Author Topic: Blog Post: Automated Buying of Notes on Lending Club Folio Secondary Market  (Read 18325 times)

AnilG

  • Hero Member
  • *****
  • Posts: 1123
    • View Profile
    • PeerCube
Automated Buying of Notes on Lending Club Folio Secondary Market
https://www.peercube.com/blog/post/automated-buying-of-notes-on-lending-club-folio-secondary-market

Why should you buy notes from secondary market?

- Returns from Day One
- No Waiting for Loans to be Issued
- Minimize Early Prepayments
- Avoid Straight-Rollers
- Purchase Seasoned Notes
- Buy Notes of Shorter Term
---
Anil Gupta
PeerCube Thoughts blog https://www.peercube.com/blog
PeerCube https://www.peercube.com

Rob L

  • Hero Member
  • *****
  • Posts: 2133
    • View Profile
Not to mention that the number of fractional loans presently being released by LC four times a day is quite small.

AnilG

  • Hero Member
  • *****
  • Posts: 1123
    • View Profile
    • PeerCube
I was just trying to list reasons that apply across different situations. You are correct, right now the lack of new loans on primary platform, greater possibility of loans not getting sufficient funding to be issued, and borrower not wiling to wait long enough are good reasons to consider purchasing notes on secondary market.

Not to mention that the number of fractional loans presently being released by LC four times a day is quite small.
---
Anil Gupta
PeerCube Thoughts blog https://www.peercube.com/blog
PeerCube https://www.peercube.com

fliphusker

  • Sr. Member
  • ****
  • Posts: 463
    • View Profile
    • Email
I think what you are doing is a great service Anil.  Anyone who has done anytime digging through FOLIO knows how bad the site is to deal with. 
The question I have though, is why not set filtering with more options?  Will that come in the future?  I have pretty stringent filters that I use, and do not deviate from them.  DTI, income and public records, just to name a few. 

AnilG

  • Hero Member
  • *****
  • Posts: 1123
    • View Profile
    • PeerCube
Staleness and unavailability of credit data such DTI, income and public records is the main reason such data doesn't work well with automation. Despite my best attempts to reconstruct credit data for recently issued loans from other sources, I still miss out a large fraction. Using credit attributes will restrict the filters to picking loans that are at least one quarter+ old. Majority of Folio activity is with loans issued within last 0-6 months. Also, once loans have been issued and paid for a while, application level credit data becomes much less relevant. The loan status change, payment history and FICO change becomes more important.

BTW, I do provide manual filtering using credit data but not automated filtering (too many things can go wrong).
https://www.peercube.com/filter/filtercreate

I think what you are doing is a great service Anil.  Anyone who has done anytime digging through FOLIO knows how bad the site is to deal with. 
The question I have though, is why not set filtering with more options?  Will that come in the future?  I have pretty stringent filters that I use, and do not deviate from them.  DTI, income and public records, just to name a few.
---
Anil Gupta
PeerCube Thoughts blog https://www.peercube.com/blog
PeerCube https://www.peercube.com

fliphusker

  • Sr. Member
  • ****
  • Posts: 463
    • View Profile
    • Email
Staleness and unavailability of credit data such DTI, income and public records is the main reason such data doesn't work well with automation. Despite my best attempts to reconstruct credit data for recently issued loans from other sources, I still miss out a large fraction. Using credit attributes will restrict the filters to picking loans that are at least one quarter+ old. Majority of Folio activity is with loans issued within last 0-6 months. Also, once loans have been issued and paid for a while, application level credit data becomes much less relevant. The loan status change, payment history and FICO change becomes more important.

BTW, I do provide manual filtering using credit data but not automated filtering (too many things can go wrong).
https://www.peercube.com/filter/filtercreate

I think what you are doing is a great service Anil.  Anyone who has done anytime digging through FOLIO knows how bad the site is to deal with. 
The question I have though, is why not set filtering with more options?  Will that come in the future?  I have pretty stringent filters that I use, and do not deviate from them.  DTI, income and public records, just to name a few.
Can you clarify your statement?  "Staleness and unavailability of credit data such DTI, income and public records is the main reason such data doesn't work well with automation."  Are you saying that they are irrelevant, not enough data to support the assumption or that once you reach certain points in a note's payment status (let's say 12 months) these factors no longer apply?  Or am I misreading your comment?
I disagree though that the longer a note is paid upon the less relevant the the borrower's history is.  Leopards do not change their stripes, often.  CC and consolidation notes are a prime example.  Higher DTI people are more likely to default then lower DTI, in backtesting.  I really believe that these people are not going to change their spending habits overnight.  Public records is a huge red flag for me.  They have had themselves buried in a bad hole once before, not saying it was their fault, and have either had a lien or BK.
One giant suggestion I would give though is on the "fico change", if you can make it more recent it would be of greater use.  One of the first notes I bought on FOLIO, technically was an UP FICO, so I did not look at the score fluctuation, sloppy on my part, I know.  This note just went into IGP, so I looked at it.  It is still above the initial FICO, but the FICO had actually dropped 60 points prior to me buying the note.  That is what would scare me from having notes auto picked. 

AnilG

  • Hero Member
  • *****
  • Posts: 1123
    • View Profile
    • PeerCube
The historical loan data file is not updated and released until a few weeks after a Quarter has completed. This file includes the information about loans that were issued in the last Quarter. Without this file you don't have access to loan and credit attributes for a large number of new loans. As this file will be used to filter loans for credit and loan attributes, you will miss out on the loans issued in last Quarter. This is what I mean by "unavailability" of credit data for a large segment of new loans. Similarly, by the time historical loan data file is released, some of the time-sensitive data such as loan status, recent FICO score etc. are already stale (old) and might have changed. Any filtering based on stale data will result in erroneous automated purchase. By using manual filtering, you are able to review individual loan and note data for any such discrepancies but you don't have that luxury with automated filtering without adverse impacts.

The loan and credit data provided at the time of loan application loses its "relative" value with time. Older the loan, less likely this application data will help you make better note selection. Borrower payment history, change FICO and loan status become much more valuable once loan has been issued. The payment history, FICO change and loan status will help you more in identifying leopards who have changed their "stripes" than the information on stripes 12 months ago. If we had access to updated credit data, nothing will be better than that but we don't have access to updated credit data. We have to make the best decisions based on what information is available and updated unless we want to take our ball and go home because we don't like the playing field.

BTW, have you read my blog post  "FICO Score Trends and Defaults for Lending Club Loans" https://www.peercube.com/blog/post/fico-score-trends-and-defaults-for-lending-club-loans? 99+% of loans with FICO trend of FLAT and UP from application FICO fully pay the loans. The odds are in your favor that most borrowers with Up FICO trend will repay their loans.

Can you clarify your statement?  "Staleness and unavailability of credit data such DTI, income and public records is the main reason such data doesn't work well with automation."  Are you saying that they are irrelevant, not enough data to support the assumption or that once you reach certain points in a note's payment status (let's say 12 months) these factors no longer apply?  Or am I misreading your comment?
I disagree though that the longer a note is paid upon the less relevant the the borrower's history is.  Leopards do not change their stripes, often.  CC and consolidation notes are a prime example.  Higher DTI people are more likely to default then lower DTI, in backtesting.  I really believe that these people are not going to change their spending habits overnight.  Public records is a huge red flag for me.  They have had themselves buried in a bad hole once before, not saying it was their fault, and have either had a lien or BK.
One giant suggestion I would give though is on the "fico change", if you can make it more recent it would be of greater use.  One of the first notes I bought on FOLIO, technically was an UP FICO, so I did not look at the score fluctuation, sloppy on my part, I know.  This note just went into IGP, so I looked at it.  It is still above the initial FICO, but the FICO had actually dropped 60 points prior to me buying the note.  That is what would scare me from having notes auto picked.
---
Anil Gupta
PeerCube Thoughts blog https://www.peercube.com/blog
PeerCube https://www.peercube.com

Rob L

  • Hero Member
  • *****
  • Posts: 2133
    • View Profile
One of the first notes I bought on FOLIO, technically was an UP FICO, so I did not look at the score fluctuation, sloppy on my part, I know.  This note just went into IGP, so I looked at it.  It is still above the initial FICO, but the FICO had actually dropped 60 points prior to me buying the note.  That is what would scare me from having notes auto picked.

First let me agree with Anil regarding the "at loan issue time" data relevance. My perspective is skewed a bit since I'm selling notes that I bought in the first place (ie they met my purchase criteria). So, I know they were "good" from the start per my definition of good (your mileage will vary). That's not the case when buying Folio notes in general. One doesn't know what the owner's criteria for buying was. But clearly as time passes the loan performance to date must become much more significant than the initial conditions. The timeframe for the overtaking of the relevance of recent performance vs at loan issue data to my knowledge is unknown and probably worthy of some serious study. How many months ?? Very good question. My GUESS is that it's not that many (<12 or < 6); maybe even less. Early defaults are killers. FWIW the portfolio liquidation software I've been working on over a month now has no filters for "at loan issue time" data. My how time flies.

The data available to buyers and sellers are different, but it works both ways. Sellers don't have "never late" unless they have recently listed their notes. Buyers know "trend" but don't "easily" know the magnitude of FICO point change (+5 is the same as +500; up and -5 is the same as -500; down) as you mentioned in your post (unless major screen scraping is involved). I've said it before; it's really wacko.

Meanwhile, Anil may be in a growth market. Sellers who want to sell and buyers who want to buy. Good place to be.
 


GregTexas

  • Newbie
  • *
  • Posts: 17
    • View Profile
    • Email
Anil I'm curious how your system handles if multiple buyers' filter settings result in customers vying for the same note(s)? Which customer gets the note(s)? Do you process with the customers randomly sorted for each session?

AnilG

  • Hero Member
  • *****
  • Posts: 1123
    • View Profile
    • PeerCube
Yes, random.

Anil I'm curious how your system handles if multiple buyers' filter settings result in customers vying for the same note(s)? Which customer gets the note(s)? Do you process with the customers randomly sorted for each session?
---
Anil Gupta
PeerCube Thoughts blog https://www.peercube.com/blog
PeerCube https://www.peercube.com

fliphusker

  • Sr. Member
  • ****
  • Posts: 463
    • View Profile
    • Email
The historical loan data file is not updated and released until a few weeks after a Quarter has completed. This file includes the information about loans that were issued in the last Quarter. Without this file you don't have access to loan and credit attributes for a large number of new loans. As this file will be used to filter loans for credit and loan attributes, you will miss out on the loans issued in last Quarter. This is what I mean by "unavailability" of credit data for a large segment of new loans. Similarly, by the time historical loan data file is released, some of the time-sensitive data such as loan status, recent FICO score etc. are already stale (old) and might have changed. Any filtering based on stale data will result in erroneous automated purchase. By using manual filtering, you are able to review individual loan and note data for any such discrepancies but you don't have that luxury with automated filtering without adverse impacts.

The loan and credit data provided at the time of loan application loses its "relative" value with time. Older the loan, less likely this application data will help you make better note selection. Borrower payment history, change FICO and loan status become much more valuable once loan has been issued. The payment history, FICO change and loan status will help you more in identifying leopards who have changed their "stripes" than the information on stripes 12 months ago. If we had access to updated credit data, nothing will be better than that but we don't have access to updated credit data. We have to make the best decisions based on what information is available and updated unless we want to take our ball and go home because we don't like the playing field.

BTW, have you read my blog post  "FICO Score Trends and Defaults for Lending Club Loans" https://www.peercube.com/blog/post/fico-score-trends-and-defaults-for-lending-club-loans? 99+% of loans with FICO trend of FLAT and UP from application FICO fully pay the loans. The odds are in your favor that most borrowers with Up FICO trend will repay their loans.
Thanks for clarifying especially with the historical data.  I did not know about this, appreciate you teaching me something new today.  But like you said in your article Anil, is that the larger the FICO swing the more likely a default will occur.  With auto investing, what prevents it from not picking a note with a huge downswing but still has a overall gain from the original FICO?

I had a week from when I opened up my LC account that I did a lot of reading and PC with your blogs was on that I gleamed a lot of info from.  As a very small investor, I get to be extremely picky on what notes I buy. 

Rob, correct me if I am missing your point.  For me, when I filter notes from NSR (Sorry Anil, have not bought prem yet, I am just do not invest enough) I use my filter that I normally use to buy new notes.  So what the logic for the the original FOLIO note, does not come into play.  Just like I do not care what their reasoning is for selling the note, as long as I see nothing wrong with the note. 
Why LC and FOLIO do not use a chart on main page of a note that shows it, like the LC Greasemonkey script does.  Or just show the number and not arrows.  A -5 drop in FICO score drop after the first month or two is completely expected, and is meaningless. 

Rob L

  • Hero Member
  • *****
  • Posts: 2133
    • View Profile
Rob, correct me if I am missing your point.  For me, when I filter notes from NSR (Sorry Anil, have not bought prem yet, I am just do not invest enough) I use my filter that I normally use to buy new notes.  So what the logic for the the original FOLIO note, does not come into play.  Just like I do not care what their reasoning is for selling the note, as long as I see nothing wrong with the note. 
Why LC and FOLIO do not use a chart on main page of a note that shows it, like the LC Greasemonkey script does.  Or just show the number and not arrows.  A -5 drop in FICO score drop after the first month or two is completely expected, and is meaningless.

Just saying that as a borrower continues to make on time payments there's a point at which that fact becomes the dominant predictor regarding loan performance going forward. Couple that with FICO change and in a rather short time (a few months) the original data at loan application time becomes much less important. However, I don't have any numbers to back that up; could be dead wrong.

AnilG

  • Hero Member
  • *****
  • Posts: 1123
    • View Profile
    • PeerCube
Lets do a thought exercise.

Q1a: Will you buy loan ID 6744587 from primary platform? Why or why not?
Q1b: When would you have sold this loan if you bought from primary platform? Why?
Q2: Will you now buy a note from Loan ID 6744587 on Folio and what premium/discount you will be willing to pay? Why or why not?
Q3: If you hold a note from this loan in your portfolio, will you sell this note now? If yes, at what premium/discount and why? If not, why not?

Loan Details on PeerCube
https://www.peercube.com/comment?loanid=6744587

A few notes from this loan listed for sale on Folio
https://www.lendingclub.com/foliofn/browseNotesLoanPerf.action?showfoliofn=true&loan_id=6744587&order_id=76770933&note_id=28750753
https://www.lendingclub.com/foliofn/browseNotesLoanPerf.action?showfoliofn=true&loan_id=6744587&order_id=66076167&note_id=28749212
https://www.lendingclub.com/foliofn/browseNotesLoanPerf.action?showfoliofn=true&loan_id=6744587&order_id=16022463&note_id=28749207
https://www.lendingclub.com/foliofn/browseNotesLoanPerf.action?showfoliofn=true&loan_id=6744587&order_id=10391995&note_id=28749387
https://www.lendingclub.com/foliofn/browseNotesLoanPerf.action?showfoliofn=true&loan_id=6744587&order_id=12117700&note_id=28747801

Thanks for clarifying especially with the historical data.  I did not know about this, appreciate you teaching me something new today.  But like you said in your article Anil, is that the larger the FICO swing the more likely a default will occur.  With auto investing, what prevents it from not picking a note with a huge downswing but still has a overall gain from the original FICO?

I had a week from when I opened up my LC account that I did a lot of reading and PC with your blogs was on that I gleamed a lot of info from.  As a very small investor, I get to be extremely picky on what notes I buy. 

Rob, correct me if I am missing your point.  For me, when I filter notes from NSR (Sorry Anil, have not bought prem yet, I am just do not invest enough) I use my filter that I normally use to buy new notes.  So what the logic for the the original FOLIO note, does not come into play.  Just like I do not care what their reasoning is for selling the note, as long as I see nothing wrong with the note. 
Why LC and FOLIO do not use a chart on main page of a note that shows it, like the LC Greasemonkey script does.  Or just show the number and not arrows.  A -5 drop in FICO score drop after the first month or two is completely expected, and is meaningless.
« Last Edit: June 14, 2016, 01:47:33 AM by AnilG »
---
Anil Gupta
PeerCube Thoughts blog https://www.peercube.com/blog
PeerCube https://www.peercube.com

fliphusker

  • Sr. Member
  • ****
  • Posts: 463
    • View Profile
    • Email

Lets do a thought exercise.

Q1a: Will you buy loan ID 6744587 from primary platform? Why or why not? 
No, has 3 issues that would kick it from my filter.  Income, inquiry and BK.   
Q1b: When would you have sold this loan if you bought from primary platform? Why?
I would have posted it immediately if accidentally bought.  For the same reasons I would not have bought it above. 
Q2: Will you now buy a note from Loan ID 6744587 on Folio and what premium/discount you will be willing to pay? Why or why not?
I still would not buy this note on FOLIO, due to not meeting my initial criteria.  I did not break it down and taking a rough guess to keep the YTM at or above its current value, would be -6%.  I am assuming I am high,but would be what I would be willing to pay for it if I was going to buy it. 
Q3: If you hold a note from this loan in your portfolio, will you sell this note now? If yes, at what premium/discount and why? If not, why not?
If this was an accidental buy, only reason I would end up with it, I would sell it. 
I sit on over 70 notes similar to this (not same number of payments obviously) because of a bad filter on LR when I initially started.  70 is not much to most here, but that is a big chunk for me.  And I do list them on FOLIO.  Normally for 1% simply to cover the fee.  Right now I have them at 0, as most of them would just make a bit.  While I would feel more comfortable in loans I know have a smaller chance, as a whole, not to default.  Sure, most of the notes are probably just fine, but I see them as ticking time bombs. 
Loan id: 73955953 this is one bomb that has already gone off.  1 payment then BK.  You see any reason why a sane person would invest in that note?

Lets do a bit of backtesting and get specific on this note. 
40-45k income and more then 1 inq. on an E grade note.=6.95 ROI and 13.42% loss. 4,213 total notes. 
Let's toss in some of my filters.
E notes-11.7 ROI and 8.19 loss on 1700 notes.  Relevant count?
C notes-10.3 and 3.6 loss on 12.6k notes. 
Hope I am not comparing apples to dolphins here.  But just by cutting off a couple of simple criteria points, I have significantly increased earnings and decreased risk. 
And yes, I see the listings you have there for the same note, and how they are broken.  Just another reason to not use FOLIO for searches. 

Rob L

  • Hero Member
  • *****
  • Posts: 2133
    • View Profile
Lets do a thought exercise.

Q1a: Will you buy loan ID 6744587 from primary platform? Why or why not?
Q1b: When would you have sold this loan if you bought from primary platform? Why?
Q2: Will you now buy a note from Loan ID 6744587 on Folio and what premium/discount you will be willing to pay? Why or why not?
Q3: If you hold a note from this loan in your portfolio, will you sell this note now? If yes, at what premium/discount and why? If not, why not?

Looks like a perfect example of on time payment history becoming more important that credit history at loan origination.
Q1a) I never bought 60 month loans so can't say if I would have bought it. Since it took 458 lenders to fund then it probably was not a popular bet.
Q1b) Up until a few weeks ago I was only buy and hold, so definitely would not have sold it.
Q2) Except for the fact I'm not buying anything then sure I'd buy it for the right price. Perfect on time payments, 33, with 27 to go. FICO up 20 AND it is an E5 22.7% loan. In my book the 33 on time payments have shown me the current risk is much lower than it was at origination. One of your example notes was priced at 20% markup giving a 3% YTM; a little high (LOL). One was at a 2.8% markup and 18.6% YTM which "seems" quite good.
Q3) Yes, I'm in the sell mode of course. I'd have trouble listing this one for less than a 2-3% premium. 1% to simply break even and liquidate if very motivated. That would be almost giving it away.

What are your answers Anil?