Over the past twenty years, the awareness and adoption of alternative data has grown substantially as financial institutions seek sustainable ways to grow and expand their customer base. Alternative data provides lenders with an opportunity to innovate and differentiate from competition while maintaining desired underwriting standards and risk levels.
Today, “mainstream” alternative data includes public records (e.g., bankruptcies, liens, and judgments), asset ownership history, rental and utility payments, occupational licenses, and education data. More recently, other information which is being collected about consumers on a regular basis has emerged as a potential way to improve credit scoring models. Examples include social media activity, internet browser history, mobile device data, location (GPS) data, and loyalty/purchase data. While companies have considered using these sources, a more thorough examination of their efficacy and ability to stand up to regulatory scrutiny will be required before such data becomes widely accepted.
When evaluating data for inclusion into our risk scoring models, we primarily look at five categories to determine whether or not a content source should be included:
1. Compliance and Transparency
All data used in consumer credit scoring must comply with the Fair Credit Reporting ACT (FCRA). Consumers must have the ability to request disclosure of all data on their consumer credit file. They must also be afforded the opportunity to dispute incorrect information.
2. Reliability and Accuracy
Content for credit scoring purposes must come from sources with a track record of accuracy and consistency. Owners or distributors of FCRA data must endeavor to frequently update consumer information and uphold the highest levels of security and privacy.
3. Fair Lending
Information which directly or indirectly discriminates against underserved groups or protected classes runs afoul of The Equal Credit Opportunity Act (ECOA) and is carefully excluded from our scoring models.
4. Predictive Power
New data should provide unique insights which directly reflect on a consumer’s creditworthiness. If the data is highly correlated to existing data or is not orthogonal or accretive, it will hold limited value in a risk model.
5. Coverage Breadth
While many data sets may prove predictive of credit behavior, a major obstacle is a lack of coverage of the broader population. Data that is specialized in nature will have limited applicability. A more serious concern is the potential for niche data sets to disproportionately impact one group over another.
The Era of Big Data has brought about a proliferation of intriguing and exciting possibilities for credit scoring purposes. We have touched on the prospects for some of these previously, including social media data. While forward thinking lenders and data providers should continue to evaluate the effectiveness of these sources, they must adhere to core statistical principles, support sound risk strategies, and comply with all relevant regulations.
Click here to read more about Alternative Data and credit risk at our LexisNexis® Risk Solutions Credit Risk blog.
