Zusammenfassung:
Spatial dependence results from the existence of spillover effects such as the impact of the price of one housing unit on the price of its adjacent neighbors. One way to account for spatial dependence is to specify spatial lag models in which a spatially lagged variable is assumed to play a role in explaining the variation of the original dependent variable. Most studies use a priori non-sample information in the construction of the spatial weights matrix which serves as a spatial lag operator. In contrast, this study assumes no a priori value for the spatial weights matrix in the estimation of spillover effects. We adopt a classical maximum likelihood approach and also a Bayesian Sampling-Importance-Resampling (SIR) procedure to estimate the weights matrix and the significance of spatial dependence. We apply the two estimation procedures to data on housing prices in the city of Belo Horizonte, Brazil, and compare the results obtained with these two techniques with the one derived by a priori fixing the weights. The analysis shows that the likelihood function of the weights matrix parameters has a well-defined peak, and the estimated distance-decay parameter is quite different from the standard a priori assumptions such as the 'all-or-nothing' decay within the cut-off distance or the "inverse distance" adopted in the empirical literature.