Factor Based Clustering

Abstract

We propose a novel approach to cluster funds based on their factor exposures. The approach uses investment returns as input data and calculates similarity scores across funds, which are then used to form clusters. The derived clusters avoid common pitfalls that correlation based or other cluster methods fall into. They can be used as peer group alternatives to what vendors provide or to further refine existing categories that might be too obscure to make sense of. When tested against long/short equity funds, we find that we can form clusters with relatively high levels of stability across time.