| Article ID: | iaor19942535 |
| Country: | United Kingdom |
| Volume: | 13 |
| Issue: | 2 |
| Start Page Number: | 69 |
| End Page Number: | 90 |
| Publication Date: | Mar 1994 |
| Journal: | International Journal of Forecasting |
| Authors: | Aoki M. |
This paper discusses two alternative innovation representations (forward and backward) of a state space model of a multivariate weakly stationary time series, and suggests estimators of system matrices and innovation noise covariances which are alternative to these based on stochastic realization theory. A scheme for iterative improvements of the initial estimates is also suggested.