Singular spectrum analysis for time series. (Q690695)

This book is fully devoted to the methodology of a technique for time series analysis and forecasting called singular spectrum analysis (SSA). This technique combines elements of classical time series analysis, multivariate statistics, multivariate geometry, dynamical systems and signal processing. SSA aims at decomposing the original series into a sum of a small number of interpretable components such as a slowly varying trend, oscillatory components and a `structureless' noise. It is based on the singular value decomposition of a specific matrix constructed on the time series. Neither a parametric model nor stationarity-type conditions have to be assumed for the time series. This makes SSA a model-free method and hence enables SSA to have a very wide range of applicability. The authors of the book are well-known statisticians, and specialists in time series analysis. Their previous book joint with \textit{V.V. Nekrutkin} [Analysis of time series structure. SSA and related technques. Boca Raton, FL: Chapman and Hall (2001; Zbl 0978.62073)] has achieved very good reviews and has become a classical reference in the field. Unlike their previous book, the present book does not consider theoretical aspects and only deals with methodology of SSA. It consists of three chapters: (i) An extended introduction with detailed description of the book structure, its topics, and literature reviews, (ii) a chapter devoted to the application of SSA to time series analysis and decomposition of series into separate components, and (iii) a chapter discussing a variety of topics such as forecasting, missing data imputation, parameter estimation and filtration. The main methodological concept within the book is that of separability (that is, possibility of extraction of components from their observed sum). The book summarizes the current state of the art of SSA. In comparison with the previous book of the authors mentioned above, many methodological aspects are extended and several entirely new topics are included. New aspects of separability including the use of Independent Component Analysis are considered. The methods of automatic identification of SSA components for extraction of series components such as trends and periodicities are described. Fast implementation of SSA is also considered, since the quality of implementation strongly influences the methodology. One of SSA's capabilities is its ability to be a frequency filter. The book describes the relation between basic SSA and filtration in a general (not just spectral) form and also considers specific filters generated by basic SSA. In particular, causal SSA, which can also be called last-point SSA, is considered from the viewpoint of causal filters. Moreover, the book considers several new issues related to models of series governed by linear recurrence relations including sums of exponentially-modulated (damped and undamped) sinusoids. It is shown how SSA analysis can be supplemented by series modeling including frequency estimation. SSA forecasting (extrapolation) and missing data imputation (interpolation) are also considered based on the use of the model mentioned above but without direct estimation of the model parameters. Extended recommendations on the choice of SSA forecasting methods and parameters are provided. The circle of potential readers of this book includes professional statisticians and econometricians, specialists in any discipline where problems of time series analysis and forecasting occur, specialists in signal processing and those who need to extract signals from noisy data, Ph.D. students and students working on topics related to time series analysis. I believe that this book is a valuable addition to the literature on time series analysis and will therefore be well received by statisticians and specialists in many other fields interested in the analysis of time series data.