本文将现有的众多矩阵更新技术纳入约束最优化的数学建模框架中,并统一进行了若干扩展,特别地,在约束条件存在冲突时通过引入误差精度调整项,使得原本不可行的矩阵更新问题能够求解,以扩大其适用范围。然后使用中国2002年和2007年的投入产出表对6种扩展修正后的矩阵更新方法进行了实证比较,结果显示无论是基于流量还是基于系数形式,或是对矩阵进行聚合处理,更新矩阵时ERAS的表现始终是最优的,ENSD其次,两者差距较小;而EAD方法表现较差。
This paper develops a unified mathematical constraint optimization model of matrices updating techniques, and then some extensions and necessary modifications are made. Empirical comparisons by updating China's Input-Output matrix from 2002 to 2007 indicate that- ERAS approach performs best, while ENSD second but with higher efficiency. This conclusion is consistent whenever data subject to different levels of aggregation or updating in either transaction or coef- ficient form, so we suggest that ERAS rather than other approaches should be used for updating IO matrix or SAM.