Unpredictability creates alpha: Stock picking

Regular readers of these missives know that I sometimes come across papers that are so full of important insights that I discuss them in several posts. A new NBER Working Paper from Lauren Cohen and his collaborators, which tried to mimic fund managers with AI, is such a case. Today, I will discuss what this study can teach us about finding outperforming stocks. Tomorrow, I will focus on what this study can tell us about how to find outperforming funds.

The researchers collected quarterly fund holdings of all US-domiciled US equity funds from 1990 to 2023. Then they trained a machine-learning programme to forecast which stocks a fund manager will buy, hold, or sell in the next year. The training data were the fund holdings of each fund, as well as macro data like interest rates at the time, etc.

As you may have guessed, some fund managers are more predictable in their behaviour than others and tomorrow, I will dive into what the results tell us about that. But on average, AI was able to correctly predict 71% of all stock trades of fund managers.

These stocks, which can be better predicted as targets for fund investments, in my view, are the ones that fit neatly into a specific investment style. For example, they are obvious value stocks and fund managers looking for a cheap stock to add to the portfolio will likely buy these value stocks rather than sell them.

Because these stocks are a clear-cut case for a value, growth, quality, etc. investor, they become crowded trades for the funds managers following these different investment styles.

The other 29% of less predictable stocks are stocks that are not so clear-cut. They may be borderline cheap, or they may be cheap but also low quality or distressed, so that many value managers may stay away from these stocks because of the high uncertainty around them.

Below is a chart that shows the excess return over Treasury Bills and the 4-factor alpha (taking into account a company’s market risk, valuation, size, and momentum) in the year following an assessment of stocks as more or less predictable.

Future stock returns and the predictability of being bought or sold by funds

Source: Cohen et al. (2026)

As you can see, the stocks that are most predictable as targets of fund managers are the ones that underperform going forward. These are the crowded trades where every fund manager is already invested or where everybody knows that it is a cheap/fast-growing/quality stock (delete depending on fund style). Meanwhile, the stocks that are less predictable as fund investments or divestments – the stocks that aren’t clear-cut cases – are the ones that perform the best in the future.

The lesson for fund managers and investors in general from this insight should be clear: Stay away from the clear-cut cases and the crowded trades. The alpha is where there is uncertainty and unpredictability.