The thesis is based on three papers. Each paper investigates different aspects of pretesting in regression analysis. In addition, the introduction contains a review of the literature and a discussion of recent developments. The first paper studies the effects of ignoring the effects of pretesting in estimation. It is shown that not reporting the correct moments leads to a significant distortion of the accuracy of many widely used pretest estimators. The second paper concerns forecasting. It derives the unconditional moments of the one-step ahead WALS forecast and applies these results to stock market forecasting. The third paper investigates properties of the neutral Laplace WALS estimator in the important case where the variance of the innovations is not known.
|Qualification||Doctor of Philosophy|
|Award date||7 Feb 2003|
|Place of Publication||Tilburg|
|Publication status||Published - 2003|