Credit scoring is an important finance activity. Both statistical techniques and Artificial Intelligence (AI) techniques have been explored for this topic. But different techniques have different advantages and disadvantages on different datasets. Recent studies draw no consistent conclusions to show that one technique is superior to the other, while they suggest combining multiple classifiers, i.e., ensemble learning, may have a better performance. In this study, we conduct an empirical evaluation of the performance of three popular ensemble methods, i.e., bagging, boosting, and stacking, based on four base learners, i.e., Logistic Regression Analysis (LRA), Decision Tree (DT), Artificial Neural Network (ANN) and Support Vector Machine (SVM). The experiment uses the credit dataset including 239 companies’ financial records from China, collected by the Industrial and Commercial Bank of China. Results reveal that ensemble learning can substantially improve individual base learners. Stacking, in our experiments, gets the best performance in terms of all six performance indicators, i.e., type I error, type II error, average accuracy, precision, recall, and F-value.
TopIntroduction
The recent world financial tsunami arouses unprecedented attention of financial institutions on credit risk. Especially for any credit-granting institution, such as commercial banks and certain retailers, the ability to discriminate good customers from bad ones is crucial to their profit (Yu, Wang & Lai, 2008). These institutions must balance risks as well as returns. A good credit risk assessment method can help them to grant loans to more creditable applicants, thus increase profits; it can also deny credit for the non-creditable applicants, so decrease losses. In recent years, credit scoring has been become one of the primary ways for credit-granting institutions to assess credit risk, improve cash flow, reduce possible risks and make managerial decisions (Huang, Chen & Wang, 2007).
The purpose of credit scoring is to classify the applicants into two types: applicants with good credit and applicants with bad credit. Applicants with good credit have great possibility to repay financial obligation; applicants with bad credit have high possibility of defaulting. For credit scoring, the accuracy is quite significant to financial institutions’ profitability. For example, the accuracy of credit scoring of applicants with bad credit increases only 1% may retrieve a great loss for the financial institutions (Hand & Henley, 1997).
Many credit scoring models have been developed by credit-granting institutions and researchers for the credit admission decision. Credit scoring was originally evaluated subjectively according to personal experiences, and later it was based on 5Cs: the character of the consumer, the capital, the collateral, the capacity and the economic conditions. But with the tremendous increase in the number of applicants, it is impossible to conduct the work manually. Two categories of automatic credit scoring techniques, i.e., traditional statistical techniques and Artificial Intelligence (AI) techniques, have been studied by prior researches (e.g., Huang, Chen, Hsu, Chen & Wu, 2004).
Some traditional statistical techniques have been widely applied to build the credit scoring models, such as Logistic Regression Analysis (LRA) (Thomas, 2000; West, 2000), Linear Discriminant Analysis (LDA) (Reichert, Cho & Wagner, 1983; Karels & Prakash, 1987), Multivariate Adaptive Regression Splines (MARS) (Friedman, 1991). Although these methods can be used to assess credit risk, the ability to discriminate good customers from bad ones is still not satisfactory (Yu, Wang & Lai, 2008).