Does segmentation always improve model performance in credit scoring?

Abstract

Credit scoring allows for the credit risk assessment of bank customers. A single scoring model (scorecard) can be developed for the entire customer population, e.g. using logistic regression. However, it is often expected that segmentation, i.e. dividing the population into several groups and building separate scorecards for them, will improve the model performance. The most common statistical methods for segmentation are the two-step approaches, where logistic regression follows Classification and Regression Trees (CART) or Chi-squared Automatic Interaction Detection (CHAID) trees etc. In this research, the two-step approaches are applied as well as a new, simultaneous method, in which both segmentation and scorecards are optimised at the same time: Logistic Trees with Unbiased Selection (LOTUS). For reference purposes, a single-scorecard model is used. The above-mentioned methods are applied to the data provided by two of the major UK banks and one of the European credit bureaus. The model performance measures are then compared to examine whether there is improvement due to the segmentation methods used. It is found that segmentation does not always improve model performance in credit scoring: for none of the analysed real-world datasets, the multi-scorecard models perform considerably better than the single-scorecard ones. Moreover, in this application, there is no difference in performance between the two-step and simultaneous approaches.

A window of opportunity: Assessing behavioural scoring

“…The findings reported in this work are based on the use of static characteristics in behaviour scorecards. As we are using real world data from a credit bureau our work focuses on a special type of behavioural scoring called credit bureau scoring (see Bijak and Thomas, 2011), however our findings are applicable to the many types of behavioural scoring. Figure 1 illustrates the longitudinal aspect to the data used in behavioural scoring.…”

“…logistic regression) are used by application scoring and behavioural scoring that best classify customers into one of two categories: goods and bads . Behavioural scorecard modellers encounter many of the same scorecard development and implementation issues as with application scoring, such as: identifying and adjusting for different segments of the population (see Bijak and Thomas, 2011), ensuring the optimal correlation between features (see Tsai, 2009), handling class imbalance (see Burez and den Poel, 2009), identifying the correct sample size (see Crone and Finlay, 2011), to name but a few challenges.…”

“…As credit scoring is essentially a discrimination problem (good or bad), one may resort to the numerous classification techniques that have been suggested in the literature (see Lee et al, 2005). Based on both the task and data used, credit scoring is traditionally divided into two broad types (Bijak and Thomas, 2011). The first, application scoring, is used at the time an application for credit is made and estimates an applicant's likelihood of default in a given time period.…”

A Soft Intelligent Risk Evaluation Model for Credit Scoring Classification

“…Bijak and Thomas [29] have proposed a two-step and simultaneous approach, in which both segmentation and scorecards are optimized at the same time by using Logistic Trees. Chi and Hsu [30] have combined a bank's internal behavioral scoring model with the external credit bureau scoring model to construct the dual scoring model for credit risk management of mortgage accounts.…”

Investigating the Performance of Smote for Class Imbalanced Learning: A Case Study of Credit Scoring Datasets

“…Traditionally, CS is categorized into two types on the basis of the data used and task assigned (Bijak & Thomas, 2012) i.e., application scoring and behavioral scoring. Application scoring will estimate the probability of applicants to default for some given time interval.…”