Title: Community detection in large vector autoregressions
Authors: Gudmundur Gudmundsson - Aarhus University (Denmark) [presenting]
Abstract: A family of vector autoregressive models of order 1 (VAR(1)) are introduced where the coefficient matrix is based on an underlying random network. Real-world networks are frequently endowed with a community structure, where the vertices form natural groups within which the frequency of linkages between vertices is higher than without. We therefore focus on the case where the network underlying the model has a community structure and introduce an algorithm to detect the communities consistently. The algorithm is based on spectral clustering and uses the eigenvectors of the coefficient matrix of the VAR(1) for detection. We apply the methodology to study clustering in industrial production and ETF volatility.