Multiscale correlation networks analysis of the US stock market: a wavelet analysis

Abstract

We investigate the interaction among stocks in the US market over various time horizons from a network perspective. Unlike the high-frequency data-driven multiscale correlation networks used in previous works, we propose method-driven multiscale correlation networks that are constructed by wavelet analysis and topological methods of minimum spanning tree (MST) and planar maximally filtered graph (PMFG). Using these techniques, we construct MST and PMFG networks of the US stock market at different time scales. The key empirical results show that (1) the topological structures and properties of networks vary across time horizons, (2) there is a sectoral clustering effect in the networks at small time scales, and (3) only a part of connections in the networks survives from one time scale to the next. Our results in terms of MSTs and PMFGs for different time scales supply a new perspective for participants in financial markets, especially for investors or hedgers who have different investment or hedging horizons.