A Measure of Pure Causality: Application to the CAPM Framework

Abstract

An observed correlation between two variables, unless there are some special conditions, includes the possibility of multiple relations among the two and a third variable. The correlation coefficient is therefore not sufficient to measure the pure relation between two variables. I propose a measure of pure causality (pure predictability) and a linear regression model to estimate it. I apply this method to the capital asset pricing model (CAPM) framework, and analyze the pure predictability of asset returns.