University of Ioannina
The asymptomatic refinement of the Linear Model with Non-Scalar Error Covariance Matrix
Abstract
dc:descriptionIn the linear regression model with non-scalar error covariance matrix, Refined Generalised Least Squares (RGL) estimators are developed, with better small-sample properties compared to the feasible FGL estimators. The RGL estimator of a structural parameter approximates the benchmark GLS estimator better than the corresponding FGL estimator. The RGL estimator has better bias, variance, and mean square error, in small samples, compared to the corresponding FGL estimator. The RGL estimator results in small-sample size corrected t and F econometric tests. The RGL size corrections and the corresponding Edgeworth and Cornish-Fisher size corrections, used for the t and F tests based on the FGL estimator, are second-order equivalent. For all these theoretical reasons, we suggest the use of the RGL estimator instead of the FGL estimator. In the linear regression model with AR(1) random errors, the RGL size corrected t and F tests are preferable for large absolute values of the autocorrelation coefficient, while the corresponding FGL size corrected econometric tests are preferable for small and intermediate absolute values of the autocorrelation coefficient. In practice, since the true autocorrelation coefficient is unknown, the researchers' choice of suitable estimation and testing techniques is based on the estimated autocorrelation coefficient. Although the Edgeworth, Cornish-Fisher, and RGL size corrections have an error of the same order of magnitude, the corresponding t and F size corrected tests may differ considerably in small samples. Given that these three size corrections can be compared only by means of simulation experiments, there is no guarantee that the simulation results hold in every practical situation. For this reason, in each particular problem, the researcher should apply both the FGL and RGL estimators, and all three small-sample size corrections, to decide the most suitable one.
Degree
thesis:*- Grantor dc:publisher
- University of Ioannina
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Papachristodoulou, Theodosia
- Παπαχριστόδουλου, Θεοδοσία
Subjects
dc:subject × 12Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/58560