1. Course wrap-up
Congratulations! You've now mastered some basics in the credit risk modeling field. Before finishing the course, I'd like to make some closing remarks. Obviously, there is a lot more to credit risk modeling than discussed in this course.
2. Other methods
In fact, other methods for classifying defaults versus non-defaults exist as well. One of them is discriminant analysis. In discriminant analysis, a discriminant function is used to assign data to one group or another, default and non-default in the credit risk modeling case. Another very popular and effective method is a random forest. Random forests are based on decision trees, using an ensemble of trees to classify cases. Other methods include neural networks and support vector machines. Whereas the latter two techniques are extremely flexible and effective, the disadvantage is that these methods are complex and black box, which makes it hard to understand how certain variables affect the outcome.
All models we refer to here are focusing on classification.
3. But... very classification-focused
An important shortcoming of these models is that the timing of default is completely neglected, so no distinction is made between a debtor that defaults in earlier stages of the repayment phase and one that defaults after many months of completed repayments. A method that overcomes this and has become increasingly popular in credit risk modeling is survival analysis. Using survival analysis, one can obtain probabilities of default that change over time. Another advantage of these models is that variables that change over time can be included, which enables you to include factors that reflect the state of the economy, such as house prices and the employment level. As the number of defaults typically goes up in hard economic times, including these factors could significantly improve predictive power.
4. Congratulations!