Tuesday, December 18, 2007

New And Shiny Collections Statistics

I had proposed a new way to look at collections progress back in early December. This method took to snapshots of Prosper's nightly data export, one month apart, and looked at how the lone status changed over time. I took my first snapshot on December 1, but by some fluke of luck, it turns out that I had the Nov 1 download laying around a second computer, so now we get an early look. The underlying assumption is that if a loan's state does not get worse (as in, going from 1 month late to two months late), some payment must've occurred. So, into the data.


TotalGot BetterThe SameGot Worse
Payoff in progress 28 26
92.8%
0
0.0%
2
7.1%
Current 11697 154
1.3%
11238
96.0%
305
2.6%
Late 252 66
26.1%
40
15.8%
146
57.9%
1 month late 326 42
12.8%
24
7.3%
260
79.7%
2 months late 263 13
4.9%
9
3.4%
241
91.6%
3 months late 231 5
2.1%
9
3.8%
217
93.9%
4+ months late 618 6
0.9%
607
98.2%
5
0.8%
The interesting figures are that between Nov 1 and Dec 1, it looks like 20.3% of "1 Month Late" loans, 8.4% of "2 Month Late" loans, and 6.1% of "3 Month Late" loans had some kind of payment activity that kept things from falling further behind. The blood from a turnip category ("4+ Months Late") had 1% improve, though this is hardly definitive because payments that do not bring the borrower to 3 months late or better are not visible. I am hesitant to infer things from the "Late" category because it is not on a clean month to month boundary.

These statistics imply better collection performance than the official Penncro numbers that Prosper publishes, but a one month snapshot is insufficient to rule out a statistical anomaly.

Some other notes from the statistics include the 5 "4+ Late Loans" got worse by folks entering bankruptcy and 1 "1 Month Late" loan that was repurchased.

No comments: