Applications of Differential Geometry to Econometrics (Q4526787)

This book is a collective monograph devoted to connections between differential geometry and econometrics. The basic principle here: geometry is mathematically fundamental, and frequently a geometric understanding of problems highlights the essential issues which can be hidden in an algebraic development of the same issues. The first chapter surveys the mathematics needed to understand the methods of differential geometry and their application to econometric inference. Ch. 2 considers several aspects of the important issues of encompassing and nesting in model selection. Here the conditions are examined under which a general model encompasses simpler ones independently of their relationship to the data-generation process. It is shown that nesting implies more than one model being an algebraic simplification of another. Encompassing hypotheses are considered as mappings in Hilbert spaces and a sufficient condition for a model to encompass models nested within it is that these mappings are orthogonal projections in the relevant spaces. In Ch. 3, a recently developed and geometrically motivated technique [\textit{G. Hillier} and \textit{M. Armstrong}, Econometrica 67, No. 6, 1459-1470 (1999)] is applied for obtaining exact distribution results for implicitly defined estimators to study exact properties of maximum likelihood estimators (MLE) in exponential regression models. This method provides an integral formula for the exact density of the MLE, expressed as a surface integral over the manifold in the sample space upon which the MLE is defined. A detailed discussion of the applications of this approach to exponential regression models is given. Ch. 4 presents a semi-parametric method for the study of empirical likelihood estimation and inference. Ch. 5, on the usage of Hilbert space techniques, develops a general geometric basis for the analysis of efficiency and robustness for estimators defined by estimating equations. Ch. 6 provides a discussion of Rao distances and associated tests in certain two-parameter models. In Ch. 7 the problem of determining an optimal predictive density is considered. It is shown how this predictive density may lie outside the class of the original parametric model describing the variable of interest and how geometrically a global solution can be constructed by considering formally all possible directions in which the original model may be extended. Ch. 8 provides a detailed and powerful analysis of the effects of statistical curvature on the classical tests used in econometrics. Ch. 9 considers the question of testing for unit roots in autoregressive and moving-average models. The last Ch.10 gives very general issues of the geometry of transformations and the role of parametrization in econometric models.