A0393
Title: Nonlinear factor analysis for large sets of macroeconomic time series
Authors: Diana Mendes - ISCTE-IUL (Portugal)
Vivaldo Mendes - ISCTE-IUL (Portugal) [presenting]
Abstract: Dynamic factor models are frequently used for empirical research in macroeconomics. Several papers in this area have argued that the co-movements of large panels of macroeconomic and financial data can be captured by a relatively few common unobserved (latent) factors. The main purpose of this paper is to analyze and compare the transmission mechanism of monetary policy in the USA, by using a FAVAR (Factor Augmented Vector Auto Regression) model based on two different approaches, in order to include information from large data sets. The first approach consists of a classical linear methodology where the factors are obtained through a principal component analysis (PCA) and Granger causality, while the second one employs a nonlinear factor algorithm based on independent component analysis (ICA), nonlinear PCA, and cross-entropy. In comparison to PCA, the factors extracted by nonlinear methods provide a better performance in the Factor Augmented VAR model, which can be illustrated by Impulse Response Functions and forecasting. We perform the dynamic inference on the model by using a dynamic Bayesian estimation.
