About
Leonard Yang Liu is a Ph.D. candidate in Accounting at the University of Maryland's Robert H. Smith School of Business, advised by Musa Subasi and Michael Kimbrough. His research studies how new technologies, especially artificial intelligence and generative AI, are changing the way financial information is produced and used in capital markets. He is also interested in financial intermediaries and behavioral finance. His research has been recognized as the best paper in Fintech at the 2024 SFA Annual Meeting, and he has received multiple awards, such as the Frank T. Paine Award for Academic Achievement and the Long Jiang Graduate Student Fellowship.
Before the Ph.D., he worked as an investment manager in Beijing and as an auditor at Ernst & Young. He holds an M.S. in Management from Peking University and a B.S. in Accounting from Huazhong University of Science and Technology.
He is on the 2026–27 academic job market.
Research
-
Synthesizing the Consensus: Generative AI, LLM Visibility, and the Integration of Multi-Provider Consensus ForecastsDissertation
Status: Dissertation, draft May 2026.
Dissertation committee: Musa Subasi (chair), Michael Kimbrough (co-chair), Rebecca Hann, Nick Seybert, Jingyi Qian, Erkut Y. Ozbay.
Abstract
This paper studies how generative AI (GenAI) tools facilitate financial information processing in the setting of the market's integration and use of multi-provider analyst consensus forecasts. Using S&P 1500 firm-quarters over 2021–2025 covering all five major forecast data providers (FDPs), I first document a wider, more selective use of consensus after ChatGPT: prices respond more to more-accurate providers and reweight from I/B/E/S where it loses relative accuracy. I then construct an accuracy-weighted consensus, and announcement returns load on it increasingly over two heuristics: I/B/E/S and the equal-weighted average across the five FDPs. This shift strengthens with GenAI's retrieval capability and concentrates where providers disagree most, especially where that disagreement most likely reflects methodological differences across FDPs. Measuring the visibility of each firm's consensus forecasts to GenAI tools using the GPT-4o model, I find the synthesis benefit concentrates in high-visibility firms. The results show how GenAI powers investors' processing of multi-provider consensus.
-
AI-Powered Analysts
Status: Revise & resubmit, The Accounting Review.
Best Paper in Fintech, SFA 2024.
Abstract
We examine how brokerages' institutional adoption of artificial intelligence (AI), measured from AI skill requirements in research-related job postings, shapes sell-side analyst work. AI adoption varies across brokerages and is predicted by peer adoption, organizational capacity, analyst human capital, and institutional demand. Comparing the same analyst-firm pair over time, analysts at higher-adoption brokerages issue more accurate earnings forecasts. The gains are larger when forecasts rely more on public, machine-readable information: early in the cycle, for firms with lower information asymmetry and more readable disclosures, and among technically trained analysts. Using intraday decision fatigue, we show that AI substitutes for analysts' cognitive effort within the forecasting task: it attenuates fatigue-related accuracy losses and herding. Freed capacity is redeployed toward relationship building with management and investing clients. Forecasts by analysts at higher-adoption brokerages elicit stronger price reactions, and adopters shift hiring toward senior analysts. Institutional AI thus benefits analyst research by conserving the scarce cognition it runs on, with gains that surface in forecast quality, market pricing, and the composition of the analyst workforce.
-
When Does Generative AI Level the Playing Field? Evidence from Crowdsourced Earnings Forecasts
Status: Preparing for submission to a journal.
Abstract
Whether generative AI (GenAI) levels the playing field or reinforces existing advantages depends on which constraint it relaxes. We argue that GenAI substitutes for institutionalized analytical resources but complements human expertise, and we test this distinction using crowdsourced earnings forecasts from Estimize. Before ChatGPT's release, non-professional contributors persistently underperform professionals; afterward, the already-narrowing gap closes. Inferring contributors' GenAI reliance from forecasting behavior during ChatGPT service outages, we find that adoption improves the accuracy and independence of non-professionals' forecasts while leaving professionals largely unaffected, consistent with GenAI relaxing a binding resource constraint. Within non-professionals, where resources vary little, only experienced contributors improve; inexperienced adopters show no gains and, if anything, herd more. Cross-sectional evidence suggests that the benefits concentrate where disclosures are difficult to process and, separately, where information asymmetry is low. Reflecting the leveling, earnings announcement returns respond more strongly to non-professional forecast surprises post-ChatGPT. These findings reconcile conflicting evidence on GenAI's distributional effects: the technology democratizes information production along the resource margin while amplifying the returns to expertise.
-
Stepping Into the Spotlight and Staying in the “Comfort Zone”: CFO Debuts in Earnings Conference Calls
Status: Preparing for submission to a journal.
Abstract
This paper investigates whether CFOs who debut on earnings conference calls stay in their “comfort zone” by emphasizing numbers and accounting jargon rather than qualitative strategic information to maintain credibility under public scrutiny. Hard information is the content CFOs know best, and it is verifiable and objective. In a large sample of earnings call transcripts, debuting CFOs rely significantly more on quantitative language and accounting jargon at the expense of softer strategic disclosures, and these tendencies decline nearly monotonically over subsequent calls. The debut effect is strongest where pressure and uncertainty should be most acute: when it is the CFO's first time in the C-suite, when the firm is high-growth, and, more weakly, when the firm reports a loss. We also find that CFOs engaging in greater comfort-zone behavior are less likely to be subsequently promoted to a CEO role or to move to a more senior role at another firm, while comfort-zone communication is not measurably associated with market or analyst reactions to the calls. Taken together, our results suggest that debut pressure encourages CFOs to favor hard information over soft on early calls, a behavior associated with worse career outcomes and no measurable offsetting benefit.
Teaching
BMGT 220 — Principles of Accounting I: Financial Accounting. Instructor of record, University of Maryland, Robert H. Smith School of Business.
| Term | Course | Instructor |
|---|---|---|
| Summer 2025 | 3.82 / 4.0 | 3.89 / 4.0 |
| Winter 2025 | 3.77 / 4.0 | 3.85 / 4.0 |
Other Teaching Experience
Representative PhD student speaker, GenAI and Business Research, UMD Smith (scheduled, Sep 2026).
Guest speaker, Fundamentals of Business Research, UMD Smith (Sep 2025).
Teaching assistant, Peking University HSBC Business School (2018–2020).
Contact
- yliu337@umd.edu
- Office
- 3330F Van Munching Hall, Robert H. Smith School of Business, College Park, MD 20742