Busan National University study rethinks how artificial intelligence supports investment decisions

Artificial intelligence (AI) is rapidly transforming modern finance, powering applications ranging from stock market prediction to investment advice. But does making more accurate forecasts necessarily lead to better investment decisions? The answer may be no, according to two recent studies by researchers at Busan National University and international collaborators.

Just as a weather app can accurately predict tomorrow’s temperature but tell you to leave your umbrella at home before a storm, financial AI systems can make highly accurate market predictions but still make poor investment decisions. Researchers argue that AI should not only be evaluated on how accurately it predicts markets, but also on how effectively it supports real-world financial decision-making.

To address this challenge, researchers led by Professor Yoontae Hwang of Busan National University, in collaboration with Professor Stefan Zohren of the University of Oxford, UK, developed a new artificial intelligence framework called Signature-Informed Transformer (SIT). Rather than focusing only on where prices will ultimately go, the model learns how market prices change over time and how assets affect each other, allowing you to directly optimize investment decisions while accounting for risk. This research Minutes of the 43rd meeting^rd International conference on machine learning Professor Huang served as the study’s lead author.

The framework was evaluated on three major stock markets: the US and China. Compared to traditional prediction-based approaches, decision-focused models achieved stronger risk-adjusted performance and more robust asset accumulation.

“Our findings indicate that future financial AI systems may need to shift their focus from maximizing predictive accuracy to optimizing decision quality.” points out Professor Hwang.

The second study looked at another fundamental question: can reports of financial AI success be trusted? Researchers reviewed 164 studies on large-scale language models (LLMs) in finance published between 2023 and 2025 and identified recurring biases that can inflate reported performance, including unintended use of prospective information, survivorship bias, unrealistic evaluation settings, and overlooked real costs. This research Minutes of the 43rd meeting^rd International conference on machine learning Professor Huang served as co-lead author of this study.

“We observed several biases, including unintended use of forward-looking information, exclusion of failed companies from the dataset, unrealistic valuation targets, and omission of practical constraints such as transaction costs. ” Professor Hwang explains:

To address these issues, researchers proposed a construct validity framework. This is a practical checklist for assessing whether financial AI systems have been tested under realistic conditions and whether the reported performance is likely to hold up beyond the laboratory.

Together, the two studies have a common message: train AI to make important decisions and evaluate them under real-world conditions. Looking to the future, researchers are envisioning an AI-powered “flight simulator” for financial markets. There, virtual investors can help institutions and regulators test policy, product, and market shocks before real people’s savings are at risk, ultimately leading to more transparent financial advice and more trustworthy AI.

reference
Original paper title: Signature-Informed Transformer for Asset Allocation
journal: Proceedings of the 43rd International Conference on Machine Learning
DOI: https://doi.org/10.48550/arXiv.2510.03129

Original paper title: Position: Evaluation of LLMs in finance requires consideration of explicit bias
journal: Proceedings of the 43rd International Conference on Machine Learning
DOI: https://doi.org/10.48550/arXiv.2602.14233

About Busan University
Busan National University, located in Busan, South Korea, was founded in 1946 and is currently South Korea’s number one national university in terms of research and teaching capabilities. This multi-campus university also has smaller campuses in Yangsan, Miryang, and Ami. The university prides itself on the principles of truth, freedom, and service and has approximately 30,000 students, 1,200 professors, and 750 faculty members. The university is made up of 14 colleges (schools) and one independent department, with a total of 103 departments.

Website: https://www.pusan.ac.kr/eng/Main.do

About Professor Hwang Yun-te
Yoontae Hwang is an assistant professor at the Graduate School of Data Science at Busan National University, South Korea. Prior to joining the company in September 2025, he was a Sejong Science Fellow and Postdoctoral Research Fellow at the University of Oxford, collaborating with Professor Stefan Soren. He received his Ph.D. in Industrial Engineering from UNIST in 2024. His research focuses on AI-driven asset allocation, large-scale language models of finance, and agent-based market simulation. His lab aims to transform rigorous research into practical tools that have meaningful real-world impact.
Laboratory: https://tsi-yoontae.github.io/#home
ORCID ID: 0000-0002-5856-1914