A matricial extension of the Helson-Szegö theorem and its application in multivariate prediction (Q1069253)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | A matricial extension of the Helson-Szegö theorem and its application in multivariate prediction |
scientific article |
Statements
A matricial extension of the Helson-Szegö theorem and its application in multivariate prediction (English)
0 references
1985
0 references
Let \(\{X_ n\}\) be a q-variate weakly stationary stochastic sequence with the spectral density W. Let \(\hat X_ v\) denote the linear predictor of \(X_ v\), \(v\geq 1\), based on \(\{X_ k\), \(k\geq 0\}\). An important problem in the prediction theory of stationary sequences is to find conditions on W which are necessary and sufficient for the existence of a mean-convergent autoregressive series representation for \(\hat X_ v\) in the form (1) \(\hat X_ v=\sum^{\infty}_{k=0}A_ kX_{-k}\), and to devise algorithms for the determination of the coefficients \(A_ k\) in terms of W. Under the conditions \(W\in L^{\infty}\) and \(W^{-1}\in L^ 1\), it is known [\textit{P. Masani}, Acta Math. 104, 141-162 (1960; Zbl 0096.115)] that (1) holds and that the \(A_ k's\) are computable by the following scheme: Let \(\Phi\) be the optimal factor of W and let \(C_ k\) and \(D_ k\) be the kth Fourier coefficient of \(\Phi\) and \(\Phi^{-1}\), respectively. Then (2) \(A_ k=\sum^{k}_{n=0}C_{v+n}D_{k-n}\). Even when W is not in \(L^{\infty}\) it is possible that the representation (1) could hold with \(A_ k's\) as given in (2) for certain sequences as shown in the recent work by \textit{A. G. Miamee} and the reviewer, On explicit representation of the linear predictor of a weakly stationary stochastic sequence. Bol. Soc. Mat. Mex., II. Ser. 28, 81-93 (1983). Guided by this and other results, the present author proceeds to examine the connection between the validity of the autoregressive representation (1) and the positivity of the angle between the past and future subspaces of the sequence \(\{X_ n\}\). To give the reader the flavor of the type of the results proved in this paper, we state the following: representation (1) holds with \(A_ k's\) as in (2) if the angle between the past and future subspaces of \(\{X_ n\}\) is positive. Reviewer's remark: the positivity of the angle between the past and future for a multivariate \(\{X_ n\}\) is more geometric than analytic in nature. The analytic conditions of Masani, cited above, and the positivity of the angle between the past and future of \(\{X_ n\}\) are not equivalent, although each implies the representation (1). On page 272 of the article under review it is claimed that for \(W(\theta)=| \theta |^{\lambda}\), 1/2\(\leq \lambda \leq 1\), the angle between the past and future is zero. This is not true, as the discussion on p. 133 of \textit{H. Helson} and \textit{G. Szegö}, Ann. Mat. Pura Appl., IV. Ser. 51, 107-138 (1960; Zbl 0178.500) demonstrates the fallacy of this claim.
0 references
Helson-Szegö theorem
0 references
multivariate stationary process
0 references
prediction
0 references
Fourier series
0 references
weakly stationary stochastic sequence
0 references
spectral density
0 references
linear predictor
0 references
mean-convergent autoregressive series representation
0 references
angle between the past and future subspaces
0 references
0 references
0 references