Title: FAVAR revisited: A sparse dynamic factor approach
Authors: Simon Beyeler - University of Bern (Switzerland) [presenting]
Sylvia Kaufmann - Study Center Gerzensee (Switzerland)
Abstract: Extracting relevant information from many different time series measuring different aspects of an economy and compressing it using factor analysis is a neat way to handle large amounts of data and to circumvent the curse of dimensionality problem without ignoring possibly important features. To do so, we combine the FAVAR framework with recently developed estimation and identification procedures for sparse dynamic factor models. Our estimation procedure allows us to explicitly discriminate between zero and non-zero factor loadings. This provides one solution to the identification common to all factor models. Further, the identified unobserved factors get a meaningful economic interpretation due to the structure of non-zero loadings. An additional distinction to traditional factor models is that we work with correlated factor innovations allowing us to implement different strategies to identify structural shocks used in the literature and to perform traditional structural VAR type analysis. Applying our methodology to US macroeconomic data (FRED QD) reveals indeed a high degree of sparsity in the data. The proposed identification procedure yields seven unobserved factors that together account for about 55 percent of the variation in the data. All of them are clearly related to an economic concept.