Author: CHEUNG Ying Lun, Goethe University Frankfurt
Abstract: This paper considers the estimation of the time-varying factor (TVF) model. We assume the TV-loadings to be Hölder continuous, and by applying appropriate identification restrictions, we show that the latent factors can be consistently estimated without rotation. We propose to approximate the factor loadings with B-spline, and estimate them by sieve estimation. Asymptotic behavior of the estimator is studied. Finally, we propose an information criteria approach to test for time variations in loadings. Simulation shows good finite sample properties of our estimation and testing procedure. We also apply the test on a large dataset of US macroeconomic variables, and find evidence of structural instability in 80% of the series.