AI is already in

AI is making strides at many levels in the world of investment management. Investors may already be riding the wave of artificial intelligence, unaware of the many ways they’ve been integrated. Is it helping? And how?

There are three main levels where AI is making a mark, says Amit Gupta, a managing director in Accenture's capital market industry group. At the first level, firms are using AI in back-office administrative tasks like net asset value calculations, reconciliation, settlement operations. At the second level, they use it in front-office tasks like client targeting and management, profiling of clients, personalization of service. Finally, at the third level, investment processes themselves are integrating AI in trading, stock selection, risk management, prediction models and alpha generation.

Many techniques are called upon, all revolving around the central kernel of AI: machine learning, data mining, deep learning, neural nets, natural language processing, knowledge representation. MAN AHL, in London, uses mostly machine learning in developing trading strategies (what to buy or sell, and when), and in improving the execution of such trades, explains a study by the CFA Institute.

Some applications can be unexpected. For example, American Century analyzes the language coming from management in quarterly conference calls. The application serves to evaluate a company’s prospects from information that isn’t fully reflected in its financial statements.

Data - massive data sets – are central in propelling AI ahead in portfolio management. At first, analysts and portfolio managers called on “standard” sets, like historical stock price behaviour or short selling trends. Then “alternative” sets came into play, like satellite imagery to predict the direction of agro-commodity prices and retail trends tracked through cellphones.

Previously, models designed by quants on limited data sets were static and needed to be updated by humans to fit changing market conditions. A key advantage of the new AI-augmented models hinges on their ability to define their own rules and update them independently to changing market environments. The best AI models pick up trends and patterns that previously evaded perception by humans.

How Good is AI?
Studies that look into performance don’t abound. One analysis by London-based Preqin compares the performance of all hedge funds to a set of 152 AI-assisted hedge funds that Preqin tracks.

From August 2016 to April 2018, the AI fund returns trailed their traditional counterparts by about two percentage points. But from that date forward, AI funds took the lead: at the end of June 2019, the AI funds showed a three-year return of 26.96% against an all-hedge fund index return of 23.87%. The AI funds also showed better metrics at other levels. Their 3-year volatility stood at 3.2% versus 3.87% for the regular guys; and their Sharpe ratio stood at 1.96 versus 1.4 (a higher Sharpe ratio indicates a better return for the level of risk taken).

Hard to Spot
How prevalent are AI-augmented funds? That’s a tough nut to crack. If we consider all three levels of AI applications (back-office, front-office, investment processes), Gupta roughly evaluates that, out of a global US$100 trillion assets under management, 60% integrates AI at one level or another. But doing AI-assisted back-office settlements is very different from AI-augmented stock selection and doesn’t particularly throttle alpha. How many funds actually use AI to drive results. No one can really tell, though all the behemoth investment managers like BlackRock, State Street and Fidelity invest significantly in the technology.

In Canada, that search is even harder. “Large firms are investing heavily in AI to gain an edge, but they tend to be tight-lipped about it,” claims Gaetan Ruest, vice-president of data science and advanced analytics at IGM Financial. One key reason, believes AI specialist and University of Montreal professor of mathematics Manuel Morales, hinges on the fact that banks, which dominate the investment landscape in Canada, maybe fear a backlash on the part of an AI-weary  population if something went wrong with AI-augmented operations.

Among hedge funds, the AI contingent seems to be still microscopic. Preqin tracks a total of 305 such funds with only USD 17 billion in assets. Total hedge fund assets stood at USD 3.2 trillion in August 2017, according to the Hedge Fund Association, spread across 10,000 active funds.

Challenges Ahead
The road to AI implementation is rife with obstacles and risks, and rewards at the outset are far from obvious. “It’s difficult to determine the precise impact of AI – this is why adoption is slow, says Larry Cao, senior director of industry research at the CFA Institute and co-author of the study. You don’t really know until you finish an implementation. You need to get a lot of things right for it to be a contributor.”

Data sets themselves present risks. For example, a passive model designed by Vanguard to set up an ESG index ended up with gun manufacturers among the holdings. And results presented by AI models can be far from obvious. To begin with, many models remain black boxes: they don’t explain how or why they have reached their results. As Robert Pozen and Jonathan Ruane write in a Harvard Business Review article: “Many of the patterns machine learning identifies in large data sets are often only correlations that cast no light on their underlying drivers, which means that investment firms will still need to employ skilled professionals to decide if these correlations are signal or noise. According to a machine learning expert at a large U.S. investment manager, his team spends days evaluating whether any pattern detected by ML meets all of four tests: sensible, predictive, consistent, and additive.”

One thing is quite sure: AI is here to stay and destined to become the new norm. When that happens, adding alpha with AI will become more difficult because all AI models will be competing against each other. In the meantime, machine learning “surely has the potential to allow early adopters to find new sources of alpha and outperform the indexes,” says the Harvard study. For a while, the chase for outperforming portfolio managers could become a chase for outperforming AI-augmented funds.