Statistical Inference And Applications Of Spatial Dynamic Durbin Panel Data Models

The most significant feature of spatial econometrics is the spatial interaction effects among variable observations across different spatial units.This spatial dependence is used to construct spatial econometric models,then to describe,explain and predict things.Spatial econometric models have successively added the spatial,temporal and spatiotemporal dependencies of dependent variable,independent variables and random error terms into regression models.Their specifications,theories and methods have been continuously innovated and perfected,and have been widely used in economics,management,environmental science and other fields.The statistical inference of spatial econometric models is usually based on traditional mean regression estimation.On one hand,traditional mean regression estimation can just obtain the mean information of dependent variable conditional distribution,while cannot fully describe its scale and shape,and the random error term is constrained by the distribution assumption.To overcome the disadvantages mentioned above,a natural idea is to model separately at different quantiles of response distribution such that the impacts of independent variables can be differentiated,and have a much richer view in applications.Therefore,the study of quantile regression estimation for spatial econometric models has important significance.On the other hand,a prerequisite for regression estimation is a correct model and suitable variables.It is one of hot topics for spatial econometrics to screen out significant independent variables from numerous variables and establish a optimal spatial econometric model.Therefore,the study of variable selection of spatial econometric models also has great value.Based on the above considerations,this thesis studies quantile regression estimation method and variable selection method of the spatial dynamic Durbin panel data model,and empirical applications of these two estimation methods.The main research contents and conclusions are summarized as follows:First,we study the quantile regression of spatial dynamic Durbin panel data model.Conventional estimators are usually inconsistent in the presentation of endogeneity.Under quantile restrictions,we propose the instrumental variable quantile regression estimation method by using instruments to eliminate the endogeneity.A grid search is used to implement the estimation procedure and lead to the final estimators.Under certain regular assumptions,the consistency and asymptotic normality of estimators are derived.Monte Carlo simulation show that the instrumental variable quantile regression estimators not only have excellent finite sample performance at different quantiles,but also have robustness for different random error distributions and different spatial weight matrices.Second,we study the variable selection of high-dimensional sparse spatial dynamic Durbin panel data model.Classical penalized least squares estimators are usually inconsistent due to the endogeneity.For high-dimensional sparse data,to overcome the adverse effects of endogeneity,we use Helmert orthogonal transformation to eliminate fixed effects,and propose the penalized focus generalized method of moments estimation method based on the restrictions of conditional moment and over-identifying to realize parameter estimation and variable selection.Under certain regular assumptions,the Oracle property and asymptotic normality of estimators are derived.Monte Carlo simulation show that the penalized focus generalized method of moments estimators perform well in finite sample.Comparing two different variable selection procedures based on SCAD and LASSO penalty functions,the SCAD outperforms the LASSO in estimation accuracy.In addition,the estimators have sensitivity to different random error distributions and have robustness for different spatial weight matrices.Third,we study the influencing factors of inbound tourism development in China.Based on the relevant provinces data of domestic from 2011-2019,we analyze the driving factors of inbound tourism development according to the proposed quantile regression estimation technique of spatial dynamic Durbin panel data model.The empirical results show that there are significant positive spatial and serial dependencies in inbound tourism development level under different quantiles.The short-term and long-term effects of transportation convenience,residents’living standards and trade openness are all positive.The effect of tourism resource value in short-term span is significant different from that in long-term span.Therefore,improving transportation convenience and living standards of residents,promoting trade openness,and paying attention to the joint governance in space and the cyclic accumulation in time should be relied on to promote the development of inbound tourism.Fourth,we study the influencing factors of economic growth in China.Based on the relevant provinces data of domestic from 2017—2019,we analyze the driving factors of provincial economic growth,according to the proposed variable selection technology of high-dimensional sparse spatial dynamic Durbin panel data model.The empirical results show that there exist high spatial correlation and positive spatiotemporal dependence in economic growth.Whether short-term effect or longterm effect,fixed asset investment growth,CPI index and tertiary industry added value are significantly positive,while import trade is significantly negative.The fixed asset investment growth has a huge impact on economic growth in short-term span.Therefore,maintaining an appropriate growth of fixed asset investment,ensuring that the CPI index rises moderately within a reasonable range,expanding the scale of import trade appropriately,raising the proportion of the tertiary industry in the national economy,and paying attention to joint governance in space should be relied on to promote the economy healthy and sustainable development.