Rice Researcher Measures Creditworthiness through Prosper Faces
A team of researchers headed by Jefferson Duarte from Rice University’s Jones School of Management has published a study entitled “Trust and Credit“. In their research, the team took photos posted on Prosper loan listings, and had 25 users from Amazon’s Mechanical Turk site evaluate each picture for “trustworthiness”.
Each of the 25 users was presented a picture of a Prosper borrower, and asked to rate their trustworthiness on a scale from 1-5. They were also asked to predict the probability that that borrower would repay a loan of $100.
The researchers found that the R2, or coefficient of determination, between perceived borrower trustworthiness (as determined by humans looking at their pictures) and the borrower’s credit grade (as determined by the Experian credit reporting agency) was between 1% and 1.7%, which seems incredibly low. This means that your first impression of a borrower’s trustworthiness may be just as valuable as your peek at their credit score. Since Prosper doesn’t validate borrower photos, it’s unclear how much value this information has on such a platform. But in a real-world situation (such as a bank lobby, or if Prosper were to build brick & mortar locations), there may be an opportunity to put this information to use.
How did this play out in actual listings funded and loans made? Not surprisingly, borrowers who looked un-trustworthy were less likely to be funded. Even among those who did get funded by Prosper lenders, un-trustworthy borrowers paid on average 1.82% more interest per annum for their loan.
The researchers also made a couple of interesting points about Prosper, citing it as a unique example for their research. First, photographs to accompany loan information are unique. Second, behavioral psychology rarely gets to measure the input of thousands of unbiased actions; in the case of Prosper, thousands of lenders express their desires through their bids, and because Prosper provides all of their data through a downloadable file and web services API, researchers and academics have the ability to do any analysis they want. Third, economics rarely has the ability to see failed outcomes – typically, only successful matches can be measured; on Prosper, even failed listings are saved and provided through the API.
Here is the full text of the paper: (Source)
Related posts:
- Lending Club vs. Prosper, One Week In
Last Monday, after the SEC approved Prosper’s registration filing, the peer to peer lender opened its doors to borrowers in 47 states and lenders in 16 states. Lending... - Prosper Relaunches in California Without SEC Approval
It has been a little over six months since Prosper shut down its social lending marketplace under threat (and later consummation) of a cease and desist action from... - LendingClub introduces mapping, top lenders list
In an email announcement that went out to active lenders last week, LendingClub previewed a couple of new upcoming features.... - IOU Central becomes Canada’s first P2P Lender
Canada’s IOU Central launched yesterday, becoming the first person to person lending site in the great white north. CommunityLend, which had made their intention to launch public months... - Prosper Announces Major Changes to Groups Concept
In an email dated today, Prosper announced that they would be removing the so-called “Group Leader Rewards” option which allows Prosper group leaders to collect a portion of their borrowers’...