Beyond the Numbers: Asset Pricing with Graphical and Alternative Data

• Applicant: Professor Dr. Christian Fieberg
• Subject Area: Accounting and Finance
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 563675652

This project aims to explore the potential of alternative data, particularly image-based representations, in asset pricing and return predictability. Building on recent advances in machine learning, our research will challenge traditional asset pricing models by incorporating non-traditional data types. OBJECTIVES: The primary goal is to assess whether combining visual data with machine learning techniques can uncover new insights into global asset pricing. The specific goals of the project will investigate: 1) the information content of graphical data for return predictability; 2) enhancements in image analysis methodologies for financial forecasting; 3) methodological uncertainties in image-based analysis; 4) the applicability of image-based methods to assets with limited fundamental data; 5) the comparative predictive importance of technical, fundamental, and alternative data; 6) the information content for asset returns at different horizons; 7) cross-market and temporal variations in return predictability; 8) the feasibility of developing an asset pricing model incorporating graphical data. METHODOLOGY: The research will utilize a global dataset covering 45-49 developed and emerging markets (1990-2024), incorporating data sources like CRSP, Compustat, and Bloomberg. It will combine machine learning models with asset pricing frameworks to explore the predictive potential of graphical data. Methods such as convolutional neural networks (CNNs), which capture visual patterns in charts, and long short-term memory networks (LSTMs), adept at analyzing sequential data, will be employed to uncover trends. The outputs of these machine learning models, such as directional forecasts or predicted returns, will be integrated into traditional asset pricing tests to evaluate their economic value. Portfolio construction based on these predictions will be assessed using metrics like Sharpe ratios, alphas, and factor regressions. EXPECTED CONTRIBUTIONS: This project is expected to significantly contribute to financial sciences in several key areas. It will enhance the understanding of asset pricing by exploring new ways to explain the cross-section of returns, both locally and globally, through alternative data sources and machine learning. These insights will address challenges posed by the "factor zoo" and provide a foundation for future studies in underexplored areas. The project will also develop innovative tools and methodologies, integrating machine learning and generative artificial intelligence to transform how return patterns are analyzed in international markets. These advancements will deepen the understanding of market efficiency and predictability. Additionally, the research will offer practical tools for strategic and tactical asset allocation, inspiring new investment strategies and improving performance measurement.

• DFG Programme: Research Grants
• International Connection: Poland
• Cooperation Partner: Professor Dr. Adam Zaremba