Research On Several Types Of Semi-parametric Panel Data Models

When dealing with practical problems,the observations of panel data always be combined with spatial information.Based on this,spatial panel data model has received rising attention from theoretical researchers and practical problem researchers in recent years.At present,the researches of spatial panel data model mostly assumes that the relationship between dependent variable and the independent variables is strictly linear,and the individual effect and time effect are set in the traditional one-way or two-way form.But in many cases,the assumption is not accurate enough for the actual situation.The wrong model form setting would make the inference results to deviate from the actual situation.In order to explore the complex relationship between dependent variable and independent variables and improve the flexibility of the model,a variety of non-parametric and semi-parametric modeling techniques have been proposed and applied to various complex situations.Naturally,using semi-parametric modeling methods to analyze spatial panel data has also attracted attention.Recently,many types of semi-parametric spatial panel data models have been proposed and studied.Based on the existing models,this paper has proposed several new semi-parametric panel data models and did some researches about the models.In the first part of the paper,based on the common one-way time effect panel data model,a spatially heterogeneous time effect panel data model is proposed,which sets time effect to be combined with spatial heterogeneity and changes with location.The paper gives the model estimation based on the Profile least squares method and the geographically weighted regression technique.The hypothesis-testing problem of the model is considered based on the generalized F test.Then,this paper validates the effectiveness of the proposed method by numerical simulation.Based on this method,the problem of provincial unemployment rate in China is studied.Finally,the model is extended to the case where all or part of the regression coefficients are spatially variable coefficients.In the second part of the thesis,a kind of semi-parametric spatial panel data model is proposed based on a common stochastic frontier panel data model.In this model,time-varying individual effect is set into a spatially varying coefficient with time polynomial form.The model estimation and the hypothesis-testing problem of the model are solved then.Next,the paper validates the effectiveness of the proposed method by numerical simulation and uses the method to analyze the relationship between the agricultural added value of each province in China and some factors like the climate change factor and mechanization degree of each region.Finally,the paper promotes the model into two more broad forms.With the help of semi-parametric technology,the paper proposes several new panel data models based on the existing models.The research results enrich the research content of semi-parametric model and spatial econometrics,and provide more flexible statistical modeling method for the analysis of spatial panel data.