We conducted research on ESG investment based on the paper, The Effect of Socially Responsible Investing on Portfolio Performance, by Alexander Kempf and Peer Osthoff (2007).
Sustainable, Responsible and Impact Investing (SRI), also known as ESG investing, is an investment discipline that considers environmental, social and corporate governance (ESG) criteria to generate long-term competitive financial returns and positive societal impact.
In the present socio-economic environment, where more and more investors have awareness towards issues viz. climate change, global warming, gun violence, tax evading etc. They apply socially responsible screens when building their stock portfolios. It raises the question whether these investors can increase or maintain profitability of their portfolio performance by incorporating such screens into their investment process.
We approximately replicated the investment strategy in the paper, based on ESG ratings: Buy stocks with high socially responsible ratings (ESG score) and sell stocks with low socially responsible ratings (ESG score). As we did not have access to the KLD analytics data used by the paper, we used ESG rating by Sustainalytics, available on Bloomberg, as a substitute.
We found that replicating this strategy with the available data leads to high abnormal returns of up to 5% per year. Long short portfolio with extreme ESG score stocks gives better 4 factor alpha. The extra returns remain significant even after considering reasonable transaction costs.
Data Cleaning Procedures
As we had no access to the SRI ratings of KLD Research & Analytics that Kempf and Osthoff (2007) used, we used Sustainalytics Rank data from Bloomberg as the ESG indicator. For return data, we used monthly return data from CRSP. Finally, we used factors data (MKT, SMB, HML, MOM, RF) from French’s website for estimation of alphas and betas.
All US stocks in BESGPRO INDEX in Bloomberg, which are around 600 companies per year that have ESG data.
We were only able to find Sustainalytics Ranking data starting from 2014-02-28 to 2018-4-30. The monthly stock return data were from 2014-01 to 2017-12. As a result of these two limitations, We used ranking data from 2014-03-31 to 2017-9-30 and apply it to monthly returns from 2014-04-01 to 2017-12-31.
We used RET from CRSP as the simple returns. Non-numeric values in the dataset are regard as NA’s.
We used DLRET as the delisting return. Any non-numeric values are treated as NA’s. The final output returns are Ret in the data table. If RET is missing, use DLRET as return; if DLRET is missing, use RET as return; if both are not missing, used (1+RET)(1+DLRET)-1 as the return.
Market Capitalization = SHROUT * PRC Any NA’s or non-numeric values are ignored, and negative PRC are converted to be positive.
We chose the positve screening discussed in the paper, which was to construct portfolios by soritng the stocks on Sustainalytics Rank. Higher ranking means better ESG performance. We did analysis on four thresholds: 5%, 10%, 25% and 50%. For the 5% threshold, we categorized the top 5% stocks into the High portfolio, the bottom 5% stocks into the Low portfolio, and ignore all others in the middle; do the same for 10%, 25% and 50% (or median) breakpoints; for the 50% threshold, no stock is left in the middle. Long-short portfolios are constructed by longing the High portfolio and shorting the Low portfolio.
In terms of the rebalancing frequency, we were not able to follow the annual rebalancing in the paper as we only had 4 years of data, which limited the back-testing period to be 3 years. To approximate the 16 periods of data used in the paper (from 1992 to 2004), we used quarterly rebalancing (2014-04 to 2017-12), which gave us 15 periods to analyze. Such approach also allowed us to explore if increasing the trading frequency would produce a higher abnormal return.
Consequently, all of our portfolio sortings in quarter t are based on Sustainalytics Ranking at the end of quarter t-1.
Stocks within each portfolio are value-weighted based on its lag Market Capitalization. The stocks need to have non-NA market capitalization on the last day of the previous month in order to be included in any portfolio this month.
Weights of a stock within its portfolio at time t, when there are stocks in this portfolio (for example, High portfolio):
k marks the low, high, long-short portfolio. Portfolio returns are the sum of the weights times simple stock returns.
Based on the procedures we discussed above, we plotted a graph of the cumulative return of long-short ESG portfolios with different breakpoints.
Cumulative Return of ESG Portfolio using Different Breakpoints
We tested if the portfolios constructed based on Sustainalytics Rank have Carhart alpha - i.e. abnormal returns relative to Carhart’s four-factor model. According to the paper, market excess return, size, book-to-market and the momemtum factor all have significant coefficients. Therefore Carhart would be a more appropriate than Jensen’s alpha or FF-3 alpha as the measurement of the strategy’s abnormal return.
Table 1 is the regression of 10% breakpoint. Alpha’s were annualized by . As expected, high portfolio, composed of stocks with high Sustainalytics Rating, has a positive abnormal return of 1.56, while the low portfolio with low rating stocks gives a negative return of -3.68. Consequently, the long-short portfolio has a 5.01 alpha, and a larger part of it comes from shorting the low portfolio. In other words, stocks that do not care about ESG have worse returns. We suspect that this alpha is from the increasing awareness of ESG investing. It encourages people to buy high ESG rating stocks instead of the low, leading to a tilt in demand-supply of these stocks, and consequently increasing the return in high ESG rating stocks and decreasing that of the low ones.
In terms of factor loadings, both high and low portfolio have close to 1 market beta, resulting in a market-neural long-short portfolio with . High portfolio has -0.23 loading on SMB whereas the low portfolio has barely any loading on SMB, leading to long-short portfolio’s small negative loading on SMB of -0.2. In addition, both high and low portfolio have slightly negative and close to zero . Therefore, the long-short portfolio has little exposure to HML and MOM.
As for , both high and low portfolio have relatively high , meaning that the factor model explains the variations in their returns well. The long-short portfolio has small . This is consistent with the result in the paper.
In this table, we are analyzing how the profitability of the long-short strategy depends on the cut-off chosen. The high-rated portfolio consists of stocks with top x% ESG score and the low-rated portfolio consists of stocks with bottom x% ESG score. The four-factor alpha has been calculated for top 5%, 10%, 25% and 50% (median) ESG score percentile and bottom 5%, 10%, 25% and 50% (median) ESG score percentile and also for long top percentile - short bottom percentile portfolios.
We can observe in the table that there is significant positive alpha in some of the long-short strategy. Alpha of long-short strategy is decreasing as we go from 5% quantile long-short to 50% percentile long-short which is due to less ESG contrast in long-short strategy.
The t-stat is weak for a few alpha’s which can be attributed to lack of access to data in our analysis*. we used only 4 years (2014-2017) quarterly time series for our analysis while the paper used 1991 to 2003 data for analysis.
*Bloomberg student account provides only 4 years of ESG data.
In Table 3, we can see that 5%, 10% and 25% breakpoints all have good sharpe ratio, and they are not too fat-tailed. Both 5% and 10% have positive skewness. In other words, they do not have much downside or tail risk. Up to this point, we consider 5%, 10%, and 25% to be good breakpoints.
Table 4 shows that we still have positive abnormal returns under 5% and 10% breakpoints after transaction costs, whereas 25% and 50% breakpoints are not profitable if transaction costs are high. We estimated the transaction costs based on the change in weights of stocks each quarter, with 0 - 200 bps unit costs. The result indicates that investors should use 5% or 10% breakpoints if they use this ESG investment strategy - long the true top tier and short the real bad ones.
- The high-rated portfolio performs better than the low-rated portfolio.
- A long-short strategy (long in the high-rated stocks, short in the low-rated stocks) yields a positive four-factor alpha of up to 5% per year.
- The maximum alpha is obtained when choosing stocks with extreme ESG scores.
- The alpha remains significant even after controlling for transaction costs.
- Test the strategy in different countries
The strategy might perform differently in different countries. For example, in developed countries, companies with high ESG rating might have better return as more investors are aware of the importance of ESG and are willing to invest more in them. On the contrary, in developing countries, companies might not have the technology to sustain a high growth while meeting a high ESG standard. Therefore, it is possible that strategy has opposite performance in developing vs developed countries.
- Customize the ranking criteria for different industries.
The paper used a “best-in-class” approach to pick the top and bottom stocks in each industry. We could further explore the differences in various industries by finding out the best ranking criteria for each industry. To achieve this, we might sort stocks in each industry based on specific ESG ranking - for example, water usage and conservation, instead of the overall rating we have now. This would help us understand the most relevant and influential ESG factors in each industry.
- Regress on different factor models
The paper and our replication regressed the returns on ff-3 factors and momentum factor. We could further the analysis by regressing the returns on ff-5 factor model and find if the strategy still produces an abnormal return.
Kempf, A., & Osthoff, P. (2007). The Effect of Socially Responsible Investing on Portfolio Performance. European Financial Management, 13(5), 908-922.
Sustainalytics. (2018) Sustainalytics Rank from Mar. 2014 to Dec. 2017. Retrieved May 26, 2018 from Bloomberg database