NBER-NSF Time Series Conference; University of California at Davis, USA

Abstract for "Factor Models and VARMA Processes" by Dalibor Stevanovic

In this paper we generalize the existing approximate factor model analysis by specifying vector autoregressive moving average (VARMA) dynamics for latent factors. We show that when factors are obtained as linear combinations of observable series their dynamic process is generally a VARMA. Moreover, this generalization can be motivated by the usual arguments of parsimony, invertibility and marginalization issues in which VARMA models outperform the VAR representations. We apply our approach in two pseudo-out-of-sample forecasting exercises using an U.S. monthly balanced panel and a Canadian mixed-frequencies monthly panel taken from Boivin, Giannoni and Stevanovic (2009, 2008) respectively. We find that considering VARMA representation for factors help in predicting several key macroeconomic aggregates relatively to standard factor based forecasting models.