Whole loans

[brycemason]

My analysis was extremely simple (for the mass audience on Peter's blog), but it did look at a number of borrower traits. If there are other traits you think are key, I can add them in. Given your work above, I may also play with the defaults a bit. But, I think from a disclosure standpoint, all that we need worry about is whether the characteristics that are known at the time of loan listing are balanced across whole vs. fractional.

[rawraw]

My analysis was extremely simple (for the mass audience on Peter's blog), but it did look at a number of borrower traits. If there are other traits you think are key, I can add them in. Given your work above, I may also play with the defaults a bit. But, I think from a disclosure standpoint, all that we need worry about is whether the characteristics that are known at the time of loan listing are balanced across whole vs. fractional.

Yea.  Unless LC has data we don't have access to, which then makes it interesting to use defaults.  Right?

And looking forward to the blog Bryce

[brycemason]

That's right, Rawraw. We know they keep some facets private. However, I think it would be exceedingly hard to have so many factors apparently equal, and then partition along the remainder and achieve a big default shift. It's worth a critical look, but we're only a year into the default curve.

[AmCap]

I don't think I agree.  As a securities matter, the only thing LC has told us about the risk of a note is their assigned letter grade.  If it is true that, within notes that LC has told us should have equal risk characteristics,  the notes offered only to large investors are less risky, I have to think that's a disclosure item.  I agree that it may be too early to make any broad conclusions, though.

[Peter]

And here is Bryce's analysis:
http://www.lendacademy.com/lending-club-whole-loan-program-one-year-later/

[AmCap]

And here is Bryce's analysis:
http://www.lendacademy.com/lending-club-whole-loan-program-one-year-later/

I appreciate this!  But Bryce, do you agree that there *is* a material difference in the default rates as a historical matter?  If so, and accepting the point you made about borrower attributes being the same, do you have any theories as to why the difference exists?

[brycemason]

With the understanding that these loans are still quite early in repayment, my own table construction from the same data extract that I used for the article yields the following. I do not see any cause for concern here, so I guess I disagree with however the data to which you refer was prepared. I aggregated the codes some for simplicity. It's just a loan count, so not weighted by loan amount or amount in the various categories, but I just don't see it mattering.


. tab status whole if policy_code==1, col

+-------------------+
| Key               |
|-------------------|
|     frequency     |
| column percentage |
+-------------------+

            |         whole
     status |         0          1 |     Total
------------+----------------------+----------
Charged Off |       153         52 |       205
            |      0.24       0.22 |      0.23
------------+----------------------+----------
    Current |    62,359     22,323 |    84,682
            |     95.88      96.13 |     95.95
------------+----------------------+----------
   Impaired |       732        199 |       931
            |      1.13       0.86 |      1.05
------------+----------------------+----------
       Paid |     1,793        647 |     2,440
            |      2.76       2.79 |      2.76
------------+----------------------+----------
      Total |    65,037     23,221 |    88,258
            |    100.00     100.00 |    100.00

[AmCap]

Bryce - what happens to your analysis when we look at lower grade loans issued b/t October 2012 and, say, May 2013?  I'm wondering if that might ferret out any significant gap in the default rates...

[brycemason]

I doubt the time restriction will matter, because the entire whole loan program is basically October 12 to July 13 (in that data), but for the hell of it, here it is. D-G. Still no discrepancy on the surface. At this point, it might be easier to provide the code on which the other analysis is based and I'll try to find the error. The assignment is random; there aren't going to be any differences with 90,000 loans.


. tab status whole if policy_code==1 & inrange(list_month,10,17) & inlist(grade,"D","E","F","G"), col

+-------------------+
| Key               |
|-------------------|
|     frequency     |
| column percentage |
+-------------------+

            |         whole
     status |         0          1 |     Total
------------+----------------------+----------
Charged Off |        61         17 |        78
            |      0.53       0.43 |      0.50
------------+----------------------+----------
    Current |    10,790      3,708 |    14,498
            |     93.15      93.26 |     93.18
------------+----------------------+----------
   Impaired |       289         91 |       380
            |      2.50       2.29 |      2.44
------------+----------------------+----------
       Paid |       443        160 |       603
            |      3.82       4.02 |      3.88
------------+----------------------+----------
      Total |    11,583      3,976 |    15,559
            |    100.00     100.00 |    100.00