A0366
Title: Weak factors are everywhere
Authors: Philipp Gersing - University of Vienna (Austria) [presenting]
Christoph Rust - IREEN GmbH (Austria)
Manfred Deistler - Vienna University of Technology (Austria)
Matteo Barigozzi - University of Bologna (Italy)
Abstract: There are two different approaches to time series factor models: a) The approximate factor model, where the factors are loaded contemporaneously to the common component. b) The generalised dynamic factor model (GDFM), where the factors are loaded with lags. By introducing the canonical decomposition of factor models, it is shown how both approaches are related, and their conceptual difference is clarified: Both models entail two different types of common/idiosyncratic components, respectively. The canonical decomposition includes what is called the weak common component, which is the difference between the dynamic- and the static common component. It is driven by potentially infinitely many non-pervasive factors, i.e., weak factors (not to be confused with factors being pervasive but at a slower rate). It exemplifies why these types of weak factors should not be neglected in theory and practice. Furthermore, a simple estimator for the canonical decomposition is proposed and applied to US macroeconomic data. The estimates reveal that the weak common component can account for up to 20\% of the total variation of individual variables.
