Sunday, March 9, 2008

Prosper.com - late loan stats update

Here's the 03/01/08 update to my late loan statistics charts.

These charts show statistics for the performance of all prosper.com loans. Each curve represents the set of loans that were created in one calendar month. The vertical axis is the fraction of those loans that have "gone bad", in other words are 1 month late or worse (up to and including default). The horizontal axis is the observation date. All data comes from Prosper.com's performance web page.

A larger, more readable version of that chart can be found here

prosper late 2008-02-15 small

Here's a version with the curves slid to a common origin. The horizontal axis is now days since month of loan origination.

2008-03-01 slid

Prosper threw us a curve this month. You may recall that a group of loans were partitioned off a few weeks back into a program called "new agency test". Its actually a set of very late loans on which has engaged some lawyers to take legal action against deadbeat borrowers. For techncal legal reasons, prosper bought these loans back from lenders for either $0.00 or something very close to zero (depending on whether lenders opted-in to the program ... a detail that doesn't matter here) prior to initiating legal action. After buying the loans back, prosper changed the state of these loans in their database to "repurchased". I know thi sounds straightforward, but it is not. The "repurchased" state was created early in Prosper's history for an entirely different purpose. At that time, Prosper would occasionally acknowledge that a loan was a result of identity theft, at whcih time Prosper would buy back the loan from lenders at 100% of face value. Prosper argued that by buying back the loan at face value, they restored lenders to a condition as if the loan had never existed. Therefore, they chose to exclude these "repurchased" loans from loan performance statistics. The prosper performance web page, for example treats "repurchased" loans as if they had never originated. I believe thta most 3rd party statistics web sites (ericscc, lendingstats,...) do the same.

Now however, Prosper has used the "repurchased" state for a new and different purpose. Loans that have gone bad, and for which the lender has not been reimbursed, have been put in the "repurchased" state, and eliminated from loan performance statistics. This error makes Prosper's ROI calculations look better than they should, and similarly boosts the lender ROI calculations at ericscc's and lenginstats' web sites. Bummer. This is database pollution.

A note for the conspiracy theorists: I don't think Prosper did this on purpose. I think they were lazy, and didn't think thru the implications of their action. The loan "states" in the database have meaning. Software depends on that meaning, even software written by Prosper itself. If you're gonna use an existing state for a new purpose, make sure you understand the meaning first.

I have carefully adjusted each month's stats in the curves above so that these repurchased loans are counted in both the numerator and denominator of the bad loan ratio. I urge others who use Prosper's data to do the same, at least until Prosper fixes this problem.

Discussion of methodology can be found in earlier posts in this blog, and the late loan thread on the old prosper forum, now found here: http://www.prosperreport.com/prosper/forums/archive/threads/1/1/2/11290.0.HTM