Mathematics and Statistics Colloquium – Dynamic Factor Models in the Functional Domain: Methods for the Determination of Factor Cardinality
Friday, April 6, starting at 3:30 pm, Davis 201
Refreshments at 3:00 pm, outside of Davis 216
Recent developments in applied statistics have combined research in the area of functional data analysis to the analysis of time series, resulting in the field of functional time series analysis. Problematic in multivariate time series modelling is parameter proliferation. Dynamic factor analysis is a method that has served well in terms of decomposing a multivariate series into a smaller set of explanatory though latent factors. With the increasing availability and frequency of data collection, it is natural to consider the collection of that data as functional components rather than a collection of discrete processes. The application of dynamic factor analysis to the functional domain has its own challenges; and still retains the artifact of its discrete counterpart: how many factors are to be chosen? In some cases, the determination of the number of factors is dictated via an underlying theoretical model. Latent by their design, the determination of the number of factors in many more applications has not been thoroughly resolved and further not applicable to the functional domain. Presented herein are methods and applications for the selection of the number of factors to be selected in dynamic factor models with functional factors loadings.