Abstract
The first phase of transfer function model identification is preliminary estimation of transfer function weights. Previous studies by the author have shown that ordinary least-squares estimates in most cases can be improved significantly by using ridge regression techniques. In particular the use of the RIDGM and Lawless and Wang estimators are recommended. The main object of this paper is to compare the ridge estimators to other relevant estimators on real data. For this purpose business cycle data from Sweden have been collected and transfer function models have been identified by ridge regression, prewhitening and cross-spectral analysis. The results indicate that the ridge estimators compare favorable to the prewhitening estimator. The prewhitening technique advocated by practitioners seems to generate too many significant weights which complicates the search for a suitable model. The theoretically correct cross-spectral estimator gives too few significant weights and fails to identify two of the three models.