Elements of multivariate time series analysis (Q5894742): Difference between revisions

The field of statistical time series analysis has grown vigorously in the last three decades. Until the sixties, there were few researchers in the area, a limited number of published textbooks and monographs was available, and the computational aspects associated with the empirical analysis of time series had not been solved to the point of allowing the routine use of the methods. Beginning in the seventies, important developments took place, and we now find the field strong among statistical areas, with many qualified researchers in both theory and applications, good books with various approaches, and a great deal of experience in the application of the methods to empirical analysis. These remarks, however, apply more fully to the univariate analysis of time series, since ``the situation is not so complete for the multivariate case'', as the author writes in his preface. The present book is devoted to contribute one approach to filling in the indicated gap. In the author's words, ``it is designed to introduce the basic concepts and methods that are useful in the analysis and modeling of multivariate time series, with illustrations of these basic ideas. The development includes both traditional topics such as autocovariance and autocorrelation matrices of stationary processes, properties of vector ARMA models, forecasting ARMA processes, least squares and maximum likelihood estimation techniques for vector AR and ARMA models, and model checking diagnostics for residuals, as well as topics of more recent interest for vector ARMA models such as reduced rank structure, structural indices, scalar component models, canonical correlation analysis for vector time series, multivariate unit-root models and cointegration structure, and state-space models and Kalman filtering techniques and applications.'' In terms of these ideas, the seven chapters of the book can be grouped as follows: traditional methods occupy most of four chapters, two dealing with probabilistic analysis, namely, Chapter 1, ``Vector time series and model representations'', and Chapter 2, ``Vector ARMA time series models and forecasting'', and two dealing with statistical analysis, Chapter 4, on ``Initial model building and least squares estimation'', and Chapter 5 on ``Maximum likelihood estimation and model checking''. Chapters 1 and 4 have short appendices on technical material: matrices, multivariate normal, multivariate linear regression, etc. Topics of current interest occupy most of three chapters, Chapter 3 on ``Canonical structure'', Chapter 6 on ``Reduced-rank and nonstationary co-integrated models'', and Chapter 7 on ``State-space models, Kalman filtering, and related topics.'' About the mathematical level and requirement of the book, the theorem- proof format has not been used, ``elaborate and detailed mathematical developments and arguments are generally not emphasized, although substantial references are usually provided for further mathematical details.'' The reader of the book is assumed to have some knowledge of matrix algebra methods, and a working background in time series modeling techniques. More precisely, the author's statement is that ``the book could serve as a graduate-level textbook on `multivariate time series' for a second course in time series as well as a reference book for researchers and practitioners in the area of multiple time series analysis''. The presentation does a good job in collecting and reviewing a sizable portion of the recent literature on multiple time series, with a selection of technical material. As such, the text, together with a reasonable use of the original literature cited when appropriate, can be a valuable asset for those interested in the theoretical developments associated with these topics. At the end of the book there is a collection of 33 exercises and problems, without solutions. A special word can be said about data sets and numerical examples. An appendix contains the listing of 7 data sets in 2, 3 or 4 variables, of various lengths, some of which have been analyzed in the recent literature. The series deal with business and economic problems, and serve as illustrations in various parts of the book. The analyses in the corresponding sections are quite brief, little motivation is given about the multivariate approach that is used, but references to other sources are often provided. In Chapter 1 the author states that one objective of multivariate time series analysis is ``to improve accuracy of forecasts for individual series by utilizing the additional information available from the related series in the forecasts for each series.'' However, no such analysis is pursued in the examples presented in the text. Also little mention is done of the computational problems; the computer packages SAS and SCA are mentioned only briefly in pages 143 and 215, respectively, and in both cases cautionary notes are inserted about the relation with the author's text. The packages are not included in the index, and SAS is also missing among the references. Some misprints were noted, for examples, Figure 4.3 is misplaced. The author has written a useful monograph on multivariate time series, with emphasis on the probabilistic and statistical theory, and on recent developments in the area. The formal (i.e. mathematical) level of the book corresponds to the graduate level. The book should prove useful for researchers, and advanced students interested in the field; they will find a treatment of many important papers written recently with novel proposals. The book does not stress computational aspects, and those willing to reproduce the computations suggested or presented in the book, will have to resort to the references or other material. Numerical examples are provided, with references to original sources.