Date
12-2020
Department
Graduate School of Business
Degree
Doctor of Business Administration (DBA)
Chair
David Bosch
Keywords
Auto Insurance, Criminal Background, Segmentation, Loss Cost
Disciplines
Business | Insurance
Recommended Citation
Whipple, Krista Malmquist, "Predictive Power of Criminal Background on Losses" (2020). Doctoral Dissertations and Projects. 2761.
https://digitalcommons.liberty.edu/doctoral/2761
Abstract
Product and data science teams for the auto insurance industry have been trying to increase pricing segmentation with validated rating variables to decrease rate subsidization. The criminal background data availability provided a new behavior variable to test against insurance-based credit scores as a potential predictive variable in the generalized linear rating model. Criminal background was analyzed using a Poisson Log Linear model and other key insurance rating variables for predicting loss costs. The study supported the inclusion of the criminal background data in combination with insurance-based credit score as the variable’s addition could improve the overall fit of the predictive model. The study also acknowledged there was a statistically significant association between criminal background and insurance-based credit score, but the overall size of the effect was small and weak. The overall contribution of value criminal background variable needs to be considered with a full rating dataset to determine if other, less powerful variables could be removed from the generalized linear to reduce the overall model complexity.
