All Times EDT
Keywords: Multivariate,Functional Time Series,Hilbert Space,Singular Spectrum Analysis,SVD,Functional Data Analysis
In this work, we develop multivariate functional singular spectrum analysis (MFSSA) which is the functional extension of multivariate singular spectum analysis (MSSA). The algorithm is based on ideas from univariate functional singular spectrum analysis (FSSA) and is flexible in the sense that it can be used to decompose a multivariate functional time series (MFTS) that is comprised of covariates taken over different dimensional domains. We also present work done with horizontal MFSSA (HMFSSA) which is the functional extension of horizontal MSSA (HMSSA). We provide a simulation study showcasing the improvement in reconstruction accuracy of a MFTS signal found using MFSSA as compared to other approaches including HMFSSA. We also provide a real data study that illustrates how MFSSA behaves similarily to its MSSA counterpart in providing a richer analysis of multivariate data.