Something strange is going on with Prosper’s Credit Segment default projections and the Bidding Guidance which utilizes them.
Prosper clearly defines that estimated losses are created from very specific data:
“Estimated average annualized loss rate based on the historical performance of Prosper loans for borrowers with similar characteristics, originated between Jun-01-2006 and Jan-31-2008, measured as of Mar-10-2008. Actual performance may differ from estimated performance due to many reasons, for example, worsening economic conditions.”
When the latest Credit Segments were released last Monday (4/14/08), I naturally saved a copy, and then proceeded to recreate each segment on Prosper’s Performance Page so that I could generate the historical loss rate for each slice. * On a side note, I wish Prosper would show the loss rate in the Credit Segment table provided on the website. This would save me a TON of time. *
Prosper’s Bidding Guidance uses the loss rates for each Credit Segment to display to the lender on the bidding page. For example:
But when I recreate this specific Credit Segment on Prosper’s Performance Page, I get a much different result:
Prosper is displaying a historical default rate of 4.98% (net defaults + adjustment), but I can only recreate a total loss of 2.84%, nearly half of what is displayed.
This particular listing is of interest to me because the borrower is the husband of the forum regular mhs505. I had the chance to meet mhs505 at Prosper Days 08, and would feel comfortable bidding on this listing. Actually, I already have, and might just bid more.
Unfortunately for those that do not know this borrower, all they will see is the projected loss provided by Prosper of 4.98%. This projected loss is more than 2% higher than what I believe should be displayed. Of course, this could impact the final rate this borrower sees by as much as 2%. Ouch!
But wait, there’s more.
I took it upon myself to (randomly) check some other default projections shown on the Bidding Guidance and compare them to the default projections I recreated on the Performance Page. Some match exactly, but often there is a difference, although not as drastic as the case above.
In the list below, the first number is the Credit Segment (#), the next is number shown on the Bidding Guidance page (BG) , and the third is what is shown on Prosper’s Performance page (PP), using the appropriate date ranges:
#...............BG.................PP
01...........0.19%...........0.19%
03........... 1.53%...........1.29%
06........... 4.98%........... 2.85% -- example shown above
07........... 5.74%........... 4.50%
13........... 0.61%........... 0.58%
16........... 1.25%........... 1.25%
19........... 4.39%........... 3.85%
20........... 5.24%........... 6.53% -- the only one I found that was less
22........... 4.47%........... 4.47%
30........... 2.80%........... 2.80%
33........... 3.27%........... 3.05%
34........... 11.21%........... 11.21%
39........... 7.56%........... 7.56%
50........... 4.38%........... 4.38%
51........... 7.49%........... 7.49%
60........... 17.95%........... 17.95%
71........... 5.57%........... 5.57%
74........... 6.61%........... 6.61%
75........... 15.22%........... 15.22%
It seems from the quick sample that I took that there were no problems in the C and D grade loans, 1 problem in the B grade, and 3 problems in each the A and AA grade.
Most of the variation is pretty small. I’ve noticed these discrepancies in the past, but because they are so small I’ve never put much weight on the differences. Plus it seems there has always been some slight differences shown on the performance page depending on what day you use it and at what phase the moon is in. If that was the case here, I would expect all the figures to show some variation, but that is not the case.
While I hope I’m not screwing something up on the Performance Page because then I’ll feel really dumb, I kind of hope that I am. (It’s pretty easy to make mistakes on the Performance Page, as I had to make corrections to a recent blog post regarding some results.) But I’ve checked these several times and have a high level of confidence on the figures shown above. Please correct me if I’m wrong.
But, if I have not made a mistake, then Prosper has some explaining to do (especially to those AA borrowers that fall in Credit Segment #6 - 21 listings, at the moment) Why are there differences between what is shown on the Bidding Guidance page and the Performance Page?
Prosper clearly defines that estimated losses are created from very specific data:
“Estimated average annualized loss rate based on the historical performance of Prosper loans for borrowers with similar characteristics, originated between Jun-01-2006 and Jan-31-2008, measured as of Mar-10-2008. Actual performance may differ from estimated performance due to many reasons, for example, worsening economic conditions.”
When the latest Credit Segments were released last Monday (4/14/08), I naturally saved a copy, and then proceeded to recreate each segment on Prosper’s Performance Page so that I could generate the historical loss rate for each slice. * On a side note, I wish Prosper would show the loss rate in the Credit Segment table provided on the website. This would save me a TON of time. *
Prosper’s Bidding Guidance uses the loss rates for each Credit Segment to display to the lender on the bidding page. For example:
But when I recreate this specific Credit Segment on Prosper’s Performance Page, I get a much different result:
Prosper is displaying a historical default rate of 4.98% (net defaults + adjustment), but I can only recreate a total loss of 2.84%, nearly half of what is displayed.This particular listing is of interest to me because the borrower is the husband of the forum regular mhs505. I had the chance to meet mhs505 at Prosper Days 08, and would feel comfortable bidding on this listing. Actually, I already have, and might just bid more.
Unfortunately for those that do not know this borrower, all they will see is the projected loss provided by Prosper of 4.98%. This projected loss is more than 2% higher than what I believe should be displayed. Of course, this could impact the final rate this borrower sees by as much as 2%. Ouch!
But wait, there’s more.
I took it upon myself to (randomly) check some other default projections shown on the Bidding Guidance and compare them to the default projections I recreated on the Performance Page. Some match exactly, but often there is a difference, although not as drastic as the case above.
In the list below, the first number is the Credit Segment (#), the next is number shown on the Bidding Guidance page (BG) , and the third is what is shown on Prosper’s Performance page (PP), using the appropriate date ranges:
#...............BG.................PP
01...........0.19%...........0.19%
03........... 1.53%...........1.29%
06........... 4.98%........... 2.85% -- example shown above
07........... 5.74%........... 4.50%
13........... 0.61%........... 0.58%
16........... 1.25%........... 1.25%
19........... 4.39%........... 3.85%
20........... 5.24%........... 6.53% -- the only one I found that was less
22........... 4.47%........... 4.47%
30........... 2.80%........... 2.80%
33........... 3.27%........... 3.05%
34........... 11.21%........... 11.21%
39........... 7.56%........... 7.56%
50........... 4.38%........... 4.38%
51........... 7.49%........... 7.49%
60........... 17.95%........... 17.95%
71........... 5.57%........... 5.57%
74........... 6.61%........... 6.61%
75........... 15.22%........... 15.22%
It seems from the quick sample that I took that there were no problems in the C and D grade loans, 1 problem in the B grade, and 3 problems in each the A and AA grade.
Most of the variation is pretty small. I’ve noticed these discrepancies in the past, but because they are so small I’ve never put much weight on the differences. Plus it seems there has always been some slight differences shown on the performance page depending on what day you use it and at what phase the moon is in. If that was the case here, I would expect all the figures to show some variation, but that is not the case.
While I hope I’m not screwing something up on the Performance Page because then I’ll feel really dumb, I kind of hope that I am. (It’s pretty easy to make mistakes on the Performance Page, as I had to make corrections to a recent blog post regarding some results.) But I’ve checked these several times and have a high level of confidence on the figures shown above. Please correct me if I’m wrong.
But, if I have not made a mistake, then Prosper has some explaining to do (especially to those AA borrowers that fall in Credit Segment #6 - 21 listings, at the moment) Why are there differences between what is shown on the Bidding Guidance page and the Performance Page?
2 comments:
Hi LoanChimp - Andrew from Prosper here. What you're seeing is expected – on some of the credit grade segments, we manually adjust the default rate up or down. We do this in one of three cases: 1) if the segment is very thin, and the performance page doesn't provide any data at all, 2) if the segment is thin, and the performance page uses default roll rates, but we think the actual roll rates will be higher or lower, or 3) in the case of E and HR borrowers, where the credit grade definition changed in Feb 2007.
As time goes on and these segments fill out, the estimated default rates on the manual bidding page and the marketplace performance page will come closer and closer. Until then, there will be variations between the data.
I agree with your suggestion about posting the estimated default rates and adjustments in that table - we'll try to add that soon.
Hi Andrew, thanks for dropping by and provding an explanation.
I posted a follow up to this here:
http://warmnfuzzyprosperblog.blogspot.com/2008/04/bidding-guidance-variation-explained.html
thanks again, LC
Post a Comment