Non-parametric sign prediction of high-dimensional correlation matrix coefficients

Abstract We introduce a method to predict which correlation matrix coefficients are likely change their signs in the future high-dimensional regime, i.e. , when number of features is larger than samples per feature. The stability signs, two-by-two relationships, found depend on three-by-three relationships inspired by Heider social cohesion theory this regime. apply our US and Hong Kong equities historical data illustrate how structure matrices influences sign its coefficients.