Introduction to Credit Scoring Using R

Motivations

I have been planning to write a book on credit scoring using R for quite a while. Credit scoring systems generally aim at the identification of loan borrowers who default on their loans. There are a lot of reading materials available on credit scoring. But to my best knowledge, most of these materials focus on basic logistic regression methods or SAS programming. SAS is great but it is not free. Logistic regression is powerful and widely used in practice because of its great interpretability, but if the prediction accuracy is the first priority we always can do it better using some modern machine learning methods such as gradient boosting machine and factorization machine. If we are more interested in the question when the defaults occur but not if the borrowers default or not, survival modeling techniques should be used. Survival models are especially useful in case you want to forecast the cashflow of a loan.

Audience

The targeted audience of this book include, but not limited to, people who are

• interested in learning R with limited previous programming knowledge and experience
• working on credit scoring in banks, funds, etc.
• looking for some real world applications of machine learning and statistics in Finance

Topics

• Random variable, conditional probability and Bayes' theorem

• Distributions and sampling distributions

• Maximum likelihood estimation and hypothesis testing

• Logistic regression

  • Logistic regression
  • Logistic regression with regularization
  • Logistic regression with spline
  • Logistic regression with (nonlinear) interactions

• Decision tree and Random forest

• Gradient boosting machine

• Survival modeling

  • Poisson process and intensity function
  • Continuous-time and discrete-time survival models
  • Proportional hazards models
  • Parametric proportional hazards model
  • Cox proportional hazards model

• Mixture cure models

  • EM algorithm
  • Monte Carlo simulation methods

• Gradient boosting survival method

• Competing risks

  • Multiclass classification
  • Survival models under competing risks

• Something else