This article is derived from “Ensemble Active Management – AI’s Transformation of Active Management” and “Methodology, Design, and Data Integrity Validation Study of Turing Technology’s 2024 Ensemble Active Management” white paper.
Numerous studies have evaluated active US equity managers’ ability to outperform index funds and exchange-traded funds (ETFs). While time horizons vary, the results tend to converge on the same result: Active managers outperform standard benchmarks less than half the time. Adding to the headwinds, active investments are structurally more expensive than their passive counterparts.
Investors don’t want to pay more for equivalent returns and are voting with their wallets. As of year-end 2023, actively managed US equity funds have experienced 18 consecutive years of net outflows totaling more than $2.5 trillion, according to Morningstar Direct.
The required leap for active to once again outperform passive cannot be accomplished through incremental gains. The gap is simply too large. For active management to acquire sufficient alpha to achieve a step-change improvement, a paradigm shift driven by new technologies and new methods is required.
That’s where Ensemble Active Management (EAM) comes in. EAM is built on critical new technologies and employs a stock-selection approach mirroring other industries’ best practices for conducting complex decision making. It pivots from a single manager to a multi-manager approach. In short, EAM represents the paradigm shift necessary to revitalize active management.
EAM is not an academic concept. It was first introduced in 2018 and EAM portfolios launched later that year. There are now dozens of EAM track records that range in age from two to five years.
This paper lays out EAM’s construction mechanics and presents three critical validation pillars that support EAM’s results to date and explain its future potential.
The data shows that as of year-end 2023, live EAM portfolios represented the country’s strongest lineup of actively managed US equity portfolios.
![Graphic showing Three Pillars of Ensemble Active Management Validation](https://i0.wp.com/blogs.cfainstitute.org/investor/files/2024/03/EAM-Three-Pillars-Validation-Graphic-2.png?resize=640%2C408&ssl=1)
Ensemble Active Management Defined
EAM must harness substantial added alpha to outperform both traditional active and passive management. To accomplish that, we apply the proven mathematics of Ensemble Methods to portfolio management.
Ensemble Methods feature a multiple-expert system that improves the accuracy of single-expert predictive algorithms or engines. This is accomplished by mathematically integrating multiple predictive models based on consensus agreement. The end result is a stronger predictive engine. Ensemble Methods are thus an artificial intelligence (AI) version of the “wisdom of experts.”
For clarity, EAM does not employ Ensemble Methods to design a “smarter” portfolio manager. In fact, a defining principle of Ensemble Methods is its use of multiple predictive engines. Instead, EAM generates active security selection by integrating a multi-investment-manager platform through Ensemble Methods.
Actively managed mutual funds work within an Ensemble Methods environment because they effectively operate as predictive engines wherein managers try to “predict” which stocks will outperform. Further, substantial research shows that managers’ highest conviction stock picks do reliably outperform.
EAM’s breakthrough came from the discovery of how to extract a fund’s “dynamic predictive engine” from its real-time holdings and weights. Turing Technology accesses this data through its machine learning-based fund replication technology, Hercules.ai. Launched in 2016, Hercules.ai provides real-time replication of actively managed funds. It houses data representing more than $4 trillion in assets and achieves a 99.4% correlation between the replicated fund returns and the actual fund returns.
To build EAM portfolios, 10 to 12 quality mutual funds are selected from a similar investment category. Turing extracts each fund’s predictive engine by accessing its real-time holdings and weights, and then maps that data against the benchmark’s weights. The relative over- or underweight positions reflect the funds’ predictive engines.
Turing then deploys these extracted predictive engines within the Ensemble Methods mathematical “engine” to generate the EAM portfolio. The final result is a portfolio of up to 50 stocks, with no derivatives, no leverage, and all holdings represented in the benchmark. EAM therefore constitutes the “consensus top picks of a dozen quality managers.”
![Ensemble Methods to EAM Portfolio Diagram](https://i0.wp.com/blogs.cfainstitute.org/investor/files/2024/03/EAM-Ensemble-Methods-Diagram-2.png?resize=640%2C339&ssl=1)
Further Understanding of Ensemble Methods
The subset of machine learning known as Ensemble Methods is the key to creating new sources of alpha. Ensemble Methods are integral to nearly every major computational challenge in the world, and Giovanni Seni and John F. Elder have described them as “the most influential development in Data Mining and Machine Learning in the past decade.” There are more than 250,000 published applications of Ensemble Methods, including facial recognition, early autism detection, MRI-based tumor detection, cyber threat detection, and many more.
Scaled Research: 2024 EAM White Paper
The following data are excerpted from “Ensemble Active Management – AI’s Transformation of Active Management,” the largest study ever conducted to measure the performance potential of EAM.
- 60,000 randomly constructed portfolios of 12 funds each were built.
- 60,000 EAM portfolios were constructed based upon the sets of 12 underlying funds.
- Results were evaluated over 2016 to 2022.
- 333 underlying funds were used from more 140-plus fund companies representing more than $3 trillion in AUM. These funds account for more than 60% of the assets of the active US equity universe.
- The study covered Large Value, Large Blend, Large Growth, Small Value, Small Blend, and Small Growth style boxes, or 10,000 EAM portfolios per style box.
To put the scale of this research effort into perspective, 420,000 discrete calendar year performance returns were generated (seven years each, from 60,000 portfolios). This is 20 times larger than the number of discrete calendar year returns delivered by the entire active US equity industry for the past 25 years.
The results are statistically significant, and were subjected to an independent academic revie, verifying the study’s methodology and results.
Performance Comparison vs. Standard Benchmarks
The study compared the performance of the 60,000 EAM portfolios versus their corresponding benchmark (the Russell Indexes), based on rolling one-, three- and five-year periods, as well as the full seven-year window. The results, derived from more than 560 million total data points are presented in the chart below.
Two of the key metrics were Success Rates and average annual excess returns. The former measures the percentage of rolling time periods that the EAM portfolio outperformed the benchmark, with the average annual excess return reflecting the average of all rolling period relative returns.
EAM Success Rates vs. Benchmark, Average of All Style Boxes
![Bar chart showing EAM Success Rate vs. Benchmark, Average of All Style Boxes](https://i0.wp.com/blogs.cfainstitute.org/investor/files/2024/03/EAM-Success-Rates-vs-Benchmarks.png?resize=640%2C472&ssl=1)
An important aspect of the Success Rates is that EAM outperformed across all six of the evaluated style boxes:
EAM Success Rates vs. Benchmark by Style Box
![Pie charts showing EAM Success Rates vs. Benchmark by Style Box](https://i0.wp.com/blogs.cfainstitute.org/investor/files/2024/03/EAM-Success-Rates-vs-Benchmark-by-Style-Box-2.png?resize=640%2C382&ssl=1)
Independent Review and Validation
David Goldsman, Coca-Cola Foundation Professor and director of master’s programs at the Georgia Institute of Technology’s School of Industrial and Systems Engineering, conducted an independent review to validate our research. His academic team had full access to the study’s methodology, stated biases, input and output data, and even code. Their review took several months to complete.
Key Excerpts
“We found that the underlying methodology is sound. Standard sampling/randomness protocols were followed, appropriate randomness protocol for the underlying POF [proof of funds] construction was carried out properly, EAM analytics and construction methodology was performed properly, and EAM and POF performance has been properly interpreted by Turing, including bias analysis and mitigation.”
“Across all portfolio fund style boxes the EAM portfolio has an overall expected performance benefit of 400 to 500 bps when compared against the corresponding [. . .] benchmark.”
“Our summary conclusions are that EAM and POF performance has been properly interpreted by Turing, including bias analysis and mitigation. Turing’s claims that EAM performance is comparatively better than traditional active management and standard industry benchmarks were also substantiated.”
Use and Impact of Fees
The study measured benchmarks as publicly reported, without fees or transaction costs, and calculated the EAM Portfolio performance in the same manner as the indexes.
As a reference, a simulation calculation was done of the Large Blend category wherein the EAM Portfolios’ returns were reduced by the maximum fee charged by Turing (25 basis point (bps)). The success rate of EAM vs. the underlying Portfolio of Funds was 71.5% without any added fees and fell slightly to 69.7% with the maximum fee factored in.
EAM Portfolios Live Returns
Several dozen EAM portfolios have been built by Turing clients and are commercially available for use by outside investors. All performance track records are independently verifiable.
A useful approach to evaluating EAM strategies is to establish a “lead” portfolio for each of the nine standard US equity style boxes. The results as defined by average annual excess return from the inception date of each strategy through year-end 2023 versus the corresponding benchmark, as a percentage in green, are presented below.
EAM Portfolio Performance Relative to Benchmark
![Chart showing EAM Portfolio Performance Relative to Benchmark](https://i0.wp.com/blogs.cfainstitute.org/investor/files/2024/03/EAM-Portfolio-Performance-by-Style-Box-1.png?resize=640%2C285&ssl=1)
Two key conclusions emerge:
- All nine lead EAM portfolios outperform their benchmarks. The average age of the EAM strategies is three years.
- The average annual excess returns equal 516 bps.
To put these results into context, we evaluated all actively managed fund families in the industry, focusing on mutual funds that fall into a standard US equity style box. Nine “lead” funds were identified for each fund family, reflecting the best-performing fund for each style box based on three- and five-year relative performance compared to the corresponding benchmark.
No other fund family had nine out of nine “lead” funds outperform the benchmark and average annual excess returns exceed 500 bps.
Thus, in just a few short years, EAM portfolios, powered by Ensemble Methods and a multi-manager platform, have emerged as the strongest lineup of actively managed US equity funds in the country.
Conclusion
Eighteen consecutive years of net outflows demonstrate that active management has long stood on the wrong side of history due to its chronic underperformance. And yet the status quo remains the defining state of the industry. Skeptics rail against “yet another” new idea, but skepticism does not need to ignore provable facts. EAM’s investment design and its application of enabling advanced technology and hard performance data should make EAM the guiding light for active management.
The stakes are high for the industry. Given long-term negative flows and now weakening company valuations, the adage “adapt or die” should be the rallying cry of the industry. AI is providing a helping hand, and active management would be wise to embrace the assist.
If you liked this post, don’t forget to subscribe to Enterprising Investor and the CFA Institute Research and Policy Center.
All posts are the opinion of the author. As such, they should not be construed as investment advice, nor do the opinions expressed necessarily reflect the views of CFA Institute or the author’s employer.
Image credit: ©Getty Images / Olemedia
Professional Learning for CFA Institute Members
CFA Institute members are empowered to self-determine and self-report professional learning (PL) credits earned, including content on Enterprising Investor. Members can record credits easily using their online PL tracker.