Overview
Most banks want to understand what makes one loan riskier than another. Many banks use expert based Credit Scoring to assign a credit score to each loan. However, it is often useful to find a quantitative assessment of the factors affecting loan default.
One of the primary tools used to assess the relationship between default and underlying factors is Logistic Regression. This example demonstrates using logistic regression ot estimate the sensitivity of loan default to the chosen factors.
Logistic Regression App
A simple way to run the regression is to use the logistic regression app. The app provides a graphical user interface for running logistic regressions.
Scripting
Logistic regressions can also be run by writing a script and utilizing a logistic regression library. The following demonstrates a script to run the regression. An alternative to using the logistic regression library given above, one could also evaluate a logistric regression using tensor flow, although it will run much slower. See Running Factors with Tensor Flow for the alternate script.
Next Step
In order to measure total portfolio risk, we must have some measure of the correlations of default of each loan.