使用多元序列的Jackknife模型平均(JMA)方法平均 向量自回归模型, 并将该方法用于预测中国六大港口的集装箱吞吐量. 由于 JMA方法在自相关异方差结构下的渐近最优性, 因此更适用于具有大的波 动性、复杂性和不规则性的港口集装箱吞吐量的预测. 另外, 相比单序列, 多 元序列的JMA平均方法也考虑了港口之间的相关影响因素. 比较发现, 在大多 数案例中, 此方法比常用模型选择和模型平均方法具有更高的预测精度.

This paper applies the Jackknife model averaging (JMA) method to average the multivariate vector autoregressive model. This method is utilized to forecast six Chinese ports container throughputs. Due to the asymptotic optimality of the JMA estimator under a setting that assumes exogeneity of regressors but allows for both serial correlation and heteroscedasticity, it is more suitable for forecasting ports container throughputs which have large volatility, complexity and irregularity. Besides, compared to univariate vector autoregressive model, JMA with multivariate vector autoregressive model considers the related influencing factors among ports. The empirical analysis shows that the JMA method performs better than some commonly used model selection and model averaging methods in most cases.