Mean-variance optimization and pair-wise strategies

This paper analyzes quantitative active strategies based on forecasting models and mean-variance optimization. The central concept in our analysis is pairwise strategy. We demonstrate that the correct metric for assessing forecast quality is pair-wise information coefficient. We prove that a general active strategy using mean-variance optimization is equivalent to a linear combination of pair-wise strategies and the relative weights of the pairs in this combination are directly connected to the covariance matrix used in the optimization. In addition, we derive the expected long-term performance of an active quantitative strategy and optimal choice of pair-wise combination that attains maximum information ratio.