Researeh On The Application Of Bootstrap Methods In Spatial Dependence Of Spatial Panel Data Models

The most notable features of Spatial econometric are taken into the spatial effects which relaxed independent variable assumed of traditional econometric. Obviously, the Spatial econometric can get more realistic and accurate conclusions than traditional econometric. Currently, spatial econometric study of the spatial effect is mainly focused on the spatial dependence test.At present, a lots of spatial dependence test methods have proposed by many scholars, such as Moran’s I 、LM、LR and so on. Among them,Moran’s I test and LM test arethe most commonly used test methods.Moran’s I test can only verify the existence of spatial dependence but it can’t distinguished form of the models are Spatial lag or spatial error models. LM test can be divided into LM-Lag test and LM-Error test, We can determine the spatial manifestation significant dependence based on comparison of LM-Lag test and LM-Error test.However, in theory, Moran’s I test and LM can be established when the large sample and error term must obey classical distribution.And,in the he study ofreal economy phenomena, because of data availability and other issues, our study sample was small sample and the moderate. Because of the complexity of the real economic problems, error term ofthe models are difficult to fully meet the assumptions of the classical distribution, there are often exist heteroskedasticity and serial correlation time. Therefore, in practical studies, the above stringent assumptions can not meet. Moran’s I test and LM test which depending on large sample and error term obey classical distribution will invalid.Due to the presence of individual effects(random effects and fixed effect) and cross-sectional dimensions and time-series dimension double propert in Spatial Panel Data models,the spatial dependence test of spatial panel data models areunresolved problems. Bootstrap methods does not need classical distribution assumptions,can be used as one of the choice of method to solve the problems.However, the spatial panel data models have both spatial effects and panel data characteristics, we not only to consider the existence of individual effects, but also consider the time-series and cross-sectional dimensions,which making the validity of the results of Bootstrap sampling method will doubt. The Bootstrap methods which can used in cross-sectional modles are no longer suitable for Spatial Panel Data Models.We introduced Bootstrap methods into spatial panel data model Moran’s I test and LM test to deal with the spatial denpence test problems when the limited sample or error term does not hold classic distribution assumption.In this paper, we can get the conclusions as follow: First of all, in this paper,we construct Moran’s I test statistic to three-derived Bootstrap mtheods of DB1, DB2, FDB, we take Spatial lag panel data model as an example to do Monte Carlo simulation experiments. From the size of distortions and used time we can know,FDB Moran’s I test has least size of distortion and time-consuming less than DB2. From the perspective of power, FDB Moran’s I test close to asymptotically Moran’s I test mostly. Therefore, we can believe that regardless of the size of distortion and power, FDB Moran’s I test showed the best, is the best choice. Secondly,we introduced FDB method into spatial denpence Moran’s I test of spatial panel data models,through a large number of Monte Carlo simulations we canfound: FDB Moran’s I test the existence of smaller size of distortion and it’s power closed to asymptotically Moran’s I test when the limited sample or error term does not hold classic distribution assumption. Therefore, From the size of distortion and power we can see FDB Moran’s I test is more ideal method to deal with spatial denpence of spatial panel data models. Third, Monte Carlo simulation results show: when the individual effect is random effects, under non-classical error term,there is a greater size of distortion in asymptotic LM-Lag test,and FDB LM-Lag test has less size of distortion and smaller than asymptotic test without loss of power. Thus,FDB LM-Lag is amore effectively method.of spatial denpence test in spatial panel mode. Finally, we use Monte Carlo simulation results show:when the individual effect is random effects, under non-classical error term,there is a greater size of distortion in asymptotic LM-Error test,and FDB LM-Error test has less size of distortion and smaller than asymptotic test without loss of power than asymptotic LM-Error test. FDB LM-Error is a more effectively method to test spatial denpence of spatial panel models.