Linear stochastic systems with constant coefficients. A statistical approach (Q797554): Difference between revisions

This is a book for specialists in the field of analysis and modeling of time series. It is roughly divided into three parts. The first part (Chapter 1) presents a statistical investigation of several problems of various characters. The Brownian motion, torsion pendulum and electrical circuits, rotation of the earth, measurement of performance in computer systems, round-off errors in solutions of ordinary differential equations and sunspot activity are considered from a statistical standpoint. The problem of Kalman filtering is also treated in this part, and an explicit solution of it is given in the case of continuous-time processes described by stochastic first-order differential equations. The material presented in this first part heavily relies in places on the treatment and results in the next chapters of the book. It can be best read after a first reading of the chapter 2-4. The second part (Chapter 2) contains a detailed treatment of stationary Markov processes with Gaussian distributions. These processes are called ''elementary Markov processes'' (EMP) by the author. Generation of EMP as solutions of first-order stable stochastic vector difference (in discrete- time) or differential (in continuous-time) equations is discussed and the covariance structure of EMP derived. The strong existing link between EMP and stationary Gaussian processes with rational spectral density (the so- called autoregressive moving-average (ARMA) processes) is shown. The important problem of deriving density functions and sufficient statistics for EMP is also part of this chapter. The third part (Chapters 3 and 4) presents the maximum likelihood (ML) estimation theory for one-dimensional respectively multi-dimensional EMP. ML estimators for the parameters of EMP are derived and their asymptotic distributional properties are established. Finally, the book contains two appendices with background material on linear differential equations with constant coefficients, respctively basic results on Gaussian systems and basic concepts in probability theory. It should be pointed out that there are several recent developments in the field of estimation of parameters of ARMA processes (possibly with exogeneous inputs) which are not even mentioned in the book. Most of these results have appeared in the control literature and may thus be less well-known to statisticians. Another minor criticism concerns the fact that only some few references from the extensive list of books and papers included are cited in the text. The aforementioned drawbacks may be compensated by the advantage to have a book treating together continuous-time and discrete-time EMP and revealing the interplay between these two treatments. This is indeed a special feature of the present book.