科研成果详情

Despite the widespread practice of pre-testing, the unfavourable risk properties of pre-test estimators are well-documented; hence, researchers have been searching for alternatives that present less risk. The weighted averaged least squares (WALS) estimator with weights based on a Laplace prior density, proposed by Magnus (2002), has attractive risk performances in that it has lower risk than the ordinary least squares (OLS), restricted least squares (RLS), and pre-test (PT) estimators around the region where the risk functions of the OLS and the RLS estimators intersect. However, it is only shown in the literature that the WALS estimator is better when the error structure is correctly specified. Thus, this paper aims to examine the robustness of the WALS estimator under various non-ideal conditions and compare its performance with the traditional estimators, as well as such shrinkage estimators as the generalized ridge (GR) and least angle (LA) estimators. Through Monte Carlo simulations, we show that although the WALS estimator is not the best in terms of the risk performance, it is the most stable estimator under various conditions. Hence, in situations when we are uncertain about what estimator to choose, the WALS estimator provides a good alternative.