Validation for Logistic Regression in Teradata


Validation for Logistic Regression in Teradata

We are working in a Credit Risk Model Exercise for a Bank. We would like to show the feasibility for doing directly in Teradata: the development of a logistic model, the using of this model for scoring of the customers, and the validation (assessment) of the discrimination power & calibration of the model.

For the Calibration & Scoring, we could use TWM directly in Teradata. But I am not very sure about what could be achieved in Teradata for the Validation Step.

Validation Criteria for a Credit Risk Model is classified in two:

• Calibration Validation: Testing if the Probability of Default estimates are accurate. Comparation between the differences of the forecast Defaults vs realised default rates must be validated.

• Discrimination Validation: Assessment of discriminatory power; finding the proper classification tool (threshold value) for the default/ non-default classification based in the score given by the model. T

I know that TWM has several statistical tools (such as Kolgomorov-Smirnov, Binomial Tests, Chi-Square Test and others), but I wonder if anyone has ever implemented these validation tools, or if these could be built in a script that could validate the models directly in Teradata