Time series and linear systems (Q1087284): Difference between revisions

This volume contains seven tutorial papers centered on time series as embedded into the linear systems theory, a standpoint quite fruitful in most applications. The contributions are intended to review the recent progress in the field. The intention of the editor that the volume can be ''used as a textbook for monographic courses'' is only partly fulfilled, because of the poor connection between the chapters and of the differences in their style. The book is valuable rather for the quick orientation in some advanced topics in the field. The second goal of the volume, i.e. to be an introduction to research directions is indeed attained. The first chapter is ''Time series and stochastic models'', by \textit{E. J. Hannan}. After contrasting the finite parameter model with known inputs to other ones, the approximation by a system of finite degree is discussed. The rational transfer function approximation is revealed to be difficultly manageable. Then one introduces the Gauss-Newton procedure, and examples are given. The remarks on both applications and references are interesting. The paper is dense and contains many valuable discussions of the literature. It can be used as a survey. The second chapter is ''Linear errors-in-variables systems'', by \textit{M. Diestler}. For linear systems with both inputs and outputs subjected to errors, the problems discussed are identifiability and the determination of observationally equivalent characteristics. The information provided by the second order and higher order moments is analysed. Problems dealt with are: the possibility of infering on the causality for rational transfer functions, the second order moments being known; conditions for unicity of the transfer function; properties of the transfer function, the spectrum and the input errors being known a.s.o. Conditions for identifiability and for uniqueness of the transfer function are also given. The third chapter is ''A new class of dynamic models for stationary time series'', by \textit{G. Picci} and \textit{S. Pinzoni}. The paper deals with a generalization of the EV model, namely the dynamic errors-in-variable models. Thus, factor analysis models are included in the discussed ones. ARMAX models and multivariate time series analysis are refered under this frame. The paper includes some entaicing results. Chapter 4 is ''Predictive and nonpredictive minimum description length principles'', by \textit{J. Rissanen}. This is a good tutorial on the stochastic complexity, minimum length principles and optimal models. Applications are illustrating the theory in an appropriate manner. Chapter 5, ''Deterministic and stochastic linear periodic systems'', by the editor, is a review of the main properties of these systems. The paper has the merit of being well structured and indeed can be used as basic material for both lectures and research. Structural properties are emphasized and deterministic periodic systems are detailed before dealing with stochastic systems. References are not just quoted, but discussed. Chapter 6, ''Numerical problems in linear system theory'', by \textit{D. Boley} and the editor, has the same clear style as the previous chapter. After a review of the basic computation methods, (LU decomposition, orthogonal decomposition), some more complex methods as Schur and singular value decompositions are discussed in detail with respect to the computational problems they pose. Chapter 7, ''Some recent developments in econometrics'', by \textit{M. McAleer} and \textit{M. Diestler} is an introduction into the topic emphasizing some new methods. A deeper discussion of the strength of the methods and of their usefulness could be expected from an applications- oriented paper.