Abstract
The search for the most appropriate spatial econometric specification has received considerable attention in the literature, with a major tension between a specific to general approach, mostly based on Rao Score/Lagrange Multiplier statistics; and a general to specific approach, based on Wald or Likelihood ratio statistics. This talk reports on work in progress to address a number of aspects that have received less (or no) attention to date. Most important of these is the use of “modern” estimation methods, i.e., IV/GMM-based methods rather than maximum likelihood estimation which is characteristic of the literature so far. In addition, several strategies are considered, including hybrid forms that move beyond the classic STGE-GETS dichotomy to reflect empirical practice. The talk sets out with a detailed examination of the typical specifications suggested to incorporate spatial effects into a linear regression model. Particular attention is paid to the type and range of interaction implied by the associated reduced forms. In the process, it revisits some familiar identification and overparameterization issues. The main part of the talk consists of a report on work in progress on an extensive Monte Carlo simulation experiment to assess the comparative performance of several specification search strategies.
There will be a reception following the seminar.