We present improved out-of-sample predictability of future fund flows using state-of-the-art machine learning methods. Nonlinear machine learning models significantly outperform linear models in terms of out-of-sample R-squared. Using interpretable ML methods, we identify past flows and the Morningstar rating as the most important predictors for netflows, while other past performance variables are of minor importance. We find that the importance of Morningstar ratings and expenses has increased over time. In addition, the interaction effect of past flows with the Morningstar rating has a substantial impact on future flows. Furthermore, our results demonstrate that machine learning-based fund flow predictions can be used to ex-ante differentiate between high and low-performing mutual funds. Finally, we provide evidence that funds whose flow predictions can be improved the most using ML reveal the worst performance, consistent with the idea that liquidity management is particularly challenging for these funds.

We use frontier advancements in Artificial Intelligence and machine learning to extract and classify the part of key economic agents’ behaviors that are predictable from past behaviors. Even the agents themselves might view these as novel (innovative) decisions; however, we show in strong contrast that a large percentage of these actions and behaviors can be predicted – and thus mimicked – in the absence of these individuals. In particular, we show that 71% of mutual fund managers’ trade directions can be predicted in the absence of the agent making a single trade. For some managers, this increases to nearly all of their trades in a given quarter. Further, we find that manager behavior is more predictable and replicable for managers who have a longer history of trading and are in less competitive categories. The larger the ownership stake of the manager in the fund, the less predictable their behavior. Lastly, we show strong performance implications: less predictable managers strongly outperform their peers, while the most predictable managers significantly underperform. Even within each manager’s portfolio, those stock positions that are more difficult to predict strongly outperform those that are easier to predict. Aggregating across the universe of fund managers each quarter, stocks whose position changes are least predictable additionally significantly outperform stocks whose position changes are most predictable across the universe. Our framework allows researchers to delineate and classify the portion of financial agents’ action sets which are predictable from those which are novel responses to stimuli – open to being evaluated for value creation or destruction.

The standard approach to portfolio selection involves two stages: forecast the asset returns and then plug them into an optimizer. We argue that this separation is deeply problematic. The first stage treats cross-sectional prediction errors as equally important across all securities. However, given that final portfolios might differ given distinct risk preferences and investment restrictions, the standard approach fails to recognize that the investor is not just concerned with the average forecast error – but the precision of the forecasts for the specific assets that are most important for their portfolio. Hence, it is crucial to integrate the two stages. We propose a novel implementation utilizing machine learning tools that unifies the expected return generation process and the final optimized portfolio. Our empirical example provides convincing evidence that our end-to-end method outperforms the traditional two-stage approach. In our framework, each investor has their own, endogenously determined, efficient frontier that depends on risk preferences, investor-specific constraints, as well as exposure to market frictions.