How We’re Using AI to Reveal How Fund Managers Really Invest

Investment research has always faced a familiar challenge: there is far more information available than there is time to analyze it.

For a single strategy, researchers may have decades of portfolio holdings, performance records, fund flows, manager commentary, meeting notes, and investment committee discussions. Historically, investigating every promising lead simply wasn't practical.

Now, AI is helping change that. Rather than replacing analysts, AI helps automate data retrieval and exploration, allowing researchers to investigate questions that previously might have required significant quantitative work. That means analysts can spend less time gathering information and more time evaluating what it means.

Of course, AI alone isn’t enough. High-quality data, carefully designed analytical frameworks, and expert interpretation remain essential.

Our new Trade Analytics toolkit brings those elements together.

Morningstar’s historical holdings database stretches back decades. Instead of viewing portfolios as isolated snapshots, Trade Analytics reconstructs how positions evolved over time, helping analysts understand the decisions behind portfolio changes.

The framework follows holdings from initial purchase through final sale, distinguishing between active investment decisions and changes driven by fund flows. The result is a richer view of manager behavior and a more powerful way to investigate new research questions.

In a separate study, we examined a small-cap strategy which we worried had grown too big. While the management team believed their process remained unchanged, Trade Analytics revealed a more nuanced picture.

By analyzing the full history of buy and sell decisions, we found that trading activity became considerably slower once assets surpassed a certain threshold. Position-building and selling periods both lengthened substantially, suggesting that an increase in scale may have influenced how the strategy was implemented.

Could we have conducted this research without AI? In theory, yes, but it would have required a lengthy bespoke analysis. AI and Trade Analytics made it practical to investigate quickly and follow up with deeper research.

Trade Analytics can also help answer another common research question: does a manager behave consistently across different strategies? 

In one case, we compared how the same manager ran both small-cap and mid-cap mandates. Looking solely at performance would have provided an incomplete picture, so we focused on behavioral characteristics instead. 

As can be seen in the table below, the differences were striking. The mid-cap strategy had around three times the name turnover, holding periods roughly half as long, and was more than three times as likely to sell and later repurchase the same companies. 

What makes the comparison interesting is that the two portfolios held a similar number of names and had similar levels of concentration. The difference was therefore less about headline portfolio construction and more about how the manager was implementing the two mandates. The data didn't explain why those differences existed, but it helped narrow the investigation. Rather than asking whether the process translated across mandates, researchers could focus on exactly where behaviors diverged and what factors might explain them.

The potential applications extend well beyond individual manager reviews. Trade Analytics can be used to compare managers across different mandates, examine how investment behavior changes over a career, identify distinct manager archetypes, or analyze how groups of investors respond during periods of market stress.

Importantly, these insights don't determine ratings. They give analysts better evidence, better context, and better questions to pursue. The final judgment remains a human one.

There’s a common assumption that access to broadly similar AI models will produce broadly similar research capabilities. In practice, the model is the most visible part of the system, but it’s also the most replaceable.

The lasting advantage comes from combining AI with proprietary historical data, carefully designed analytical frameworks, and continuous feedback from experienced analysts. Those ingredients allow researchers to investigate questions that would once have been impractical and uncover evidence that might otherwise have remained hidden.

General-purpose models will be widely available. Decades of historical evidence, proprietary analytical frameworks, and effective feedback loops will not. Organizations that use AI mainly to produce faster prose risk losing their edge, while those that use it to investigate questions that were previously impractical can materially expand what their research can uncover.