Research On The Efficiency Of Bootstrap Tests In Spatial Panal Data Pool Models

To identify the specific spatial relationships in spatial econometric analysis, it isusually to use Moran’s I test that are asymptotically normally distributed, LM-Errorand LM-Lag test that are asymptotically2(1)distributed to identify the specificspatial dependence. Moran’s I, LM-Error and LM-Lag test are only applied to largersample size and model error which are normally identically distributed. However, alarge number of economic work in the study, the sample size is usually very limited,or the model error the classical normality assumption of the model isviolated(heteroscedasticity or time series related errors, etc.), which would lead to thesituation that Moran’s I, LM-Error and LM-Lag test are questionable.Taking into account the characteristics of spatial panel data, in line with theexisting literature, the paper extends Bootstrap methods to DB (Double Bootstrap)method to solve the deficiency of the asymptotic test. For solving the problem of itslarge amount of computation, we use FDB (Fast Double Bootstrap) method tooptimize the DB method and constructed FDB statistics(so called FDB test). Thepaper takes a large amount of Monte Carlo experiment and compare the differencebetween the FDB tests and asymptotic tests to demonstrate the efficiency of FDB tests,and solve problems under small sample from the size distortion and power.This paperis divided into three parts. The first part is about the research background; The secondpart describes the main idea of section Bootstrap test and FDB test of Panel Data Poolmodel, and designs Monte Carlo simulations; The third part analyzes the Monte Carloexperiment results of section Bootstrap test of FDB test, and finally draw a conclusion,including the third and fourth section.A large amount of Monte Carlo simulation results show that, when the errorterms are independent normal distributed, both asymptotic test and FDB test havesuperior size and power performance; when errors terms are time series correlationdistributed or heteroscedasticity distributed, asymptotic test can not exactly identifythe spatial error dependence or lag dependence, exists serious size distortion, while FDB test can reduce the size distortion effectively and increase the power ofasymptotic test, which is a more effective test statistics. Generally speaking, ourresearch shows that FDB test of Panel Data Pool model is effective whether theclassical distributional assumption is violated or not.