The ESG-efficient frontier (Part II)
This week we look at some empirical results from the ESG-efficient frontier
In the previous week, we looked at the first part of a paper by AQR’s Lasse H Pedersen, Shaun Fitzgibbons and Lukasz Pomorski. This remains an influential paper in that it contains some really useful ideas on how to incorporate ESG in a portfolio optimization problem. You can read the first part here.
Just to summarize, in Part I we covered the theory behind the ESG-efficient frontier, which is a framework that starts from a standard mean-variance frontier to derive a set of optimal portfolios.
Under the ESG-efficient frontier framework, we can construct a frontier of portfolios with the best Sharpe Ratio for each level of ESG score (see figure below). This frontier then allows investors to choose the trade-off between Sharpe Ratio and ESG score.
In this part, we will look into some empirical results.
Let’s begin!
ESG Measures and Data
This study proposes some useful metrics to capture the different aspects of ESG, which allows for empirical testing:
A measure of E: low carbon intensity. This is quite widely used in the industry and is typically measured as scope 1 and 2 carbon emissions normalized by a company’s sales.
A measure of S: non-sin stock indicator. This is also a commonly used proxy for the Social aspect of ESG. Stocks in certain sin industries (such as tobacco and gambling) are shunned by some ESG-conscious investors. We wrote about the sin premium here.
A measure of G: low accruals. This, to me, is an interesting one. From the authors' point of view, a company’s governance can be measured by the level of the firm’s accruals over assets. Accruals are “essentially accounting income for which the related cash has not yet been received”. A company with good governance, in theory, would be more conservative in recognizing their revenue.
A measure of overall ESG: MSCI ESG scores. No surprises here. MSCI ESG scores have become one of the most widely used ESG scores by institutional investors.
Empirical ESG-efficient frontier
With the measures defined above, the authors then proceeded to test the ESG-efficient frontier using the E and G metric. S is not tested here because the definition of a sin stock is binary and therefore we can’t compute scores from these (as compared to E and G).
The results are certainly interesting.
For an efficient frontier constructed using the E score, the ex-ante (or perceived) frontiers for an ESG-aware and ESG-unaware investor are close together, meaning that the environmental proxy used here is not useful in explaining average returns.
But what is more stunning is in Panel B, which looks at the realized Sharpe ratio. The two frontiers practically overlap, meaning that the Sharpe ratios of the portfolios on the two frontiers are the same for any level of carbon intensity.
In other words, incorporating carbon intensity does not have a significant impact on the Sharpe ratios, which has also been covered by other papers such as this one, which shows that “it is possible to cut emission intensities in half at least with virtually no loss in Sharpe ratio”.
For the ESG-Sharpe frontiers constructed using the G proxy, the results are significantly different from the ones above.
As we can see in Panel A below, the perceived Sharpe ratios differ between an ESG-unaware and an ESG-aware investor. Just to recap, in this context our ESG-aware investor incorporates accruals (a proxy to governance) in calculating his or her Sharpe ratio. This suggests that the G proxy was able to predict returns in the sample.
Another interesting observation from Panel A is the asymmetry of the perceived frontier for the ESG-aware investor. In the authors’ words, the frontier peaks at around 2.25 (which is higher than the market’s), and the asymmetry means that it would be more costly to the Sharpe ratio if we were to decrease the portfolio’s G score from that point than to increase it.
In panel B, the realized Sharpe ratios more or less follow the same shape as the perceived frontiers, especially for the ESG-aware one. Again, this is because the G proxy positively predicts returns.
Lastly, we move on to the impact of screening on the ESG-efficient frontier.
In this set-up, the top bold line represents the unconstrained investor (which will be exactly the same as the line in Panel A above), and investors with a 10% and 20% screen respectively (removing the percentage of stocks with the lowest G score in this case).
Two observations here:
Adding a constraint reduces a portfolio’s expected performance. As we can see, the frontiers shift downwards each time we screen out more stocks.
Adding a constraint also reduces the optimal portfolio’s ESG score. Observe the optimal point of each line representing the tangency portfolio (portfolio with the best Sharpe ratio) and you will notice that for the top blue line, the best Sharpe Ratio is achieved at an ESG score of 2.25. After adding a 10% screen, the optimal Sharpe Ratio is achieved at an ESG score of 1.5. After removing 20%, the optimum is an ESG score of one. What does that mean?! The intuition here is that it is almost always better to have an unconstrained universe because these low-ESG assets can act as funding sources and be used to hedge risks to help improve the overall Sharpe ratio.
In summary, this paper is really useful in quantifying the metrics used in proxying ESG scores, which allows for empirical testing. The ESG-efficient frontier framework makes it easier to calculate the trade-off in ESG score and Sharpe ratio and communicate that to investors.
From the empirical results, we learnt that the G proxy is the only one with modest predictive powers, but that is probably because it is the one most closely tied to accounting. For the E proxy, you could practically achieve a lower-carbon portfolio without sacrificing too much in terms of Sharpe ratio. And one last thing, you can achieve a better outcome of ESG and Sharpe Ratio using an unconstrained universe compared to a screening approach.
That concludes this week’s post on the empirical results on ESG-efficient frontier.
Before we go, it is worth mentioning the other topics that the authors also covered in this paper:
Does ESG predict a firm’s fundamentals? Mostly no evidence except for accruals.
Does ESG predict investor demand (measured by institutional holdings or trading activity)? Again, only accruals show the best results, the other metrics are not conclusive.
Does ESG predict valuation and future returns? Yes, you guessed it. Accruals are the best in terms of predicting excess returns, followed by some evidence for the sin premium. Not so much for the E proxy (carbon emissions) and ESG proxy (MSCI score).
If you are interested, be sure to read the paper from the link here.
great follow up to part 1. have been waiting for this.