Sunday, December 2, 2007 - late loan statistics

Prosper no longer allows me to publish this material in the discussion forum, so I will now publish it here on .

Here's the 12/01/2007 update of the late loan chart.

late loan chart - small 2007-12-01

A larger version of this graph is available: here

The following chart presents the same data, except that the curves have been slid to a common starting point. The horizontal axis now shows days since loan origination month. The first three months Mar'06, Apr'06, May'06 have been left off of this chart, because the curves for those months (Prosper's early days) have a different shape. With those curves left out you can easily see how similar the remaining curves are.

prosperlate 2007-12-01 slid

Look about 1 year out. You see about 20% of the loans have gone bad at that point. Individual months vary, but are quite similar.

Notes on methodology:
Each curve contains data on all the loans that originated in a particular month, as identified in the legend. All data comes from Prosper's performance web page. Data points are computed from data observed the 1st and 15th of each month.

Each data point is a ratio. The numerator is the number of loans that have 1 month late or worse. The denominator is the number of loans originated. Thus this ratio is the fraction of loans that have "gone bad".

More specifically, the numberator is the sum of loans in the following states: defaulted + "1-3 months late" + "4+ months late" + repurchased. The first three numbers are read directly from Prosper's performance web page. Prosper does not explicitly display the number of loans repurchased. I calculate the # loans repurchased by tracking changes in the number prosper shows as "loans originated".

The denominator is the number of loans originally shown by prosper as having originated in the designated month. When prosper repurchases a loan found to have been fraudulent, they "unoriginate" the loan, in other words they subsequently show that a lower number of loans have originated. I use the original (higher) number in the denominator of my ratio. For example, in May of 2006, 458 loans originated, although they now show the number of originations as 446, after having repurchased 12 fraudulent loans.

Excluding the early few months of Prosper, these curves look very similar to one another. Separating loans by age allows us to see the time evolution of delinquency. Delinquency grows over time. We can also see the rate at which it grows. For the first year each group of loans shown here has gone bad at more than 20% per year. Because almost all Prosper loans that go more than 1 month late go on to default, we can also see that the default rate on the all-prosper portfolio is approximately 20%/year.

Given that this is so clear from Prosper's own data, it is sad to see the quality of Prosper loans misrepresented in the press over and over. The most recent example is a November 27, 2007 Associated Press article wherein Chris Larsen is quoted as saying:

Prosper’s default rate hovers at about 2.7 percent, Larsen said, but that figure is expected to rise as more loans mature.

Its not difficult to figure out how he comes up with 2.7%. If you take the total $ worth of loans that prosper has declared defaulted so far, and divide it by the total $ worth of loans that Prosper has originated so far, you get approximately 2.7%.

That ratio, however, is not meaningful to lenders.

One problem with the 2.7% number is found in the choice of numerator. The numerator in this fraction includes only loans that have been declared "default" by Prosper. Prosper ages highly delinquent loans quite some time before declaring them "default". Loans are subject to default at 4 months late, but Prosper holds such loans until a quarterly disposal, and only declares them default once they have passed the quarterly event. Loans >4 months late would be in the default state except for the fact that they are waiting for the quarterly auction. Today Prosper shows $2,864,899 of loans in the "default" state, and $3,694,261 in the "4+ months late" state. More loans are in this suspended waiting-for-the-auction state than have so far completed the process and been declared "default"! If the quarterly auction had been held yesterday, the calculation of Larsen's ratio would have come out 6.4% instead of 2.7%.

The denominator is also a bad choice. Itcontains many loans so young that they could not possibly have defaulted yet. (Loans must be 5 months old before they can default.) The denominator also contains many loans old enough to default, but still low on the curve. This methodology is simply defective.

Only when lenders understand the rate at which Prosper loans go bad will they have a chance to understand how selective they must be to succeed. Prosper loans have been going bad on average at about 20% per year.

There was one small surprise in this update. The Aug'07 curve took a jump. This occured because the # of originated loans in Aug'07 dropped. Was 979 originally, and was still 979 at my 11/15/07 reading. As of 12/01/07 Prosper shows only 940 loans originated. This (probably) means that some time during the last two weeks Prosper repurchased 39 loans from Aug'07. Must have discovered a big batch of fraudulent loans. If this theory is correct, it would be the largest batch of fraudulent loans ever detected. Haven't heard anything about this event, but then it seems like it may have occurred at about the same time the Prosper forums were shut down, and there has been less sharing of information in the Prosper community since that time. We should soon hear confirming information from lenders who were part of this repurchase, or else learn that it was a data glitch.

No comments:

Post a Comment