Spatial Econometric Model Selection, Estimation And Its Application Based On The Comparison Of Classical Methods And Mcmc Methods

Spatial econometric model have been increasingly widespread attention. In the real world, there are actually independent and non-independent two possible observations, the traditional econometric theory is built on the basis of the assumption of independent observations, and the presences of the spatial dependence between geographic regions and economic phenomenon to break the basic assumptions about the samples are independent in classical econometric model. Then to accurately extract the spatial relationship of these data, appropriate description and use of space characteristics of spatial interaction research is particularly necessary. In fact, some economic phenomena of spatial unit or a property always with adjacent space unit on the same phenomenon or related attributes, we must analyze the relationship through the spatial econometric model.We will be faced with the three important problems when adding spatial effects into econometric analysis framework. One is how to properly introduce spatial effects to original spatial econometric model, or to construct specific econometric models according to the particularity of the spatial effects. The second is how to select the proper spatial econometric model in specific spatial econometric analysis and how to estimate specific model. The third is how to make specification of empirical analysis be using spatial econometric model. The first question relates to reasonable set of spatial weight matrix. The second question relates to spatial econometric model selection and estimation method. In fact, the third question is how to combine the first two problems and empirical analysis.Based on the above three questions, this paper firstly studies the most important object in spatial econometric-spatial weights matrix, then analysis spatial model selection and effective estimation methods of spatial econometric model with unknown heteroskedasticity by combining the classical econometric methods and MCMC methods, and empirical analysis is given on the application. Conclusions of this study are as follows.(1) Spatial weights matrix is the link between spatial econometric model in the theoretical analysis and the real world spatial effects. Build and select the appropriate spatial weights matrix is directly related to the final estimation results and the explanatory power of the model. Different spatial weights matrix reflects the research behind the economic principles and objects of different perspectives, but also corresponds to the researcher for the spatial effects of different perceptions. The wrong choice of spatial weight matrix will seriously interfere with the spatial econometric analysis of the various tests, which have a great effect on further research of spatial econometric model.(2) Spatial econometric model selection is an important research topic in spatial econometric analysis. Moran index test, LM test, the likelihood function, three information criterion, Bayesian posterior probability, Markov chain Monte Carlo methods are very different spatial econometric model selection. Simulation analysis shows that Moran index and LM test based on the OLS residuals have a big limitation when the model selection in the expansion of spatial econometric model clusters, the maximum value of the log-likelihood principle lack of discrimination, LM test only for distinguishing SEM and SAR model is valid, information criterion is effective for most models, but there will be false choice. And because of it full uses of the likelihood function and a priori information, MCMC showed higher test validity when the M-H algorithm is more appropriate given, especially to get a more accurate judgment in the larger sample conditions. In additional, MCMC is also very effective when choosing a different stage of its adjacency matrix of spatial econometric model.(3) The difference of spatial unit size and other economic characteristics often lead to spatial variance problem. For generalized spatial model includes heteroscedasticity, estimation method is relatively complicated. This paper presents three kinds of estimation methods for heteroscedasticity problem of general spatial model. The first method is to use ML estimation to overcome the lack of degrees of freedom by parametric heteroscedastic form. When the forms of heteroskedasticity are unknown, we solve it by using the iterative GMM based on 2SLS estimation and more directly MCMC sampling method. And MCMC method is even more graceful. Monte Carlo simulation founds, the MLE is still a better estimate by parametric heteroscedastic style when the forms of heteroskedasticity is known. When the forms of heteroskedasticity are unknown, with increasing sample size, two other methods can be gradually consistent with ML estimation.(4) Based on spatial weight matrix analysis, spatial econometric model selection, estimation of generalized spatial econometric model with unknown heteroskedasticit and directional distance function GML super efficient model, we study the measurement and influence factors of China’s provincial total factor energy efficiency under the constraints of resource and environment. And the following conclusions were obtained from the process of empirical research, resource and environmental constraints should be considered when we study the measure of total factor energy efficiency, only in this way, the results obtained can be more in line with China’s actual situation. Furthermore, the spatial effects should be considered when we analysis the influencing factors of provincial total factor energy efficiency, it will come to have a bias in the estimates if ignoring the impact of spatial effect. Especially considering the spatial effect will also be required to select the appropriate spatial weights matrix and appropriate spatial econometric model based on spatial econometric theory and research methods. We can form a complete framework of total factor energy efficiency analysis only a reasonable combination of these processes. And the following conclusions were obtained from the results of the empirical analysis, China’s provincial total factor energy efficiency continued to decline under the constraints of resource and environment, and the trend is not optimistic; under conditions of resource and environmental constraints place undue reliance on coal resources will greatly reduce total factor energy efficiency, the negative impact brought about by the consumption of coal does not ignore; "pollution haven hypothesis" is established in the country; it helps to improve the overall energy efficiency when increasing the proportion of the service industry; the foreign country who will use relatively more advanced energy technology, and there is a positive impact on China’s total factor energy efficiency on the existence of a positive spillover effect of domestic enterprises.Throughout this research has some theoretical and practical value. It’s the first time to do a systematic study of the spatial weight matrix, and made a certain degree of visual analysis through graphical and improves the system of understanding spatial weight matrix. In the analysis of spatial econometric model selection method, rarely used and very effective MCMC methods is presented. In spatial econometric model estimation, a more common spatial econometric model-generalized spatial econometric models are studied, and effective MCMC estimation method is given under consideration of different conditions with unknown variance. Finally, all of the abstract theoretical analyses have carried out a Monte Carlo simulation, given the effectiveness of the comparison method, which is an important reference for spatial econometric empirical analysis. Based on this, the paper finally conducted empirical research according to the theoretical analysis, and the whole process is a normative research.