Proposing an innovative real-time monitoring approach for the parameter stability of credit-default prediction model

In real-world applications, the default prediction process generally requires periodic recalibration of parameter estimates with up-to-date data. However, few studies have investigated sudden changes or unknown structural breaks in the regularly estimated parameters, which could lead to biases in the models and thus affect the prediction accuracy. The instability of estimated parameters has been shown to under- or overestimate prediction results, but their real-time monitoring remains unexplored. This paper aims to fill this research gap by monitoring the estimated parameter stability in default prediction models in real time using statistical process control methods. We innovatively transform this issue into a research problem of profile monitoring and examine the monitoring performance of three control charts using Monte Carlo simulation. Simulation experiments demonstrate that the self-starting multivariate exponentially weighted moving average control chart outperforms other charts in detecting speed and range. It can also predict credit crises, detect more minor changes in risk, and describe their changing trends in real time. Throughout this article, we employ a forward intensity approach, which has been put into long-term applications for corporate default predictions, to illustrate the profile monitoring process.