One of Prosper’s most recent changes was to add bidding guidance on the manual bid page. See the picture below.
I think this is a great feature in theory. It makes the estimated return obvious using Prosper’s own roll rates. I am sure this will be an incredibly valuable feature for all manual bidding lenders.
However, I would caution lender to think for themselves a little bit. Speaking as one who has analyzed a bunch of Prosper data if you sliced the current loan data into 54 equal segments (see graph at the bottom of the page) you start to get some mighty fine segments. There are currently 15,832 loans on Prosper. 15,832 / 54 = 293 per segment assuming equal size buckets. I haven’t looked but I am sure that not all buckets are of equal size. Even with even distribution this is cutting it close.
And then there is Heisenberg’s Uncertainty Principle (ie now that these segments are measured they a bound to change). I would imagine the Prosper updates the various adjustment fields on a regular interval, but even so the nature of the buckets themselves might change.
Overall, I think this feature is more good than harm. In fact, I am not sure there is any harm. I just think that lenders need to think for themselves. For example, I biggest hole in the segments is that they don’t adjust for loan amount. Then again 108 segments would cut the data even thinner.
If you liked this article, vote for it on del.icio.us and stumbleupon.Categories:
Features, Prosper.com, Statistics, Tools
Related Articles Related Stores
6 comments ↓
Rather than 293 buckets, shouldn’t that be 54 buckets of 293 loans each (assuming they are equal which they probably aren’t). Where are you getting 293 buckets ?
Some will be sparse and some will hide good/bad loans, such as there is only one segment of HR with autofunding that isn’t broken down further into # of DQs, so the 0 DQs and 6+DQs share a single combined expected loss rates, but guess who carries most of that, and who is the less risky of the two.
I haven’t checked the actual numbers, but as long as people don’t get scared off by the big estimated loss numbers, there may be possiblities there.
As you mention loan amounts will be a factor as well
What amazes me after all the fuss about DTI
There are no DTI buckets!!! Higher DTI is obviously worse. Prosper data does not show that because lenders would fund every junk loan under 20% DTI. So Prosper is using numbers that were biased by Prosper.
Chrisfs:
Rateladder is saying 54 buckets, not 293:
“There are currently 15,832 loans on Prosper. 15,832 / 54 = 293 per segment assuming equal size buckets.”
Correction. A’s do have a DTI bucket
But As do not have a DQ bucket!
A’s with DQs will now be funded by Prosper Portfolios!
[...] are based on the slices of data… Which brings me to the ugly. There are almost certainly thin data issues. Readjustments can help, but Heisenberg rules this [...]
[...] the ROI would be nice, but I have already pointed out what I consider to be the major flaws: thin data and large buckets of loan criteria. Related Articles Prosper Starting To Use the Developer Mailing [...]
Leave a Comment