Authors: Savas Papadopoulos
Addresses: Department of Economics, Democritus University of Thrace, University Campus, Komotini 69100, Greece; Department of Financial Stability, Bank of Greece, Amerikis 3, Athens 10250, Greece
Abstract: Factor analysis of multivariate longitudinal data are discussed, where measurements are taken from individuals at several occasions. Unbalanced cases, in which some individuals do not appear at all occasions and the number of measured individuals may change from one occasion to another, are considered. For such cases, the full likelihood method is difficult even if a particular distribution is assumed. In this paper, a relatively simple method based on a partial likelihood is considered, and is shown to have various advantages over the full likelihood method and the time-series modelling. It is shown that the associated inference procedures, including the goodness-of-fit statistic, have a good asymptotic performance for a broad class of non-normal data having any time trend. The proposed method is compared with standard methods using real data from the banking sector.
Keywords: latent variable modelling; structural equation modelling; SEM; robust standard errors; asymptotic normality; partial likelihood; goodness-of-fit statistics; efficiency; capital adequacy; credit risks; market risks; time effects; time heteroskedasticity; confirmatory factor analysis; non-normal panel data; banking industry; inference procedures.
International Journal of Computational Economics and Econometrics, 2013 Vol.3 No.3/4, pp.124 - 145
Accepted: 08 May 2013
Published online: 31 Dec 2013 *