Sections

Is ESG inherently multi-factor?

Related Categories

Multi-factor ESG is probably more successful as a marketing term than it is at providing a coherent approach to mixing ESG and other factors in a portfolio-construction methodology

Key points

  • There is positive correlation between ESG and some factor scores, such as size (larger companies) but index providers generally view ESG as independent from factors
  • Analysing a simple ESG strategy shows ESG is inherently multi factor
  • Neutralising factor exposure decreases the performance of ESG portfolios
  • It is hard to construct a portfolio with substantial ESG tilts without also tilting on style, industry or country factors

The recent announcement by AP2 , one of five buffer funds within the Swedish pension system, that 29% of its overall portfolio will be managed to “multi-factor indices where environmental, social and governance (ESG) is the most important factor for the weighting in the indices” is just the latest indicator of the growing interest in multi-factor and ESG investing.

However, because ESG is inherently multi-factor, the term multi-factor ESG is probably more successful as a marketing and branding term than it is at providing a coherent, consistent, transparent and explainable approach to mixing ESG and other factors in a portfolio-construction methodology.

It is well known that there is substantial overlap between ESG and many traditional factors associated with risk premia indices and factors used in commercial factor risk models. For example, ESG scores are traditionally positively correlated with size – larger companies usually have higher scores; or, from the sceptic’s point of view, larger companies are more likely to have the resources to provide more information and transparency on ESG metrics. ESG scores are also strongly associated with industry and country factors. Indeed, if one were inclined to build a fundamental factor-risk model with ESG (or E, S and G) as a factor, one would first have to overcome the strong linear dependence between ESG and the traditional style, industry and country factors found in those models. This contrasts with the index providers’ approach, which views the ESG score as essentially independent from traditional multi-factor approaches.

This article illustrates the degree of overlap between ESG and traditional risk factors. It begins by back-testing a simple ESG strategy and then progressively neutralising the exposures to style factors, industry and country factors, and all fundamental risk factors. As a result, performance, not unexpectedly, is neutralised, underscoring the reality that ESG is inherently multi-factor.   

So, what does “multi-factor ESG” mean? The answer is, “it varies.” So buyer beware.

Results

We construct a series of back tests that isolate and capture different kinds of performance associated with ESG. The back-test strategy maximises the long-only portfolio exposure to an ESG score with a fixed tracking-error limit, where we perform the back test over a range of different tracking-error limits. We use Axioma’s Medium Horizon Fundamental Factor Risk Models to predict tracking error. We only hold assets in the benchmark with limits fixed as follows:

• If the asset has an ESG score, the maximum active position is ±2.5%.

• If the asset does not have an ESG score, then we hold that asset at the benchmark weight – for example, zero active weight.

By holding the uncovered assets at the benchmark weight, we reduce the potential bias in the results associated with the historical lack of coverage. The coverage is worst in the early years of the back test and then steadily improves. Nevertheless, coverage still affects the results, but with the asset limits imposed, the uncovered assets do not contribute to any active performance.

We use ESG scores provided by OWL Analytics. These scores cover worldwide equities from 30 March 2009 with monthly updates. The back test is performed from 30 March 2009 to 29 December 2017 with quarterly rebalancing. The performance results reported here are based on monthly returns.

A series of four back tests are performed:

Pure ESG: no additional exposure constraints.
Style neutral ESG: the active exposures to all style factors in the risk model are zero.
Industry and country neutral ESG: the active industry and country exposures are zero.
Factor neutral ESG: the active exposures to all risk model factors are zero.

The idea behind these four cases is as follows:

• The pure ESG results show the potential alpha associated with ESG in isolation from any multi-factor considerations.  

• The style neutral results show how the style factors contribute to the potential alpha associated with ESG. The performance reduction of the style neutral case compared with the pure ESG case is a measure of the performance associated with non-zero, active style factor exposures.

The industry and country neutral ESG results show how the industry and country factors contribute to the potential alpha associated with ESG. The performance reduction of the industry and country-neutral case compared with the pure ESG case is a measure of the performance associated with non-zero, active industry and country factor exposures.

“Neutralising factor exposure decreases the performance of ESG portfolios, in some cases considerably. That is not a criticism of multi-factor ESG products, as these products explicitly take non-neutral factor exposures that are meant to enhance performance”

In the factor-neutral case, whatever outperformance remains is not associated with any traditional risk-model factors and would be intrinsic alpha associated with ESG.

Figure 1a shows the pure ESG results for four all cap investment universes: US, EU, Asia Pacific and emerging. For sufficiently large tracking error, all four portfolios exhibit outperformance. The active returns increase roughly linearly with increases in tracking error. So, in isolation from any traditional factor considerations, the ESG signal appears to produce alpha.

Figure 1b shows performance for the same universe when the portfolio is neutralised with respect to all the style factors in the corresponding risk model. As shown, when the portfolio is style neutral, the outperformance is essentially obliterated, except for the Asia Pacific universe with large tracking error. Notice also that the curves terminate early – that is, even when we allow the portfolio potentially to have a tracking error larger than 4%, the style bounds limit the maximum achievable tracking error to a much smaller value.

Figure 1c shows performance for the same universe when the portfolio is neutralised with respect to all the industry and country factors in the corresponding risk model. In this case, some alpha remains, as all four universes have positive active returns, at least for sufficiently large tracking error. However, as with the style neutral case, the active returns are much smaller than the pure ESG case.

Finally, figure 1d shows performance for the same universe when the portfolio is neutralised with respect to all risk-model factors. In this case, not only is the alpha essentially negligible, but the maximum achievable tracking is also quite limited, usually to less than 2%. This indicates that ESG not only overlaps with traditional risk factors, but also that the residual ESG (for example, that part of ESG that is orthogonal to traditional risk factors) is quite small.

realised performance for four all cap investment universes

As expected, neutralising factor exposure decreases the performance of ESG portfolios, in some cases considerably. However, that is not a criticism of multi-factor ESG products, as these products explicitly take non-neutral factor exposures that are meant to enhance performance, as seen in figure 1a.

Figure 1b, however, shows how each of these performance curves is affected by selective neutralising against well-known risk factors. As expected, due to the known overlap, performance always decreases as factor exposures are neutralised. In the full neutralisation case (factor neutral, red), only the Swedish AP portfolios continue to show outperformance. Also, the largest achievable realised tracking error shrinks as more factors are neutralised, as the portfolio holdings reach their asset limits.

Conclusion

Our goal has been to illustrate in an easily accessible manner the degree of overlap that exists between ESG scores and traditional risk model factors. The bottom line is that it is hard to construct a portfolio that substantially tilts on ESG without also tilting on style, industry and country factors. ESG is inherently multi-factor.

Ian Webster is the chief of staff and head of corporate strategy and Anthony Renshaw is director of applied research at Axioma

Have your say

You must sign in to make a comment