This thesis considers two aspects of statistical inference associated with the linear regression model set in an economic context. The implications of replacing the conventional normality assumption with the broader assumption that the disturbances follow an elliptically symmetric distribution, are investigated, and three features of the problem of detecting serial correlation in elliptically symmetric disturbances, are studied.

An examination of the conventional justification of the normality assumption in econometrics, conducted in Chapter 2, provides motivation for the study of regression analysis under the elliptical symmetry assumption.

In Chapters 4 and 5, properties of estimators and tests associated with the linear regression model are investigated, assuming elliptically symmetric disturbances. This broadening of the normality assumption is found to have few practical consequences for classical regression analysis. The usual least squares estimators are shown to satisfy a stringent optimality property. Conditions are determined for weak consistency and for strong consistency of these estimators. Distributions of statistics invariant to the disturbances' scale are found to be unaffected by the broadening of the normality assumption, while the distributions of arbitrary statistics can be viewed as mixtures of their distributions for different scales of the disturbances under normality. The implications of these results for hypothesis testing, are explored.

Chapter 6 attempts to find an