Research On Inflation Rate Prediction Based On Bayesian Vector Autoregressive Model

Inflation has an important impact on output,employment and wealth distribution.The forecast of inflation rate is helpful for the country to formulate monetary policy and for investors to make decisions.Commonly used inflation rate forecasting models such as the Phillips curve model and the factor model usually involve many potential economic variables.These economic variables often have strong correlations,resulting in poor prediction results of these models.To this end,this paper combines Bayesian vector autoregressive model(BVAR)with Markov chain Monte Carlo algorithm(MCMC)and sequential Monte Carlo-approximate Bayesian calculation(SMC-ABC)respectively,and proposes MCMC-BVAR model and SA-BVAR model,which are applied to the empirical study of China ’s inflation rate prediction.The prediction performance of the above two models is compared by error evaluation index.The expression of the posterior density distribution of the parameters is derived by combining the prior distribution of the parameters and the likelihood function.The posterior values of the parameters of the BVAR model are sampled by the MCMC algorithm to construct the MCMC-BVAR model.The approximate expression of the posterior density distribution of the parameters is derived by combining the prior distribution of the parameters and the approximate likelihood function.The posterior values of the parameters of the BVAR model are estimated by the SMC-ABC algorithm to construct the SA-BVAR model.From the aspects of production,consumption and financial market,12 major macroeconomic variables are selected to construct the inflation rate forecasting index system,and a total of 182 monthly index data from January 2008 to February 2023 in the China Economic Network database.The MCMC-BVAR model and the SA-BVAR model are applied to the empirical study of forecasting the year-on-year growth rate of CPI and the year-on-year growth rate of PPI,so as to reflect the trend of inflation rate in a more comprehensive way from the field of consumption and production.The results show that : for the prediction of CPI year-on-year growth rate,except that the MAPE value of the SA-BVAR model is slightly larger than that of the MCMC-BVAR model,the values of MSE and SMAPE are smaller than those of the MCMC-BVAR model;for the prediction of PPI year-on-year growth rate,the error index values of the SA-BVAR model are smaller than those of the MCMC-BVAR model;the D-M test results show that the SA-BVAR model shows good performance in the prediction of inflation rate,and the effect is obviously better than the MCMC-BVAR model.Therefore,the SABVAR model proposed in this paper is effective in predicting inflation rate.