No label defined (Q3314788)
From MaRDI portal
!
This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:
scientific article; zbMATH DE number 3844867
| Language | Label | Description | Also known as |
|---|---|---|---|
| default for all languages | No label defined |
||
| English | No label defined |
scientific article; zbMATH DE number 3844867 |
Statements
1983
0 references
strong law of large numbers
0 references
central limit theorem
0 references
law of iterated logarithm
0 references
Toeplitz quadratic forms of time series
0 references
martingale methods
0 references
stationary processes
0 references
model identification
0 references
0.7681514620780945
0 references
0.7575679421424866
0 references
The main result of this paper relates to the law of the iterated logarithm for expressions of the form \((1/2\pi)\int^{\pi}_{-\pi}| \sum^{n}_{1}X(t)\exp it\lambda |^ 2g(\lambda)d\lambda,\) which is called \(X^ n(g)\), where g is continuous and X(t) is stationary and purely nondeterministic, with a representation \(X(t)=\sum^\infty_{0}\alpha_ j\epsilon(t-j),\) converging in mean square, the \(\epsilon\) (t) being martingale differences with finite fourth moment and, additionally, \(E\{\epsilon(t)^ 2| \epsilon(s),\quad s\leq t\}=\sigma^ 2=E\{\epsilon(t)^ 2\}\), a.s. This result is then used to study the behaviour of estimators of a parameter point \(\theta\in \Theta\), obtained by minimising \(L_ n(\theta,\sigma^ 2)=\sigma^{-2}X^ n(g_{\theta})-n/2 \log \sigma^ 2,\) where now \(g_{\theta}=(-1/2)| \Phi_{\theta}(\lambda)|^{-2},\quad \Phi_{\theta}(\lambda)=\sum \alpha_ j\exp ij\lambda.\) It is required that \(\Theta\) be compact and \(\theta_ 0\), the true point, be interior and that \(\Phi_{\theta}=\Phi_{\theta '}\), a.s. with respect to \(\lambda\), should imply \(\theta =\theta '.\) A main application considered is to a family of closed subsets, T, of \(\Theta\) and to the consistent estimation of a minimal intersection of these containing \(\theta_ 0\). The estimation is by choosing \(T_ n\) to maximise \({\mathcal L}_ n(T)=\sup L_ n(\theta,\sigma^ 2)-\delta_ n(T),\) where the sup is over T and \(\delta_ n(T)\) satisfies lim sup \(\delta {}_ n(T)/n=0\), \(\theta_ 0\in T\); \(\lim \inf \{\delta_ n(T)- \delta_ n(T_ 0)\}/(\log \log n)>1,\) a.s., \(\theta_ 0\in T\), \(T\neq T_ 0.\) It must be mentioned, that in spite of contrary statements in the paper, this last result does not deal with the case of the determination of the order of an ARMA process (i.e. the degrees of the polynomials of which its transfer function is a ratio) and such results are more difficult to obtain. This is because if \(\Theta\) is taken to be defined in terms of the coefficients in the polynomials of the transfer function, and the true order is less than the maximum order considered, then \(\Phi_{\theta}\) will be constant along an affine subspace where the true point lies while if, more relevantly, \(\Theta\) is defined to be the union of spaces of transfer functions of orders up to the true order then \(\theta_ 0\) will not be interior, if its order is less than the maximum order.
0 references