Research

Working Papers

My research interests include AI in finance, textual analysis, institutional investors, and investments.

What Makes a Star Analyst? Evidence on Industry Expertise from 1.2 Million Analyst Reports

with Alejandro Lopez-Lira, Yuehua Tang, and Yuan Wang

We develop novel text-based measures of sell-side analyst industry knowledge and report quality by applying a large language model evaluation pipeline to 1.2 million industry reports from 40 major brokerage houses between 2012 and 2024. Each report is evaluated across 18 dimensions capturing general research quality and specific forms of industry expertise.

Our measures are substantially more predictive of Institutional Investor All-Star status than traditional forecast-accuracy metrics. Analysts with higher knowledge and quality scores also produce more accurate earnings forecasts, and their forecast revisions and recommendation changes elicit stronger market reactions. At the firm level, coverage by more knowledgeable analysts is associated with lower information asymmetry and a lower implied cost of capital.

Presented at: 2026 FMA Annual Meeting, 2026 NFA Annual Meeting.

The Memorization Problem: Can We Trust LLMs’ Economic Forecasts?

with Alejandro Lopez-Lira and Yuehua Tang

Large language models cannot be trusted for economic forecasts during periods covered by their training data. Under black-box access, counterfactual forecasting ability is not identified when a model has already seen the realized values: any observed output is consistent with both genuine forecasting skill and memorization.

We show that LLMs have memorized economic and financial data, including exact values from before their knowledge cutoffs. Instructions to respect historical information boundaries do not prevent recall-level accuracy, while masking fails because models can reconstruct entities and dates from limited contextual information. We observe no comparable recall after the models’ knowledge cutoffs. Memorization also extends to embeddings.

Presented at: 2026 EFA Annual Meeting, 2026 Wolfe Research 8th Annual AI in Finance Conference, 2026 AFA Meeting, 2025 GSU–MS AI & FinTech Conference, 2025 Journal of Accounting, Auditing and Finance Symposium, Applied Machine Learning, Economics, and Data Science Webinar.

Reaching for Synthetic Yield

with Wei Jiang, Yuehua Tang, and Yanbin Wu

We examine “reaching for synthetic yield” through covered-call option writing by investment companies. Using return and holdings data, including short positions, from a comprehensive sample of U.S. options-writing funds, we find that funds become more aggressive in synthesizing income when interest rates, term premiums, and default premiums are low.

Investors reward high-yield funds with inflows, conditional on total performance, without fully distinguishing between asset-generated yield and yield manufactured through option writing. However, yield-manufacturing funds experience greater total-return losses, particularly during bull markets. Their aggregate option-writing activity also suppresses option prices and short-term implied volatility.

Presented at: 2026 ICI Summer Research Workshop, 2025 FMA Meeting, 2023 University of Florida Warrington Seminar.

Technology Adoption in Mutual Funds

with Beatrice Chang

We study technology adoption by mutual fund families using the hiring of employees in data- and engineering-related roles. Fund families with larger average fund sizes are more likely to hire for these positions, suggesting that scale facilitates investments in technology.

Families that hire data and engineering employees subsequently have lower expense ratios, are more likely to launch new funds, and experience higher overall flows. Textual analysis of job descriptions and responsibilities supports the interpretation that these positions contribute to cost reduction, product development, and investor acquisition.

Presented at: 2024 University of Florida Warrington Seminar.