基于贝叶斯方法的低违约组合违约率估计

现代信用风险建模的核心是估计违约率,违约率估计是否准确将直接影响信用风险建模的质量。在估计违约率的众多文献中,频率法或logistic回归等统计方法的运用非常广泛,此类统计模型的基础是大样本,它客观上需要最低数量或最优数量的违约数据,而低违约组合(LDP)是指只有很少违约数据甚至没有违约数据的组合,如何估计LDP的违约率、反映违约率的非预期波动是一个值得关注的现实问题。本文针对银行贷款LDP缺乏足够历史违约数据的情况,采用贝叶斯方法估计LDP的违约率,并进一步探讨了根据专家判断或者根据同类银行LDP违约数量的历史数据来确定先验分布的方法。在贝叶斯估计中,通过先验分布的设定,不仅可以实现违约率估计的科学性和合理性,而且可以反映违约的非预期波动,有助于银行实施谨慎稳健的风险管理。

Bayesian Estimation of Default Probability for Low Default Portfolios

Estimation of default probability is the key issue in modeling the credit risk,and accuracy of estimation will directly affect the quality of model.In many literatures about estimation of default probability,the frequency method or logistic regression methods is widely used,but it is based on a large number of samples and requires a minimum or optimal number of default data.Low default portfolios (LDP) has only little relevant historical data information,therefore,how to estimate the LDP's default probability and reflect the non-expected volatility are the concern problems.As an attempt to overcome this issue,we suggest Bayesian approach.In order to give effect to estimation of default probability and reflect the prudential risk management style of banks,we discuss in this paper how to choose prior distribution appropriate to Bayesian estimation.Incorporating expert opinion is a feasible plan and using historical data from analogical banks for reference is an attractive option.