Sensitivity of model parameter estimates in stress testing the probability of default: evidence over the financial crisis
扩展贝叶斯方法预测信用风险VaR，纳入参数不确定性和不稳定性，发现考虑这些因素会使银行所需资本增加超过170%。
We extend a Bayesian method for predicting Value at Risk (VaR) for credit risk by incorporating not only macroeconomic uncertainty, but also parameter uncertainty (including parameter instability over time and parameter estimation uncertainty). We apply this method to estimate the capital required by banks to cover credit risk. By simulating default rate distributions based on models built on crisis and tranquil periods, we explore how shifts in model parameters across economic conditions influence stress testing results. We use parameter posterior distributions from the Bayesian framework in VaR computation to mitigate estimation risk arising from reliance on point estimates. In an illustration for mortgages our results show that accounting for parameter instability and estimation uncertainty increases the required capital by over 170%.