基于改进BVAR模型对通货膨胀率的预测分析
Forecast and Analysis of Inflation Rate Based on Improved BVAR Model

针对向量自回归模型(VAR)的高维估计问题，本文结合贝叶斯理论提出了一种融合正态–逆Wishart共轭先验分布的估计方法。该方法引入了Metropolis-Hastings (MH)算法，从数据集中确定先验分布的超参数，并通过设定与模型规模相关的系数进行估计。与传统VAR模型相比，基于贝叶斯理论的估计方法在保留相关样本信息的同时，能够有效控制过度拟合，表现出较好的稳健性和有效性。最后，本文以湖南省为例，利用所提模型对通货膨胀率进行了分析与预测，并提出了可行的建议。
Aiming at the problem of high-dimensional estimation of vector autoregressive model (VAR), this paper proposes an estimation method combining normal-inverse Wishart conjugate prior distribution with Bayesian theory. In this method, Metropolis-Hastings (MH) algorithm is introduced to determine the hyperparameters of prior distribution from the data set, and estimate them by setting the coefficients related to the model size. Compared with the traditional VAR model, the estimation method based on Bayesian theory can effectively control over-fitting while retaining the relevant sample information, showing better robustness and effectiveness. Finally, taking Hunan Province as an example, this paper analyzes and forecasts the inflation rate by using the proposed model, and puts forward feasible suggestions.