基于DREAM算法的DSGE模型参数估计精度研究OACHSSCDCSTPCD

This paper applies DREAM algorithm of artificial intelligence algorithm to the parameter estimation of dynamic stochastic general equilibrium model for the first time.Taking LS(2005)as an example,the paper systematically analyzes the esti-mation accuracy of this algorithm.The results go as below:According to the boxplot of random sampling sequence of parameters to be estimated,the box length of random sampling sequence of parameters to be estimated generated by DREAM algorithm is longer than that generated by RWMH and IMH algorithm,indicating that the dispersion degree of parameter estimation sequence gener-ated by DREAM algorithm is larger than that generated by RWMH and IMH algorithm.The results show that DREAM algorithm has a wider range of traversal and a stronger ability to escape from local optimal values.In addition,from the perspective of the median value in the boxplot,the median value of the parameter estimation sequence generated by DREAM algorithm is closer to the real data generation process than RWMH and IMH algorithms.Because DREAM algorithm removes useless chains according to IQR statistical method,the outliers of parameter estimation sequences generated by DREAM algorithm are also significantly re-duced,and the outliers of parameter estimation sequences generated by RWMH algorithm are particularly serious.Seen from the 90%maximum posterior density confidence interval of parameters to be estimated,the 90%maximum posterior density confidence interval of parameters to be estimated generated by DREAM algorithm contains truth values except for three parameters,while the traditional RWMH and IMH algorithms only have seven and one intervals containing truth values respectively,indicating that the estimation uncertainty of DREAM algorithm is much less than that of traditional RWMH and IMH algorithm.Finally,according to the invalid factor of the parameter to be estimated,the correlation between the sequence of parameters to be estimated generated by DREAM algorithm is weaker than the traditional RWMH and IMH algorithm,that is,the invalid factor value is smaller,which further verifies that DREAM algorithm has a stronger ability to traverse the whole parameter space.