This thesis consists of two chapters on time series modelling. The first chapter introduces a class of vector autoregressive (VAR) models with a community structure for large panels of time series. In the model, the series are partitioned into latent groups such that spillovers are stronger within groups than between them. We then propose an algorithm that uses the eigenvectors of a function of the estimated autoregressive matrices to recover the communities. We study the properties of the procedure and establish its consistency. The algorithm motivates us to suggest a regularised VAR estimator, which performs favourably relative to a number of alternatives in a forecasting exercise. The methodology is applied to study clustering in industrial production for a set of major economies. The second chapter introduces a class of partial correlation network models with a community structure. The series form unknown groups, where correlation is higher within groups than otherwise. We propose an algorithm that consistently detects the communities using the eigenvectors of the sample covariance matrix. The procedure is used to analyse real activity clustering in the U.S. and Europe.
© 2001-2024 Fundación Dialnet · Todos los derechos reservados