PanAgora Asset Management has developed a new way to measure and scale sustainable investing using an unsupervised machine learning model, according to a recently released academic paper from the group.

The research, which allows investors to algorithmically map a company’s self-reported activities to the United Nations’ Sustainable Development Goals framework, represents a way for investors to continually verify that their portfolios are actually aligned with the sustainability goals they set at the fund’s inception.

The model analyses company reports and uses natural language processing techniques to classify the firm’s activities relative to the UN’s 17 SDG goals, allowing investors to see which companies are pursing clean water policies, for example. The model uses a binary 0 or 1 rating system — once the algorithm identifies words and phrases about a company’s services that align with one of the SDGs, it informs the investor.

Sustainable investing has long been plagued by concerns that sustainability judgements are subjective, which has stymied adoption by some firms. Mike Chen, director of equity investments at PanAgora and co-author of the study, noted that developing a systematic, verifiable framework can address some of those concerns. “Human readings [of the SDGs] are somewhat subjective,” he explained. “I think you get more variation through human analysis [than you would though machine analysis].”

He also noted that human assessments aren’t scalable, which can create problems for asset managers with many investments. “With a machine, you let it run,” Chen said. “If you don’t like the result, you can trace it in the algorithm and figure out what part you don’t like, so it’s verifiable.”

So far, the research hasn’t been deployed in any of PanAgora’s active models, but the researchers plan to expand the universe of data inputs from a company’s self-issued reports to include data from NGOs and news sources.

Despite not yet being productionalised, Chen noted that the authors were surprised at the high F1-score — one measure of NLP model accuracy used in testing — that they obtained in the initial test. “This is an initial foray into creating something that’s verifiable [and transparent],” Chen explained.

PanAgora has actively pursed ESG quant research over the past few years, with an aim toward adding to the body of research on the topic as well as generating models to use in-house. Last year, the firm told Neudata that it continued to see strong performance of its quantitative portfolio focused on ESG factors, three years after the strategy was launched. At the time, Chen said he believed that ESG would become so integrated into the investment process that it would become part of everyday analysis over the next 10 years.

Photo by Scott Webb on Unsplash