Correcting the Factor Mirage: A Research Protocol for Causal Factor Investing

Abstract

P-hacking is a well-understood cause of false positives in factor investing. A far less studied cause is factor model specification choices encouraged by the current econometric canon. We prove that specification errors cause factor strategies to underperform and potentially yield systematic losses, even if all risk premia remain constant and are estimated with the correct sign. Unlike the p-hacking–driven factor zoo, where noise is mistaken for signal through repeated testing, we identify a distinct phenomenon—the factor mirage—in which canonical econometric practices systematically reward misspecified models that appear statistically strong but are structurally flawed. We show that these practices, especially over-controlling for colliders, increase the likelihood of few-shot p-hacking and adverse outcomes. The implication is that specification errors are a more insidious and underappreciated threat to investors than previously recognized. To our knowledge, this is the first study to connect factor model selection, collider bias, underperformance, and systematic losses through a unified causal framework. These findings challenge the scientific credibility and long-term viability of the current associational (non-causal) multi-trillion-dollar factor investing industry. To address these risks, we propose adjustments to the econometric canon, informed by recent advances in machine learning and causal inference.

Keywords: Causal inference, causal discovery, confounder, collider, factor investing, p-hacking, underperformance, systematic losses

JEL Classification: G0, G1, G2, G15, G24, E44.