Community detection for New York stock market by SCORE-CCD

摘要

Abstract To study the stock market networks, we propose a community detection procedure, Spectral Clustering On Ratios-of-Eigenvectors with Composite Coefficient of Determination. This method has two main contributions. First, it is simple and fast with small error rates. Second, it addresses the problem that sample correlations are hard to capture the structural characteristics of financial markets. Simulation results show the effectiveness of this method. We build the stock network of the 500 largest companies listed on the New York stock exchanges. We show the rationality of community detection on the stock market network. The detection method explores five communities of the market network. We then provide detailed discussions on the most influential companies. For example, we find the pair features from the largest community. Clarify the difference between near-location pairing and distant-location pairing. Our study obtains some financial meaningful results. We also provide an insight into the internal structure of the stock market.