Race is on to grab most possible from machine learning

The investment management industry is moving full tilt to capture the promise of alpha and efficiency offered by the application of machine learning to managing money.

Keen competition, slower growth, intense fee pressure and the resulting need for differentiation is driving active, passive, quantitative, fundamental, traditional and alternative managers alike to figure out how machine learning can improve all aspects of their businesses — from investment management to back-office reconciliation, observers said.

Putnam Investments, Boston, for example, is utilizing machine learning not only to automate operational tasks, but also to make the analysts on its actively managed, fundamental equity team more efficient through quicker access to better data, said Robert L. Reynolds, president and CEO.

"We're using machine learning to enhance returns, control risk and arbitrage information about stocks at a much faster pace," Mr. Reynolds said, noting "speed is essential" in using data effectively.

Putnam hasn't quantified the impact of machine learning on its equity investments, but Mr. Reynolds said performance has exceeded expectations.

The positive results of using machine learning in managing equities convinced the firm to begin applying the practice to its credit and quantitatively managed fixed-income strategies, global asset allocation approaches (including multiasset strategies) and target-date funds.

Putnam managed a total of $173 billion as of Feb. 28.

"The use of technology is a huge issue for money managers looking for ways to drive scale, efficiency and cost savings," said Jeffrey Levi, principal at Casey Quirk, a practice of Deloitte Consulting LLP, in New York.

The upper echelon of the world's largest money managers is in various stages of incorporating machine learning into their investment strategies. They include BlackRock Inc., Fidelity Investments, Goldman Sachs Asset Management LP, J.P. Morgan Asset Management and The Vanguard Group Inc.

Observers noted the machine-learning capabilities of traditional money management firms significantly trail the technological prowess of quantitatively managed hedge fund firms such as AQR Capital Management LLC, D.E. Shaw Investment Management LLC, GAM Systematic, Man Group PLC, Renaissance Technologies LLC, Two Sigma Investments LP and Winton Group Ltd., which collectively pioneered the use of artificial intelligence-fueled investment processes over the past two decades.

Critical advances over the past few years in computing power, newly available vast pools of data, and "very clever mathematical algorithms" are reasons traditional money managers and companies in other industries are rushing to adopt machine learning, said Matthew Killeya, head of research, Cantab Capital Partners LLP, Cambridge, England.

He stressed, however, the application of machine learning processes to investment management is very different from other industries because of the "very noisy" attributes of financial markets.

"There is much less agreement about structural patterns in global markets. It's much harder to recognize patterns than in other industries," Mr. Killeya added.

Cantab is part of GAM Systematic, which used machine learning to manage $4.7 billion in systematic quantitative strategies as of Dec. 31.

The common definition of machine learning is a computerized system that uses algorithms to detect and act on repeatable patterns it learns from large sets of data without human instruction about what patterns to find.

In the investment management context, at the simplest level, managers want to access, scrub and wrangle huge pools of data and let their algorithms spot patterns that will lead to better predictive pricing of stocks and bonds, said observers.

The massive data sets available for analysis span the spectrum from basic market and economic data to data from social media sources, satellite imagery, written documents and weather forecasts.

"We find that machine learning is very powerful in taking large unstructured data and simplifying it into something with more structure that we can more easily interpret," said Osman Ali, managing director and a portfolio manager on the quantitative investment strategies team of Goldman Sachs Asset Management, New York.

"You need a foundational, economically motivated belief for why something makes sense to invest in, and then you can see how analysis of big data can help in confirming or disproving that belief," Mr. Osman said.

GSAM managed a total of $1.29 trillion as of Dec. 31, of which $135 billion was managed by QIS in quantitative equity strategies and $35 billion of that was managed using machine learning.

Sources said the primary use of machine learning by traditional managers is for stock and bond analysis that provides portfolio managers with more precise information about securities from a much larger data universe than was previously available at a much faster pace than human analysts could ever achieve.

However, sources said only a handful of money managers are using machine learning meaningfully in portfolio construction, including Acadian Asset Management LLC, AlphaSimplex Group LLC, AQR Capital Management LLC, BlackRock, Cerebellum Capital LLC, GAM Systematic, Man Group, J.P. Morgan, Two Sigma Investments and Vanguard.

Casey Quirk's Mr. Levi observed that for most firms, machine learning is still in the early stages of being applied at the portfolio management level, noting "new data requires looking at portfolio construction through a new lens."

He did stress, however, that machine learning applications for true portfolio construction are on managers' radar, given "the huge opportunities" for managers in solutions-based strategies.

For example, forward-looking manager research now is focused on using machine-learning processes to break down return streams into their underlying components and recombine them in customized portfolios focused on specific factors to meet specific investor needs, Mr. Levi said.

At New York-based Two Sigma — a self-described algorithm-driven firm — management of the firm's hedge fund and long-only strategies is a synthesis of hypothesis-driven research and machine learning, said Alfred Z. Spector, managing director, chief technology officer and head of the engineering organization.

"Hypothesis-driven strategies assume an outcome while machine learning produces an idea for an investment," Mr. Spector said, noting the investment process is completely automated.

That said, "the combination of people and machine learning will always be essential" because humans with experience in financial markets must be in control, assessing the investment process, said Carter Lyons, managing director and Two Sigma's director of investor relations.

A huge infrastructure of massive computing power, multiple teams of human researchers, coders, modelers and tech support staff is required maintain Two Sigma's automated investment system $52 billion of assets under management and to "stay ahead of the competition," Mr. Spector said.

The adoption of machine learning "comes down to a belief that computers are more intelligent than people," Mr. Spector said, adding "it's really fun to be in a front seat watching this machine learning experiment evolve. It's really not known where this all will go."

Machine learning is a top priority for New York-based BlackRock across all functions needed to run $6.3 trillion of assets, including investments and operations. BlackRock's systematic active equity team ran $110 billion as of year-end 2017 using machine learning.

The SAE strategy uses machine learning to mine vast data sets to "get a better read" to predict returns, said Jeffrey Shen, managing director, chief investment officer of the active equity unit and co-head of the SAE strategy.

"Machine learning lets you ask about individual features of 15,000 stocks of different weightings in the global equity universe. Previously, analysts asked about a combination of factors of an individual stock, but machine learning now can take you from a view of the forest to a view of an individual tree," Mr. Shen said.

SAE team members are researching better portfolio optimization in terms of risk, returns, transaction costs and portfolio construction, Mr. Shen said, through use of a much broader set of portfolio parameters than previously were available.

The Vanguard Group, Malvern, Pa., like BlackRock, is employing machine learning across the company to "serve clients better," said John T. Marcante, managing director and chief information officer. Mr Marcante also oversees Vanguard's Information Security Team.

The firm is using machine learning to develop robotic automation "to do the repetitive, menial tasks people don't need to do," Mr. Marcante said. He said Vanguard expects to save 50,000 human hours a year by automating routine tasks throughout the firm.

Specific uses of machine learning for the firm's $1.2 trillion active internally and externally managed equity and fixed-income strategies include signal pattern recognition to "transform data patterns into investment insights" and speech recognition to search themes from investor calls, Mr. Marcante said.

The latter process, for example, "turns voice patterns into digital data that can be instantly analyzed," he said, noting analysis of client phone calls after the June 2016 British referendum that mandated the country leave the European Union uncovered concerns about the impact of the action on investments. Vanguard acted on the discovery and sent messages to all investors explaining potential global market reaction.

Mr. Marcante said Vanguard also is using machine learning to create customized portfolios for institutional and individual investors, as well as for defined contribution plans.

Vanguard ran a total of $5 trillion in internally and externally managed strategies as of Feb. 28.

J.P. Morgan Asset Management, New York, also is applying machine learning across its investment strategies, including the development of customized target-date funds for defined contribution plans, said Jed Laskowitz, head of intelligent digital solutions.

JPMAM manages a total of $120 billion in target-date funds and has been working with an external record keeper, which Mr. Laskowitz declined to name, to use machine learning to analyze investment behavior of participants in individual DC plans.

JPMAM portfolio managers then use the data analysis to create customized target-date funds using active and passive funds or a combination to meet the specific needs of the plan's participant population, Mr. Laskowitz said.

"Target-date funds are managed in five-year increments and it is absolutely essential to develop the right strategic asset allocation for the fund. Machine learning analysis of participant data hugely informs the development of the glidepath for these funds," Mr. Laskowitz said.

So far, JPMAM has analyzed data for 2 million individual plan participants for 400 companies. Mr. Laskowitz said he could not identify the defined contribution plans.

JPMAM managed $1.7 trillion as of Dec. 31.