Credit risk analysis for financial corporations


Student thesis: Master's Thesis

View graph of relations


  • Chi Hang WONG

Related Research Unit(s)


Awarding Institution
Award date3 Oct 2006


In this study, several statistical models including multiple discriminant analysis, ordinal logit and probit regression analysis, cluster analysis, neural network analysis, and decision tree were employed to predict the credit ratings of US financial corporations. It was found that principal components analysis and factor analysis provided additional power for developing the prediction models using the above analysis. Due to the data limitations, only multiple discriminant analysis and ordinal logit and probit regression analysis were found to have distinguishing power in predicting credit ratings. Among these models, logit regression analysis performed the best, and can correctly predict 70% of corporations. Moreover, it was revealed that although the data structure of financial corporations were different from non- financial corporations, multivariate analyses were still useful and applicable. Other than statistical models, mathematical programming model was also employed in this study to classify corporations into different credit rating classes, and provided satisfactory performance. Furthermore, another commonly used credit evaluation tool – KMV model which was originally developed by KMV Corporation using the firm’s current market information to assess the default risk was used in this study. In addition to the original Black-Scholes-Merton’s KMV model, two modified models have been developed and found to have better performance than the basic one. When compared to the agency’s rating, it was found that all KMV models have stronger power to detect default for “Unhealthy” corporations. In addition, recovery rate based on three approaches, namely long-term average recovery rate, correlation between default rate and recovery rate, and relationship between probability of default and recovery rate, were calculated to assess the potential credit losses of US financial corporations. It was found that probability of default and recovery rate were strongly inversely correlated. Besides, using the modified models to estimate the recovery rate produced a more reasonable and superior result to the original model. Moreover, models’ performances have been evaluated by using non-parametric bootstrap, parametric bootstrap, and empirical simulation techniques. The findings were consistent with those in the original sample, and thus they have provided evidence to support that all developed models in this study were valid and applicable in other situations. Lastly, the models developed in this study are applied to Asia rated and Hong Kong unrated financial corporations. It was found that all models provided satisfactory performance, which indicated that they remained useful in evaluating creditworthiness of financial corporations in other parts of the world.

    Research areas

  • Credit, Commercial finance companies, Management, Risk management