Diversification Is Causing Active Managers to Underperform. AI Could Fix It.

Human portfolio managers — who have a limited number of good ideas — have a better chance of outperforming benchmarks when they run concentrated portfolios. But their efforts to manage risks by diversifying into hundreds of stocks are diluting their returns.

Data science techniques known as ensemble methods, which have long been used in other industries, could help preserve active managers’ alpha by providing a different approach to diversification, argue the authors of a paper that will appear in an upcoming edition of CFA’s Enterprising Investor. Using ensemble methods, which rely on artificial intelligence and machine learning, firms could use the highest conviction picks of multiple managers, rather than the single manager employed by most active funds, to provide downside protection.

“Concentrated portfolios have a statistically better chance of outperforming. Your best thought is valuable, but your fiftieth — not as much,” said Alexey Panchekha, a PhD in math and physics who co-founded Turing Technology Associates to focus on volatility management. “But you have to manage the risks of owning fewer securities, otherwise your firm could be killed during times of underperformance.”

Panchekha is the technology guru for a group called the Ensemble Asset Management Research Consortium, whose goal is to raise awareness about the benefits of applying ensemble mathematical techniques to active management.

Active managers as a whole have long turned to diversification as their primary risk management tool. The concept is simple: The effect of a few losing stocks is muted by holding a large portfolio. But a manager’s good ideas are also muted by diversification. 

“The very structure designed to protect managers actually hurts them,” said Matthew Bell, one of the paper’s authors and president of Bell Family Interests, a family office and consultant. 

[II Deep Dive: The Costly Line Between Winning and Losing in Active Management]

Ensemble methods, by contrast, diversify the so-called alpha engine by taking multiple managers’ best investment opportunities, removing the need to add extra stock picks to create a diversification cushion. Such a technique could be applied to the existing portfolio managers of a fund company or pension plan.

According to the paper, “the integration of multiple independent investment strategies through the application of ensemble methods techniques allows diversification of individual managers’ biases, and substantially reduces the potential for toxic tail events.”

As one example, the authors analyzed funds rated by Morningstar as gold, its top rating, in January 2017. The gold funds returned on average 0.9 percent annually after fees. When the authors adjusted the funds so they were concentrated versions of the same portfolios, the annual outperformance of the group was 4.2 percent, but the gain came with significantly larger risks. When ensemble techniques were applied, the funds returned 5.5 percent.

“As a board member I would listen to these managers and the approach has been, ‘I’m taking a bet, but not too big a bet; I don’t want to miss my passive benchmarks,” said Bob Tull, one of the principals behind the development of WEBS, the precursor to the iShares exchange-traded funds family now owned by BlackRock. “Here’s an option to reverse the asset flow out of active.”

Still, advisor Bob Willis warned that investors should be wary of innovations applied to back-tested data. Willis, the founder of Georgia-based advisory firm Willis Investment Counsel, noted that the study’s own footnotes caution that the portfolios were constructed on a historical and hypothetical basis and the time period analyzed was limited, dating back to July 2007. 

“We should all be appropriately skeptical of any purported breakthrough or silver bullet, and remember how often they look impressive on paper with their elegant mathematical equations, but frequently do not work in practice,” he said.