Logistic违约率模型的最优样本配比与分界点研究

The Optimal Sample Pairing and the Critical Value of Logistic Default Risk Model: the China Case

Logistic模型是研究违约率的主流方法之一,但目前的研究未能对最优样本配比与分界点这两个基本问题给予足够的重视。文章就此展开研究,设计了15种典型的样本配比—临界点的情景,通过实证比较的方法得出1:3的样本配比与0.647的临界点比较适合我国的情况,而常用的1:1的样本配比可能并不适用。

Logistic model is becoming more and more popular in default risk modeling. But the literatures by now have not paid enough attention to two kinds of fundamental questions concerning Logistic default modeling, that is, what is the optimal pairing structure of the training sample and what is the optimal cutoff point？ This paper carries out a research into these problems. We design 15 typical scenarios of different sample pairing structure and cutoff point combination. Comparing the estimations of Logistic model and prediction efficiency in these different scenarios, it concludes that 1 ： 3 of sample pairing and 0. 647 of cutoff value fit with China＇s data set, but the widely used 1 ： 1 structure may not be suitable for China case.