Model structure selection for multivariable systems by cross-validation methods
The authors describe an application of the cross-validation criterion for finding multi-variable ARMA models for vector processes. It is well known that such models pose a particular difficulty for estimation, because the models appear in several structures. A major result is to show that the criterion is asymptotically equivalent to Akaike's criterion. This discussion may be regarded as an elaboration of a related one by \textit{M. Stone}, J. R. Stat. Soc., Ser. B 36, 111--147 (1974; Zbl 0308.62063). The paper is well written, and it describes the criterion with a sound criticism; for example, the criterion fails to estimate the number of parameters consistently.
- A new look at the statistical model identification
- Convergence analysis of parametric identification methods
- Kronecker products and matrix calculus in system theory
- Model-structure selection by cross-validation
- On line structure selection for multivariable state-space models
- On non-singular information matrices and local identifiability
This page was built for publication: Model structure selection for multivariable systems by cross-validation methods
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3802457)