Title: Detecting granular time series in large panels
Authors: Geert Mesters - Universitat Pompeu Fabra (Spain) [presenting]
Christian Brownlees - UPF (Spain)
Abstract: In large panels of economic and financial time series there are series that have a pervasive influence over the entire panel. We call such series ``granular'', in the sense that as the cross section dimension increases such series do not become negligible. We introduce methodology for detecting and testing granular series. We introduce a signaling device that ranks the series in order of granularity. The tests that we develop are designed to infer which of the series form the granular series. The tests have power against a variety of other popular structures for the covariance matrix, such as factor, sparse and block structures.The methodology is examined in a large Monte Carlo study as well as for empirical applications in macroeconomics and finance.