The systems and methods described herein can identify meaningful relationships between variables, such as particular investments or general asset classes. Unlike conventional correlation analysis, these systems and methods provide an improved technique of comovement analysis that implements a threshold to eliminate data “noise” and then discretizes the remaining observations to normalize any outliers. Such comovement analysis has numerous advantages over known techniques for characterizing relationships between variables. In some embodiments, this improved comovement analysis can be used to calculate a covariance matrix for purposes of mean-variance optimized portfolio construction.