Forecast Research Based On Random Coefficient Panel Data Model

In recent years,best linear unbiased prediction theory,as an important conclusion in prediction analysis,is getting more and more attention.In theoretical analysis,the derivation of best linear unbiased prediction expression based on various panel data model has been deeply studied and popularized.In practical applications,the method has been widely used in financial economy,social science,agriculture and other fields.The panel data model includes static panel data model and dynamic panel data model,while the commonly used static panel data model can be divided into variable intercept model and variable coefficient model.The variable intercept model accounts for individual effect by intercept item,and the variable coefficient model changes the coefficient along with the individual cross section.The variable coefficient model is split into fixed effect model and random effect model.In fixed effect model,the coefficient vector is set as the constant vector varying with the cross section,while in random effect model,the coefficient vector is set to random,which can be divided into the average component and the random component with variable coefficient.This dissertation mainly studies the panel data model of random effect variable coefficient.In existing studies regarding best linear unbiased prediction,most of them concentrated on the theoretical derivation and case analysis of the panel data model with variable intercept terms,and discussed the best linear unbiased prediction problem with serial correlation or spatial autocorrelation residual disturbance panel data model.However,in practical applications,different social background and economic structure can lead to the changing of the parameter of affecting factor along with the sectional individual term,thus based on the previous studies,we extend the research object to the random coefficient panel data model.In order to predict the model more accurately and effectively in practical problems,this dissertation considers the model parameters vary with individual items.Firstly,the random coefficient panel data model with AR(p)disturbances is established in chapter 2;then the spatial random coefficient panel data model and the model with AR(1)remainder disturbances,which are the generalized form in existing studies,are established in chapter 3,considering the spatial autocorrelation between the sectional individual terms.On this basis,the model establishment and parameter specification are introduced in each part,the parameter estimation is conducted by the maximum likelihood estimation method and limited maximum likelihood estimation method,the expressions of best linear unbiased prediction are derived,the accuracy and stability of the proposed prediction quantity are verified through Monte Carlo simulation,and the feasibility and precision of the forecast quantity are further verified by combining with per capita GDP of 31 provinces,cities and autonomous regions.In this dissertation,the best linear unbiased prediction theory of panel data is extended,and the related conclusions about best linear unbiased prediction of panel data model with various random coefficients are obtained,which have important practical significance in common life.