Quantitative Monographs: AI in Finance Forum, Singapore 2026

Research evidence excerpt

Quantitative Monographs: AI in Finance Forum, Singapore 2026

When machines disagree: evidence from large language

models

Presenter: Si Cheng, Syracuse University

Discussant: Byoung-Hyoun Hwang, Nanyang Technological

University

Abstract:

Using six leading large language models to extract sentiment from news headlines and

predict stock returns, we document substantial variation across model providers and

investment horizons. This model disagreement reflects systematic differences in causal

reasoning rather than noise. Stocks with higher cross-provider and cross-horizon

dispersions earn lower future returns, especially among firms with more opaque

information environments and greater operating uncertainty. Model disagreement

amplifies post-earnings-announcement drift for small firms, while delaying the

immediate price reaction to earnings announcements for large firms. Model

disagreement also increases both overall and retail trading volume. Evidence from

exogenous ChatGPT outages reinforces our conclusions. Overall, our findings highlight

the growing role of AI in shaping investor beliefs, stock price dynamics, and price

informativeness.

Generative AI and knowledge-informed deep learning for

asset pricing and investment management

Lin William Cong, NTU President’s Chair Professor in Finance,

Computing and Data Science; Associate Dean, Nanyang Business

School

Summary:

Professor William Cong presented a comprehensive review of three cutting edge

working papers on generative AI in asset pricing, including how to use generative

modeling for general portfolio management and Robo-Advising, how to use tree-based

interpretable AI models for asset pricing and heterogeneity analyses, and how to use

Structured-Knowledge-Informed-Neural-Networks (SKINNs) for option pricing and