DMCA
On Least Squares Optimization of Linear Dynamic Systems (1993)
BibTeX
@MISC{Bijleveld93onleast,
author = {Frits D. Bijleveld and Catrien C. J. H. Bijleveld and Jan de Leeuw},
title = {On Least Squares Optimization of Linear Dynamic Systems},
year = {1993}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
We discuss two estimation methods for fitting linear dynamic systems. The first is the existing DYNAMALS algorithm, that uses Alternating Least Squares and a majorization substep. With this method, it is difficult to ensure that the latent states are completely in the space of the predictor variables. We propose an alternative method that uses a single step algorithm. After direct implementation, the latent states are in the space of the predictor variables. The proposed method can also estimate intercepts in the system and measurement equations. The porposed method is compared with the existing DYNAMALS method using a real-life example. Keywords: Longitudinal reduced rank regression, state space modelling, optimization methods. 1 Introduction. Recently, Bijleveld and De Leeuw proposed an algorithm for fitting the longitudinal reduced rank regression or state space model (Bijleveld & De Leeuw 1991) The algorithm used least squares and majorization substeps. Its main advantage over ex...
Keyphrases
linear dynamic system square optimization latent state predictor variable rank regression single step algorithm alternating least square main advantage alternative method dynamals algorithm bijleveld de leeuw state space model measurement equation optimization method dynamals method state space modelling majorization substep real-life example estimation method direct implementation de leeuw porposed method majorization substeps
