Financial Thought Exchange Podcast Causality in Factor Investing: Marcos López de Prado, PhD & Vincent Zoonekynd, PhD
Jan 20, 2026
In this enlightening discussion, Marcos López de Prado, a professor and expert in machine learning for finance, joins Vincent Zoonekynd, a leader in quantitative research. They unveil the pitfalls of factor models, including the perils of p-hacking and collider bias. Listeners discover how causal graphs can enhance factor attribution and improve model design. The duo emphasizes that factor investing should focus on causality rather than mere statistical correlations, urging researchers to clarify their assumptions in model reporting.
AI Snips
Chapters
Transcript
Episode notes
Factor Zoo Often Means False Discoveries
- The factor zoo reflects many spurious, statistically significant variables that fail out of sample due to p-hacking and false discoveries.
- Marcos López de Prado warns these are statistical flukes rather than genuine risk premia.
Forecasting Vs Explaining Demand Different Variables
- Forecasting and explaining require different variable choices; predictors useful for forecasting can be poor for causal explanation.
- Vincent Zoonekynd stresses that consequences of the target should not be used when explaining why something works.
Confounders Versus Colliders — Key Difference
- Confounders are common causes and should be included; colliders are common effects and should usually be excluded in explanatory models.
- Vincent Zoonekynd highlights that colliders mislead causal interpretation despite improving fit.