Granger causality networks of S&P 500 stocks

Abstract

The aim of this study is to investigate weighted time-varying causal graphs based on daily log-return and realized volatility time series of the S&P500 stocks. In order to construct causal graphs for each quarter between Q1 2005 and Q2 2020, we compute cross-correlations between series and conduct pairwise Granger Causality tests for lead-lag relationships. We take individual stocks as nodes and significant causal relationships as weighted edges. Our analysis is based on a comparison of network topology and properties for different periods, including the 2007-2010 subprime crisis, the European sovereign debt crisis, and the coronavirus pandemic, as well as tranquil periods. In addition to edge weights, our study also takes into account the estimated lag of the causal interactions and uses network level measures to identify meaningful patterns over time. We also conduct an analysis of the industrial sectors interconnectedness via smaller graphs derived from the stock-level causality graphs. The results of the work reveal that network topology changes during periods of market stress and crisis and this can be captured by using convenient network construction methods and measures. Our results also show that the patterns of network topology transformation and magnitude of spillovers transmission during the coronavirus crisis and accompanying events differs from previous crisis periods.