Focused Model Averaging Estimator:An Application In Credit Scoring

Credit scoring models are useful instruments in credit risk management and play pivotal role in credit decisions of banks and other financial institutions.How to improve the prediction accuracy of a credit scoring model and reduce the cost of misclassification are the core research topics in credit scoring models.Logistic regression is one of the most commonly used model in credit scoring.When a researcher has different candidate models at hand,the usual practice is to determine an appropriate model first and use the selected model to conduct credit scoring.As a natural extension of model selection,model averaging is deemed to be more accurate in prediction.However,existing model averaging methods are not very appropriate for the purpose of credit scoring.The reason is that these methods are not misclassification cost focused.In fact,the cost of misclassifying a good customer as bad customer is usually smaller than that of misclassifying a bad customer as good customer.To handle these concerns,in current thesis,I construct the loss function for misclassification cost within the framework of model averaging and propose a new model averaging method(focused jackknife model averaging,FJMA)based on this loss function for logistic regression.Monte-Carlo simulation shows that the proposed method outperforms the existing model averaging and model selection methods in terms of misclassification cost.The FJMA is adopted to analyze a set of real data and the results also support the use of FJMA in practical situations.