C. Geissler, V. Margot et al. The 7th Public Investors Conference, October 22nd, 2018, Rome, Italy.
Abstract: We designed a machine learning algorithm that identifies patterns between environmental, social and governance (ESG) profiles and financial performance for companies in a large investment universe. The goal of the algorithm, which falls in the category of supervised machine learning, is to predict the (conditional) excess return of each company over the benchmark, given the specific values taken by some of its ESG indicators (the features). In other words, the algorithm identifies regions in the high-dimensional space of ESG features that are statistically related to financial outperformance or underperformance. The final aggregated predictions are transformed into scores, which allow us to design simple strategies that screen the investment universe for stocks with positive scores. By linking ESG features with financial performance in a non-linear way, our strategy is shown to be an efficient stock picking tool, outperforming classic strategies that screen stocks according to their ESG ratings, such as the popular best-in-class approach. Our paper introduces new ideas into the growing field of financial literature investigating the links between ESG behaviour and the economy. We show, indeed, that there is clearly some form of alpha in the ESG profile of a company, but that this alpha can be accessed only with powerful, non-linear techniques such as machine learning.