Title: FAVAR models for mixed-frequency data
Authors: Franz Ramsauer - Technical University of Munich (Germany) [presenting]
Michael Lingauer - Technische Universitaet Muenchen (Germany)
Abstract: The previous Factor-Augmented Vector Autoregression (FAVAR) Model is extended to mixed-frequency and incomplete panel data. Within the scope of a fully parametric two-step approach, the alternating application of two expectation-maximization algorithms jointly estimates the model parameters and missing observations. Furthermore, it addresses the selection of the factor dimension and autoregressive order. In contrast to non-parametric two-step estimation methods comprising principal component analyses and linear regressions, we use maximum likelihood estimation. Thereby, we derive equations for the Kalman filter and smoother, which explicitly take into account that the factors consist of latent and observed components. To eliminate any identification problem of the model parameters we constrain the loadings matrix. Our empirical study applies the presented framework to U.S. data for measuring the effects of the monetary policy on the economy and the financial markets. In this context, the consequences for the quarterly growth rates of the Gross Domestic Product (GDP) are of particular importance.