Saturday, July 26, 2008

Prosper.com - collections - you have forsaken lenders

Collections on late prosper.com loans are getting worse. Prosper simply doesn't take their responsibility to lenders seriously. Along the way there have been words from Prosper saying they were going to improve this or that, but they've all been hollow. Ding ding .. nobody's home.

Here's the latest data on collections performance, direct from prosper.com. Most late Prosper loans are sent to the collection agency Amsher. To keep things simple, I've graphed only Amsher's performance here. I've shown the fraction of loans sent to Amsher that have been cured. These numbers come directly from Prosper's web site, with no manipulation whatsoever. Prosper breaks this statistic down into three categories, according to credit grade, and that's what you see in the graph.

Prosper/Amsher late loan "cured" Fraction

Prosper switched to Amsher, and dumped a bunch of loans on them, on 2/23/08, so of course the fraction of those loans that Amsher had cured on that date was zero. It took a couple of months for the stats to recover from that initial impulse, and to climb the learning curve, in other words learning how to collect Prosper loans. Well, according to this data, the honymoon is over. Since 6/1/08 Amsher's stats have been going down.

It wouldn't be so bad if there had ever been any good performance here, but Amsher never did anything good (according to the stats Prosper.com has published).

In fact, according to Prosper's stats, Amsher has never done as good as Penncro, the collection agency it replaced. Now how the heck can that be? In some of my earlier posts on this subject I've included a graph showing combined Penncro+Amsher performance. This is the all-time performance of loans worked on by either or both. This representation eliminates the startup impulse problem in the above graph. However, it involves some manipulation of the data, which confuses some folk. Here's an update on that graph:

Prosper collections rates 07/26/08

This curve changes from blue to red when the handoff from Penncro to Amsher occurred. As you can see, Penncro got up to over 17% cured, and Amsher has been mostly downhill from there. As the number of loans handled by Amsher has increased, their lower cure rate has started really pulling down the combined curve. The explanation of how the data is combined can be found here.

How (expletive deleted) could things be this bad? Prosper management has made comments over the past few months about how collection statistics are improving. The stats Prosper.com gives us, on the other hand, are getting worse. What's the deal?

I put this question to Prosper management in mid-June. They said they'd have to look into the numbers and get back to me. I'm sure they ment to do just that, but no answers have been forthcoming. Perhaps higher priorities intervened. Ain't it always the way.

One possibility is that the numbers Prosper displays on the web site may be faulty. It wouldn't be the first time. Prosper used to display a statistic they called "net collected". There's still a row with this label on the web site, but they no longer fill in the numbers. I used to graph this statistic vs time, and as a result, I knew it was ... well ... I suppose the polite word is "bogus". This was discussed at length in the old prosper.com forum. Here's a chart I made about a year ago:

Prosper collections rates 2007-07-02

The red curve jerked up and down faster than was physically possible. From that I knew it was bogus. Prosper never did admit there was a problem. They just deleted the statistic from their web site. Well maybe that is an admission.

With that precedent in mind, is it possible that the collection statistics they provide for us now are just wrong? I don't know. I look forward to some feedback from inside Prosper.

If the numbers correctly reflect collection activity, then how the hell could it be that they are so horrible? When a loan goes 1 month late, I'd expect it to recover a significant fraction of the time. Amsher's 6% is not my idea of a significant fraction of the time. In fact, I'd expect more than 6% of 1-month-late loans to recover all by themselves, without any collection activity at all.

Something is horribly wrong.

Nobody seems to be doing anything about it. Seems like nobody cares.

Prosper, you have forsaken us (lenders).

See my prior writings on this subject, including:
Written 05/06/07: Collections is broken
Written 05/04/08: Collections is not improving

Update 08/08/2008: Just got a note from Prosper saying they have found and corrected a major bug in the way they were calculating the collection statistics they present on the web site, and believe that there is an additional problem, for which they are continuing to search. So... It looks like the data they were giving us was indeed faulty.

Friday, July 4, 2008

Prosper.com - 07/01/08 late loan stats update

Here's the July 1st update to 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

06/01/08 small late loan chart

Here's a chart of the same data in which each curve has been slid to the left to a common origin. The horizontal axis is now days since loan origination month.

5/15/08 slid

Explanation of methodology can be found in my prior postings in this blog.

An important factor in prosper.com loan results is how well Prosper collects the payments due on these loans once have become late. Please see my prior writings on that subject, including:

Written 05/06/07: Collections is broken
Written 05/04/08: Collections is not improving

PS: The best discussion among Prosper.com lenders can be found at http://prospers.org/