Research On The Nonnested Tests Of Spatial Econometric Models And Applications

After giving a survey to classical methods of spatial econometrics, we point out that the traditional nested tests will give seriously wrong conclusions about the spatial structures, and we give a nonnested test method of the spatial weights. Meanwhile, we show that the unaccepted matrix after testing can be used as the instrumental variable in the following estimation. With the Monte Carlo simulation results and two comparative analyses of the Changsanjiao, Zhusanjiao and Huanbohai districts, we show that the nonnested method works well.With a Monte Carlo simulation, we show that there are misuses of spatial structures in the analyses using spatial econometric models. The usual nested tests often lead us to inappropriate spatial matrix. When the misuse happens, we often get the wrong result about the influences of the spatial effects.Based on the pseudo nested method, which is one of the three methods of nonnested test, we give a nonnested test of the spatial structure. The use of the new test is not restricted by the specification of spatial models and it is very easy to be employed in the spatial model analyses. The Monte Carlo simulation shows that the nonnested test works well in the comparison of different spatial matrix.The unaccepted matrix after the test can be used as the instrumental variable for the accepted spatial matrix. The estimator of either generalized spatial 2 stage least squares or generalized method of moments using the unaccepted matrix as instrumental variable is as efficient as those using the usual instrumentals. In addition, the new instruments do not need the condition that the corresponding matrix has its inverse exist.The new nonnested test is employed in two empirical studies in this paper about Changsanjiao, Zhusanjiao and Huanbohai districts. One is about"industrial development strategies"and"spatial convergence", and the other is"urbanization","urban concentration"and"spatial structures".