Research

Working Papers

Empirical Asset Pricing Using Explainable Artificial Intelligence, with Umit Demirbaga

Abstract: This paper applies explainable artificial intelligence in empirical asset pricing to explain the reasoning behind return predictions made by various complex machine learning models. We use two state-of-the-art explainable AI methods, LIME and SHAP. Our findings indicate that the primary drivers in our model predictions are stock-level characteristics such as momentum, 52-week high, and volatility. We demonstrate large improvements in predictive power and investment performance when incorporating insights from explainable AI into model refinement, surpassing the performance of machine learning models without such explanations. Additionally, we use a variety of data visualization methods within explainable AI to help institutional investors interactively communicate the inner workings of these models to stakeholders.

Machine Learning Execution Time in Asset Pricing,  with Umit Demirbaga

Abstract: In the fast-paced world of finance, where timely decisions can yield substantial gains or losses, machine learning models with time-consuming training and prediction may miss crucial market timing opportunities. This study examines the machine learning model execution time including both training and prediction phases, in empirical asset pricing. We conduct a comprehensive analysis of machine learning execution time, examining ten models and introducing two strategies to save time: feature reduction and the reduction of time observations. Our findings reveal that XGBoost stands out as a top performer, demonstrating relatively low execution times compared to other machine learning models, with exceptional accuracy, boasting an out-of-sample R-squared of 0.78 and a Sharpe ratio of 1.76. Furthermore, feature reduction and shorter time observations reduce execution time by as much as 18 times while also slightly enhancing investment performance. This research underscores the vital interplay between model accuracy and execution time to make accurate and prompt investment decisions in practice.

Spillovers of Senior Mutual Fund Managers' Capital Raising Ability

Abstract: This paper documents a sizeable spillover effect of senior mutual fund managers' capital raising ability on their colleagues. I find that when a junior fund manager has new senior colleagues in a fund, the junior manager’s other funds also have substantial capital inflows. To identify the cause of these capital inflows, I extend the active investment skill in the Berk and Green model with capital raising ability. Empirical evidence shows that a fund manager's performance in other funds (measured by net or gross alphas) decreases significantly after having new senior colleagues, and value added from the active investment does not increase. This is consistent with the spillover effect of senior managers' capital raising ability rather than active investment skill.

Origins of Mutual Fund Skill: Market versus Accounting Based Asset Pricing Anomalies,  with Charlotte Christiansen and Ran Xing

Abstract: This paper decomposes active U.S. equity mutual funds’ value added using 234 public asset pricing anomalies. We find that these funds on average lose substantial value through their high exposure to the short legs of accounting anomalies (e.g., investment and profitability). We identify funds’ capital flows and price pressure of flow-induced trading as important drivers of this value loss. Partially due to the high correlation between the short legs of accounting anomalies and the long legs of market anomalies (e.g., momentum and liquidity risk), mutual funds fail to keep a low exposure to the former when flow-induced price pressure increases their exposure to the latter. We confirm this channel using the 2002 Morningstar rating methodology reform as an exogenous shock. 

Reallocation of Mutual Fund Managers and Capital Raising Ability

Abstract: This paper establishes the fund manager's capital raising ability as an important managerial skill that fund firms exploit to generate higher firm revenues. Fund firms reallocate fund managers with high capital raising ability to other funds with large outflows. Investors demand the capital raising ability of managers and reward it by investing more capital despite lower future alphas. A team with a larger experience difference between reallocated managers and existing managers attracts more capital inflows, suggesting that there is a synergy effect on the fund manager's capital raising ability.