Tomasz KOSSOWSKI, Adam Mickiewicz University, Poland
Jan HAUKE, Adam Mickiewicz University, Poland
The analysis of unemployment rates across a set of regions is often affected by the underlying spatial process. This process can influence on unemployment rates through spatial dependence effect. However, the problem is how to represent this spatial dependence effect in the model. The typical solution is the use of a spatial weights matrix, which is generally unknown.
The specification of a spatial weights matrix in a spatial model is considered as a difficult taks. The researcher rarely has information about the form of this spatial weights matrix. It is why so-called exogenous approach, where the researcher builds the matrix based on a set of a prioris (proximity, similarity, etc.) is preferable. However, this is not always the best solution and the subsequent analysis becomes conditional on a non- tested hypothesis. Recent papers deliver several proposals which advocate for estimating the weights matrix directly from the data.
In this paper we introduce a new method to estimate the spatial weights matrix. The premise of our approach is that the observed covariances matrix (of the residuals, of the explained variable) contains valuable information to estimate the unobserved weighting matrix. We address this problem by using a Generalized Method of Moments approach. The computational burden is, for the moment, considerable and the assumption of symmetry plays a key role. These are some of the constraints of the procedure which, however, allows us to test for different assumptions of interest for the researcher.
The proposed method is used to identify an unknown spatial weights matrix for unemployment rates of Polish regions.
Mots clés : spatial models|weighting matrix|generalized method of moments|unemployment
A105341TK