基于Gibbs抽样的集装箱吞吐量Bayes AR(p)分析

Bayes AR(p) Analysis of Container Throughput Based on the Gibbs Sampling

摘要: AR(p)模型广泛应用于时序预测，然而传统静态模型难以处理突发事件以致模型估计偏差。鉴于突发事件对模型估计的影响，采用Gibbs抽样方法对模型进行Bayes分析，根据时序样本似然函数的统计结构构造出模型各参数的先验分布。在导出模型参数后验条件分布后给出具体抽样策略。在最小均方误差估计准则下对中小样本的模拟显示，参数估计值与真值接近。对上海港1982~2015年集装箱吞吐量数据的分析表明：借助Bayes分析，可以克服由于突发事件导致的模型估计偏差，使模型预测更加准确。

Abstract: AR(p) model is widely used for time series forecasts; however, it’s difficult for traditional static model to deal with emergencies, which lead to estimation bias. In view of the influences of emergencies for model estimation, we carry out Bayesian analysis of the model by aid of the Gibbs sampling. According to likelihood function’s statistical structure of the time series samples, the prior distribution is obtained. After getting the posterior empirical distribution of parameters, the specific sampling strategy is proposed. Under the minimum mean square error estimation criterion, the simulation experiments show that the estimates are close to the true value. The analysis for the data of Shanghai port’s container throughput from 1982 to 2015 indicates that by aid of the Bayesian analysis, the estimation bias from emergencies can be overcome so that the model prediction is more accurate.