Author Topic: Value of previously late notes  (Read 18931 times)

P2PFact

  • Jr. Member
  • **
  • Posts: 91
    • View Profile
Re: Value of previously late notes
« Reply #30 on: April 29, 2014, 11:06:57 PM »
Draw your own conclusion if anyone thinks Folio listing discount average is a good buying indicator.
I think Fred's point is not just using listing price (or mark down which leads to price). If historical data can accurately predict the default, then we can use that to calculate intrinsic price of the note and then compare with listing price. If listing price is way cheaper, likely the bet ( aka buying the note ) would end up in the money.

Even 99% of the loans in your study go bad eventually, Fred's model may choose the 1% and end up making money.


core

  • Hero Member
  • *****
  • Posts: 1784
  • Your loss is my gain
    • View Profile
Re: Value of previously late notes
« Reply #31 on: April 29, 2014, 11:56:41 PM »
If historical data can accurately predict the default, then we can use that to calculate intrinsic price of the note and then compare with listing price.

Sure, but so far all I've heard is talk about default rates or loan-go-bad rates or whatever.  Knowing the odds of a default isn't good enough.  As a BJ player I don't want to know what the odds are that I'm going to win every hand in a shoe, or even the odds of winning any given hand.  I want the EV.  The expected cash value, not the %chance.

You show me a note that has a 100% chance of default and I might still pay you a decent price for it, assuming that the odds favor me getting my investment back plus profit through normal payments before that default happens.  Default rate is meaningless in this case.

Fred

  • Hero Member
  • *****
  • Posts: 1421
    • View Profile
Re: Value of previously late notes
« Reply #32 on: April 30, 2014, 12:03:56 AM »
I want the EV.  The expected cash value, not the %chance.

Expected Value = (1 - defaultRates) * (outstandingPrincipal + accruedInterest)?

Just make sure you define (and calculate) the defaultRates correctly.

core

  • Hero Member
  • *****
  • Posts: 1784
  • Your loss is my gain
    • View Profile
Re: Value of previously late notes
« Reply #33 on: April 30, 2014, 12:49:17 AM »
Just make sure you define (and calculate) the defaultRates correctly.

That statement can hide a multitude of sins.  Yes, I suppose IF you calculate defaultRate "correctly", your calculation would sorta work.  I wonder how many people here calculate default rates in such a manner.  They definitely should.  How many here do?  *crickets*

However your offered calculation assumes an equally distributed chance of complete default throughout the life of the loan (with respect to time, not dollars), and it also assumes the notes will pay up to the EV and stop at about the same time which of course is not correct.  One could argue it does not matter, but I say it does:  Two notes may have the same exact EV but one note is certain to reach that EV in 2 months, where the other one might take 4+ years before you receive the calculated EV before the default.  That's a big difference in return.

Fred

  • Hero Member
  • *****
  • Posts: 1421
    • View Profile
Re: Value of previously late notes
« Reply #34 on: April 30, 2014, 01:32:37 PM »
Just make sure you define (and calculate) the defaultRates correctly.

That statement can hide a multitude of sins.  Yes, I suppose IF you calculate defaultRate "correctly", your calculation would sorta work.  I wonder how many people here calculate default rates in such a manner.  They definitely should.  How many here do?  *crickets*

My remarks were due to numerous definitions of "default rates" in the prior threads.  Some used the term "probabilities", "odds", "rates".  Some defined it as transition from some status to a worse status; some defined it as transition to a Default status, some defined it as transition to Charged-Off status.  Some defined it as starting from Issued status; some from Current status; some from what-status-the-loan-is-now.

No meaningful conversations can be had when the same term refers to different semantics.
« Last Edit: April 30, 2014, 01:36:38 PM by Fred »

rawraw

  • Hero Member
  • *****
  • Posts: 2776
    • View Profile
Re: Value of previously late notes
« Reply #35 on: April 30, 2014, 04:26:35 PM »
Just make sure you define (and calculate) the defaultRates correctly.

That statement can hide a multitude of sins.  Yes, I suppose IF you calculate defaultRate "correctly", your calculation would sorta work.  I wonder how many people here calculate default rates in such a manner.  They definitely should.  How many here do?  *crickets*

My remarks were due to numerous definitions of "default rates" in the prior threads.  Some used the term "probabilities", "odds", "rates".  Some defined it as transition from some status to a worse status; some defined it as transition to a Default status, some defined it as transition to Charged-Off status.  Some defined it as starting from Issued status; some from Current status; some from what-status-the-loan-is-now.

No meaningful conversations can be had when the same term refers to different semantics.
I know a lot of what you use in your model you keep close to your chest.  But I'd be curious how you define default, if that isn't part of the secret sauce.  We can call it the Fred Default so no one conflates the terms!

Fred

  • Hero Member
  • *****
  • Posts: 1421
    • View Profile
Re: Value of previously late notes
« Reply #36 on: April 30, 2014, 10:28:17 PM »
I know a lot of what you use in your model you keep close to your chest.  But I'd be curious how you define default, if that isn't part of the secret sauce.  We can call it the Fred Default so no one conflates the terms!

I use the definition that LC has at the bottom of this page https://www.lendingclub.com/info/demand-and-credit-profile.action, except I relaxed the following restrictions:
- "over 9 months"
- "loans for the months of August 2012, September 2012, and October 2012"

The default rates are essentially the blue portions of the pie charts.
« Last Edit: April 30, 2014, 11:36:10 PM by Fred »

rawraw

  • Hero Member
  • *****
  • Posts: 2776
    • View Profile
Re: Value of previously late notes
« Reply #37 on: May 01, 2014, 02:24:52 PM »
I know a lot of what you use in your model you keep close to your chest.  But I'd be curious how you define default, if that isn't part of the secret sauce.  We can call it the Fred Default so no one conflates the terms!

I use the definition that LC has at the bottom of this page https://www.lendingclub.com/info/demand-and-credit-profile.action, except I relaxed the following restrictions:
- "over 9 months"
- "loans for the months of August 2012, September 2012, and October 2012"

The default rates are essentially the blue portions of the pie charts.
So you use those as the POD?  I imagine you are assuming 100% LGD, since recoveries probably aren't something to count on.  Do the POD's you use get calculated per sub grade, grade, or some other segmentation?   Do you have any EAD assumptions?

Fred

  • Hero Member
  • *****
  • Posts: 1421
    • View Profile
Re: Value of previously late notes
« Reply #38 on: May 01, 2014, 05:09:50 PM »
I knew you'd ask these follow-up questions. :)

If I can explain in terms of Expected Loss (http://en.wikipedia.org/wiki/Expected_loss) = PD * LGD * EAD, the blue portions of LC pie charts are (PD * LGD).  I define the EAD simply as the dirty price (Oustanding Principal + Accrued Interest) of the notes.

Please note that the LC definition is the probability to reach the Charged-Off status; while the usual PD measures only until the Default status (i.e., one node prior to Charged-Off).

rawraw

  • Hero Member
  • *****
  • Posts: 2776
    • View Profile
Re: Value of previously late notes
« Reply #39 on: May 01, 2014, 09:31:07 PM »
I knew you'd ask these follow-up questions. :)

If I can explain in terms of Expected Loss (http://en.wikipedia.org/wiki/Expected_loss) = PD * LGD * EAD, the blue portions of LC pie charts are (PD * LGD).  I define the EAD simply as the dirty price (Oustanding Principal + Accrued Interest) of the notes.

Please note that the LC definition is the probability to reach the Charged-Off status; while the usual PD measures only until the Default status (i.e., one node prior to Charged-Off).
I must be too predictable!  I guess the real trick comes not from getting price on Folio, but if the behavior changes for the new notes.  Perhaps variables we don't know (like marketing channel used) impact those characteristics.  But I'm sure you have some quant way of looking at that, too lol

P2PFact

  • Jr. Member
  • **
  • Posts: 91
    • View Profile
Re: Value of previously late notes
« Reply #40 on: May 01, 2014, 11:07:51 PM »

I use the definition that LC has at the bottom of this page https://www.lendingclub.com/info/demand-and-credit-profile.action, except I relaxed the following restrictions:
- "over 9 months"
- "loans for the months of August 2012, September 2012, and October 2012"

The default rates are essentially the blue portions of the pie charts.
According to the footnote, the percentage in the pie chart is in % of principal. I think in your model you should be (quite likely already) using % of loans to build transition matrix, right? Since you are investing loan by loan basis.

Thank you Fred for your insight. Learned a lot in this thread!

« Last Edit: May 01, 2014, 11:15:56 PM by P2PFact »

Ran

  • Full Member
  • ***
  • Posts: 145
    • View Profile
    • Email
Re: Value of previously late notes
« Reply #41 on: June 25, 2014, 10:07:59 AM »
I took a snapshot of Foliofn listings two months ago and what I found out is that the loan default rates are MUCH higher than discount.

Care to elaborate more, specifically on how you measure default rates on FOLIO-purchased notes?  How much different are they compared to the LC-published default rates?

I will post a prilimilary statistics this weekend to show how much off listed (not Traded) Folio note discount is vs late/default rate

I collect Folio loans between 02/22 and 03/02 (I did update per day) that were LISTED at <=-5% discounts, e.g., the loans that most probably would not sell at listed discount.
I chose -5% as cut-off because my main interest was in impaired loans.
There were 3600 loans total, and 527/1067/447/1559 were Current/IGP/Late 16-30/Late 31+ separately.
As of 04/24:
527 Current Loans with average discount -7.3%
5/429/93 were Paid-off/Current/GP or worse, so a loan-go-bad rate of 17.7%

1067 IGP Loans with average discount -13.7%
647/135/285 were Current/GP/Late or worse, so a loan-go-bad rate of 26.7%

447 Late 16-30 Loans with average discount -23.12%
165/32/250 were Current or GP/Late 16-30/Late 31+ or worse, so a loan-go-bad rate of 55.9%, a loan-healing-rate of 36.9%

1559 Late 31-120 Loans with average discount -47.33%
152/671/736 were current or GP/Late 31-120/Default or Charged off, so a loan-go-bad rate of 47.2%, a loan-healing-rate of 9.7%

Draw your own conclusion if anyone thinks Folio listing discount average is a good buying indicator.

I guess it worth to post a follow up on the statistics.
The Folio loans discussed in this post were snapshot of Folio list between 02/22 and 03/02 at <=-5% discounts. There were 3600 loans total, and 527/1067/447/1559 were Current/IGP/Late 16-30/Late 31+ separately. As of 6/15, here is how the loans status has changed in 3 months:
Current loans: 527 Total, Average discount -7.3%
14 (2.6%) is full paid, 58(11%) is IGP, 5(1%) is Late 16-30, 65(12%) is Late 31+, 4(1%) is defaulted
13% Current loans are Late 31+ in 3 months

In Grace Period loans: 1067 Total, Average discount -13.7%
50 (5%) is full paid, 621(60%) is Current, 28(3%) is Late 16-30, 219(20%) is Late 31+, 111(10%) is defaulted
30% IGP loans are Late 31+ in 3 months

Late 16-30 loans: 447 Total, Average discount -23.1%
18 (4%) is full paid, 143(32%) is Current, 26(6%) is IGP, 60(13%) is Late 31+, 197(44%) is defaulted
57% Late 16-30 Loans are Late 31+ in 3 months

Late 30-120 loans: 1559 Total, Average discount -47.3%
19 (1%) is full paid, 109(7%) is Current, 16(1%) is IGP, 20(1%) is Late 16-30, 1195(76%) is defaulted
76% Late 31-120 loans are defaulted in 3 months

I sort of anticipate the sloppiness of Folio loans that stay in list, but did not expect the quality is so bad (even Current loans) compared to their average discount.
Folio Sellers are really enjoying a festival throughout the years :-[ :-[ :-[ :-[ :-[!!! 

« Last Edit: June 25, 2014, 10:13:26 AM by Ran »

rawraw

  • Hero Member
  • *****
  • Posts: 2776
    • View Profile
Re: Value of previously late notes
« Reply #42 on: June 25, 2014, 12:10:00 PM »
So the sweet spot is buying grace period notes?

Ran

  • Full Member
  • ***
  • Posts: 145
    • View Profile
    • Email
Re: Value of previously late notes
« Reply #43 on: June 25, 2014, 12:20:38 PM »
So the sweet spot is buying grace period notes?

I am not sure 30% Late 31-120 & Default in 3 months can justify ~15% discount, but may be you can try it out :) Do not forget even the worst grade (F/G) in primary market only has 7~10% default rate in a whole year.

rawraw

  • Hero Member
  • *****
  • Posts: 2776
    • View Profile
Re: Value of previously late notes
« Reply #44 on: June 25, 2014, 09:28:32 PM »
So the sweet spot is buying grace period notes?

I am not sure 30% Late 31-120 & Default in 3 months can justify ~15% discount, but may be you can try it out :) Do not forget even the worst grade (F/G) in primary market only has 7~10% default rate in a whole year.
Well given the average, I assumed there is some variation around the mean.  Seemed like the group that this dispersion would be able to lead to profitable strategies (without actually knowing the dispersion). Of course you wouldn't buy the mean ones with those PDs