The development, evaluation, and monitoring of scorecard models play a vital role in credit risk management within the financial industry. For researchers and academics in South Africa, STATISTICA Scorecard offers a comprehensive and reliable software solution to cater to these needs. This article aims to provide an informative overview of the capabilities and workflow of STATISTICA Scorecard, focusing on its modules and functionalities.
Data Preparation
The first step in scorecard model development is Data Preparation, which involves gathering and organizing the initial set of characteristics. STATISTICA Scorecard offers a Feature Selection module that utilizes Information Value (IV) and Cramer’s V to identify significant variables impacting credit risk. Additionally, the Selecting representatives option enables the identification of redundancy among numerical variables using factor analysis with score rotation.
Attributes Building
In the Attributes Building module, researchers can prepare risk profiles for each variable. Using the CHAID method or manual mode, variables are divided into homogenous risk classes or attributes. To ensure proper risk profile construction, the module generates statistical measures, such as Weight of Evidence (WoE) and Information Value (IV), while also offering the option to adjust attributes based on business and statistical criteria.
Modeling and Scorecard Building
The Credit Scorecard Builder module combines attributes prepared in the previous step with the logistic regression model. Users can choose from various model-building methods, including Forward entry, Backward elimination, and Bootstrap for all effects. The module provides a set of statistics and reports, aiding in model evaluation and selection. After model creation, the logistic regression algorithm is used to prepare the scorecard format, which can be saved in different file formats.
Survival Models
Survival Models allow researchers to build scoring models using the Cox Proportional Hazard Model, incorporating additional information about the time of default. This facilitates the calculation of the probability of default within specific time frames, aiding in risk assessment.
Reject Inference
When credit applications are rejected, information about the output class (good or bad credit) is missing. The Reject Inference module in STATISTICA Scorecard addresses this issue by utilizing the k-nearest neighbors method and parceling method to garner the required information, resulting in a new data set with complete information.
Model Evaluation
The Model Evaluation module enables researchers to compare and assess different scorecard models using statistical measures such as Information Value (IV), Gini index, ROC curve analysis, and Hosmer-Lemeshow statistic. The module also provides additional reports for a comprehensive evaluation of generated models.
Cutoff Point Selection
Cutoff Point Selection determines the optimal value of scoring to differentiate between accepted and rejected applicants. Users can manually set cutoff points based on custom misclassification costs and bad credit fraction using ROC analysis. The module also allows for the assessment of the selected cutoff point through various reports.
Score Cases
The Score Cases module facilitates the scoring of new cases using the selected model. It calculates overall scoring, partial scorings for each variable, and the probability of default, adjusted by an a priori probability of default for the whole population.
Population Stability
The Population Stability module provides tools for comparing two data sets to detect any significant changes in characteristics or applicant populations. This analysis helps researchers identify the need for re-estimating model parameters.
Comments from STATISTICA Scorecard Users
STATISTICA Scorecard has garnered praise from users, such as Millennium Bank, who find the software to be comprehensive and user-friendly. Stefczyk Credit Union and SKOK im. M. Kopernika also share their positive experiences with the software, highlighting its effectiveness in credit risk model development and management.
Conclusion
STATISTICA Scorecard is an indispensable software solution for researchers and academics in South Africa seeking to develop, evaluate, and monitor credit scoring models. Its diverse modules and functionalities empower users to create robust and reliable scorecard models for effective credit risk management. To experience the benefits of STATISTICA Scorecard firsthand, interested parties can download a trial version from the official website.