AI-based ESG-scores as a source of alpha

Intro

One of the Master Thesis conducted in collaboration between Sanctify and Lund University is titled “Examination of AI-based ESG-scores as a valid source of alpha in the Swedish investing landscape”. As the title suggests, the thesis is about finding alpha using machine learning in combination with Sanctify’s proprietary ESG data. Several different models have been evaluated in combination with different processing applied to the ESG data such as simple moving average, momentum etc.

The thesis was limited to the Swedish market and the index OMXS30GI was used as reference when comparing the results.

Below follows a summary of the thesis and the main results.

Summary

Investing based on environmental, social, and governmental (ESG) criteria has grown rapidly in recent years. The trend has been driven by both an increased interest in sustainability, and the fact that ESG related corporate events have been shown to influence stock prices. In tandem with this development, a number of companies has pioneered methods to quantify a firm’s ESG performance. One of these companies is Sanctify Financial Technologies.

In this thesis, numerous machine learning models are used to try and predict stock prices. ESG scores from Sanctify are then incorporated in some of the models, these models are then compared to identical models without access to these scores. The predicted stock prices are then inserted into a custom-made trading algorithm that creates daily investment portfolios that maximizes the expected Sharpe ratio. The benchmarks used for evaluating the models are the Sharpe ratio, the Sortino ratio and gross returns.

Of all the models evaluated, a random forest regressor using various moving averages of the ESG scores ends up performing the best. With a Sharpe ratio of 1.185, a Sortino ratio of 1.658 and gross returns of 60.5%, it outperforms the OMXS30GI index on all three benchmarks during the tested period 2020-2021.

Excess returns are achieved, and the results indicate predictive performance for the Sanctify scores on stock prices.

The full paper is published here Examination of AI-based ESG-scores as a valid source of alpha in the Swedish investing landscape .

Results

Gross returns of the random forest model compared to the reference index OMXS30GI. Each different lines represents the same model but fed with different data, both with different processing applied to the ESG data, and without any ESG data at all.

Results of the best performing model with Sanctify ESG data compared to the same model but without the ESG data and the reference index. The values for the models are the average over several iterations.