AI and ML are Outperforming Traditional Models for Asset Pricing, Fraud Detection
This report from Columbia University Professor Agostino Capponi explores how artificial intelligence (AI) and machine learning (ML) are revolutionizing the financial services industry, outlining key ML methodologies like supervised and unsupervised learning, and discussing major applications of AI and ML across finance.
In asset pricing, ML enables the use of high-dimensional data to develop flexible models that outperform traditional linear approaches. In risk management, AI enhances scenario analysis and stress testing by capturing nonlinear relationships between macroeconomic variables and risk factors. ML tools like XGBoost and LightGBM have been shown to improve the accuracy of loan default predictions by over 20% compared to traditional credit risk models, as demonstrated in an evaluation of 800,000 loan records.
AI robo-advisors have significantly evolved, using reinforcement learning to infer a user’s risk tolerance based on real portfolio choices over time, outperforming static questionnaire-based systems. In fraud detection, AI models—particularly those using natural language processing—analyze sequences of user actions rather than individual transactions, offering a more advanced approach than traditional methods. Capponi also highlights the growing use of alternative data, including customer reviews and product photos, to enhance demand estimation.
While ML techniques have delivered more accurate, adaptable, and faster financial models, the report stresses the importance of responsible implementation, rigorous validation, and profound understanding of financial market dynamics to ensure long-term sustainability and competitive advantage.