No label defined (Q4516211)
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 1535932
| Language | Label | Description | Also known as |
|---|---|---|---|
| default for all languages | No label defined |
||
| English | No label defined |
scientific article; zbMATH DE number 1535932 |
Statements
27 November 2000
0 references
minimax inhomogeneous linear estimators
0 references
regression model under restrictions
0 references
minimax mean squared error
0 references
minimax bias
0 references
0.8789333701133728
0 references
0.851013720035553
0 references
0.8494362831115723
0 references
The authors consider a linear regression model of the form \({\mathbf Y}={\mathbf X}\beta+\varepsilon\), where \({\mathbf Y}=(Y_1,\dots,Y_{n})\) are observations of a random response variable, \({\mathbf X}\) is a given matrix, \(\beta=(\beta_1,\dots,\beta_{m})\) is a parameter vector, and \(\varepsilon=(\varepsilon_1,\dots,\varepsilon_{n})\) is a vector of random errors with zero mean value and with the unknown covariance matrix \(\text{Var}(\varepsilon)={\mathbf W}\). It is assumed that the unknown parameter \(\varepsilon\) belongs to a set \({\mathcal G}\subset{\mathbf R}^{m}\) and the unknown matrix is from a specific set \(\mathcal R\) of nonnegative definite matrices. The problem is to estimate a linear parameter function \(\gamma={\mathbf c}'\beta\) with the help of inhomogeneous linear estimators \(\hat\gamma={\mathbf u}'{\mathbf Y}+a\), where \({\mathbf u}\in{\mathbf R}^{n}\) and \(a\in{\mathbf R}\). Let \({\mathbf a}=(u_1,\dots,u_{n},a)\). An inhomogeneous linear estimator \(\gamma^{\ast}={\mathbf u}'_0{\mathbf Y}+a_0\) is called a minimax inhomogeneous linear estimator (MILE) if \[ \inf_{{\mathbf a}\in{\mathbf R}^{n+1}} \sup_{\beta\in{\mathcal G},{\mathbf W}\in{\mathcal R}} {\mathbf E}(\gamma-\hat\gamma)^2= \sup_{\beta\in{\mathcal G},{\mathbf W}\in{\mathcal R}} {\mathbf E}(\gamma-\gamma^{\ast})^2=\sigma^2<\infty, \] where \(\sigma^2\) is the minimax mean square error of the MILE \(\gamma^{\ast}\). First the authors investigate the problem of existence of MILE. Then the minimax mean square error of the MILE is found using the Bayes approach. Finally, exact formulas for calculation of the MILE under elliptical restrictions on the parameter space and for two special classes of possible covariance matrices \(\mathcal R\) are derived. It is shown, for example, that a special choice of a constant \(q_0\) and a matrix \(W_0\) which determine one of the discussed classes \(\mathcal R\) leads to the well-known admissible estimator of \textit{J. Kuks} and \textit{V. Ol'man} [Izv. Akad. Nauk Ehst. SSR, Fiz. Mat. 21, 66-72 (1972; Zbl 0227.62019)] with known covariance matrix \(W_0\).
0 references