhavan et al. (2021) investigated the relationship between fund’s ESG scores, factor loadings, and alphas of 1,312 active US equity mutual funds and found that funds with high ‘E’ scores have strong exposure to quality and momentum factors. However, the researchers did not find a link between fund alphas and active returns to ESG components unrelated to stylized factors. This means that the link between high ESG ratings and alpha is only through the correlation with factor components. Since the analysis covers only a small period, namely from 2014 to 2019, the researchers acknowledged that for larger or future periods, funds with high ESG scores may deliver positive alpha as the flows increase. Pástor & Vorsatz (2020) analyzed the performance and flows of U.S. active equity mutual funds during the COVID-19 crisis. They found that most active funds underperform passive benchmarks, contradicting the hypothesis that active funds outperform in recessions. The underperformance is large when measured against the S&P
500 index, but it is also observed relative to style benchmarks. However, sustainable funds perform better in periods of market crisis. This indicates that investors shift to ESG investing when there is market turmoil. Furthermore, Nofsinger & Varma (2014) find that socially responsible mutual funds tend to outperform during market crises. Their sample consists of 240 U.S. equity mutual funds and covers two crisis periods, namely 2001 and 2007–2009.
However, the researchers show that both SRI funds and conventional funds deliver negative risk-adjusted alphas during the period starting in 2000 until 2012. At last, Vandekerckhove et al. (2011) found significantly higher returns in socially responsible funds as opposed to the MSCI world index by analyzing financial returns of 151 SRI funds during phases of market turmoil between December 2001 and June 2009. The researchers conducted correlations, statistical tests, and multivariate regression models using, among other things, sustainability ratings. However, the research of Vandekerckhove et al. (2011) possibly lacks precision as the outperformance of SRI funds may be subject to luck which questions the reliability of the results.
ESG investing has become popular and the literature has shown that sustainable funds outperform in crisis periods (Pástor & Vorsatz, 2020) (Nofsinger & Varma, 2014) and can possibly outperform their passive benchmark (Vandekerckhove et al., 2011). On the contrary, Madhavan et al. (2021) showed that higher alphas are not necessarily related to ESG ratings.
Because the literature has previously shown that active fund managers fail to outperform their benchmark and are to a large extent not skilled (Jensen, 1968) (Fama & French, 2010), this research proposes to evaluate the performance of ESG mutual funds and investigate whether their managers are skilled. This research is different from Pástor & Vorsatz (2020), Nofsinger
& Varma (2014), Vandekerckhove et al. (2011), and Madhavan et al. (2021) because it covers a larger timeframe starting from 2000-01-01 until 2022-05-01, a larger sample of 12,461 funds, and conducts various methods to evaluate skills which will be outlined in the next section. The choice behind a larger timeframe is also relevant because in recent times ESG investing has become increasingly popular.
2.4 Performance measurements and controlling for luck
There seems to be a division in the literature on how the performance of active fund managers should be evaluated. The methodology of Fama & French (2010), who evaluate the skills of active fund managers by conducting net alphas, differs from Berk et al. (2020), who argue that the net alpha is not the appropriate measurement for managerial skill. To establish the
appropriate performance measurement which will be conducted in this research, the findings of different researchers using these methods will be outlined.
Lückoff (2011) argues that for a long-only investment product that holds equities and bonds and does not use derivatives or leverage, the performance can be evaluated by the alpha.
Mutual funds do not use complex investment products and rarely go short (Cremers & Petajisto, 2009), which indicates that the performance of mutual funds can be evaluated by the alpha.
Fama & French (2010) measure skill by evaluating net alphas and show by using a bootstrap technique that only a few mutual funds have enough skills to cover the trading costs, with some funds performing extremely well and some funds doing very poorly. Barras et al. (2010) examine the performance of 2,076 actively managed U.S. open-end mutual funds between 1975 and 2006, the monthly net returns adjusted for trading costs show that only 0.6% of fund managers are skilled. By measuring the alpha against the factor model benchmark, the researchers show that roughly 13 actively managed funds out of the total sample have managers with some stock-picking ability. When evaluated over a shorter period of 5 years, the results show that 2.4% of fund managers are skilled, delivering a positive net alpha. This phenomenon is moderately due to flows that followed the past performance. Newer and smaller fund managers have some skill before investors learn about their abilities while old funds are typically distributed among the non and zero-skilled funds. Overall, by controlling for false discovery, the researchers show that the proportion of unskilled fund managers has increased over the last 20 years. Moraes et al. (2021) investigated cross-validation-based algorithm performances by conducting gross and net alphas, which led to the finding that, 95% of the active mutual funds have neither the necessary skills nor the ability to generate value for their investors, based on an assessed set of five ETFs.
On the other hand, Berk et al. (2020) argue that the alpha is not an appropriate skill measurement. The skill of fund managers can not be evaluated by analyzing alphas because they fail to achieve superior performance in the long run. This argument is also used by Barras et al. (2010) who do not prefer to measure the performance of funds with the long-term net alpha because flows compete away any alpha surplus. Thus, the return the fund makes is not informative about the skill because when a stock has a positive risk-adjusted return, investors find out about this and bid up the price until the risk-adjusted return is zero. This means that negative net alphas indicate that investors are suboptimally allocating capital and are willing to accept a negative-NPV investment and positive net alpha indicates that capital markets are not competitive. Barras et al. (2010) argues, however, that there exists evidence of funds with superior short-term alphas before investors become fully aware of such outperformers due to
search constraints. Berk & van Binsbergen (2015) shows that the skill of the manager can be measured by the amount of money extracted from financial markets, which can be assessed by taking the product of the size of the fund and the gross alpha, also known as the value-added.
Their research evaluates skills in mutual fund performance over the period 1962 until 2011.
They find that the average fund persistently adds 3.2 million US dollars to the market per year.
According to the researchers, their measure of managerial skill is more accurate than what has previously been used in the literature. At last, Kooli & Stetsyuk (2021) evaluate whether hedge fund managers are skilled by analyzing the value-added following Berk & van Binsbergen (2015). By using a bootstrap technique to control for luck, the researchers find that hedge fund managers are skilled. When evaluated against the S&P 500 and eight Vanguard indices, hedge fund managers add, on average, 3.24 million US dollars per year. However, hedge funds have different benchmarks and use different trading strategies than mutual funds, as they can make use of short selling and derivatives. This makes the comparison between hedge funds and mutual funds more challenging.
Because both viewpoints are valid but lack superior persuasion, the research question,
‘Are managers of ESG mutual funds skilled?’, will be answered by evaluating both the net alpha and value-added. A fund is considered to be skilled when it can generate a positive risk- adjusted return net of trading costs, fees, and other expenses (Barras et al., 2010) or adds a positive value to the market (Berk et al., 2020), in other words, the dollar value the fund extracts from the markets (Berk & van Binsbergen, 2015). However, Berk & van Binsbergen (2015) argue that there is no evidence that this value results from human capital alone, which weakens the validity of the results. To adjust the returns for risk, the appropriate benchmarks should be used. This is important because if the model fails to capture certain risks related to the fund, the results may be biased. After all, outperforming capacities of fund managers may be related to specific risk factors rather than their stock-picking abilities. The benchmarks used in this research stem from prior literature, namely the four and six-factor model following Fama &
French (2010), similarly conducted in Barras et al. (2010), and a combination of four random ETFs following Moraes et al. (2021). The importance of passive ETFs inclusion is, among others, motivated by Cremers & Petajisto (2009) who argue that mutual funds are required to declare a benchmark index and the purpose of an active manager is to beat that benchmark.
Furthermore, ETFs are easily tradable which puts them in direct competition with active managers in terms of attracting assets.
At last, a common problem in active funds’ performance measurements is to distinguish between skill and luck as fund returns are subject to random fluctuations in the underlying
securities. For a fund manager to be skilled, their performance should positively differ from what we would expect from random chance (Jensen, 1968). This leads to several techniques being adapted in the literature, such as Bayesian estimations, bootstrapping, and using daily returns instead of monthly (Lückoff, 2011). To control for luck, the net alphas will be corrected for false rejected hypotheses, similarly conducted in Moraes et al. (2021) and Barras et al.
(2010), and the value-added will be corrected for luck using a bootstrap simulation, similarly conducted in Kooli & Stetsyuk (2021).
3. Data & Descriptive Statistics
This section covers the data used to conduct the research and displays summary statistics of the final dataset. At first, the source of the data utilized will be outlined. Second, the summary statistics display key features of the data which will be discussed.
3.1 Data
The data is gathered from Refinitiv Eikon Datastream, according to Refiniv Eikon, ESG mutual funds can be identified in their database by filtering for ‘yes’ on Alternative Energy or Ethical or Green or Water (Refinitiv, 2021). Commodities, bonds, money market funds, real estate, and alternatives are removed, such that the data only contains equity and mixed assets. The funds are geographically globally focused, such that the utmost markets are captured. Funds are either liquidated or active, this makes sure the analysis lacks survivorship bias. Variables included are ESG scores, historical expense ratios, historical TNAs denominated in US dollars, and historical NAVs denominated in US dollars. For the ESG funds, there are 122 funds with no reported expense ratio, this is replaced with the average expense ratio of all ESG funds. This method is also used in Moraes et al. (2021). Similarly for non-ESG funds, there are 361 funds with missing expense ratios which are replaced by the average expense ratio of all non-ESG funds. Historical data is monthly, and the starting period is 2000-01-01 until 2022-05-01. The reason behind starting in 2000 is due to the availability of ETFs (Pástor & Vorsatz, 2020).
Funds that have less than 30 months of NAV data are removed, which is equivalent to Moraes et al. (2021). The final dataset consists of 6,602 ESG mutual funds and 5,859 non-ESG mutual funds.
Monthly historical risk factors are captured from the Kenneth R. French Library. ETF data is gathered from FactSet and Refinitiv Eikon, ETFs that have less than 15 years of historical price data are removed from the sample, to reduce the risk of deleting unnecessary
mutual fund NAV data when merging. By doing so, the performance evaluation against ETFs may lead to survivorship bias because the benchmark is only focused on the survivors and not the failures. However, if ETFs that only have a few observations are included and merged with a fund, then the eventual subsample will be small and the regression results which are based on this subsample will lack econometric validity. Furthermore, the selection procedure of the benchmark is not different from other researchers who, for instance, only use the S&P 500 or Vanguard ETFs as a benchmark, because these ETFs are also considered to be survivors. The final dataset contains 488 ETFs that are globally focused. The reason for ETF inclusion is motivated by Cremers & Petajisto (2009), and Moraes (2021), who verified that ETFs can be used as benchmarks to assess active fund performance, and Pástor & Vorsatz (2020), who used ETFs as benchmarks to evaluate active fund performance.
3.2 Descriptive Statistics
The table below shows the descriptive statistics of the final dataset, variables included are the annual average excess return, annual Sharpe ratio, average size, average expense ratio, and length of the data.
The average ESG fund reports a higher annual excess return than the average non-ESG fund, respectively of 3.306% versus 0.037%. Remarkably, Nofsinger & Varma (2014) report an average return of 2.6% for conventional funds, and 2.56% for SRI funds, which is not in line with these results, because they induce that the conventional funds have a higher return than the SRI funds. When observing the outliers for ESG funds, two remarkable values can be noted. The minimum reported average annual excess return is -106.588%, which is exceptional.
The fund that reported this value is Neptune Green Planet A Acc NAV, which likely has been liquidated in August 2013 after it stopped reporting NAV information. Furthermore, their NAV dropped from 74.11 US dollars in July 2009 to 0.813 US dollars in September 2009. The fund has been trading under one US dollar ever since this large drop until the liquidation. The maximum average annual excess return of ESG funds is 36.589%, the fund that reports this value is Wealth Invest Akl Seb Globalt Aktieindeks Dkk I, which has risen significantly over time. Starting from a net asset value of 105.70 US dollars in 2014 to its peak in 2020 of 1110.28 US dollars. Likewise, for non-ESG funds, Berkshire Technology Fund reports an average annual excess return of -104.686%. This fund likely has been liquidated in 2002, after it dropped from a net asset value of 13.59 US dollars in 2000 to 0.8 US dollars in 2002. At last, Jyske Invest Japanese Equities CL, reports an average annual excess return of 73.064%. This
fund has risen significantly, starting from a net asset value of 11.892 US dollars in June 2007 to its peak in May 2015 of 9410.77 US dollars. This indicates nearly a 90-fold price increase.
According to Refinitiv Eikon, the fund stopped reporting NAV information in March 2016. All in all, funds that report an average excess return below -100% likely do so because their NAV has dropped significantly over time, which indicates large negative returns, and these returns become even more extreme when adjusted for the risk-free rate.
The annual Sharpe ratio of the average ESG fund is 0.158, this is higher than the annual Sharpe ratio of the average non-ESG fund, which is 0.017. This indicates that the average ESG fund manager delivers a higher marginal return per unit of risk than the average non-ESG fund manager. Moreover, the 95th percentile shows that the top ESG fund managers perform well since they report an average annual Sharpe ratio of 0.548. This is higher than the 95th percentile of non-ESG funds, which is 0.407, and all the discussed benchmarks. For instance, the ETFs report an average annual Sharpe ratio of 0.446 in the 95th percentile, and CMA which is the best performing risk factor, reports an average annual Sharpe ratio of 0.507. This result is in line with Dai (2021), who showed that ESG investing leads to higher Sharpe ratios. However, these differences have not been statistically tested because it is not the purpose of this research to evaluate the significance of Sharpe ratios.
ESG and non-ESG funds are comparable in size, as the average total net assets of ESG funds is 229 million, while this is 700 million for non-ESG funds. The difference is likely because of non-ESG funds in the 95th decile, which report an average TNA of 2.732 billion US dollars as opposed to the 95th decile of ESG funds, which report an average TNA of 949 million.
The expectation is that the difference in TNA between non-ESG funds and ESG funds does not have a large effect on the results, as for most of the distribution, the average TNA is relatively close to each other. Nevertheless, the dataset contains funds where the AUM is smaller than 5 million US dollars which may lead to incubation bias (Moraes et al., 2021). These funds, which start privately, may have upward-biased returns during the incubation phase.
Furthermore, the average ESG fund is more expensive because they charge a higher fee to its investors, respectively of 1.448%, while this is 1.134% for the average non-ESG fund.
Similarly, Geczy et al. (2006) found that the average expense ratio of SRI funds, which is 1.33%, is higher than that of non-SRI funds, which is 1.10%. Perhaps, since ESG funds possibly deliver higher excess returns than non-ESG funds, they charge a larger fee to their investors. At last, the time series length shows that there is sufficient data to analyze because the average number of NAV data is 142 months for ESG funds and 161 months for non-ESG funds.
Table 1: Descriptive statistics
This table presents the descriptive statistics of ESG funds (A), non-ESG funds (B), ETFs (C), and risk factors (D). Av. Excess returns refers to the average annual return net of the risk-free rate, Av. S.R. refers to the average annual Sharpe ratio, Av. TNA measures the average total net assets, and Av. Expense Ratio measures the average expense ratio. The total sample consists of 12,461 funds with at least 30 months of data and 488 ETFs from 2000-01-01 until 2022-05-01.
Description Mean Std. Min P05 Median P95 Max
Panel A: ESG funds
Av. Excess return (in %) 3.306 6.690 -106.588 -3.632 2.976 12.722 36.589 Av. S.R. (ex-post) 0.158 0.219 -1.334 -0.148 0.130 0.548 1.125 Av. TNA (Bi $) 0.229 0.707 0.000 0.001 0.064 0.949 35.960 Av. Expense Ratio 1.488 0.678 0.000 0.366 1.509 2.479 8.070
Time Series Length 142 85 30 34 125 265 265
Total Number of Funds 6602 Panel B: Non-ESG funds
Av. Excess return (in %) 0.037 7.137 -104.686 -11.388 0.747 7.921 73.064 Av. S.R. (ex-post) 0.017 0.271 -3.686 -0.405 0.034 0.407 1.363 Av. TNA (Bi $) 0.700 3.363 0.000 0.001 0.055 2.732 91.433 Av. Expense Ratio 1.134 0.878 0.000 0.106 1.100 2.206 30.841
Time Series Length 161 87 30 38 155 266 266
Total Number of Funds 5859 Panel C: ETFs
Av. Excess return (in %) 3.894 4.350 -17.658 -3.282 3.951 10.435 21.181 Av. S.R. (ex-post) 0.145 0.164 -0.319 -0.087 0.117 0.446 0.662 Av. TNA (Bi $) 2.479 6.259 0.016 0.044 0.521 11.548 67.161
Time Series Length 208 24 178 179 202 249 265
Total Number of ETFs 488
Panel D: Risk Factors
Description Av. Excess Return (in %) Av. S.R. (ex-post)
Mkt 6.942 0.444
SMB 3.114 0.290
HML 2.423 0.205
MOM 4.374 0.306
CMA 5.200 0.507
RMW 3.800 0.529
4. Methodology
The methodology covers the empirical framework of the research. At first, the framework behind the analysis of the net alphas will be outlined, second, the value-added will be discussed.
At last, several procedures will be reviewed to control the results for luck.
4.1 Net alphas
To assess whether ESG fund managers are skilled, the alphas are computed concerning factor models and combinations of four random ETFs. Pástor & Vorsatz (2020) found that sustainable funds performed better during the COVID-19 crisis. Their analysis included alphas evaluated against the factor model benchmark. Furthermore, Fama & French (2010) measure skills by evaluating fund performance against risk factors. In this research, the alphas will be measured by conducting time-series regressions. The four-factor and six-factor models are given below, where the four-factor model covers the equation excluding 𝛽5 and 𝛽6.
𝑅𝑖,𝑡 = 𝛼𝑖 + 𝛽1(𝑀𝑘𝑡 − 𝑅𝑓) + 𝛽2𝑆𝑀𝐵 + 𝛽3𝐻𝑀𝐿 + 𝛽4𝑀𝑂𝑀 + 𝛽5𝐶𝑀𝐴 + 𝛽6𝑅𝑀𝑊 + 𝜀𝑖,𝑡 The 𝛼𝑖 is the intercept of the time series regression, the gross 𝛼𝑖 is computed by running a time- series regression of the excess return of fund i on the excess return of the benchmark, whereas the net 𝛼𝑖 is computed by running a time-series regression of the excess net return of fund i on the excess return of the benchmark. The net return in period t is calculated by subtracting 1/12th of the expense ratio of t-1 from the gross return (Fama & French, 2010). The time-series regressions in this research are conducted by using the Statsmodels package in Python.
Furthermore, according to Moraes et al. (2021), a skilled fund is defined as any fund capable of delivering returns after it is controlled by a set of selected ETF excess returns.
The researchers evaluated the active fund’s return against a set of random ETFs using a P vector dimension, which will also be conducted in this research. The equation is given below.
𝑅𝑖,𝑡 = 𝛼𝑖+ 𝛽𝑖𝐸𝑇𝐹 (𝑝)𝑏,𝑡+ 𝜀𝑖,𝑡
The gross 𝛼𝑖 is the intercept of a time series regression of the excess return of fund i on the excess returns of a vector of four random ETFs, and the net 𝛼𝑖 is the intercept of a time series regression of fund i’s excess net return on the excess returns of a vector of four random ETFs.
The choice of four ETFs is because Moraes et al. (2021) have found that the optimal vector should not stray away from three to five ETFs. Furthermore, the researchers argue that random selection is the most straightforward technique applied to establish sets of ETFs and it provides an average measure of the outperforming capacity of a fund on a random group of ETFs.
However, according to the researchers, the model should be repeated multiple times to obtain
a precise estimate. Therefore, in this research, the procedure is repeated 500 times, and the estimates are computed by taking the subsample average, the equations are given below.
𝛼̂ = 1
𝑁∑𝑛=1𝑁 𝛼(𝑝),(𝑛),𝑖 𝑅̂2 = 1
𝑁∑𝑛=1𝑁 𝑅2(𝑝),(𝑛),𝑖 𝑇̂ = 1
𝑁∑𝑛=1𝑁 𝑇(𝑝),(𝑛),𝑖
From the aforementioned analysis, three skill groups can be selected. Namely, unskilled funds, zero skilled funds, and skilled funds, this is similarly defined in prior research.
These groups respectively have an alpha of significantly smaller than zero, equal to zero, and significantly greater than zero. Based on prior literature, the expectancy is that ESG fund managers are not skilled. This leads to the following hypothesis:
𝐻0: 𝛼 ≤ 0 𝐻1: 𝛼 > 0
The performance of a fund manager cannot always be related to skills, sometimes a fund is lucky enough to obtain a positive result or unlucky and obtains a negative result due to random fluctuations of the securities held in the portfolio. To address the problem of lucky and unlucky funds, the groups are controlled for false rejected hypotheses, similarly used in Barras et al. (2010) and Moraes et al. (2021). The false rejected hypotheses can be conducted using the False Discovery Rate which is based on Benjamini & Hochberg (1995). Their approach is the desirable control against errors originating from multiplicity and will be measured using the Statsmodels Multitest package in Python. This test also poses the benefits of not requiring prior assumptions about alpha distributions and has been widely used in active fund literature.
4.2 Value-added
Berk & van Binsbergen (2015) and Berk et al. (2020) unfolded the debate on assessing active managerial skills. According to the researchers, the net alpha is not an appropriate measure of skill, since investors typically find funds with positive risk-adjusted returns and bid up the price until the risk-adjusted return is zero. Therefore, the return is not informative about the quality of the fund, but the value-added is. In this research, the method of Kooli & Stetsyuk (2021), who evaluated the skills of hedge fund managers, will be used. Following Berk & van Binsbergen (2015) the below equation will be conducted to compute the value-added.
𝑆𝑖 = 1
𝑁∑𝑛=1𝑁 𝑞𝑖,𝑡−1(𝑅𝑖,𝑡− 𝑅𝑏,𝑖𝑡)
Where 𝑞𝑖,𝑡−1 measures the size, which is the total net assets, of fund i in period t-1. 𝑅𝑖,𝑡 is the excess return of fund i in period t and 𝑅𝑏,𝑡 is the sum of the excess returns j = 1, …., N of the benchmark n in period t multiplied by the estimated slopes from the time-series regression of fund i excess return on the excess returns of the benchmark. The equation of the benchmark return is given below:
𝑅𝑏,𝑖𝑡 = ∑𝑗=1𝑁 𝛽𝑖𝑗𝑅𝑖𝑗
Since both Berk et al. (2020) and Kooli & Stetsyuk (2021) argue that Vanguard ETFs are the appropriate benchmark for the measurement of managerial skill, because they are popular and provide the least costly method of diversification, this research will conduct the value-added using the factor models and an equally weighted portfolio of eight Vanguard ETFs. Thus, whereas a sample of four random ETFs is used to conduct net alphas, the value- added is calculated using a portfolio of eight Vanguard ETFs as a benchmark. Another reason for Vanguard ETFs inclusion instead of random ETFs is because of the divergent procedure in which the value-added is calculated. Furthermore, the researchers who outlined the method of using random ETFs have only done this to compute alphas. The Vanguard portfolio selection is similar to Kooli & Stetsyuk (2021) who selected a sample of eight Vanguard ETFs that are globally focused and either small, mid, or large blend following Berk et al. (2020).
Furthermore, this research proposes a bootstrap analysis of the value-added to distinguish between skill and luck which is equivalent to Kooli & Stetsyuk (2021) following Elyasiani (2018). The reason for the bootstrap simulation is because applying the false discovery rate is complicated since it is difficult to generate p-values when computing the value-added and bootstrap simulations seem to be frequently used in the literature to control for luck, such as Fama & French (2010) who controlled the net alphas for luck using bootstrap analysis. The below equation of the bootstrapped value-added stems from Kooli & Stetsyuk (2021) and will be conducted in this research.
𝑆𝑖 = 1
𝑁∑𝑛=1𝑁 𝑞𝑖,𝑡−1(𝛼̂𝑖 + 𝜀̂ ) 𝑖,𝑡
Where the 𝛼̂𝑖 and 𝜀̂𝑖,𝑡 is the intercept and residual captured from a time-series regression of the excess fund return on the excess return of the benchmark for funds i = 0, …, N. The null hypothesis follows that the value-added is equal to zero, this can be tested by running a time series regression for each fund to capture the residuals 𝜀̂𝑖,𝑡. To ensure that the expected value- added is equal to zero, the estimated abnormal return is subtracted from the excess fund returns.
This method is similarly used in the bootstrap analysis of the 𝛼 of Fama & French (2010), where the 𝛼̂𝑖 is set to zero by subtracting the simulated 𝛼𝑖 from the excess return of fund i.
After the time series regression model is estimated, 1000 bootstrapped samples of the value- added are computed. The 𝜀̂𝑖,𝑡 will be replaced by 𝜀̂𝑖,𝑡𝑏 which is randomly drawn with replacement from the sample 𝜀̂𝑖,𝑡, t = 0, …. , T. Because 𝜀̂𝑖,𝑡𝑏 is randomly drawn, this ensures that it is uncorrelated with the total net assets of fund i. Berk & van Binsbergen (2015) argue that the t-statistic of the value-added may be overstated because correlation in the value-added likely exists and the value-added distribution features excess kurtosis, therefore the sample may not be ensured to be t-distributed. Because in this bootstrap analysis the error term is uncorrelated with the TNA, it is more likely that the value-added is t-distributed, which ensures that the t-statistic is not overstated. In the end, for each fund and bootstrapped sample, the average bootstrapped value-added is formed which will be tested against the null hypothesis.
Since it is not possible to measure all funds in the population, the results of the sample in this research may not be representative. The bootstrapped sample gives reason about the population. By drawing test statistics of the bootstrap sample, it is possible to test the results of the sampled data against the resampled population data and hereby gain credibility.
5. Results
In this section, the performance of ESG mutual fund managers is evaluated by analyzing net alphas, corrected for false rejected hypotheses, and the value-added, controlled for luck by bootstrapping. Furthermore, a comparison is made with non-ESG funds to assess whether these fund managers perform better or worse.
5.1 Net Alphas
Performance in the asset management field is evaluated by measuring the manager’s skill.
According to Barras et al. (2010), Moraes et al. (2021), and Fama & French (2010), skill can be measured with the net alpha. As a result, Figure 1 shows the net alphas of all 12,461 mutual funds evaluated against the factor models and combinations of four random ETFs. If a fund reports a positive significant alpha, it can be defined as skilled. Most funds report an alpha that is below 0, or close to 0, with a few outliers. At first glance, this indicates that very few funds deliver a positive risk-adjusted net return to their investors. This result is in line with Fama &
French (2010) who state that for most actively managed funds, the alpha in net returns is negative. On the contrary, more positive alphas can be observed in the top side of figure 1,
Figure 1: Annual net alphas
This figure displays the annual net alphas of ESG funds (A) and non-ESG funds (B) measured against the three discussed benchmarks. Net alphas are displayed on the y-axis and funds are displayed on the x- axis. Net alphas are the intercepts captured from a time series regression of fund i excess net return on the excess return of the benchmark. Net alphas of the random ETFs are computed using the equation of Moraes et al. (2021). The four and six-factor models are based on Fama & French (2010). The net alphas are sorted by ticker in all plots.
which may indicate that ESG mutual funds provide superior risk-adjusted returns to their investors over non-ESG funds. This makes it likely that ESG fund managers have better stock- picking abilities than non-ESG fund managers which means that the securities in the portfolios of ESG fund managers are mispriced or the expense policy of ESG funds is less severe than for non-ESG funds. This means that the skill of the non-ESG fund manager is diminished by the expense policy of the fund, something he or she has little to no influence on. However, it is also possible that the models fail to capture certain risk factors related to ESG, which means that ESG fund managers are not necessarily skilled but compensated for taking on ESG risks.
Overall, the figure shows that very few funds deliver a superior alpha. The funds that do deliver a large positive alpha are notably small to mid-cap, such as Northern Trust UCITS FGR Fund - Pacific Custom ESG Equity Index and Wealth Invest AKL SEB Globalt Aktieindeks DKK I, or liquidated, such as Jyske Invest Japanese Equities. This is related to Pastor et al.
(2017) who found stronger support for a positive relationship between an active fund’s turnover and its subsequent risk-adjusted return for smaller funds and Barras et al. (2010) who argue that smaller funds may exhibit some skills before investors learn about their superior ability.
B A
Table 2 shows the statistics of the net alphas measured across different mutual funds.
The results show that the average ESG and non-ESG fund delivers a negative net alpha. Moraes et al. (2021), Pástor & Vorsatz (2020), and Berk et al, (2020) argue that passive index funds should be used when assessing manager’s skills since the factor portfolios are not investable opportunities and do not replicate transaction costs which make it a less reliable comparison.
However, even when measured against random ETFs, ESG and non-ESG funds, on average, fail to deliver a positive risk-adjusted net return to their investors. If the true alpha has a normal distribution with mean zero and standard deviation 𝜎, the table indicates that fewer than 16%
of ESG funds have an alpha greater than 6.622% per year. Fama & French (2010) report a 𝜎 of 1.25% and argue that this implies actively managed funds are not skilled.
However, ESG funds report a higher average net alpha, -3.566%, than non-ESG funds, -4.344%. This shows that, on average, ESG fund managers are more skilled than non-ESG fund managers. The average t-statistic of ESG funds is -0.973, this is less extreme than the average t-statistic of non-ESG funds, which is 1.074. This shows that, on average, there is a greater probability that the net alpha which is reported by ESG funds will not be rejected against the null hypothesis than for the average non-ESG fund. Furthermore, the adjusted R squared indicates that the variance of the net returns of the ESG funds is less predicted by the independent variables than for non-ESG funds. Perhaps, this is an indication that for ESG funds the model used fails to capture more factors that predict the variability of the dependent variable than for non-ESG funds.
Fama & French (2010) argue that risk factors are an appropriate performance measurement of fund managers. The four and six-factor models display a higher average net alpha for ESG funds relative to non-ESG funds. However, when performance is evaluated against factor models, both non-ESG and ESG funds are not skilled on average. The results of panels B and C are more negative than the results of panel A. When the skills of a fund manager are assessed against the factor models, there are on average greater negative values reported than for the measurement against random ETFs. Overall, the results are equivalent to Nofsinger
& Varma (2014) who report an annual Carhart four-factor alpha of -0.75% for socially responsible funds, and -0.86% for conventional funds during the period 2000 until 2012.
There is not only an interest in the average performance but also in the performance of the top fund managers. Panel A shows that the top 5% performing ESG fund managers deliver an annual net alpha of 3.335%. Panel B and C shows that even the top 5% of ESG fund managers deliver a negative alpha, respectively of -1.966% and -1.692%.
Table 2: Net annual alphas
The table below reports the descriptive statistics of the annual net alphas measured against the discussed benchmarks. Net alphas are the intercepts from time series regressions of funds excess net return on the excess return of the benchmark. For Panel A, the variables are estimated averages based on the model which was run 500 times for each fund following Moraes et al. (2021). The factor models of panel B and C are based on Fama & French (2010).
ESG Mutual Funds Non-ESG Mutual Funds
Net 𝛼̂ T-stat̂ Adj. 𝑅̂ Net 𝛼̂ 2 T-stat̂ Adj. 𝑅̂ 2 Panel A: Random ETFs
Mean -3.473 -0.973 0.670 -4.478 -1.074 0.688
Std. 6.682 1.154 0.164 7.659 0.916 0.153
Min -150.419 -6.099 0.004 -194.801 -11.533 0.010
P05 -10.747 -2.730 0.351 -12.994 -2.319 0.397
Median -3.041 -1.056 0.710 -3.479 -1.129 0.716
P95 3.335 1.022 0.867 1.455 0.366 0.884
Max 126.944 4.390 1.000 131.870 9.086 0.982
Panel B: Four-factor Model
Mean -8.372 -2.993 0.340 -9.526 -2.680 0.482
Std. 6.947 1.572 0.167 5.979 1.343 0.214
Min -175.734 -11.283 0.002 -85.586 -10.529 0.006
P05 -16.651 -5.645 0.073 -18.428 -5.134 0.158
Median -7.692 -2.934 0.336 -8.775 -2.538 0.456
P95 -1.966 -0.614 0.625 -2.996 -0.755 0.898
Max 131.114 3.677 0.930 109.365 2.154 0.995
Panel C: Six-factor Model
Mean -8.023 -2.956 0.351 -8.985 -2.585 0.495
Std. 7.099 1.641 0.167 6.187 1.482 0.215
Min -183.450 -12.739 0.005 -113.417 -12.001 0.015
P05 -16.111 -5.779 0.086 -18.137 -5.327 0.173
Median -7.341 -2.865 0.347 -8.226 -2.382 0.467
P95 -1.692 -0.516 0.637 -2.415 -0.530 0.909
Max 139.351 4.252 0.933 93.591 3.147 0.996
Figure 2: Distribution of t-statistics
This figure displays distributions of t-statistics of ESG Mutual Funds (A) and non-ESG Mutual Funds (B) measured across three skill groups. Namely, unskilled, zero skilled, and skilled funds. T-statistics are captured from a time series regression of fund i excess net returns on the excess return of the benchmark.
Skilled and unskilled funds report a significant alpha above and below zero, zero skilled funds report an unsignificant net alpha. The significance level is 5% and the skill cutoff is based on the critical t-value calculated using the Scipy.stats package in Python.
Figure 2 displays the distribution of the t-statistic across three skill groups, unskilled, zero-skilled, and skilled in the fund population. At the critical t-value, the probability of a fund being unlucky or lucky can be observed when a distribution crosses this cutoff. Performance measurements against ETFs show a greater proportion of (un)skilled funds being (un)lucky since these respective distributions have a greater surface that crosses the critical t-value.
Overall, this shows that ESG fund managers are more often to be considered lucky when they deliver a positive risk-adjusted net return to their investors and are more often considered to be unlucky whenever they deliver a negative risk-adjusted net return to their investors.
Notably, the middle figure on the downside of the graph, shows that for the four-factor model, there are close to 0 skilled funds observed. This indicates that out of more than 5,000 non-ESG funds, there is not one manager which reports a significant positive net alpha.
A
B
Table 3 displays the proportion of funds across the skill groups. In panel A, it can be observed that 1.51% of ESG fund managers are considered to be skilled when measured against ETFs. However, after this proportion is corrected for false rejected hypotheses, only 0.14% of ESG fund managers are considered to be skilled. To put this in perspective, out of the 100 fund managers which deliver a positive risk-adjusted net return to their investors, 91 fund managers delivered those returns because they were lucky in a market of random fluctuations.
Furthermore, 0.097% out of all mutual fund managers, both ESG and non-ESG, are considered to be skilled when controlled for luck. This result is relatively similar to Barras et al. (2010) who found that 0.6% of active U.S. equity fund managers are skilled when corrected for false rejected hypotheses.
Table 3: Unskilled, zero skill, and skilled groups
This table divides the empirical annual net alphas, which are the intercepts of time series regressions of funds excess net return on the excess return of the benchmark, of mutual funds in three groups.
Namely, unskilled, zero skill, and skilled. Unskilled funds report a significant net alpha lower than zero, zero skill funds report an unsignificant net alpha, and skilled funds report a significant net alpha above zero. Significance level is at 5%, and the critical t-value is conducted using the SciPy stats package in Python. The False Discovery Rate is based on Benjamini & Hochberg (1995) which is conducted using the Statsmodels Multitest package in Python.
ESG Mutual Funds Non-ESG Mutual Funds
Description Count of Count of Count of Count of
Panel A: Random ETFs
Unskilled funds 1290 7 784 25
Zero skill funds 5212 4938
Skilled funds 100 9 23 3
Panel B: 4-Factor model
Unskilled funds 4878 3380 3990 2001
Zero skill funds 1718 1720
Skilled funds 5 1 1 1
Panel C: 6-Factor model
Unskilled funds 4794 3325 3697 1882
Zero skill funds 1800 2006
Skilled funds 6 3 5 3
N 6602 5745
False Discovery Control No Yes No Yes
