Quick Buzz Feed
Artificial intelligence is increasingly being used in finance, from investment management to financial analysis. However, recent research suggests that highly accurate market predictions do not necessarily result in better investment outcomes. In two complementary studies from Pusan National University, researchers investigated how financial AI can be improved to make better decisions and how the performance of these systems can be evaluated using a proposed framework. The findings support the development of reliable financial AI.
Artificial intelligence is rapidly transforming modern finance, enabling applications from stock market forecasting to personalized investment advice. However, recent research challenges the conventional wisdom that higher prediction accuracy automatically leads to better investment outcomes. Similar to a weather forecast that accurately predicts temperature but fails to warn of a storm, financial AI can make precise market predictions while still leading to suboptimal investment decisions. This highlights a critical need to evaluate AI systems not just on their predictive power, but on their effectiveness in supporting real-world financial decision-making.
To overcome the limitations of prediction-focused AI, Professor Yoontae Hwang from Pusan National University, in collaboration with Professor Stefan Zohren from the University of Oxford, developed a novel AI framework called the Signature-Informed Transformer (SIT). This model diverges from traditional approaches by learning from the intricate evolution of market prices and the interdependencies between assets over time, rather than merely predicting final price points. By directly optimizing investment decisions while inherently accounting for risk, the SIT framework demonstrated superior risk-adjusted performance and more consistent wealth accumulation across major equity markets in the United States and China, compared to conventional forecasting methods. This study suggests a crucial shift for future financial AI systems: prioritizing decision quality over prediction accuracy.
A second, complementary study by the researchers investigated the trustworthiness of reported successes in financial AI. After reviewing 164 studies on large language models (LLMs) in finance published between 2023 and 2025, they uncovered common biases that could artificially inflate performance metrics. These biases included inadvertently using future information, survivor bias (excluding data from failed companies), unrealistic evaluation objectives, and neglecting practical constraints such as transaction costs. To counter these issues and promote more reliable evaluations, the researchers proposed a Structural Validity Framework. This framework serves as a practical checklist to ensure that financial AI systems are tested under realistic conditions and that their reported performance is genuinely reproducible beyond controlled laboratory environments.
Both studies collectively emphasize a core principle: financial AI should be developed and evaluated with real-world decisions and conditions in mind. Looking forward, the researchers envision the creation of AI-powered 'flight simulators' for financial markets. These virtual environments would allow institutions and regulators to safely test new policies, financial products, and responses to market shocks without risking actual investor savings. Such innovations promise to foster greater transparency in financial advice and cultivate more trustworthy AI systems in the financial sector.
Pusan National University, established in 1946 in Busan, South Korea, is recognized as the country's leading national university in terms of research and educational competency. It operates multiple campuses, including locations in Yangsan, Miryang, and Ami. Upholding principles of truth, freedom, and service, the university serves approximately 30,000 students, supported by 1,200 professors and 750 faculty members across 14 colleges (schools) and one independent division, encompassing 103 distinct departments.
Professor Yoontae Hwang is an Assistant Professor at the Graduate School of Data Science, Pusan National University, South Korea. Prior to this role, which he commenced in September 2025, he held a position as a Sejong Science Fellow and Postdoctoral Researcher at the University of Oxford, where he collaborated with Professor Stefan Zohren. He earned his Ph.D. in Industrial Engineering from UNIST in 2024. His research interests primarily lie in AI-driven asset allocation, financial large language models, and agent-based market simulations. His laboratory is dedicated to translating rigorous academic research into practical tools that deliver tangible real-world impact.