Finance has developed a conscience – or at least a red-hot investing opportunity set with a conscience.
The market for doing well financially and doing good for the world at the same time continues to grow. Strategies that include environmental, social, and governance factors hit $22.9 trillion last year, up 25% year-over-year, according to the Global Sustainable Investment Alliance.
As ESG looks hotter than ever, the formerly sleepy strategy has seen a rush of funds launching products to address all corners of a portfolio. To stand out, they’re turning to advanced technology, including artificial intelligence and machine learning, into their investment processes.
But as some managers are starting to find, artificial intelligence is no cure-all for the problems plaguing ESG. Data sets often don’t line up, and without careful human management, computers’ seemingly best investments omit huge swaths of the world.
From plastic pollution to corporate scandals
One of the simplest applications of big data to ESG mimics how investors are already using machine learning — working with computers to scan 10-Ks and other company forms like disclosure documents and proxy statements. Index and analytics provider MSCI, for example, looks for phrases relating to health and safety procedures, risk oversight, and other governance terms.
“That’s almost basic these days,” said Linda-Eling Lee, the New York-based company’s head of ESG research.
To go beyond simple scanning, MSCI now looks at changes in keywords over time to understand when companies start paying attention to ESG issues. Between 2017 and 2018, for example, mentions of plastic pollution jumped 300%.
MSCI is also using machine learning to draw out mentions of controversies, such as budding corporate scandals that could expand to take down a company.
This approach differs from sentiment analysis, which tallies how often a controversy or other topic is mentioned. In one example, Lee highlighted the difference between a factory accident in a remote corner of the world that injures 1,000 workers, but is only covered once in local media, compared with a mishap close to home that injures one person and is widely reported. MSCI’s method pays closer attention to the first issue, because a local problem, even if it’s not all over the news, could signal underlying business problems that might hurt the stock.
“We’re trying to find all those cases because our assessment is based on the scale, the impact, and the severity of the case itself,” Lee said. “It’s that combination of having humans to set the parameters of what we’re searching for, and having humans look at the cases to apply the methodology. Everything in between, more and more, you can apply machines to.”
Advanced analytics can be particularly helpful for ESG, because the investment strategy lacks data that’s consistent across countries and industries, unlike the typical financial data that publicly-traded companies are required to regularly report. Often, companies have only recently started to report limited ESG metrics.
Debbie McCoy, a portfolio manager and head of ESG for BlackRock’s systematic active equity team, said her team approaches companies’ self-reported ESG numbers skeptically – why would companies choose to report what they do, if they’re not required to do so? She focuses on finding areas of sustainability to research holistically, gathering data both from a company and other sources, rather than on creating overarching ESG scores for companies.
Recently, for example, her team is going beyond measuring carbon emissions – now common in ESG investing – to truly understand how companies are or aren’t preparing for climate change. They do this by breaking down and analyzing individual parts of the company such a different business lines or subsidiaries.
“We think it’s discernible information, but not obvious,” she said. “This is the subtlety of working around an investor thesis instead of a score.”
Artificial intelligence is no panacea for ESG, McCoy cautioned, particularly with the risk of a computer reflecting humans’ implicit biases. McCoy has seen models that recommend avoiding emerging markets, for example, because analysis suggests that companies adhering to ESG standards can’t exist outside of developed markets.
“We think that’s a pretty biased thing, ultimately,” she said. “We don’t want ESG investing to become ‘you can invest in North America and Europe and nowhere else.’ That would be awful.”
‘Better with age’
Asha Mehta, a portfolio manager and the head of responsible investing for Acadian Asset Management, uses some elements of machine learning in her investment process at the Boston-based investment firm, which oversees $95 billion. As a quantitative investor, Mehta said one of her biggest challenges is understanding how a computer using artificial intelligence arrives at an outcome.
“We pride ourselves in having a fundamental rationale to our model – it’s not black box,” she said. Machine learning makes it more difficult to tease out a causal relationship, a concern echoed by other investors.
While that’s not likely to change, Mehta sees future opportunities for machine learning to assist with filling in gaps of existing data, and to create more ways to measure ESG.
Smaller managers that lack Acadian or BlackRock’s ability to integrate advanced technology into ESG investments in-house are increasingly looking to outsource the effort to companies like Truvalue Labs, a six-year-old data provider. The company pulls information from 150,000 sources, ranging from watchdog groups to social media, into a computer that’s been taught to account for both positive and negative ESG news.
Sign up here for our weekly newsletter “Wall Street Insider,” a behind-the-scenes look at the stories dominating banking, business, and big deals.
Artificial intelligence, aided by machine learning and natural language processing, kicks in from there. Truvalue has taught the computer about hundreds of risks and opportunities, ranging from river pollution to data breaches, with thousands of reference points to help the machine notice patterns in how things are described. The ensuing data sets are quantified, with a score measuring how positive or negative a certain issue is to a company. Some managers use the software for ongoing portfolio monitoring, picking up small signals before a company makes news for the wrong reasons.
In a 10-year test of Truvalue’s scoring, the company said its portfolio outperformed the Russell 1000 Index by 3.5%.
Truvalue’s founder Hendrik Bartel cautioned that it’s early days to use artificial intelligence and ESG together.
“Any good venture capitalist and any good software engineer will tell you it takes three to five years to get an AI to do somewhat of the right thing,” he said. “It gets better with age, just like an analyst gets better with seniority and a wine gets better over the years.”
While artificial intelligence cannot replace an analyst’s intuition, Bartel thinks that its next frontier is predictability, using the technology to forecast events and behaviors for individual stocks and entire sectors.
“If you look at how much data we’re sitting on, we’re really starting to scratch the surface,” he said.