Estimating spatial models bi generalized maximum entropy or howq to get rid of W

Laburpena

The classical approach to estimate spatial models uses a spatial weights matrix to measure spatial interaction between locations. The rule followed to choose this matrix is supposed to be the most similar to the "true" spatial effects. Literature shows clearly the negative effects of the choice of a wrong matrix. The main problem is the lack of knowledge about which is the true specification. Furthermore, a single parameter is estimated and it should be seen as an average spatial effect among locations. In this paper we propose the use of maximum entropy econometrics to estimate spatial models. This method allows the estimation of a specific spatial parameter for each pair of regions and, hence, the spatial lag matrix is not chosen but estimated. We compare by means of Monte Carlo simulations classical with maximum entropy estimators in several scenarios on the true spatial effect. The results show that maximum entropy estimates outperform the classical estimates when the specification of the weights matrix is not similar with the true.