How might AI impact investment management?

The adoption of Artificial Intelligence (AI) is not a new phenomenon for the investment industry – but the accelerated and profound recent developments suggest it can be in a powerful, business-changing tool.

1. The AI acceleration can transform laborious, yet fundamental tasks such as processing company data. This has implications for the types of roles required,
2. ‘Terminator’ scenarios, where AI tools wreak havoc on markets are overstated but not impossible. It’s likelier that AI tools will be able to extract deeper insights and correlations.
3. Product innovation is likely as AI can assimilate complex multiple risks and make counterbalancing trades to both sidestep losses and take advantage of opportunities.

Chat GPT is not even a year old, but these kind of AI tools have gone from quirky apps which help lazy students shortcut essay-writing, to potentially capitalism-transforming tools.

For those of us in the investment industry, the wheels began to turn immediately as to the many potential areas of the business where large-language-driven AI models could have an impact. Generative-AI models are in the early days of scaling, but we’ve started to see the first batch of applications across various industry functions:

• Legal & Consulting. AI models can ingest vast amounts of legal documentation or corporate due diligence, and draft synthesized summaries or detailed briefs.
• Software & IT. Generative AI can write, review, and codify software code for a range of applications.
  
• Research & Development. Generative AI can expedite the research and discovery process as it relates to drug or other chemical-based relational fields.
  
• Sales & Marketing. Generative AI can develop text, images, and video used in social media campaigns, technical sales collateral, and personalized marketing.
• Manufacturing & Operations. AI models can identify critical paths and develop optimal processes in various manual and specialized tasks.

These tasks represent real jobs and not just menial, laborious roles. GenAI can handle the time-consuming tasks that also involve a deep understanding of the topics at hand and high levels of cognitive reasoning to best organize and synthesize. This includes many of the entry level analyst and associate jobs one starts their careers with after leaving university.

When properly “trained” on the appropriate data to avoid “hallucinations” (AI output errors that appear legit), AI can consume large quantities of content and create lucid summaries and even forward-looking strategic suggestions. It has the ability to even the playing field between firms with large teams of analysts and smaller shops that have equally talented senior partners and strategists and/or proprietary IP but lack the staffing size.

In the investment industry, bigger has never necessarily meant better. There are many examples of smaller investment and advisory shops that stick to their core strengths and serve niche markets very well. When you find a fund manager who is really good and keeps their assets under management low so they can maximize their picks and not dilute their performance, you stick with them. Unfortunately, there aren’t many of them out there, they can only take on limited new investors, and they tend to get bought out by the bigger shops over time. AI can help asset and wealth managers of all sizes—in fact, it has been to varying degrees for years—but we’ll likely start seeing more of it in new roles.

Analysts are likely to benefit in a number of ways as the NLP and generative capabilities of AI will be able to handle many of the market, sector, and industry level reports that both sell-side and buy-side analysts create. They can also compile the company level reports and suggest potential investment theses based on the investment philosophy and process on which it was trained or “tuned.” The analyst can then run through the suggestions to ensure it’s reading the right signals, but the rote exercises of compiling the data and populating valuation models can be largely automated.

Theoretically, this can give the analyst more time to conduct broader and deeper coverage. Investing purists will argue that the exercise of up-and-coming analysts creating these reports enables the deeper understanding of how the parts fit together to form an investment narrative—and they’re not wrong. But once an analyst moves on to be a PM, they have that understanding and, as a PM, they don’t often crank out reports anymore. Therefore, one can see how AI can level the playing field between asset managers with large analyst staffs and smaller shops with experienced PMs.

We talked about analysts, so how about PMs? Is it possible to reduce the dependence on one of the most expensive parts of the asset management value chain? For purely passive products, PMs have always had more of an operational oversight role, but active PMs get paid handsomely for results that have a middling performance record versus their passive peers. That said, the few that are good tend to be consistently good and worth every penny, but even they could benefit from scaled research. Active PMs must articulate their process, philosophy, and buy/sell disciplines—particularly for their institutional clients—and, one could argue, those disciplines could be coded to “train” a LLM on a large data set of companies. Theoretically, that “AI PM” could cover substantially more companies in more depth than the largest asset management shops. AI could also effectively crunch decades of data, then assimilate it over time, market cycles, trends, and vis a vis macro events. With this historical relational perspective in hand, it can then suggest trading and investing ideas “tuned” to current market dynamics as well as early warning signs of risk. It could also find opportunities humans don’t have the time or stamina to find.

• Security selection
  
• Tactical and strategic portfolio construction
  
• Risk management
  
• Assessing sentiment
  
• Tactical trading strategies (for entering and exiting)
  
• Auto and algo driven trading (for additional alpha)
  
• Sales and marketing

To be clear, I am certainly not advocating that we are at a point where active portfolio managers should be worried about their jobs. In fact, effectively weaving AI into an active PM role would best be done with the live manager overseeing the outputs to ensure it doesn’t stray off message as it continues to “train.” That said, it’s arguable that AI can take the value-add PMs bring to security selection and broaden the universe from which they draw. It’s easy to understand why active managers are already incorporating AI across these use cases.

Potential downsides of AI on investments are at the core of the iconic movie “The Terminator”—i.e. machines behaving badly. One reason why all the aforementioned positives of AI in investing include some form of human oversight is that these self-learning models can hit a weird fork in the road of their cognitive development and potentially go wildly off track. We’ve already seen examples of how bad data can train a GenAI model to produce believable but patently wrong outputs.

In a world of corporate financial reporting and disclosures where accounting is somewhat of an art and fluid in its application from company to company, the chances of an AI engine to “hallucinate” from these data vagaries are very real. Toss into the mix “irrational” human beings and you have some real hurdles for training an AI model. Human behaviors confound models based on historical patterns routinely. Also, a model’s “learning” could innocently navigate it down a risky path that no human has ever conceived of before, so it may be hard to spot it happening. Consider that few investment and market professionals necessarily saw the magnitude of the chain of events leading to major market shocks such as the financial crisis, Long-Term Capital Management, Enron etc. These are the very real downsides of AI in investment markets and will require guardrails built into training models and, importantly, ongoing human oversight of outputs.

Using AI for predictive or persuasive content intended to drive client activity may also be problematic and is a chief concern of the Securities and Exchange Commission (SEC) in the U.S. Clearly a generative AI foundation model tuned to the investment perspectives of a given firm or entity can likely make biased, forward-looking market calls that may seem innocuous coming from a chatbot. It can occur continuously and even be customized to clients at scale to prompt activity self-serving to the advisor, fund manager or other perpetrator—recall the YOLO/MEME stock trading fad of 2021 and consider how an AI engine could create similar herd mentalities.

If generative AI gets deployed more widely in the markets and is actually making or inspiring decisions via AI/algo-driven “active” products covering broader security universes, it’s arguable that markets could become even more efficient.

Second and third derivative valuation dislocations may be more readily identified and traded upon. It could almost entirely automate passive management and bring such personnel and resource efficiencies into active management so as to drive down the fee structures even more, which can help even nuanced alpha capture be more meaningful for investor returns.

Product complexity may also rise as AI can assimilate multiple risks and make counterbalancing trades under certain market scenarios to both sidestep losses and take advantage of opportunities. One can foresee new thematic fund products tied to strategies and predictive analytics that only an LLM can decipher.

The potential benefits and pitfalls of AI in the investment industry are endless and the only thing we can say for sure is it’s going to happen. With proper guardrails and human oversight, we can help ensure a safe and beneficial journey into this exciting new world.