Welcome to my personal website!

Ge Yu余舸

Ph.D. Candidate in Finance
Swedish House of Finance, Stockholm School of Economics

I am a Ph.D. candidate in finance at the Swedish House of Finance, Stockholm School of Economics. My research interests include asset management, machine learning and AI, empirical asset pricing, and FinTech.

My current works focus on the applications of artificial intelligence methods on mutual fund holdings and textual data, as well as the impact of artificial intelligence on financial markets.

I am on the academic job market in 2026–2027.

Portrait of Ge Yu

Contact

671/674
Sveavägen 65
113 50, Stockholm, Sweden

ge.yu@hhs.se

Research

Job Market Paper

From Words to Portfolios Disentangling Narrative and Reality in Mutual Fund Differentiation

Best PhD Paper Award, 3rd AI in Finance Conference

Abstract

This paper studies the different implications of portfolio and narrative differentiation for mutual fund fees, capital flows, and future performance. Developing a transformer-based portfolio embedding model and combining it with pre-trained large language model embeddings of fund prospectuses, I construct two novel measures of mutual fund differentiation from portfolio holdings and prospectus narratives. The two measures are only weakly correlated, indicating that portfolio distinctiveness does not necessarily coincide with narrative distinctiveness. Prospectus uniqueness is associated with higher fees and, when past performance is strong, greater capital flows, while portfolio uniqueness predicts future alpha among funds with a strong track record. Together, these findings suggest that narrative differentiation in fund disclosures serves a marketing and pricing role, whereas portfolio differentiation is informative about future performance primarily among funds with strong prior records.

Latest Draft (PDF)

Works in Progress

Generative AI, Labor, and Financial Market

with Mariassunta Giannetti

Attractive Curves in ETF Prices

with Riccardo Sabbatucci

Generative AI and the Bundling of Work Requirements

with Tianxiang Zheng and Haofeng Zhou

Abstract

This paper examines how generative AI reshapes the division of labor by changing how clients bundle work for the market. We develop a model of how clients partition work across job postings when AI lowers the cost of executing broader task bundles. We test its predictions using over two million Freelancer.com postings around the release of ChatGPT, measuring the breadth of each posting’s skill bundle with embeddings. We find that occupations more exposed to large language models experience a decline in postings, while the remaining postings combine more disparate skills as narrowly specialized packages become rarer. Our findings reveal an important reorganization of outsourced work: clients absorb narrow tasks and bring broader residual bundles to market. Generative AI thus redraws job boundaries and coarsens the division of labor in freelance markets.

When AI Gets Nerfed LLM Quality and Financial Markets

Abstract

This paper studies the financial consequences of silent degradation in large language model (LLM) performance, or “nerfing.” Exploiting social media feedback, daily LLM benchmarks, and Polymarket transaction data, I examine account-level trading around a major, documented decline in answer quality. I find that accounts with greater historical sensitivity to AI disruptions exhibit reduced trading participation and lower returns during the nerfing episode. Return decompositions attribute these losses to weaker performance within existing trading roles, though strategic reallocation partially offsets the decline. These findings demonstrate how unannounced shifts in AI quality directly affect investor behavior, with important implications for the resilience of financial markets.

Background

Education