Dynamics of market correlations: Taxonomy and portfolio analysis

TL;DR

This paper analyzes the dynamic structure of stock correlation networks using a minimum spanning tree approach, revealing market states, robustness of topology, and implications for portfolio diversification.

Contribution

It introduces the concept of the asset tree to study correlation dynamics, identifies a scale-free structure, and links tree topology to market crashes and portfolio optimization.

Findings

Market crashes correspond to low mean occupation layer (MOL).

The asset tree maintains a scale-free structure with different exponents during crashes.

Classic Markowitz portfolio assets are located on the outer leaves of the tree.

Abstract

The time dependence of the recently introduced minimum spanning tree description of correlations between stocks, called the ``asset tree'' have been studied to reflect the economic taxonomy. The nodes of the tree are identified with stocks and the distance between them is a unique function of the corresponding element of the correlation matrix. By using the concept of a central vertex, chosen as the most strongly connected node of the tree, an important characteristic is defined by the mean occupation layer (MOL). During crashes the strong global correlation in the market manifests itself by a low value of MOL. The tree seems to have a scale free structure where the scaling exponent of the degree distribution is different for `business as usual' and `crash' periods. The basic structure of the tree topology is very robust with respect to time. We also point out that the diversification aspect of portfolio optimization results in the fact that the assets of the classic Markowitz portfolio are always located on the outer leaves of the tree. Technical aspects like the window size dependence of the investigated quantities are also discussed.