Speaking at the Allianz Global Investors 2021 Asia conference in June, Kunal Ghosh, CIO Systematic Strategies at AllianzGI, explained how we use artificial intelligence (AI) to uncover Asian investment opportunities related to environmental transformations. Our AI Environment Model incorporates an AI “machine” that sifts through thousands of news stories as they appear, aiming to unearth shifts in environmental performance far more quickly and accurately than humans can. The goal is to discover early evidence of environment-related transformations at a range of companies in Asia, including the identification of environmental leaders and laggards.

Of course, keeping track of such a huge universe of securities is a colossal challenge. Using the entire global MSCI All Country World All Cap Index for illustration, Mr Ghosh pointed out that it encompassed about 15,000 companies. Yet just one of these firms, a multinational beverage company, was the subject of 1,300 environmental stories in a single month, each one of which could affect its environmental rating.

“Front-running” the ratings agencies

“In certain ways, you can think of our model as effectively allowing you to front-run the MSCI rating,” said Mr Ghosh. “As MSCI changes ratings and investors shift their allocations, you have the opportunity to take advantage by getting ahead of the environmental premium.”

The model has been developed over the past two years, using natural language processing and machine learning techniques to uncover the “unstructured data” in environmental news stories. The news flow comes from 12,000 sources, covering the companies and countries behind 35,000 to 40,000 assets. Any environmental-related news associated with an asset – whether a sovereign bond, corporate bond or stock – can be identified, mapped and classified.

And because the environmental score is awarded by a machine, it is naturally objective. This contrasts with the ratings agencies, where subjective opinions can lead to inconsistency. For instance, MSCI and Sustainalytics recently reached different opinions about an electric vehicle manufacturer’s environmental performance, despite having the same information.

Valuable disagreement

Often, the rating agencies and our AI model reach similar conclusions, but the model’s judgments are most valuable when opinions differ. For instance, our AI Environment model has awarded high scores to a Chinese real estate developer since its recent creation of a pioneering, ecologically sustainable model in Malaysia. More recently, in 2021, this developer has been in discussions to bring similar sustainable ideas to Japan. But MSCI has been slow to catch up, only giving the firm a high rating in March 2021.

“We identified the ecologically sustainable city model that the developer was building in Malaysia and Japan,” noted Mr Ghosh. “MSCI was simply lagging; for a long time they gave the developer a poor rating. They have been known to be more than a year late in identifying positive environmental change.”

Mr Ghosh also provided the example of a Singaporean multinational bank, which received a high AI Environment score for its carbon efficient operations and environmental lending activity. However, in June 2020, this opinion was at odds with the far more critical ESG (environmental, social and governance) rating from MSCI, given the bank’s history of lending to hydrocarbon mining projects. But MSCI had not registered that the bank had started improving its environmental financing activities. Nine months later, in March 2021, MSCI caught up and upgraded its view of the bank..

AI unlocks an untapped opportunity

AllianzGI’s AI Environment model can react to news much more quickly and accurately than ESG ratings agencies, which depend on human analysts. Unstructured, text-heavy data in news reports represents an untapped opportunity for asset managers with the technology to gather insights ahead of the market. Having a model that is broad, fast and objective offers an advantage when incorporating environmental factors into investment decisions.

Download PDF