The use of Biweight Mid Correlation to improve graph based portfolio construction

Abstract

An analysis of the correlation between the returns of different securities is of fundamental importance in many areas of finance, such as portfolio optimisation. The most commonly used measure of correlation is the Pearson correlation coefficient; however, this suffers from several problems when applied to data from the real world. We propose an alternative estimator — the Biweight Mid Correlation (Bicor) — as a more robust measure for capturing the relationship between returns. We systematically evaluate Bicor empirically using data from the FTSE 100 constituents, and show that it is more robust when compared with the Pearson correlation coefficient. Finally, we demonstrate that Bicor can be used to improve a graph-based method of portfolio construction. Specifically, we show that when treating the correlation matrix as an adjacency matrix for a graph and using graph centrality to construct portfolios, the use of Bicor leads to better performing portfolios.