This thesis considers the estimation of static and dynamic exact factor models inter alia on data sets with an arbitrary pattern of missing data. We also outline the probabilistic principal component analysis of Tipping and Bishop [1999] to estimate static factor models with an isotropic idiosyncratic component, whereby the underlying data is fully observable. Thereafter, the parametric estimation approach of Banbura and Modugno [2010] and Ban ́bura and Modugno [2014] for static and dynamic exact factor models, which can handle an arbitrary fraction of missing data in the observation matrix, is studied step-by- step. The dynamics of the factors is described by a vector autoregressive model of order p. The estimation framework combines the expectation maximisation algorithm with the Kalman filter and the Kalman smoother to obtain the maximum likelihood estimates of the unknown model parameters and the moments of the latent factors. The estimation methodology allows us to handle data sets of economic indicators characterised by different frequencies, publication delays, and sample lengths. To verify the estimation procedure for static and dynamic factor models, a Monte Carlo analysis is implemented. Finally, an empirical application study to forecast the EURO STOXX 50 index is performed.
«
This thesis considers the estimation of static and dynamic exact factor models inter alia on data sets with an arbitrary pattern of missing data. We also outline the probabilistic principal component analysis of Tipping and Bishop [1999] to estimate static factor models with an isotropic idiosyncratic component, whereby the underlying data is fully observable. Thereafter, the parametric estimation approach of Banbura and Modugno [2010] and Ban ́bura and Modugno [2014] for static and dynamic exac...
»