Words of Win

Let's take a look at the 10 most likely words to have been used in a loan that repaid:

lender
risk
card
rate
fund
investment
rates
0
minimum
ratio

These words range from lender, at 8% more likely to have Paid than average, to ratio, at 4% more likely to pay than average.

Interesting here is the discrepancy between the words of loss -- maxing out at 23% less likely to pay than average -- and the words of win -- maxing out only 8% more likely to pay than average. It could be that the words a borrower chooses may only help in identifying lenders who are less likely than average to repay, but not show borrowers who are more likely to repay.

Update Aug 30: I have added a post on the methodology used to find these words.

Words of Loss

I took a look at all loans that were made on Prosper up through the end of 2007 to see what words people who failed to repay their loans used most often. Looking only at words that were used more than 1000 times, the bottom 10 list is quite interesting:

payday
chance
behind
son
daughter
mother
children
child
track
deleted

Requests where the word 'payday' was used were 22% less likely than average to repay their loans. Requests containing the word 'deleted' were 13% less likely than average to repay their loans.

Most interesting, in my mind, is all of the words that invoke family. Son, daughter, mother, children, child -- half of the bottom 10 words are family members. Husband, at 10% less likely than average to pay, is 24th from the bottom.

In the next post I'll take a look at the words which were associated with a positive loan outcome.

Update Aug 30: I have added a post on the methodology used to find these words.

Need Series: Correlation Matrix

A correlation matrix tells us how much the change in various factors relate to one another. We're looking here to see if using the word "need" is strongly associated with some other variable, such as credit grade, Debt-To-Income Ration (DTI), or any other factor. If there is a strong correlation with some other factor then we know that we're not on to anything new here and can move along.

Generally it is my understanding that a correlation (or negative correlation) over .5 is strong and over .1 is weak. A correlation of 0 would mean that there is no relationship between the values at all.

Data set: Loans started between 2005-2008 which were not Cancelled
Credit Grade was scored as: AA=10, A=9, ..., HR = 4
Loan Outcome was scored as: Paid = 3, PaidInFull and RecoveredInFull = 2, Everything Else = 1
Title has need was scored as: Yes = 1, No = 0
Description has need was scored as: Yes = 1, No = 0
DTI and Amount Requested were as published by Prosper

	Credit Grade	Description Has Need	Debt To Income	Amount Requested	Title has Need	Loan Outcome
Credit Grade	1	-.17	.03	.41	-.11	.30
Description Has Need	-.17	1	.01	-.04	.16	-.11
Debt To Income	.03	.01	1	.09	0	0.04
Amount Requested	.41	-.04	.09	1	-.05	-.06
Title Has Need	-.11	.16	0	-.05	1	-.06
Loan Outcome	.30	-.11	.04	-.06	-.06	1

As we would expect, we see some stronger correlations, like the .3 between credit grade and loan outcome. We also see a decent correlation between the description and the title having the word "need" in them.

It is interesting to see a -.11 correlation between loan outcome and the word "need" in the description of the loan -- this is stronger than the -.06 correlation when we look at the title. In a future post we'll do a t-test to see if these results are statistically significant.

---

All Articles in the Needs Series
An Introduction
Initial Findings
Correlation Matrix
Comparing to Lending Club
What We Fund

Need Series: Initial Findings

Let's look at my initial findings:

2005-2006 Loans	Total Loans	Never Recovered	Recovered	Paid
Title contains "need"	652	46.9%	52.8%	51.2%
Title does not contain "need"	5321	37.0%	62.9%	61.4%

2007 Loans	Total Loans	Never Recovered	Recovered	Paid
Title contains "need"	1092	47.7%	52.3%	50.6%
Title does not contain "need"	10387	37.3%	62.7%	61.5%

2008 Loans	Total Loans	Never Recovered	Recovered	Paid
Title contains "need"	861	34.8%	48.9%	48.6%
Title does not contain "need"	10704	30.3%	55.0%	54.3%

It's worth pointing out that not all 2008 loans will have completed yet, so we expect the percent Paid to increase--for both "need" and not "need" loans--as the year goes on. Still, for loans which have reached their conclusion (2005-2007 loans) we see roughly a 10% difference in number Paid between the groups.


Now 312lender and frinxor pointed out on the prospers.org forum that use of the word "need" could be directly correlated with credit score and, hence, non-payment rate. Let's take a look at those numbers for all loans which originated between 2005-2007:

Credit Score	Paid With "Need"	Paid Without "Need"	Difference
AA	79.6%	87.0%	7.4%
A	76.3%	76.2%	-0.1%
B	60.5%	69.3%	8.8%
C	57.1%	62.3%	5.2%
D	53.0%	60.0%	7.0%
E	45.4%	49.3%	3.9%
HR	32.3%	37.0%	4.7%

So it would seem that almost every credit grade has a difference between the groups. Now it's interesting to note that only 5% of A loans used the word "need" in the title and nearly 15% of HR loans used the word "need" in the title. This may, in fact, be relevant when we do more in-depth statistical analysis later on.

Initially, however, it still looks like we're on to something.

---

All Articles in the Needs Series
An Introduction
Initial Findings
Correlation Matrix
Comparing to Lending Club
What We Fund