Prosper Days 2008 Recap - Part 4

Let’s take a look at Prosper's data regarding endorsements and bids from friends. Where do I begin?

First I thought, ok, lets look at all loans in 2007 that could have endorsements from friends. Since the endorsement feature was introduced in February 07, I began the search on March 1, 07, with an end date of December 31, 07. All of the data presented will be from this date range with an observation date of March 2, 2008.

Next, I thought I should at least put in some sort of qualifier to only show the types of loans that I would find myself bidding on. As it turns out, Prosper created what they call the “Prosper Select Index”, which is loans in grades AA-E with 0 current delinquencies, 0-3 inquiries, and a DTI less than 40%. Perfect!

Using Prosper’s very cool Performance Page, I was able to generate the following Estimated ROI table (3431 loans):

Query 1

I then added a little more criteria to the first search, giving me only the performance where there were 0 endorsements and 0 bids from friends. I call these the lonely borrowers (2888 loans):

Query 2

Wait a minute. This isn’t the result I was expecting at all.

One part makes sense. The average interest rate for the lonely borrowers (0 bids or endorsements from friends) is slightly higher compared to the entire set in Query 1. That means the balance of loans between the two queries (3431 loans vs. 2888 loans = 543 loans) has either endorsements from friends and / or bids from friends. Endorsements and bids from friends are supposed to help the borrower get a lower interest rate. Since the average interest rates of the lonely borrowers are slightly higher than the entire data set (Query 1), I should be able to determine that borrowers with endorsements or bids from friends wound up with a lower interest rate than the entire data set (Query 1). So this part makes sense to me, you dig?

Now look at the Net Default + Adjustment figures comparing Query 2 to Query 1:

AA – equal

A – basically equal

B – about 1% lower for the lonely

C – about 1% lower for the lonely

D – about 1% higher for the lonely

E - about 2% higher for the lonely

Hold on a second here. The lonely borrowers (Query 2) actually wind up with either a better or worse performance than the entire data set (Query 1) depending on the credit grade? There is nothing conclusive here at all. I was not expecting that!

Query 3: For the heck of it, I ran the performance again, but this time showing all Prosper Select Index loans with at least 1 endorsement and at least 1 bid. There were 232 loans in the set, so I could not generate the Estimated ROI table. Looking thru the performance, I could see that out of these 232 loans, only 3 were more that 1 month late (1 is an HR). 0 defaults. Hmmmm... Interesting – let’s keep digging…

Query 4: Let’s now take a look at the Prosper Select Index loans again, but now only those with at least one endorsement yet 0 bids from friends. I still can’t get that lovely Estimated ROI table to fire up, so let’s just look at the raw numbers again. 292 loans in the set, 3 defaults (B,C,D) and 9 1+ month late loans (1 is an HR).

Big difference!

Summary so far for those folks glancing over the numbers: From the first two queries, it seems that lonely borrowers will perform slightly better than borrowers with endorsements and/or bids from friends. But that just can't be, can it? Query 3 shows some promise that bids and endorsements have some sort of impact, but what remains to be seen is if an endorsement without a bid is actually an indicator of possible lower performance. The data is just too thin to really know yet.

Ok, bear with me here. If you’re still reading at this point, that means that you’re at least somewhat interested in this stuff. Luckily, we’re halfway through.

Let’s broaden the scope to look at more data. No more Prosper Select Index. Let’s go wide open with no criteria (9,544 loans):

Query 5

(Wow. Notice the difference between this query and Query 1 (Prosper Select Index)? Those criteria sure are important.)

Now, let’s look at the lonely borrowers from this set of data (8,022 loans):

Query 6

Not a whole lot different than all the data in Query 5, huh? That’s because the lonely make up 84% of all the data in Query 5, so it’s a little weighted in their favor.

By comparison, now lets look at all the data again, but limited to those borrowers with at least 1 bid by a friend (683 loans):

Query 7

In every credit grade, the average interest rate is lower, which makes sense. And look at this: All credit grades (except those pesky A loans) showed a marked reduction in expected loss. Despite the lower average interest rates, the resulting ROI’s are much higher (except those A loans).

One last query for fun. Let’s look at all the loans again, but only where an endorsement has been made but there were 0 bids from friends (839 loans):

Query 8

Comparing these results with the wide open Query 5 table, there doesn’t seem to be much value in an endorsement without a bid.

Summary

There doesn’t seem to be much value in an endorsement without a bid. However, a bid from a friend does seem to indicate that the loan might perform better than a loan from lonely borrower. If you’ve been around enough and read the forums, the general consensus has always been “an endorsement without a bid is meaningless”. Thankfully we have the data now to show this, or at least somewhat show it.

It is also difficult to tell if there is any impact of friend endorsements and bids on borrowers falling within Prosper's Select Index. Perhaps borrowers with decent credit details don't need endorsments or bids. But maybe they do. Who knows. So far the data looks promising, but there just isn't enough of it to tell or not.

For the most part, the data is still very thin in some areas and not very mature yet. It will be interesting to watch this for a while and see how this develops. Will borrowers learn of this correlation (kind of) and get their buddy, significant other, or mother to bid on their loan, which in turn might influence the data? Probably, but nonetheless it will be interesting to see how this develops.

Thankfully we now have the tools to find listings where a bid has been made by a friend and a way to look at the real data.

So how should a lender use this data. How can a lender tap into the Social Capital of others? I’ll cover it in Part 5.

(and yes, I promise that Part 5 is the last of the Social Capital crap and I’ll move on to other Prosper Days 08 stuff. As long as I can remember what I happened. It was a long time ago now, and I was pretty groggy most of the time. Oh well, you should have been there…)

Part 5 - Tapping into Social Capital