AI going to head of the class

Enticed by the prospect of better outcomes from deeper analysis of vast amounts of newly available data, hedge fund managers are turning en masse to artificial intelligence and its subset, machine learning, to fine-tune their investment processes.

Long the tools of the largest, most sophisticated and well-heeled systematic quantitative hedge fund managers, AI/ML processes are making their way down to smaller quantitative and fundamental firms at a rapid rate, sources said.

Data-driven approaches “are still are in their infancy in many industries, not just investment management,” said Daniel Connell, managing director and head of market structure and technology at manager consultant Greenwich Associates, Stamford, Conn.

But based on recent advances in the investment arena, such as robo-advisers that provide asset allocation advice to individual investors based on machine-learning processes, Mr. Connell predicted “the growth in the adoption of AI and machine learning by money managers will be faster than we've ever seen before.”

A very small cadre of well-established managers dominate the ranks of quantitative hedge funds because of their strong record of harnessing big data with the use of machine learning. Observers said even quick implementation now of AI/ML will not help later-adopting hedge fund firms catch up.

Members of this rarified group include PDT Partners LLC, Renaissance Technologies LLC, D.E. Shaw & Co. and Two Sigma Investments LLC, sources said.

“Neither the technology nor the math behind artificial intelligence and machine learning is new. What's new is the vast amount of data that's available now,” said Anthony K. Caruso, vice president and head of quantitative and macro strategies, Mesirow Advanced Strategies Inc., Chicago.

By way of example, researchers at BlackRock Inc. noted that “where the methodology for incorporating analysts' views into the investment process once involved portfolio managers reading a small number of reports from their favored analysts that corresponded with their areas of investment responsibility, today's technology uses multilingual text mining to search the entire universe of analyst reports for new and potentially market-moving information,” in its report, “Finding Big Alpha in Big Data.”

Jeffrey Shen, the San Francisco-based managing director, co-CIO of active equity and co-head of BlackRock's scientific active equity strategy, was a co-author of the report and said in an interview that “an explosion of data sources and advances in computational power and speed allows for far deeper analysis for investment managers than was available one or two years ago.”

Mr. Shen noted, for example, that valuation signals can be collected for every single stock in the world, the analysis of which would be “impossible for any human, no matter how intelligent, to perform. But machines are really good at this.”

The granularity of the resulting analysis allows hedge fund managers to get both macro and fundamental insights about the data, Mr. Shen said, stressing AI/ML “allows you to be both broad on a macro level and deep on a fundamental level. You used to be able to be only macro or only quant.”

BlackRock's SAE unit manages $7 billion in hedge funds and liquid hedge fund strategy mutual funds using AI/ML techniques.

Hedge fund managers are expressing more interest in data-driven approaches than any other category of active money manager, said Greenwich Associates' Mr. Connell.

“It's down to the nature of the beast,” Mr. Connell said, adding “hedge fund managers are keenly interested in the ways they can add alpha to improve their tepid performance.”

Alpha hunting grounds for hedge fund beasts have expanded to alternative data categories including public website data (social media, applications and search); location tracking; consumer transaction data; and satellite imagery, according to research from capital markets researcher and consultant Tabb Group LLC.

Mesirow's Mr. Caruso said quantitative hedge fund managers equipped with AI/ML tools use them to troll diverse data sets by:

• evaluating credit card transactions to see which stocks might benefit from higher consumer spending;
• tapping satellite imagery of retail parking lots to gauge which stores are getting — or not getting — a lot of foot traffic, or to track the pace of shipbuilding in yards around the world;
• skimming data from location services in an effort to predict where people are shopping, eating or recreating; and
• applying filters to Twitter to check whether whatever's trending could result in a spike or decline in the popularity of certain goods and services.

Hedge fund managers that successfully use tech tools to produce alpha can be difficult for allocators such as institutional investors and hedge funds-of-funds managers like Mesirow Advanced Strategies to invest with, because they don't provide sufficient transparency about their investment process and their underlying holdings.

“What's really important to us is ample transparency. We like to know what we own,” Mr. Caruso said. He added that “when quantitative hedge fund managers find a really good alpha signal from their data analysis, they tend to charge a lot for it,” in ranges as high as a 2% management fee and 30% performance fee to a 3% and 30% formula.

Hedge fund seeder and funds-of-funds manager Jeffrey Tarrant, founder, CEO and chief investment officer of Protege Partners LLC, New York, spent the last three years investigating and evaluating the efficacy of AI/ML-based funds. He's now tracking 70 startup hedge fund firms that use AI/ML strategies to create what Mr. Tarrant calls automated investment learning strategies.

“These are truly new process-driven approaches that combine both fundamental and quantitative insights from data analysis. It's about machines teaching each other and teaching the human,” Mr. Tarrant said.

These quantitative AI/ML specialist hedge funds generally have not publicized their existence and are “very tiny little firms that use unconventional investment processes and are definitely not ready for prime time with regard to institutional investment,” Mr. Tarrant said.

Protege Partners manages or advises on $2.8 billion of assets invested in hedge fund seed funds or hedge funds of funds.

Veteran industry watchers like Mr. Tarrant wonder — as he said — whether “old quant and fundamental discretionary managers are going to survive” or whether the hedge fund industry will suffer from disruptive new strategies. Mr. Tarrant pointed to the way the medallion taxi business has suffered since Uber Technologies Inc. popped up and began to steal business as a potential model for the hedge fund industry.

The future will favor data-driven hedge funds, said Norman Kilarjian, head of macro and quantitative strategies at hedge fund consultant Aksia LLC, New York.

“I deeply believe that in 10, 20 or 30 years, we will realize that computers are better than humans at allocating capital. It's just a matter of time,” he said, warning, “if they're smart, today's stock-pickers are going to incorporate more artificial intelligence and big-data analysis into their processes. The lines between the quants and fundamentally oriented managers will be less distinct.”

Mr. Kilarjian and others warned of the challenges faced by the firms that are playing catch-up, seeking to add AI/ML to their existing strategies.

“This is one area of the hedge fund industry that has significant barriers to entry when it comes to the financial engineering, infrastructure and technology needed to succeed. This is where the larger, longer-established quant managers do have an edge,” said Kevin Lenaghan, managing director and head of market-neutral strategies for specialist hedge fund consultant Cliffwater LLC, Marina del Rey, Calif.

Because of the requirements of these strategies, including statistical arbitrage, market-neutral and systematic managed futures approaches, quantitative hedge fund managers were among the first — and likely will remain the largest group — to use machine learning “to reduce the reliance on human influence” in investment management, Mr. Lenaghan said.

He stressed that as efficient as AI/ML processes are, “investing still requires an element of skill and luck and human intuition and oversight still is necessary for processes such as awareness of the factor exposures of a portfolio. A human hand is needed to maintain dynamic risk management systems.”