Today I have some market intel to share with those money management firms that are building or running investment strategies that incorporate, or are completely based on, using Artificial Intelligence models to assemble and manage the basket of holdings that comprise their portfolios. Just as effective use of AI for running an investment strategy is in its infancy so too is marketing such products, particularly when it comes to selling to sophisticated investors such as family offices, endowments, foundations and institutional plan sponsors.
Here is some food for thought that may give you ideas as to where you might improve your firm’s communications with prospective investors.
It doesn’t take an economist to recognize this, but
As The Economist magazine wrote in a recent article: “Machine-learning funds have been around for a while and appear to outperform human competitors, at least a little. But they have not amassed vast assets, in part because they are a hard sell. After all, few people understand the risks involved. Those who have devoted their careers to machine learning are acutely aware of this…There was a time when everyone thought the quants had figured it out. That is not the perception today.”
And The Wall Street Journal, in an article about current AI-based investment strategies that use Big Data and natural language processing (NLP) went a step further, titling an article AI Can Write a Song, but It Can’t Beat the Market. [Note: That article was published after all but the previous sentence in this column was already written.]
Scuttlebutt from an investor conference / A discussion among investors
At a family office investor conference that I recently attended there was some interesting public discussion going on among the investors regarding AI-based investment strategies. It had to do with conducting due diligence on such quant portfolio managers and a shared skepticism among the prospective investors.
The issues weren’t that these investors were against the idea of AI-based investment processes across the board. They were fine with a manager incorporating Big Data and the analysis thereof as part of an investment process. They had no qualms about portfolio managers who might blend AI-based modeling with more traditional technical analysis, momentum or mean reversion-based investing. They took no issue, from a philosophic point of view, with a portfolio model based solely on AI.
Rather, these sophisticated investors were sharing frustrations with each other on how many of the pitching portfolio managers and their salespeople thought it was sufficient to explain how their AI-based portfolios worked in only a very generalized way — something no savvy equity portfolio manager or PE firm would ever try to get away with.
The problem had to do with blind faith. No, not the English supergroup from 1969 with Steve Winwood, Eric Clapton, Ginger Baker, and Ric Grech.
Instead, it was a ‘trust us’, ‘trust our model’ investment pitch. Question: Could this have been unintentional?
Shared skepticism among these investors was being expressed about the evangelizing portfolio managers and their salespeople and distributors who had been pitching them over the past year on their firms’ machine learning based AI strategies.
Observation: It is important to mention that there are other approaches to using AI-based modeling within an investment strategy in completely different, and interesting, ways; but none of that entered into the conversation this day and none of that has yet found its way into the press.
Back at the conference, one such seller at the event was telling the family office investors that his firm could build a customized, AI-run, machine learning strategy portfolio based on whatever investment goals they had. This machine learning model, he said, would do pattern recognition and would adapt as time went on to follow wherever high risk-adjusted returns were to be found.
When asked to explain some methodology detail about the process the seller came up short. It was state of the art machine learning, was essentially what the investor audience was told. (This turned out to be a failed attempt at a blind faith sell.)
Some vocal investors weren’t buying. A brief but telling dialogue among investor attendees ensued. It centered around the core beyond-the-numbers due diligence queries that family offices asked of any portfolio manager, no matter their asset class specific or investment approach focus.
It started with What is the actual investment process used in running the AI-based strategy?
This wasn’t asking for generalized ‘investment approach’ statements such as We run machine learning on Big Data or We take advantage of NLP. It was asking for workflow-like detail into how the model is constructed, what might be key factors among the data the model incorporates and analyzes, and how the model as it morphs over time is being tracked and analyzed, and its output understood by the portfolio management team.
Next came some back and forth between investors about performance attribution. What can the AI strategy’s generated returns be attributed to? One audience member raised the question of how they can make a determination about whether performance that an AI-based strategy is delivering is more likely due to luck or skill, if the portfolio manager doesn’t know and can’t explain where his returns came from.
Then someone brought up concerns about risk management. Anecdotally, it was appearing to be that any AI-based investment firms that had been having marketing conversations with some of these family office investor attendees had not communicated anywhere near the detail that stock portfolio managers, managed futures traders or PE funds had been providing when explaining managing their portfolios.
The concluding observation that was made by a very experienced family office investor elicited nods from the other investors in the room. It was directed to that seller in attendance at the event with the AI-run, machine learning strategy portfolio. To summarize the remarks, it was this: Common sense, and experience, has shown that when an investment edge comes from a portfolio manager using a new tactic, shrinking spreads and lower returns result as more investment managers get in on the act. As the machine learning strategies of multiple firms are left to their own devices, isn’t their trend seeking and calculations about things such as historical risk-adjusted returns going to start to converge on the same pattern analysis? This seems like it could become an automated way to get oneself into crowded trades and diminishing returns. How are the AI-based machine learning strategies your firm runs going to avoid falling prey to this drift towards underlying sameness with your peers?
The sophisticated investor universe knows, for example, that it has taken years of development using large language models to attempt to create a simulation of the nuance of language in AI Chatbots. So, they are going to be wondering how much more difficult it will be for this type of AI approach to identify the nuance of trading opportunities in the first place, and then, hope that a model’s output can be actioned upon quickly enough before every other similar AI strategy that has simultaneously ‘discovered’ the new trading opportunity does so too.
So, you can understand where the family office investors were coming from with their questions and skepticism.
It could be that the interactions some of these family office investors have had with portfolio managers running AI-based strategies was simply due to the investment firms not recognizing from the start that they need to provide more education and transparency into their thinking and investment processes in order to win over more sophisticated investors.
Hopefully AI-based quant managers will take these recent family officer investor anecdotes to heart and re-think how they go about educating and persuading people to understand and buy into how, exactly, their model driven strategies are being run.
Paint it blacker
But wait, there’s more.
Previously, Black Box strategies marketed to accredited investors tended to be of the type where the money management firm said, in essence the following: We know every step we are taking, every weighting we are giving every factor, and even which explicit steps in our process that are our secret sauce. We choose to keep all or some of that hidden from you and the rest of the outside world.
Now, there are some AI-based money management firms attempting to get buy-in from sophisticated investors to an even blacker black box. These are managers who are willing to let a machine learning model run a live portfolio of client money without knowing what the model is doing internally in real time, without knowing what might be making the model make the trades it is making, without knowing what the model is making risk management evaluations and decision making based on, and without knowing what to attribute the portfolio’s resulting performance to.
Such a fully ‘machine head’ focused approach for running an AI-based portfolio might produce acceptable returns for a while, but, from the way the dialogue among family office investors went recently, there will be significant concern among those prospects of there being a resulting smoke on the water risk incident with such strategies that includes loss of principal. Such money management firms will find they have even more challenging communications and sales marketing work ahead of them to win sticky asset clients.
# # #