A new type of investment model has hit the scene, and its proponents believe it can potentially beat both passive and active management on a consistent basis. That’s a bold claim to live up to, but backtested data—and a brief run as live models—suggest it might have something to add to the investment conversation.
Melius Investments has introduced a platform containing 10 model portfolios that use artificial intelligence to identify the highest-conviction picks from successful active mutual fund managers, and then puts them into tidy packages designed to generate investment alpha.
The underlying principle behind these portfolios is something called ensemble active management, or EAM, which is based on mathematical techniques known as ensemble methods that strive to improve the accuracy of predictive algorithms.
According to a 2018 white paper from the EAM Research Consortium, this methodology links together multiple, independent predictive algorithms and looks for consensus or near-consensus agreement between them. “Ensemble Methods generate ‘multi-expert’ predictive systems, which have been proven to be superior to stand-alone ‘single-expert’ predictors,” said the paper, whose authors work in the fields of technology, exchange-traded funds and financial planning.
The white paper said ensemble methods date from the late-1970s and are considered a foundational approach for most AI and machine learning applications. It further stated that ensemble methods have been successfully used in applications ranging from facial recognition and self-driving cars to weather prediction and medicine.
To test the potential usefulness of this methodology with investments, researchers used ensemble methods to scour the portfolios of actively managed mutual funds and apply predictive algorithms designed to identify securities that are likely to outperform the market. They crunched data covering the period from July 2007 to December 2017, and according to the white paper came up with the following results that compared the algos versus both actively managed mutual funds and the S&P 500 Index:
• EAM portfolios outperformed the S&P 500 72% of the time over rolling one-year periods, with an average annual excess return of 3.4%.
• EAM portfolios achieved a 94% success rate versus the S&P 500 for rolling three-year periods, with an average annual excess return of 3.8%.
• EAM portfolios outperformed traditional active management 82% of the time over rolling one-year periods, and 95% of the time for rolling three-year periods.
Meanwhile, the white paper said, the average large-cap actively managed fund outperformed the average large-cap passive fund just once out of 255 rolling periods from January 2008 to December 2017.
Those numbers excited Tim Mullaney, founder and president of Melius Investments, who believes EAM offers a new way to invest.
“EAM combines the best of active management, but with insights and tools from the world of A.I. and machine learning,” he said. “I believe this is the next generation of active management.”
He posited that active management in its present form is fundamentally flawed because only a certain percentage of a particular fund’s portfolio is designed to generate returns. The other part, he said, is designed for risk management, and having too much in that bucket can drag down a fund’s performance.
The base case for passive management is that active managers aren’t good at consistently generating excess returns against their benchmark. Mullaney stated that’s an inaccurate perception.
“The research finds that active managers are very good in the portion of the portfolio designed to generate alpha, and that entails their high-conviction stocks,” he said. “Those managers are really good with that part of their funds, which we’ll call the alpha engine.”
He said Melius’ EAM portfolios are designed to be high-octane baskets containing only the high-conviction stocks from select traditional asset managers. The portfolios hold up to 50 positions gleaned from a dozen active managers who’ve demonstrated their ability to beat their benchmarks.
“We have the ability to replicate an actively managed portfolio, and our A.I. lets us dissect those portfolios and grab only their alpha engines and then employ ensemble methods to create a diversified portfolio at the security and investment-process level,” said Mullaney, who officially launched Melius (Latin for “better”) earlier this year as a registered investment advisor based in Laguna Niguel, Calif.
He formerly was a senior managing director at the boutique investment bank CastleOak Securities LP and is currently a registered representative at CAVU Securities LLC, a privately owned boutique broker-dealer.
Nuts-N-Bolts
Melius is responsible for six of the 10 model portfolios on its platform; the other four are licensed from Pegassets, a firm founded by Robert Tull, an ETF industry pioneer and named inventor on multiple ETF patents.
The 10 equity portfolios cover traditional assets classes, some specialty portfolios and four flavors of ESG (environmental, social and governance). The two firms have slightly different approaches to constructing EAM portfolios. Pegassets, for example, bases its portfolios on managers from 15 different funds versus 12 for Melius.
The data feeds for EAM portfolios come from Turing Technology Advisors, a fintech software company whose work includes data analytics and predictions. Turing co-founder Alexey Panchekha, along with Tull, where two of the authors of the EAM white paper.
As explained by Tull, the EAM investing model works like this: Turing downloads the daily closing net asset value of mutual funds that’s provided by Fund/SERV, a system launched by the National Securities Clearing Corporation that processes and settles mutual fund, bank collective fund and other pooled investment product transactions between fund companies and distributors.
“Turing has the technology to reverse engineer the holdings daily based on NAV changes and the profiling they’ve done on prospectuses, and they’re doing more than $4 trillion a day in assets,” Tull said. “That’s the big data component.
“Because the technology has improved since 2018, we now know within about a 99% accuracy on a daily basis what’s in those funds because you pull in all of the data of all of those closing prices from the exchanges, you pull in the daily NAV changes, and from that you can discern what they hold,” he added.
The second piece of this step is pulling out the high-conviction stock picks of specific groups of mutual funds.
“The models I’ve licensed to Tim had a list of 15 mutual funds that I selected because I know the managers and the strategies very well,” Tull said. “From there, we know what they hold, so which of those holdings are greater than the benchmark index or the prospectus. Then we build a consensus across that grouping of managers. From that, every two weeks Turing produces a select group of 50 stocks that have been chosen based on the managers, their expertise . . . and we leverage that through EAM to produce these portfolios.”
So while Turing grabs a ton of mutual fund data each day, the bi-weekly feeds it sends to Melius and Pegassets for their respective EAM portfolios are focused on just the active fund managers the two firms have selected.
“The reason it’s limited in number is because when we were doing a lot of early work on this we found there’s a law of diminishing returns by going past more than 20 managers in a grouping,” Tull explained.
Live Models
Nine of the 10 portfolios on the Melius platform have been running as live models for at least 10 months, and according to Melius all of them have produced net-of-fee excess returns versus their primary benchmarks through April 30. The outperformance ranges from 0.3% to 17.9%.
The newest model, the Melius Sector Disruptors that went live on February 20 just before the pandemic hit, was down 8.6% during the next two-plus months (though it still beat its benchmark, the Russell 1000 Growth Index, by 0.5% on a net-of-fee basis).
Tull said EAM portfolios capture about 84% of the downside and 128% of the upside of equity markets.
“When you have portfolios being updated every two weeks, you can capture the shift out of falling knives and move into stocks that aren’t as volatile or aren’t in the same sector that’s falling,” he said. “That adds a risk management component to the technology.”
Mullaney says the EAM portfolios allow for customization, along with tax-management overlays. And all of the portfolios on the platform charge a fee of 0.45%.
Investing With A.I.
A.I. is no stranger to portfolio management. In addition to ETFs that invest in companies that provide—or benefit from—A.I. and machine learning capabilities, ETF sponsors including BlackRock’s iShares unit, State Street Global Advisors’ SPDR franchise and EquBot (short for equity robot) are among the companies that have launched products that use machines to pick the investment portfolios. Depending on the fund, they employ analytic models that can vary from parsing information in public filings and earnings reports to how many times a company's name appears in news articles or on social media.
These algo-driven funds first hit the market three years ago, and by and large have been greeted by investors with a collective yawn. From an asset-gathering perspective, the most successful such product is the $964 million SPDR S&P Kensho New Economies Composite ETF (KOMP). A number of these A.I.-generated products still have less than $10 million in assets.
Meanwhile, the performance of these products has been all over the map, running the gamut from abysmal underperformance to shining outperformance versus the broader market. Among the stalwarts are the SPDR S&P Kensho Clean Power ETF (CNRG) and iShares Evolved U.S. Technology ETF (IETC), which have one-year returns of 35% and 29.5%, respectively.
That said, EAM's approach differs from these A.I.-generated portfolios. "We've taken the high-convicion stock picks out of regular mutual funds, so there is a research component that we're leveraging," Tull said. "And if someone believes in it and is willing to allocate money to it, that means a lot more to me than someone who counts how many times Boeing appeared in the news today."
Kate Sullivan, managing director of capital markets at Accenture, said her firm has spent a lot of time looking at A.I. at investment management firms, and there’s a lot of experimentation in the space. She noted that some firms are looking at it comprehensively while others are making bets on certain capabilities just to see if they can get returns.
“I think the firms that have been more successful have tried to take a specific asset class and identify triggers that align to that group to make sure they can generate something insightful,” said Sullivan, who declined to comment directly about EAM.
Sullivan said A.I. can help automate components of the due diligence and investment management process, but at the end of the day she believes a hybrid portfolio management system involving both humans and machines will be the winning ticket.
“There still needs to be a level of humanity as an overlay to provide a sanity check in terms of what the data science is predicting,” she said.
Regarding the EAM portfolios, Mullaney expects to add EAM model portfolios to its platform from firms other than its own and Pegassets. And he said he has agreements with two third-party marketing firms. Last week, Melius announced its EAM portfolios have been added to Smartleaf’s tax-overlay management platform to be available to its clients.
For his part, Tull said he’s having discussions with some ETF issuers about bringing the EAM concept to that market. And he noted they’re working on bringing EAM to the fixed-income investment world.
“It’s not ready for prime time; maybe in January 2021 we’ll be past the major challenges [posed by fixed income],” he said. “There are probably 380,000 components of the debt market in the U.S. versus 6,600 listed securities.”