The impact of prediction market signals on stock returns: A comparative exploratory study of global and local modeling techniques

Details

Supervisors

Vande Kerckhove, Corentin

Faculty

Louvain School of Management

Degree label

Master [120] : ingénieur de gestion, à finalité spécialisée

Abstract

enThis thesis investigates the relationship between prediction market signals and stock returns through a comparative modeling framework. By contrasting global and local forecasting approaches, we examine whether prediction markets generate informational value for financial forecasting, and under what conditions this value is most pronounced. Us- ing data from Kalshi prediction markets and a diverse set of US equities and ETFs, we implemented a rigorous walk-forward validation strategy to evaluate model performance. Our analysis reveals that prediction market signals exhibit varying degrees of influence across different financial instruments: macroeconomic indicators such as GDP show broad market impact, while firm-specific signals demonstrate more targeted effects. The findings indicate that simpler models often outperform more complex ones. ARIMAX models show superior performance in local contexts, while in global settings, Linear Regression outperforms XGBoost. The latter struggle to effectively capture the structure and significance of prediction market signals. This study contributes to the financial forecasting literature by demonstrating that prediction markets can be more effective as complementary indicators rather than standalone forecasting tools, while highlighting the importance of model selection and data quality in capturing relationships between prediction market signals and asset returns. These insights have practical implications for data-driven investment firms, suggesting that the integration of prediction market signals should be done selectively and contextually, taking into account both the nature of the target asset and the type of signal used.