[BOOK][B] Estimation and inference in econometrics

R Davidson, JG MacKinnon - 1993 - cambridge.org
1993cambridge.org
The new book by Davidson and MacKinnon represents a valuable contribution to the
econometrics literature and is meant to serve both as a research reference and as an
advanced textbook. The book has several main themes that the authors pursue throughout
and that unify their approach to the topics they cover. The most prominent of these unifying
themes is regression, in particular, linear regression. It appears in various guises, both as an
estimation technique per se and as a computational device for deriving covariance matrices …
The new book by Davidson and MacKinnon represents a valuable contribution to the econometrics literature and is meant to serve both as a research reference and as an advanced textbook. The book has several main themes that the authors pursue throughout and that unify their approach to the topics they cover.
The most prominent of these unifying themes is regression, in particular, linear regression. It appears in various guises, both as an estimation technique per se and as a computational device for deriving covariance matrices and parts of statistics arising in other contexts; it is used as a tool for presenting and interpreting the geometry of different vectors associated with problems of interest. Regression has such prominence and underlies so many of the topics of the book because overwhelmingly the action is envisaged in euclidean spaces, most importantly with the euclidean norm in mind, so that even when nonlinear objects appear they are typically geometrically represented as smooth manifolds with tangent spaces and curvature well defined. The authors skillfully present the geometry underlying a vast majority of econometric work; they accomplish this by systematically linking estimators and statistics to related geometric constructions and through graphs. Another powerful theme of Davidson and MacKinnon's book is inference, especially specification testing. It would be hard to disagree with the authors when they state this issue as the primary concern of today's econometric analysis. The centerpiece of the discussion of inference in the book is the development of the authors' approach via drifting data-generating processes. This methodology generalizes Pitman's drift by permitting wide possibilities for the space of alternatives that may depend on exogenous variables, past values of the dependent variable, etc., all captured in a vector that corresponds to a direction in the euclidean space. The approach permits one to explore the geometry of test power in different directions through ax 2-test statis-
Cambridge University Press