A Personal Perspective on the Future of Discretionary Asset Management
By Michael Weinberg Managing Director, Head of Hedge Funds and Alternative Alpha, APG Asset Management
Published: 28 June 2021
If you were to ask a discretionary investment fund manager to draw a Venn diagram of the set notation where digitalization meets discretionary investing, many would not show the sets as overlapping. Moreover, they believe that the investment world is bifurcated and the discretionary managers are independent of the quantitative. I would argue that not only are the two sets converging but that the quantitative managers pose an existential threat to the discretionary managers, or perhaps will ‘Pac-Man’-style consume the discretionary managers.
In this paper, we will demonstrate how the ‘old school’ managers are under siege by quantitative managers by presenting the top-10 reasons discretionary investors tell me they are impervious to digitalisation, and refute them one by one.
1 – You can’t know what a black box will do
Discretionary investors will state that one can’t know what a black box, i.e. a systematic model will do. Moreover, they argue that it can be like the proverbial robot that runs amok, like HAL in the film 2001: A Space Odyssey. I would argue the contrary. Best-of-breed autonomous learning investment strategies (ALIS) managers are systematic strategies based on mathematical models, statistics and have rules and risk guidelines that are human encoded. This is no different to understanding what a discretionary investor does and why. The first ALIS fund I met over five years ago was predicated on models created by Benjamin Graham and David Dodd, the founders of value investing, philosophy of investment. Moreover, I would pose, what is a bigger black box than a discretionary investor’s brain? How do we know that person will follow rules, not deviate, make errors or lose one’s mind?
2 – AI can’t and won’t buy more down
Some discretionary investors contend that humans have the advantage of being able to buy more down if the thesis still holds, although, from my time as a portfolio manager at Soros Fund Management, I know that’s often easier said than done. Regardless, the best managers do this despite the innate wiring that makes this difficult from a behavioural investment perspective. However, if one has a systematic investment programme this can be automatic, not discretionary. Another comment I have heard is the fund’s terms will not allow this and investors will redeem at the bottom. The fund’s terms can be whatever one needs and should be cohesive with the strategy. If the strategy can suffer material draw-downs then the terms should be long enough and the liquidity staggered so that investors are not able to sell the bottom when they should be buying more. Incentives, i.e. fees, can even be structured so that investors invest more when the fund is down. Conceivably, a capital call structure could even be employed, to effectively ‘force’ investors to buy more down, when securities are most compelling. ALIS funds should be structured with terms that are cohesive with the investment philosophy just as discretionary funds should be.
3 – Systematic strategies don’t know about new product releases and related consumer sentiment
I’ve heard it argued that systematic strategies can’t know about new product releases. Well, they can and do. Systematic managers use natural language programming (NLP) to ‘learn’ about new product launches as well as how they are doing. One can also systematically monitor the consumer sentiment from new product launches. A discretionary investor may check a handful of websites, as I did while running a hedge fund portfolio. However, a systematic manager can check the sentiment on thousands (or more) of websites and aggregate that for a more accurate read very rapidly.
4– Computers can’t read newspapers
When I was running portfolios, I read major newspapers and a multitude of trade publications. Aside from the voluminous quantity of trees that were eviscerated for this, it was time-consuming and often tedious. A systematic strategy can and does effectively ‘read’ tons of papers and trade rags with NLP effortlessly and quickly without fatigue.
5 – An AI can’t judge company management quality
Discretionary investors often tell us that a system can’t judge management qualitatively as well as a discretionary manager can. I would contend a systematic manager can do that better. How is one judging management quality? How about using historic returns to shareholders as a proxy? Well, that’s easy: a systematic programme can easily look at thousands of stocks, their C-level management, using the CEO as the independent variable, and see what the returns to shareholders, the dependent variable, under their auspices. When I ran discretionary portfolios, we did this non-systematically, by noting which management teams generated excess returns for shareholders, and then investing in companies they went to subsequently, and the opposite when we were engaging in bearish investments. Instead of investing by ‘shooting from the hip’ — as my former partner and visionary Jeffrey Tarrant would say — systematising this could allow one to invest more statistically and consistently; at least ‘shooting from a brain’ even if artificial.
6 – They can’t tell if a company’s management is prevaricating
I will preface this one by saying we are not there yet and admittedly there are many assumptions built into this example, but, in my view, we could be there in no time. Through supervised learning and off-the-shelf web cameras, it is now possible to estimate heart rates from facial recognition videos. One instance where this could be applied is when a CEO reports a company’s earnings on a videocast. A ALIS manager could theoretically use that imagery-based forecast heart rate to determine if the CEO was prevaricating. This may have implications on the fund’s views on the veracity of the CEO’s statements and result in buy, sell or hold recommendations on the company. One could not only analyse the current imagery but could compare it to prior imagery by the same person for greater accuracy and determine recurring patterns on the speaker’s honesty. We are not there yet but it’s only a matter of time.
7 – A machine doesn’t have the perspective in comparing companies or industries
Despite this assertion, the ALIS fund based on Graham and Dodd’s work uses neural nets that do exactly this. Due to a large amount of structured financial data, record low processing and storage costs, the manager not only compares companies to their prior performance and their competitors, as well as companies in other industries. Many discretionary investors are too young to remember the Nifty 50, coincidentally now nearly 50 years ago. An ALIS fund might conclude that Apple is still a value stock because it is akin to Eastman Kodak during the Nifty 50 boom, or maybe the way it looked before the bust. There are very few active analysts today that could have traded through that period and remember how that stock, along with hundreds of others behaved then.
8 – Doesn’t know what metrics are important for out-performance
When I ran the portfolio at Soros Fund Management, we could only look at so many data points. Some income statement metrics, some balance sheet metrics, etc. We may think we know which metrics are important in determining the value of a company. However, I know another Graham and Dodd-style ALIS fund that looks at 10,000 data points per company and through machine learning has distilled that down to the 250 most important metrics. Again, with the confluence of record low processing and storage costs, data, data science and machine learning, this is easy for an ALIS manager.
9 – Can’t understand what is important in a 10Q or 10K
I remember printing and reading Qs and Ks, including, for example, Enron and Adelphia during the late 90s and early 2000s. It takes some time for a person to read and understand them, especially Enrons, which for those that don’t recall was hundreds of pages. For a computer using NLP, this can be done in no time and across thousands of securities. Discretionary investors may say that the computer won’t pick up on red flags. On the contrary, it’s easy to have the system pick up new terms related to litigation, reserves and increases or decreases in these numbers. Changes in tone can be picked up as well. Machines are now so widely applied to ‘reading’ 10Q and 10Ks, that company management is starting to adapt and change verbiage so that their language is ‘read’ by the machines as less negative, or even positively. For example, terms like ‘liabilities’, ‘litigation’ and other words, are now being extracted or replaced, with more euphemistic terms. This is a classic example of the iterative and evolutionary nature of markets, and why even systematic managers, must always be evolving, lest they suffer the same fate as discretionary managers.
10 – You can’t replicate a team of experts
Discretionary investors will assert that a systematic fund can’t replicate a team of analysts, traders, risk managers and portfolio managers. A discretionary investor who is long cheap stocks and short expensive stocks has done exactly this with decision trees and Bayesian techniques. Two PhDs — one of those who studied at CBS, where I teach — programmed an ALIS fund to have virtual analysts, traders and portfolio managers. The virtual analysts have the metrics they look for. The virtual traders act as risk managers and can sell the virtual analysts stocks if certain parameters are hit, i.e. the stock is within x% of the target or the company is reporting earnings and the street opinion is too one-sided. The virtual portfolio managers optimise position sizing and portfolio construction as a human portfolio manager would. As we have written in prior papers, these are all roles formerly held by humans, typically MBAs, now replaced by PhDs, coders and developers.
To end on a higher note, I will provide two constructive thoughts.
One: focus on the most inefficient and illiquid corners of the markets, such as emerging markets and non-exchange traded securities, that don’t lend themselves to these systematic techniques, yet. For example, Asian markets, various credit markets, select commodity. More illiquid markets may all have less widely disseminated and available public data, which in turn diminishes the ‘low-hanging fruit’ of systematising them.
Two: embrace digitalisation. If discretionary managers can not compete in more inefficient or illiquid markets then they must learn data science and/or coding or hire those that do. Discretionary managers should at least start to integrate alternative data into their investment processes.