The seminal work by Markowitz (1952) on portfolio selection theory laid the foundations for modern portfolio theory. It provided a framework for the creation of investment portfolios based on mean-variance analysis. The theories are based on the assumptions of rationality and risk preferences of investors. The notion of portfolio return being a function of market and firm-specific risk factors precipitated the creation of various asset pricing models.

The CAPM, as developed by Treynor (1961), Sharpe (1964), Lintner (1965), Mossin (1966) and Black et al., Jensen and Scholes (1972), was the first coherent framework for evaluating the effect of risk factors on the price of assets (Perold 2004). The CAPM modelled the return of an asset as a function of systematic risk and idiosyncratic risk. The former risk is the undiversifiable risk that is faced by all investors resulting from macro-economic and political factors. The latter type of risk, that is, idiosyncratic risk, is born out of firm-specific factors attached to that share.

The CAPM has enjoyed a great deal of popularity, both academically and in practice. It is frequently the only asset pricing model taught in most undergraduate and postgraduate finance classes. In addition to this, it is heavily used by portfolio managers and investors for a host of reasons, including estimating the cost of equity for listed securities, selection of investment projects, evaluating the performance of mutual funds and other portfolios (Fama & French 2004) and for the purposes of performing valuations. Despite the popularity of the CAPM, there is empirical evidence that challenges its ability to fully explain returns (Banz 1981; Bhandari 1988; Fama & French 1992).

One of the most prominent earlier studies to challenge the empirical validity of the CAPM was that of Basu (1977). The objective of the Basu (1977) research was to empirically research whether the investment performance of common stocks was related to their price-to-earnings ratios (hereafter P/E ratios). He found that when the stocks were sorted on P/E ratios, the stocks with higher P/E displayed higher expected future returns than what was predicted by the CAPM.

Banz (1981) also highlighted problems linked with the use of the CAPM. The study showed that the common stock of small firms, on average, outperformed that of larger firms; that is, smaller firms displayed higher average risk-adjusted returns, relative to larger firms – a concept commonly referred to as the ‘size effect’. Despite these findings, it was not until Fama and French (1992) that research challenging the role of beta and the CAPM gained traction. They examined the size and book-to-market anomalies that were discovered by prior literature and confirmed the empirical shortcomings of the CAPM. Their study found that the combination of the size and book-to-market equity absorbed the role of market beta in explaining cross-sectional returns.

Having noted the empirical contradictions of the CAPM in explaining expected returns (see Banz 1981; Basu 1977), Fama and French (1992, 1993) extended the CAPM to a three-factor model that captured the value and size premium in addition to the contribution of the excess market return. Size (measured by market capitalisation) and book-to-market equity (the value effect) had been found to capture the cross-section of average returns that were often associated with size, leverage and book-to-market equity in earlier literature. Notably, size and book-to-market ratios proxied underlying asset risks rather than mere anomalies (Fama & French 1992).

The FF5F was developed in Fama and French (2015) with the aim of improving on some of the empirical failings of the Fama–French three-factor model (FF3F), given that the model was found to not adequately capture cross-section returns. The FF5F is therefore an extension of the FF3F, with Fama and French (2015) adding an investment and profitability factor to the model in an attempt to address the empirical embarrassment of the FF3F. The FF5F is specified as:

where α, RF, bi(RM – RF), siSMB, hi HMLt and ε are defined as can be seen in the FF3F; r and c represent the loadings on the respective profitability and investment factors, and RMW and CMA represent the respective profitability and investment factors.

Fama and French (2015) studied companies listed in the New York Stock Exchange (NYSE), National Association of Securities Dealers Automated Quotations (NASDAQ) and American Stock Exchange (AMEX) from July 1963 to December 2013. From testing the FF5F in numerous sorts, they found that the FF5F showed an improvement in explaining expected returns when compared to the FF3F, as it produced lower intercepts. Fama and French (2015), however, found that there was very little difference between their FF5F and a four-factor model that excluded the HML factor. This suggested that the HML factor was superfluous in the dataset, as the HML associated returns were already captured by the other factors. In addition to this, they found that their model mispriced small stocks that had negative exposures to the profitability and investment factors. In response, Fama and French (2016) considered anomalies that were not addressed by the FF3F, such as accruals, net share issues, momentum and volatility. Their results showed that with the exception of accruals and momentum, the list of anomalies left unexplained by the FF3F shrink when the FF5F is used.

In South Africa, Charteris et al. (2018) sought to investigate whether the alternative three-factor model formulated by Chen, Novy-Marx and Zhang (2011) and a FF5F could explain momentum on the JSE. The results of the study indicated that the CAPM, FF3F and the Carhart (1997) four-factor model could not explain the momentum of past returns on the South African market. On the contrary, for the models that included the investment and profitability factors as pricing factors, as in Chen et al. (2011) and the FF5F, Charteris et al. (2018) found that the pricing errors were significantly lower than was observed for the CAPM and the classical FF3F. Profitability was found to be significantly and positively related to returns, with investment displaying a weaker relation to momentum than profitability.

It is apparent from the prior literature that the cycle of model development, empirical testing in context and model redevelopment is likely to continue in future asset pricing research endeavours. In developing the study hypothesis, the authors therefore evaluated recent applications of asset pricing models in a South African context.

Empirical testing of asset pricing models on the JSE includes tests of the CAPM (for example, Ward & Muller 2012), the FF3F model (for example, Basiewicz & Auret 2010) and the FF5F model (Cox & Britten 2019). Since Ward and Muller (2012) confirmed the findings of Van Rensburg and Robertson (2003) that returns of JSE-listed shares showed an inverse relationship with beta, subsequent research has found that the addition of factors tended to enhance the accuracy of returns predicted by the models. Notably, Cox and Britten (2019) found that the FF5F, on an overall basis, explained the cross-section of returns on the JSE with greater accuracy than the FF3F.

Yet, despite the FF5F accounting for the highest number of risk factors, recent empirical tests show that the model still has shortcomings in explaining returns of JSE-listed stocks (Mosoeu & Kodongo 2020). The authors therefore use the FF5F as a base model and consider whether the model’s predictability will be enhanced through the inclusion of additional factors. The hypothesis is developed from literature that suggests the potential explanatory power of sustainability-related metrics for returns on the JSE.

There is a rapidly growing strand of literature that focuses on the relationship between sustainability and stock returns. Yet the empirical evidence from studies, much like that of asset pricing models, tends to be inconsistent. Griffin and Mahon (1997) suggest that this may be caused by variations in methodology and the choice of measurement approximations. For search endeavours that, like this study, use stock returns as a measure of financial performance, three categories were identified by Wagner, Schaltegger and Wehrmeyer (2001), these being: portfolio studies, event studies and multivariate regression studies.

Maiti (2021) investigated the durability of the ESG factor in a portfolio study. Having formed portfolios based on the individual ESG components’ risk factors, Maiti (2021) results found evidence of an ESG risk factor at a significant level of 5%.

In an event study, Capelle-Blancard and Petit (2019) investigated the stock market’s reaction to both positive and negative ESG news for 100 firms listed on the Dow Jones. It was found that, on average, firms faced a drop of 0.1% of their market value following negative ESG publicity but gained nothing from positive publicity. Moreover, the authors found that investors have a stronger response to media reports than communication from the companies themselves and even nongovernmental organisation (NGO) disclosures.

Lastly, some studies utilise regression analysis to investigate the link between ESG and financial performance. La Torre et al. (2020) investigated how companies’ ESG performance affected their returns. The results from the study indicated that although a correlation exists between the ESG index and most stock returns, this correlation is weak and even absent for some firms. The statistical analyses also highlighted that the stock return effects arising from ESG strategies are limited and may be concentrated to only some industries, for example, the energy and utilities sector, where operating activities are likely to have a high environmental and social impact.

From an ideological standpoint, it can be argued that the utilisation of firm resources for sustainability objectives decreases shareholder value (Derwall et al. 2005). The reasoning is that the costs involved in such pursuits, such as costs of abiding by additional standards, translate into higher operation and product costs, which erode competitive advantage and subsequently profitability (Walley & Whitehead 1994). There is evidence that indicates that investors may support this view. Halbritter and Dorfleitner (2015) and Naffa and Fain (2022) find no significant relationship between ESG performance and stock returns.

Others believe that the improvement of a firm’s sustainability practices could enhance its operational efficiency or create new sales opportunities (Derwall et al. 2005). For example, Porter and Van der Linde (1995) argued that company policies that translate to improved sustainability practices could translate to cost and pricing advantages because of a more parsimonious utilisation of all forms of capital. It follows then that the benefits of sustainability-related practices should outweigh the costs, and firms that embrace this practice would be expected to report higher corporate earnings relative to less socially and environmentally responsible firms.

However, the extent to which ESG practices and policies translate to investment abnormal returns is a function of the ability of market participants to incorporate sustainability into their projections of future returns. From the review of asset pricing literature, it is apparent that market returns are proportionally related to underlying risk factors and that an ideal equity portfolio is one that is well-diversified. Thus, following this reasoning, socially responsible investors should expect to report suboptimal returns, as their portfolios would inherently lack diversity, given that such portfolios are ordinarily grouped based on firm-specific characteristics – for example, ESG scores.

Environmental, social and governance ratings and scores are used in most of the studies empirically testing sustainability performance effects on returns (Alda 2020; Derwall et al. 2005; Eccles, Ioannou & Serafeim 2014; Galema, Plantinga & Scholtens 2008; Kempf & Osthoff 2007; Lee, Faf & Rekker 2013; Statman & Glushkov 2009). The benefit of using ESG scores is that the sustainability effects on shares can be decoupled and the effect of each dimension analysed separately. This may have the added benefit of informing researchers and investors of the most critical elements of sustainability for particular firms – a concept referred to as ESG materiality.

Most of the studies on the United States (US) market find ESG performance to have a positive relationship with returns, with some studies finding that high ESG firms significantly outperform low ESG firms (Derwall et al. 2005; Eccles et al. 2014; Galema et al. 2008; Kempf & Osthoff 2007; Lee et al. 2013; Statman & Glushkov 2009). Other studies find a significant positive relationship between specific elements of ESG and returns: using ESG scores, Mǎnescu (2011) found that community relations criteria (social) displayed a significant positive influence on returns. Galema et al. (2008) also found that the employee relations indicator (social) had a significant positive relationship with the stocks’ returns.

Based on empirical evidence from prior literature, this study hypothesises as follows:

Given the goal of the study, the data analysis was structured in such a way that if the results of the research found that inclusion of ESG scores improves the predicting power of the FF5F, this would support the researcher’s proposal to extend the FF5F into a possible six-factor model that will better predict returns on the JSE. The researcher would achieve this goal by comparing the coefficient of determination (R-squared) of the existing five-factor model with the R-squared of the proposed six-factor model including ESG scores.

The sample consists of companies included on the JSE Top 40 index during the fiscal year end of 2015 to 2019, representing approximately 80% of the total market capitalisation of all companies listed on the JSE. Given the rampant effects of the coronavirus disease 2019 (COVID-19) pandemic on financial markets, the researcher sought to exclude this period from analysis in the study as this may skew the results of the research.

Both ESG combined score (ESGCS) data and share price data were obtained from the Refinitiv workspace database. One of the advantages of this is that the researcher was able to collect all the data needed for the research from one reputable source, and this adds to the reliability of the data and consistency in methodologies used in the calculation of key ratios. Over the last 16 years, Refinitiv’s ESG performance scores have been used widely in academic research (Berg et al. 2021). Furthermore, for the purposes of managing ESG investment risks, major asset managers, such as BlackRock, have chosen Refinitiv as their global ESG data provider. Refinitiv ESG data has been also referenced in ESG research conducted by the World Economic Forum (2020) and in an Organisation for Economic Co-operation and Development (OECD) report (Berg et al. 2021).

In reporting ESG data, Refinitiv makes use of percentile rank scores which range between 0 and 100, allowing empirical analysis. In addition, the database assigns letter grades (ranging from A+ to D-) to each company based on their performance. The conversion between these two scoring formats is shown in Table 1-A1, found in the Appendix 1.

Although some research relies on ESG disclosure scores in their assessment of ESG performance (Gutsche, Schulz & Gratwohl 2017), this study utilised ESG performance ratings data. Environmental, social and governance disclosure scores are usually based on self-reported company information. On the contrary, ESG performance scores use a variety of sources to evaluate a firm’s exposure to ESG risk and how effectively it manages such risk. Therefore, motivated by this, the research avoided the use of ESG disclosure scores in favour of ESG performance data in order to further improve the validity of the research.

Suitable for this objective, the ESG scores provided by the Refinitiv database provide a comprehensive evaluation of a firm’s sustainability impact and are prepared using verifiable data sourced from the public domain, such as company and NGO websites, regulatory reports, corporate social responsibility reports and media reports. The ESG data is updated by Refinitiv annually based on the company’s fiscal year and aligned with corporate reporting patterns. A combination of human and algorithmic processes is utilised, which includes independent audits, logical error checks and management reviews to ensure that data quality is as high as possible. The scores are based on the data available, account for notable industry performance measures and have minimal company size and transparency biases (Refinitiv 2020).

For the environmental component, issues surrounding the utilisation of resources, carbon emissions and operational innovativeness are considered. The social component considers the employees, human rights and community. The governance component considers factors around management, shareholders and corporate social responsibility strategy. These measures are then translated to a percentile-based score (ranging from 0 to 100) for each of the three pillars. The total ESG score is a weighted sum of the scores for each pillar. The weights vary based on the industry for the environmental and social category, while weights for the environmental pillar vary based on the governance considerations as per the firm’s country of origin (Refinitiv 2020).

The environmental pillar score measures a firm’s performance in implementing best environmental practices to minimise environmental impact of operations and to take advantage of opportunities to generate environmentally sustainable value for stakeholders (Refinitiv 2020). The social score reflects the impact of management practices on the well-being of its employees, customers and society (Refinitiv 2020). Finally, the governance score is an indication of the firm’s implementation of best practice policies that are designed to ensure that those charged with its governance act in the best interests of its long-term stakeholders (Refinitiv 2020). The overall ESG score is obtained by aggregating these pillar scores based on the 10 category weights.

For the purposes of the research, a combined ESG score was selected as a proxy for each respective company’s ESG performance. The ESGCS provides a balanced and thorough scoring of a company’s ESG performance. The score is reliable as it overlays self-reported information with information on ESG controversies from global media sources (Refinitiv 2020). The primary goal of the score is to discount the ESG performance score based on negative stories in the media, and this is done by incorporating the impact of significant and material ESG controversies in the overall ESGC score (Refinitiv 2020).

The controversy scoring is based on 23 ESG controversy topics, such as anticompetition behaviour, child labour, wage controversies, strikes, taxation fraud reports and others. When the respective company has been involved in any such controversy, the score is calculated as a weighted average of the overall ESG score and the ESGC score per fiscal period, with the recent controversies impacting the score in the latest completed period. In the event that a company is not involved in such controversy, the ESGC score is simply the overall ESG score for that company (Refinitiv 2020). Given that larger firms tend to attract relatively more media attention than smaller firms, severity weights are applied when performing the controversy scoring. For example, a firm with a market capitalisation of less than 2 billion would receive a controversy weighting of a 100%, whereas a firm with a market capitalisation more than or equal to 10 bn would receive a weighting of only 33%. Medium-sized firms, that is, those with a market capitalisation greater than or equal to 2 bn, but less than 10 bn, receive a weighting of 67%.

As this study used panel data in the form of annual returns and ESG scores data of the JSE Top 40 companies from 2015 to 2019, a statistical ordinary least squares (OLS) regression model along with the R statistics package (R Foundation for Statistical Computing, Vienna, Austria) was relied upon for the analysis in this study.

The use of statistical regression models has been extensive in seminal studies developing and empirically testing asset pricing models, with almost all of such studies relying on statistical regression models for attempts to determine the strength and character of the relationship between one or more independent variable(s) and expected returns.

For this research, two regression models were used. One was the FF5F as it exists, with factors defined as they were developed by Fama and French (2015). The other was a statistical regression model similar to the first, with the sole difference being the addition of an ESG scores factor. It was specified as:

Where all the factors – save for the proposed ESG factor – are defined and measured as per the FF5F above. The ESG factor (liLeMLa) is defined as the return spread of stocks of ESG leaders minus ESG laggards. Where ESG leaders are those firms with high ESG performance – as measured by ESG performance scores – and ESG laggards are those firms with low ESG performance. The independent and dependent variables remain as per FF5F. The ESG factor (liLeMLa) is an additional independent variable.

In the context of this research, R-squared was a statistical measure of the proportion of variation in returns (the dependent variable) that is explained by the five and six factors (the independent variables) of the FF5F and the proposed six-factor model including the ESG factor. Therefore, it follows that should the R-squared of the proposed six-factor model have exceed that of the FF5F: the researcher will accept H1. In this instance, the researcher may then be able to conclude that the inclusion of ESG scores in the FF5F would improve the model’s predictability of returns and therefore propose that the model be extended into a six- factor model incorporating an ESG scores factor.

Table 1 presents descriptive statistics as well as the regression results for the FF5F model, excluding ESG scores.

With the exception of profitability factor (RMW) and the investment factor (CMA), all the factors provided results that are significant, at least at a 5% level. It can be inferred that despite there being a premium associated with firms that invest conservatively, the relationship is insignificant. These results are similar to those of Cox and Britten (2019). The market premium factor (Rm-Rf) provided the highest coefficient, with a positive premium of 2.004. Following this was the value factor (HML), which provided a negative premium of -0.936, indicating an inverse relationship between portfolio returns, significant at a 1% level. This is contrary to the finding of Cox and Britten (2019) that high Book-to-Market (B/M) exhibit a greater premium compared to low B/M shares. Both studies, however, show significance at the 1% level, with the HML having the strongest relationship with returns. This suggests the importance of the value factor in the South African market.

The results of this study provide clear evidence of the size effect, as in Mosoeu and Kodongo (2020), because SMB was significant at the 5% level. However, the positive coefficient is inconsistent with the results of Mosoeu and Kodongo (2020). These differences may be explained by the sample of firms that were analysed in each study. While the present study chose the largest firms on the JSE, the prior study included the entire JSE, including smaller firms. The profitability factor produced a positive coefficient of 0.215. However, this factor was insignificant.

Notably, the R-squared for the unadjusted FF5F model was 0.344. This suggests that FF5F and the factors chosen in this study do not sufficiently explain the returns for the majority of shares on the JSE and are consistent with the findings of Mosoeu and Kodongo (2020).

Table 2 presents the factor summary statistic results for the six-factor model, including an ESG scores factor. All variables are as defined in the FF5F, with the exception of the ESG scores.

The market premium factor also continues to show a high correlation with expected returns, producing a positive coefficient of 2.802, significant at a 10% level. The profitability factor and the investment factor both displayed positive correlation with expected returns of 0.322 and 0.241, respectively. The authors note that the size and profitability premiums increase in the six-factor model. This is similar to the results of Cox and Britten (2019), as their size and profitability premiums increased as they moved from a three-factor to a five-factor model. As expected, the results are largely consistent with those of the original FF5F model of this study, as shown in Table 1.

Of particular interest is the correlation produced by the ESG scores factor. This factor produced a negative correlation of 0.207. Interestingly, this indicates that a negative relationship exists between the ESG scores reported by JSE-listed companies and their returns. This result, however, is not significant. The adjusted FF5F produced some correlation with JSE expected returns; however, the direction of this correlation was surprising, as ESG scores were seen to be inversely related to expected returns.

Similarly, employing panel regression on the JSE over the 2012 to 2019 period, Ball (2021) investigated the relationship between ESG and corporate financial performance and during this period also found no statistically significant evidence of a relationship between the two. However, as the main objective of the research performed in this study is the investigation of the model itself, instead of individual factors, more importance will be placed on the model statistics results and the statistical significance thereof.

Notably, the R-squared for the FF5F model with ESG scores was 0.352. This indicates that the inclusion of ESG scores does, in fact, improve the predictability of the FF5F asset pricing model (R-squared of 0.344), although only marginally.

A key delimitation of this study is that its primary purpose is not to empirically test the validity of the FF5F in explaining returns of JSE-listed companies’ shares. Instead, irrespective of the predicting power of the FF5F model as it stands, this study sought to investigate whether there is improvement in that predicting power through the inclusion of ESG scores. Therefore, although the above R-squared results from both models suggest that the models do not seem to capture most of the returns on the JSE, the results are acceptable for the purpose of answering the research question and addressing the aim of the study.

The null hypothesis of this study was that the inclusion of an ESG score factor would not improve the predictive power of the FF5F model. Based on the results discussed above, because the F- test of overall significance for the adjusted FF5F of 0.000000000176 is below alpha, that is, the significance level of 0.05, the null hypothesis is rejected in favour of the alternative hypothesis. Therefore, inclusion of an ESG scores factor to the FF5F results in a marginal improvement of the model’s predictability of expected returns on the JSE.

The results show that the inclusion of ESG scores enhances the explanatory power of the FF5F. However, this was only a slight improvement. As such, it appears that sustainability-related factors do not materially influence investment decisions and that ESG performance is not significantly priced by the market.