Diagnosis of Small-world Bias in Random Graphs

Abstract

Background: Imagine a paper with n nodes on it where each pair undergoes a coin toss experiment; if heads we connect the pair with an undirected link, while tails maintain the disconnection. This procedure yields a random graph. Now consider duplicating this network onto another paper with a slight bias-a fraction of its links (approximately 1/10) undergo rearrangement. If we shuffle the two papers, how can we distinguish the pure random graph from the biased one? Results: In response to this challenge, we propose a novel metric called Randomness Index (RI). The closer the metric to zero is, the higher degree of randomness in the graph. The RI can distinguish between dense small-world networks and dense random graphs; a distinction which is impossible by conventional small-world properties like clustering coefficient and average path length. To validate its effectiveness, we apply the RI to temporal correlation networks of stock indices. Our findings reveal a reduction in randomness during global economic recession periods. Conclusion: The RI emerges as a powerful metric capable of characterizing small-world topology, especially in scenarios where other network measures fail. Beyond its utility in network analysis, the RI is promising for change-point (anomaly) detection in dynamical systems studied by means of multivariate time series.