mothandrust's blog

I’ve been running various screens with Prosper’s performance utility using "social screens" to improve ROI and reduce lates. Do endorsements matter? Can a bid from a friend help a borrower get funded? Can social screens compete with lending strategies based on conventional screens to reduce lates and boost lender ROI?

Since we now have a year’s worth of "extended data" I looked at the loans originated between 2/12/07-2/12/08, with an observation date of 3/17/08. 

But before I show the results, let’s try a little quiz!

1. 9.66% of Prosper loans during this time are late or worse; to best improve this you should screen out:

(a) Non-Homeowners

(b) Loans with no bidding friends

(c) Autofund loans

(d) DTI > 20%

(e) Loans with 1 or more current DQ’s

2. You see a posting from a newbie D borrower who has 1-3 current DQ’s who can’t get funded asking for help. You make one suggestion that will improve the ROI for lenders (if he acts on it) which is to:

(a) Pay off the current DQ’s and then relist

(b) Get someone to endorse his listing (even if the endorser doesn’t bid)

3. B Borrowers that have 0 DQ’s have more percentage lates if they have:

(a) 1 or more bids from friends

(b) 0 Public Records + 0 DQ in last 7 years + 0 INQ + Exclude Autofund

4. If you were a lender on all loans that had two or more friends bidding a total of $300 or more you would have:

(a) Only 17 loans in your portfolio, 14 of them AA

(b) 48 loans, all current

(c) Only loans from the Malama Ohana group

(d) 4 defaults (out of 106 loans) that were GL pump-and-dumps

If you've been a lender on Prosper for over a year, you aced it, didn't you? 

Everyone knows the Autofund loans are awful, the newbie D should pay off those current DQ’s and come back, and endorsements are failed and flawed because the pump-and-dumpers bid big and leave lenders holding the bag.

But let's put our prejudices aside for a moment and look at the numbers instead of what we feel the numbers should be.  In fact, (b) is the correct answer to all the above questions.

And yes, the "bids from friends" includes bids from people that have no real-world connection with the borrowers, and that the "bids" may have been outbid. A $500 pump-and-dump appears the same in the performance data as it does if the friend is in the loan for $500.

Let's start with the baseline and tweak one variable at a time to see its effect:

(a) Social Screens Work. A bidless endorsement has no effect, but the act of having a friend bid reduces the "late or worse" percentage even more than screening for 0 DQ's, low DTI listings, Homeowners, etc.  The bid has to be more substantial than "just a snickers" ($50) to have this positive effect.

(b) But Huge Bids Not So Much.  The dollar amounts are the sums of all the bids by friends on the listing—they appear in the data even if the friend was later outbid.  At amounts of $300 and above the default rate actually creeps up slightly; this could be due to some of the big bids being pump-and-dumps or to "get the ball rolling".  The %Current looks better because it is cannibalizing from the payoffs. So large bids noticably reduce payoffs (the friend wants a good ROI) but they don’t reduce defaults.

(c) The More Friends The Better.  When we look at multiple friends bidding—the number of loans dropps off very fast! Most of these are Malama Ohana loans; one loan in that group has 20 bidding endorsers on it and from different relationship circles: family friends, eBay clients, high school buddies, etc.  Loans like this inspired me to "Save All My Lending Cash" for a borrower who could submit this type of listing.

(d) When we combine a quality listing (0 current DQ’s) with a friend bidding substantially, the default rate drops below 1%.

(e) There are very few Prosper loans where 3+ friends are all bidding large amounts; in fact, StaciM’s loan in Cubbies' group is the only of these 11 outside the Ohana.

Adopting a Social Factor Lending Strategy

The social factors that seem to be most relevant are multiple friends bidding, and/or friends who bid substantially ($300+). These factors are at least as important as autofund, low DTI, and even current DQ’s.  When we combine the fundamental and the social factors, the default rate improves dramatically.

But don’t program your Standing Orders just yet—some of these loans with just one bid could be pump-and-dump time bombs and the "friends" are not real-world relationships.  Just 2 defaults in a sample of 100 would skew this data; we really need a lot more loans with endorsements and friends bidding.

To protect yourself, demand that friends who endorse the loan stipulate at what rate they are bidding at—if they can write 3 lines of gushing accolades they can certainly include, "and I’m bidding $700 at 5%" so that lenders know whether they are committed to the loan or not. 

I will be personally reinvesting my Prosper payments into listings with 0 DQ's and where multiple bidding friends "state their rate".