Many flagship credit scoring models are created by sampling a population that represents the spectrum of credit risk within the entire US population. This strategy can work well for lenders with a nationwide footprint. However, lenders that do business in select regions can have significant aberrations in the performance of their model if the model does not consider attributes that account for regional variations.
Why would borrowers with similar trade line history perform differently depending on where they live?
Models that use only traditional credit data to score consumers can get it wrong. They can assign the same score to two consumers with similar tradeline history performance, and the loans will end up performing differently. What are these traditional credit only models missing? Among other factors, region of the country lived in, which impacts the current value of a consumer’s resident. Both are important factors that can contribute to predictive insight into potential consumer behavior.
At defi FEST 2017, LexisNexis Risk Solutions presented a case study that highlights this regional variance in credit risk profiles, using one of our lenders that was originally doing a majority of their business in a single state. As their business grew in other areas of the country, however, so did this lender’s delinquency and charge off rates. It soon became apparent there was something missing in their model when it came to the performance of loans in other regions.
By building a new model and bringing in additional attributes with the ability to identify regional variations, our lender was able to more appropriately score their applications across all regions. During this modeling process, the differences between the predictive power of custom models over flagship models became very evident.