Researchers Reveal Novel AI Model and Evaluation Framework for Financial Systems
BUSAN, South Korea, July 29, 2026 — Artificial intelligence (AI) is reshaping finance with applications in stock forecasting and investment advice. However, recent studies by researchers at Pusan National University and international partners suggest that higher prediction accuracy doesn’t equate to better investment decisions.
Professor Yoontae Hwang and his team have developed the Signature-Informed Transformer (SIT)—a model focusing on the evolution of market prices and asset relationships rather than just end prices. This approach aims to optimize investment decisions while considering risk. The research was presented at the International Conference on Machine Learning on April 30, 2026, showing that SIT outperformed traditional models in risk-adjusted performance and wealth accumulation in major equity markets.
Hwang noted, “Future financial AI systems may need to shift their focus from maximizing prediction accuracy to optimizing decision quality.”
A second study examined the reliability of financial AI reports, analyzing 164 studies from 2023 to 2025. The researchers found biases that could misrepresent performance, proposing a Structural Validity Framework to assess the realistic testing of financial AIs.
The studies highlight the importance of training AIs for relevant decisions and evaluating their performance in real-world scenarios. The researchers advocate for AI-powered “flight simulators” for financial markets, which could help test policies and products without jeopardizing actual savings.
For further information, refer to the original papers:
- Signature-Informed Transformer for Asset Allocation: DOI link
- Evaluating LLMs in Finance Requires Explicit Bias Consideration: DOI link
About Pusan National University
Website: Pusan National University
For any inquiries, contact:
Goon-Soo Kim
Tel: +82 51 510 7928
Email: [email protected]
