Research On Measurement Of China's Trade Openness Based On Bayesian Model Average Method

The evaluation and comparison of trade openness began with the index of trade dependence,which was widely used because of its simplicity.However,there are many defects in the degree of dependence on trade,and it is easily distorted by many factors such as the geographic location,economic development level,market size,domestic consumer demand,and population of a country,which can not accurately measure the degree of trade openness of a country.In recent years,scholars have done a lot of research on the measurement of trade openness.They either proposed new measurement methods or used relevant theories and variables to establish models for correction(abbreviated as model method).Because of the support of the theory and the verification of empirical data,the model method can also solve the endogenous problems of trade,so that it can be widely studied and applied.However,in the existing model method,the choice of variables is either influenced by human factors or different models are selected according to different selection criteria,resulting in inconsistent results,that is,existing methods ignore the model.Uncertainty issues.In order to solve the uncertainty problem of the model,it can be solved by Bayesian Model Averaging(BMA).The goal of this paper is to use the Bayesian model average method to improve the model measure of China's trade openness.Firstly,the relevant variables were selected according to the important literature on trade openness model method.Then based on the BMA method,the model of China's trade dependence degree was established based on the statistical data of China from 1990 to 2016,and the measure of China's trade openness was revised.Through empirical analysis,this paper obtains the following conclusions: First,after the revision of China's trade openness based on the BMA method,the correlation measure is more accurate.Second,when using a single model to measure China's trade openness,there are uncertainties.Third,there are certain similarities and differences between the results based on the BMA method and the traditional regression model method.Both energy,population and fixed capital stocks are significant;real exchange rate,Dowlas index and GDP are significant in the regression model,but not significant in the BMA method;FDI's probability of being selected in the BMA method is in the middle,while in the regression The model was not significant and it was not selected.Fourth,the BMA method also outperforms the single model in predicting performance.This paper divides the data into two parts: the training set and the forecasting set.Based on the training set data and the stepwise regression results based on all the data,the change is large,the variable selection and the regression coefficient are all changed,while the results based on the BMA method are more robust,so its The result is more reliable.