Portfolio Rebalancing with Reinvestment Based on Noise Filtered Correlation/Covariance Matrices: The Case of NASDAQ 100

Abstract

• Demonstration of outstanding investment performance due to noise filtering of covariance matrices in optimum portfolio selection exercises. This demonstration is based on the out-of-sample simulation of rebalancing trading done daily, weekly, biweekly and monthly.

• Exhibits of investment performance for a number of long-short and mostly long portfolios consisting of NASDAQ 100 names. The reported results clearly favor filtered covariance matrices. In extreme cases, 100% additional performance (annualized returns) can be derived from using noise filtered matrices in optimization rather than using unfiltered matrices. This particular portfolio selection method succeeded in "beating the index" even without filtering, in all exhibits during a 7 year period including the most recent financial crisis: 2006-2013.

Keywords: noise filtering, hidden dependencies, signal disguise, correlation matrix, covariance matrix, random matrix, Wishart matrix, random matrix theory, mean variance optimization, portfolio optimization, portfolio rebalancing, index plus, Sharpe ratio, Tracking error

JEL Classification: C1, C2, C6, G1