A Comparison Study on Criteria to Select the Most Adequate Weighting Matrix

Abstract

The practice of spatial econometrics revolves around a weighting matrix, which is often supplied by the user on previous knowledge. This is the so called W issue and, probably, the aprioristic approach is not the best solution. However, we have to concur that, nowadays, there few alternatives for the user. Under these circumstances, our contribution focuses on the problem of selecting a W matrix from among a finite set of matrices, all of them deemed appropriate for the case. We develop a new and simple method based on the Entropy corresponding to the distribution probability estimated for the data. Other alternatives to ours, which are common in current applied work, are also reviewed. The main part of the paper consists of a large Monte Carlo resolved in order to calibrate the effectiveness of our approach compared to the others. A case study is also included.