By S.P. Kothari, Gordon Y Billard Professor of Accounting and Finance, and Robert C. Pozen, Senior Lecturer of Technological Innovation, Entrepreneurship, and Strategic Management, MIT Sloan School of Management
Investors in the United States have poured trillions of dollars into actively managed funds. According to Morningstar, as of March 31, 2022, retail investors had invested $8.34 trillion worth of assets in actively managed funds, while $8.53 trillion was invested in index mutual funds.1 (Vanguard describes index investing as funds “designed to keep pace with market returns because they try to mirror certain market segments” compared to actively managed funds, which “try to beat market returns with investments hand-picked by professional money managers”.2) Finding alpha or beating the benchmark is the Holy Grail of all active asset managers. However, active managers in most asset categories have found it challenging to outperform the indexes, even when markets are falling.3 Research suggests the reason for active managers’ lackluster performances is that “when the industry grows, so does competition, especially from new funds. It is harder to outperform in a bigger and more competitive industry”.4
In their quest for alpha, asset managers are increasingly turning to artificial intelligence (AI) and machine learning (ML). In layman’s terms, AI is “the field of developing computers and robots that are capable of behaving in ways that both mimic and go beyond human capabilities”, and ML is a subcategory of AI that “uses algorithms to automatically learn insights and recognize patterns from data, applying that learning to make increasingly better decisions”.5
While AI and ML may initially improve the results of active managers, the new strategies can be easily copied, so they may not create a permanent advantage. This is the paradox: Active managers will have to devote substantial resources to AL and ML to keep up with the pack. The more sustainable advantage for active managers is top-quality service—highly valued by clients with significant assets, especially high-net-worth individuals and families. If such a service is customized for these clients, it will not be readily copied or produced as a standardized commodity based on scale.
The benefits and limits of AI for investing
The allure of AI in investing lies in its potential for finding new alpha in at least three ways beyond traditional quantitative-investment methods. (The discussion on the potential benefits of AI draws from Pozen and Ruane, Harvard Business Review, December 2019.6)
Identify new patterns in past data. Investment strategies in quantitative investing typically seek to exploit inefficiencies in the market by analyzing historical security price data and other sources of information—e.g., changes in a company’s inventories or analysts’ forecasts. To develop a quantitative-investment strategy, a researcher typically hypothesizes an inefficient economic relation that can be exploited. For example, a hypothesized inefficiency might be something as simple as the market underreacting to positive-earnings surprises.
By contrast, ML algorithms can be trained to spot historical patterns in security-price data even without specifying a hypothesized economic source of inefficiency attributable to inefficient pricing. However, the vast computing power that ML exploits to identify patterns in data is also a negative. Absent an economic hypothesis predicting a correlation in past security-price data, ML runs the risk of identifying correlations that are noise rather than signals of exploitable investment opportunities. That is, the portfolio manager must appreciate the difference between hypothesis-driven computer analysis and ML, wherein there is no hypothesis—just pattern identification that needs to be sorted out by the portfolio manager.
Analyze new forms of data. ML techniques have demonstrated their ability to analyze many forms of data that were previously beyond the capabilities of humans and quantitative-data analysis techniques. For example, ML algorithms called deep learning can readily process millions of images and sounds to detect certain patterns or categorize images according to certain similarities or dissimilarities. Natural language processing (NLP) and textual analysis (TA) methods can analyze large volumes of texts for intensity and amounts of different types of disclosures—e.g., anticipated performance, bankruptcy risk, environmental and other types of risks, the revenue potential of intellectual property and a firm’s strategic direction.
ML can be trained to capture and analyze other unconventional data, such as management’s private-jet travel to draw inferences about potential merger and acquisition (M&A) activity, customer traffic to retail outlets or ship traffic through ports to gauge a company’s or nation’s export volume.
Overcome human biases in investment decisions. Drawing on the literature in psychology, behavioral economists in finance convincingly suggest that individuals exhibit an array of information-processing biases that significantly affect their investment decisions. For example, individuals under- or overreact to new information, suffer from confirmation bias (the tendency to interpret new evidence to affirm pre-existing beliefs) or allow their behavior to be unduly influenced by their early-life experiences (imprinting phenomenon). Because a large swath of investors likely share these biases, they have the potential to influence security prices. That is, prices could deviate systematically from those warranted based on underlying economic fundamentals.
Thus, these biases create a profit-making investment opportunity that exploits the systematic mispricing of securities. ML can correct for the influences of these biases on a portfolio manager’s potential investment decisions or identify mispricing of securities resulting from biased trading decisions by other investors.
Implications for scale. While AI can enhance the investment results of an active manager, it is a technology-intensive, science-driven exercise. An asset manager dealing with thousands of securities and clients must rely on sophisticated technology-based architecture. Compliance with mandates imposed by federal and state regulators is an additional reason why reliance on technology is essential.
To finance the technological infrastructure necessary to support AI, some active managers will build scale—by gathering large amounts of assets, either organically or by acquisition. By contrast, small asset managers will not be able to afford to build the technology infrastructure needed to support AI-investment methods. Small managers will suffer from a scale disadvantage unless they outsource the technology infrastructure for AI from vendors that specialize in these services. This would level the playing field for asset management and enable a broad array of firms to obtain the benefits of AI for active investing.
AI will be commoditized. Despite the foregoing benefits of AI in making investment decisions, a highly competitive asset-management industry yields the sobering implications of using ML to uncover profitable opportunities. More and more portfolio managers are likely to employ ML to identify patterns in data as signals of investment opportunities. ML might uncover patterns in data because it is trained on new types of data sources or because ML is programmed to counter the effects of behavioral biases affecting security prices. However, there is no intellectual-property protection for ML-based investment strategies to the exclusion of others. Any success of these strategies will only induce others to copy them quickly. Moreover, the trading activities resulting from managers acting on the ML-based signals will move prices, thus eliminating any further profit opportunities. Because ML and technology also enable high-frequency trading, the profit opportunities are likely to dissipate within milliseconds, if not seconds.
To remain competitive, it is imperative that asset managers arm themselves with ML to guide their portfolio construction and trading decisions. However, in a competitive market, the benefits from ML are likely to be copied and commoditized—i.e., they are unlikely to boost their investment performance on a permanent basis. The resource-intensive use of ML as an investment strategy reminds us of the aphorism, “You have to run fast on a treadmill just to stay even”. That is another paradox: AI might enhance market efficiency, but asset managers won’t get paid for it!
Service, the savior. Fortunately for asset managers, performance is not the only attribute clients value when choosing their investment advisors. Clients, particularly high-net-worth individuals and families as compared to institutional investors, value good service, and service cannot easily be commoditized. Successful asset managers win the trust of their clients—which is the bedrock of the professional relationships between asset managers and their clients. In a high-quality relationship, the manager offers the client a broad array of customized services: advice on personal wealth management, overviews of macroeconomic conditions, timely updates and words of comfort when large shocks rattle the market, opportunities and pitfalls in various investment alternatives, strategies for retirement planning, and referrals for preparing wills and trusts. Wealthy clients receive additional benefits that often include highly personalized concierge services. The average fees for wealthy clients are about 1 percent of their assets under management (AUM), declining as the assets under management increase.7
Unlike portfolio construction and management, the service element of the package that an asset manager offers to a client is not scalable. Technology and ML can help improve the efficiency of the service component, but service to the client is often high-touch and customized to the client’s needs. In 2020, there were more than 31,000 registered investment advisors and more than 600,000 registered representatives in the United States.8 These statistics show that even if there are huge asset managers such as BlackRock, Vanguard, Fidelity, J.P. Morgan and Bank of America (BoA), the financial industry employs many professionals who offer personalized services in addition to money management.
Moreover, retail clients who receive excellent service from their asset managers seem less demanding about maximizing investment performance. The typical portfolio of a high-net-worth investor will hold 50 to 70 percent in equities and 30 to 50 percent in bonds. The bond component is usually structured as a ladder of bonds maturing at different dates, which are held to maturity. Such a bond ladder is not easy to compare to a performance benchmark. Equities are often invested in exchange-traded funds (ETFs) or indexed mutual funds, where average performance is accepted. Even when the asset manager tries to pick winners, wealthy clients seem to tolerate subpar investment performance as long as it is not too bad.
It’s true that a few robo-advisers have used technology-based service models with low fees and ETFs, which did not offer any opportunity for clients to speak with an actual service rep. However, these robo-advisers have been forced by competition from Charles Schwab Corp., Vanguard and others to offer premium services at higher fees. When clients sign up for these premium services, they can talk with financial advisers about their asset allocation, tax planning or other aspects of their investment situations.
Conclusion
While AI methods initially produce more alpha, they can be easily copied and, therefore, will not bestow a permanent investing advantage. Large asset managers will devote substantial resources to developing AI capabilities just to stay even with the competition. Small asset managers will either rely mainly on indexed products or pay for AI tools from independent vendors with sufficient scale.
To produce a sustainable advantage, asset managers should focus on delivering high-quality, customized service. Such service is not easily copied and does not scale readily. The appetite for personalized service remains strong among the most valued clients—high-net-worth individuals—who seem to tolerate middling investment performance if combined with excellent service.
References
1 Yahoo! Finance: “‘The democratization of investing’: Index funds officially overtake active managers,” Allan Sloan, May 22, 2022.
2 Vanguard: “Index funds vs. actively managed funds.”
3 Barron’s: “Active Fund Managers Are Expected to Win in Rough Markets. It Isn’t Working,” Karishma Vanjani, September 22, 2022.
4 Chicago Booth Review: “Why Active Managers Have Trouble Keeping Up with the Pack,” Ronald Fink, July 3, 2014.
5 Columbia Engineering: “Artificial Intelligence (AI) vs. Machine Learning.”
6 Harvard Business Review: “What Machine Learning Will Mean for Asset Managers,” Robert C. Pozen and Jonathan Ruane, December 3, 2019.
7 Kitces: “Financial Advisor Fees Comparison – All-In Costs For The Typical Financial Advisor?” July 31, 2017.
8 Barron’s: “Industry Consolidation Takes a Toll on Number of U.S. Brokerages,” Andrew Welsch, May 4, 2022.
Robert C. Pozen is currently a Senior Lecturer at MIT Sloan School of Management and a non-resident Senior Fellow at the Brookings Institution. Prior to his current post, Bob was the Executive Chairman of MFS Investment Management, which now manages more than $400 billion for mutual funds and pension plans. Before MFS, he was the Vice Chairman of Fidelity Investments and President of Fidelity Management & Research Company.