Optimum Instrumental Variable Methods of Time-Series Model Estimation

Occupying as it does such a curious position at the crossroads between theory and practice, the IV technique described in previous Chapters has tended to be frowned upon by certain statisticians and control theorists because it lacks the statistical niceties of its more sophisticated relations such as maximum likelihood. At the same time it is often misunderstood by many practising engineers and systems analysts, who sometimes tend not to appreciate fully the relevance of the structural model estimation problem and, at the same time, are apt to view the recursive IV algorithm as an over complicated approach to dynamic system estimation. This latter point is accentuated by the fact that many of the publications on the IV method in the control literature are couched in rather abstract mathematical terms which few practitioners find very digestible. And this is a pity since papers of this type can have practical significance, albeit obscured behind a formidable front of mathematical symbolism (see e.g. Finnegan and Rowe,1974).