经济核算类矩阵调整的JSD方法及扩展研究

JSD Method with Extensions for Solving Matrix Adjustment Problem in Economics

现有的矩阵更新、平衡等调整方法多存在两种缺陷:一是度量新旧矩阵间差异的函数形式不对称,并非严格"距离"概念;二是要求矩阵元素非负导致使用范围受限。为改进上述不足,本文提出了包括Jensen-Shannon Divergence(JSD)在内的若干基于对称距离优化的新方法,并统一进行保号、误差妥协等扩展以方便实际运用。在此基础上利用中国及其他28个国家的数据,对比分析多种矩阵调整方法的实际效果,结果发现:第Ⅰ类JSD方法表现最突出且相对稳健,值得代替现在使用较广泛的RAS或交叉熵方法。

Matrix adjustment methods are widely used in economics, but there are still some defects in most existing methods. Firstly, the ＂distance＂ measure between the initial and estimated matrices is not symmetric;secondly, the initial matrix is assumed non-negative, however sometimes empirically inconsistent with the fact. This paper proposes various new optimization methods based on symmetric distance measure, with sign-preservation and tolerance extensions. On this basis, updating the IO table / balancing SAM as the empirical comparison with 29 countries＇ data, results show that the Jensen-Shannon divergence I method outperforms the others in most cases, so we suggest this new rather than RAS or CE for adjusting economic matrices in practice. （JSD- I ） approach