Subscribe to DSC Newsletter

Your social network profile used in credit scoring models

Who you like on Facebook might end up in your loan application being denied. This article was posted in MotherJones as Your Deadbeat Facebook Friends Could Cost You a Loan.

You can get a loan to buy this house if you have the right Facebook friends

My opinion: Some lending institutions just started to test new predictive models that include your social (public) data - what you post, your friends - what the whole world can see. The risk if people going private, then the data is no longer available, creating instability in predictive scoring models.

The article:

It's already well known that Facebook and other social media networks harvest user data and sell it to companies that use that info to peddle their products to consumers. But some lenders have begun to find a new use for this information, scrutinizing Facebook, Twitter, and LinkedIn data to determine the credit-worthiness of loan applicants. It's an unprecedented practice that consumer advocates say can be unfair or discriminatory—and one that is poised to only become more prevalent in the years ahead.  

Among the US-based online lenders that factor in social media to their lending decisions is San Francisco-based LendUp, which checks out the Facebook and Twitter profiles of potential borrowers to see how many friends they have and how often they interact; the company views an active social media life as an indicator of stability. The lender Neo, a Silicon Valley start-up, looks at the quality and quantity of an applicant's LinkedIn contacts for clues to how quickly laid-off borrowers will be rehired. Moven, which is based in New York, also uses information from Twitter, Facebook, and other social networking sites in their loan underwriting process.

Several international lenders have been using similar tactics for a while. Lenddo, for example, which makes loans to folks in developing countries, denies credit to applicants who are Facebook friends with someone who was late repaying a Lenddo loan. Big banks have not yet jumped on board with this controversial credit-vetting method, but consumer advocates and financial industry experts say it's probably only a matter of time.

Read full article at

Views: 2710


You need to be a member of Data Science Central to add comments!

Join Data Science Central

Comment by Pradyumna S. Upadrashta on September 30, 2013 at 9:18pm

In some sense, my guess is that people will likely start complaining about their inability to get credit (the market will become aware of a problem brewing) well before any data miner ever sits down to actually analyze whether there is a gap that needs to be met.  I think the complaints will lead to a data science response, which will reveal the opportunity.  Inefficiencies tend to make a lot of noise, we just have to know where to listen for them and be able to turn them into opportunities.

Comment by Pradyumna S. Upadrashta on September 30, 2013 at 9:14pm

I think the assumption here is that all credit rating agencies will follow an identical model for the kinds of risks they want to take.  I would think that there are some agencies that might be as extreme as using facebook social graph to determine one's credit worthiness; but, this creates a space for competitors that may select a risk profile of "applicants who have solid credit history, but may be associated on some social graph with those who do not".  So, I think such detailed credit models may actually cause an explosion in tiered credit ratings and opportunities for specialized market makers to take certain specific risks.  Assuming that the market is highly reactive and aware of such gaps in the credit market, one might expect fast movers to take advantage of these opportunities.  On the other hand, if the market is not very efficient (i.e., companies don't do enough data mining, aka information flow is not very efficient because no one is looking for these opportunities) then we'll see a lot of people coming up against walls in their ability to get credit.  My guess is that initially as data science begins to infiltrate this market, the response to these situations will be slow; but as adaptation increases, these kinds of opportunities will be met by smart opportunists.  So, I see an opportunity here rather than a problem.

Comment by Vincent Granville on September 20, 2013 at 2:50pm

A reader mentioned that data governance and regulations would make such algorithms illegal. My opinion is that it depends what kind of metrics you use. Here are a few that seem relatively benign: 

  • Presence/absence of Facebook account (if Facebook account is found, then use algorithm to detect if it is fake, hijacked or real) 
  • Multiple accounts attached to person
  • Number of friends
  • Category attached to profile (using a taxonomy to assign categories) 
  • Account recency 
  • Account popularity (number of genuine likes) 
  • Number of posts 

Of course the content (postings) might not represent the reality, thus metrics based on content should be carefully calibrated. But if a solid, consistent pattern emerges, and has great predictive power, why not leverage it? For instance, if age provided on application does not match age on Facebook, and this turns out to be a good predictor. 

It's not any different from insurance companies calling your neighbors to learn about your lifestyle, before making a decision about your application.

© 2019   Data Science Central ®   Powered by

Badges  |  Report an Issue  |  Privacy Policy  |  Terms of Service