Wednesday, April 9, 2008

Prosper Listing Word Analysis

Word analysis has gone in and out of vogue among Prosper lenders. Right now, it's in vogue here at Prosperous Land. The premise is that you can get hints about the quality of a listing based on words that appear in the listing description. I've been messing with it to try and get a statistically significant edge (proving it provides an edge, now there's a dicey proposition), but some interesting things pop out when you mess with the datasets enough.

I have used the 2602 repaid loans and 2387 defaulted, bankrupt, or 4+ month late loans for my number crunching, as of the April 1 dataset. For each word, I look at the ratio of appearances in good loans versus the appearances in bad loans to find those words that are good indicators of loan quality. The top ten good and bad ones that have appeared in at least 100 listings are below:

Word
Appeared In Good Loan
Appeared In Bad Loan
reinvest
95
10
aa
120
14
investments
104
16
lend
114
26
investing
82
27
lending
142
50
borrower
100
37
invest
124
48
side
120
50
investment
227
99
prove
44
144
catch
34
113
file
24
82
mother
78
270
establish
25
89
behind
67
241
lost
33
131
explanation
33
132
priority
27
119
payday
43
276

No comments: