商业银行Logistic违约概率模型的样本配比问题研究

The Sample Matching problem of logistic probability of Default Risk Model in Commercial Banks

巴塞尔新资本协议要求商业银行建立内部评级体系,其主要内容之一是使用内部数据建立违约概率模型。在Logistic违约概率模型中,违约样本与非违约样本的配比是商业银行建立内部评级体系应重点关注和解决的技术难题之一。由于违约样本相对稀缺,商业银行在构建计量模型时,通常会采用一定的技术手段提高违约样本的比例,以改进模型对违约客户的预测能力。本文利用实际数据②,对比了样本配比的抽样法和加权极大似然估计法的不同表现。研究结果表明,加权估计法更适合中国商业银行的实际。

Basel Ⅱ requires commercial banks to establish internal rating based approach, and to model probability of default using internal data. When using Logistic regression model, we should focus on the ratio of default sample and to non-default sample. Because defaults are rare events, banks usually intentionally increase the proportion of default sample through econometric skills, in order to improve the model’s ability of ranking and forecasting. The paper studies the impact of different sample ratios, and compares two methods to model the ratio, i.e., sampling and weighted maximum likelihood estimation. Based on banks’ data, we recommend the latter for China’s commercial banks.