| Summary: | 碩士 === 東吳大學 === 經濟學系 === 93 === The purpose of this paper is mainly in the hope of using different credit scoring models to make cluster, and to observe the impact of credit scoring models on cluster. In this paper, we not only compare the difference between Logit model and Multiple Discriminate Analysis (MDA) model, but also compare the results acquired from two kinds of different data.
In sample data, we use the database of Taiwan Finance Database of Taiwan Economic Journal and collect 14,125 company’s’ finance data which include 153 default company’s’ data to study. We use these financial data to make Variance Inflation Factor (VIF) test, and use the result of credit scoring model to make Kolmogorov–Smirnov (K-S) test and Receiver Operating Characteristic (ROC) check. Finally, the conclusions are listed as following:
1, The ability of credit scoring model can be improved after the sample data was adjusted for extreme value data dealing, but not necessarily improve the stable of cluster.
2, The Logit model is better than MDA model in two kinds of database.
3, Different credit scoring model and data dealing will influence the results of cluster. The better the ability of credit scoring model, the larger the difference between clusters in probability of default.
4, After all of our study and test, we find the Logit model, which was adjusted after extreme value data dealing have the best ability to score and cluster.
|
|---|
