基于动态权重的信用评级

Credit Rating Based on Dynamic Weight Optimization

以已评级企业为依据,利用偏相关系数和Wald统计量遴选出一套最能代表企业信息的指标体系作为解释变量,以外部评级作为被解释变量,找出企业信息与其外部评级之间的动态权重映射关系;再将这种关系应用到未评级企业,从而评估未评级企业的信用等级。实证结果表明,动态权重模型在样本内外的评级准确度和可靠性均显著优于传统静态权重的OLS和有序Probit模型,新模型在样本外的平均误差仅为0.11,各评级档次犯第1类错误的概率比传统模型低20%。基于动态权重模型获得的新企业评级与原企业的外部评级具有可比性,可作为外部机构评级的合理补充,为市场提供更全面、及时的评级信息。

Based on rated firms, using partial correlation coefficients and Wald statistic to select a set of the most representative of the firm information indicator system as explanatory variables, the complex mappings between rated firms' information and their credit ratings are identified and are applied to unrated firms to obtain their credit ratings. Beyond the traditional linear regression, a dynamic weighted optimization model based on the genetic algorithm is proposed. The empirical study shows that the dynamic weighted optimization model performs better than the traditional linear models in both accuracy and reliability. Its average error outside the sample is only 0. 11, and the probability of type 1 error is about 20% lower than that of the other two models, The ratings of new firms based on the new model are comparable with the original external ratings, which can be used as a good supplement to external rating agencies and provide the market with more comprehensive rating information without delay.