Machine Learning for Transparent and Sustainable Investing

The past year we have worked a lot on increasing the data set that we use for our machine learning model. In addition to building signals based on traditional company-, market- and ESG-data from proprietary data vendors, we have also developed an innovative data set of macroeconomic money flows. We have tested the first versions of our ML model and going forward we will focus on developing our ML model further and using it to construct optimal equity portfolios. Our side project with Norsk Regnesentral to test new statistical causality methods applied to finance will be completed in December. We have developed, launched, and marketed an ESG equity fund based on factor models and ESG-data. In addition to signals based directly on ESG-scores, we have researched ESG-score momentum and ESG-score volatility. We have published two scientific articles about possibilities and challenges with using climate data and published a third article about how the market reacts to climate data. We expect to publish more scientific articles about how the stock market reacts to climate data and news about the climate. Results from our research project will be incorporated as they become available, in our ESG-fund, new investment strategies and our analysis tool. Our ESG-fund has been covered in several articles in mass media. We organised a webinar when we launched our ESG-fund and we organized two ESG-webinars with Norsk Forening for Kvantitativ Finans with external ESG-experts with a total of 1.500+ registered participants.

The underlying idea of ML for ESG-investing is to include environmental, social, and governance (ESG) factors, enable full transparency, and lower the cost and friction to more sustainable investing. This will be achieved by integrating these trends into a portfolio construction tool designed around interpretable machine learning and inclusion of ESG factors. To go beyond the (publicly acknowledged) commercial state of the art will require the R&D activities to address the following core issues. 1. Exploitation of advances in applying machine learning to factor investing suggests significant improvements in estimating the expected return of assets. 2. Most published papers that employ machine learning tend to omit a sensitivity analysis of the models with respect to factors. To meet our ambitions of transparency, we must extend previous work accordingly. 3. Commercially available ESG factors are not transparent about their construction and are aggregated at levels too high to use in factor investing. Construction of climate risk factors from scratch allows for more transparency and our own choice of aggregation level. We will adopt an iterative approach towards the research and implementation work. We will work towards successively more complex models and successively integrate more data only when the previous steps have been passed and verified. The anticipated R&D results of the project that will support the development of the portfolio construction tool are: 1. Sophisticated machine learning models for estimating the stochastic discount factor (SDF) based on various financial factors/features, including ESG factors. 2. Explainable AI models that explain and trace how different factors, including ESG and climate risk factors, impact the expected return of assets. 3. ESG and climate risk factors and how they influence asset performance at company, sector and market level, including construction of transparent ESG indices that allow benchmarking of portfolios.