Predicting Asset Return Shifts: Experimental Evidence on Human Forecasting Ability

Abstract

Forecasting asset return shifts under leptokurtic noise is a key investor challenge. We conduct a lab experiment to assess investors' ability in this task and explore performance improvement methods. Participants forecasted asset value shifts for significant money, with success requiring uncovering a pattern. They outperformed expectations, demonstrating surprising pattern recognition capacity. A second group, merely informed of the pattern's existence without explanation, performed comparably to a fully informed control group, indicating minimal guidance sufficed to boost performance. The findings support the "active thinker" view of decision-making and suggest potential for human-AI collaboration in finance, combining human pattern recognition with AI's strategy execution.

Keywords: Forecasting, Laboratory experiments, Knightian uncertainty, Information nudges