Research

AI Verification Costs and the Delegation of Work

This paper builds a measure that adds verification to what AI can do and what AI is doing. I code how cheaply each task can be checked and measure the loss from mistakes. Applying the measure to US occupations, I find that less than half of the work done with AI can be delegated.

Empirical Asset Pricing Using Explainable Artificial Intelligence, with Umit Demirbaga

This paper applies explainable AI in empirical asset pricing to explain the reasoning behind stock return predictions made by various complex ML models. We use two state-of-the-art explainable AI methods, LIME and SHAP. We demonstrate large improvements in predictive power and investment performance. This can help institutional investors interactively communicate the inner workings of these models to stakeholders.

Execution Time Optimization Through Feature and Temporal Reduction in Asset Pricing,  with Umit Demirbaga and Evrim Güler, Concurrency and Computation: Practice and Experience, 2026

We treat machine-learning asset pricing as a computational workload and propose a cost-aware framework that tunes feature dimensionality, training-window depth and model complexity for XGBoost. Benchmarked against ten algorithms on a high-performance computing cluster, it delivers meaningful computational savings while preserving out-of-sample accuracy and long–short portfolio performance.

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

Active U.S. equity funds create value through exposures to market anomalies such as momentum, while lose substantial value through exposure to the short legs of accounting anomalies, a by-product of chasing market-based anomalies.

Impostor Syndrome and Fund Managers' Performance, with Zhaoting Lin and Michael Guo

Using a machine learning measure of impostor syndrome among U.S. equity fund managers, we show that it erodes skill and performance as managers retreat from contrarian strategies toward market consensus. Our findings help investors identify fund managers with persistent skill.

Social Transmission Bias and the Disposition Effect, with Zhaoting Lin and Michael Guo

Investors in more socially active brokerage branches show stronger disposition effects, and the link weakens when face-to-face contact is disrupted. This suggests that investor biases are not only individual but social: they spread from person to person.

Determinants of Manager Reallocation in the Mutual Fund Industry

Fund families move managers with strong flow records to funds that are losing assets. After the move, these managers attract more flows. This shows that fund families allocate managerial talent to grow assets.

Senior Manager Connections and Flow Spillovers across Mutual Funds

When a fund manager starts working with a more experienced colleague, the manager’s other funds attract $125 million in additional flows over the next year, driven by the senior colleague’s capital-raising record. This reveals a hidden role of senior managers: their reputation moves investor money well beyond the funds they run.