This project examines whether the connections between firms reflect market conditions and tests the relation between market stability and evolution of the networks. The networks are constructed using pre-collected Reuters news data from October 2006 to November 2013. Global connectivity and individual centrality measures are computed for the networks and are studied in the context of historical events, and clusters detected in the network are compared with sector information of entities. Quantitative methods are also applied to investigate the dynamic relationship between market stability, which is proxied by volatility, and company co-occurrence. Results show that the change in network parameters is associated with the market condition, and the strength of connections is significantly correlated with pairwise stock volatility relationship during stressed periods. Additionally, the difference between cluster partition and sector classification widens and sector information is less correlated with volatility when the market is unstable. These findings suggest a novel approach of measuring market stability and systemic risk using news data, and expand the application of network science in finance.
Figure 1. Changes in the beta coefficient in a regression model where the dependent variable is the pairwise correlation between stock volatility and the independent variable is the pairwise edge weights in the normalised co-occurrence network. The 95% confidence interval found via the quadratic assignment procedure (QAP) with randomly permutated data is also presented. Colours represent the different thresholds (0.5, 0.7 and 0.9) used to prune edges in the co-occurrence network.