Author Topic: Have I found the Holy Grail of filters? (Nickelsteamroller)  (Read 12252 times)

jpildis

  • Full Member
  • ***
  • Posts: 158
    • View Profile
    • Email
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #15 on: July 27, 2014, 10:22:57 AM »
My concern with all of the back-testing is that LC is changing their credit scoring all the time.  Their objective function is to get the overall grade to NAR relationship as advertised.  That great filter that looked at a subset of D & E grades and had really low default rates, may not work with today's D & E grades... LC's smart folks did the same analysis and now those same borrowers would have gotten B & C ratings.  We're not competing against other lenders... we're competing against LC's desire to have an efficient market where each loan is properly graded based on the likelihood of default.  The vast majority of us are bringing a dull knife to the proverbial gun fight.

hoggy1

  • Sr. Member
  • ****
  • Posts: 401
    • View Profile
    • Email
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #16 on: July 27, 2014, 11:14:06 AM »
... That great filter that looked at a subset of D & E grades and had really low default rates, may not work with today's D & E grades... LC's smart folks did the same analysis and now those same borrowers would have gotten B & C ratings. ...
I share the same concern and a few more. I have developed filters on NS which reduce the loss rate on E notes to less than 4%. But the same filter has a loss rate of over 8% on F grade loans. To what can I attribute that?
Steve

rawraw

  • Hero Member
  • *****
  • Posts: 2795
    • View Profile
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #17 on: July 27, 2014, 11:38:03 AM »
... That great filter that looked at a subset of D & E grades and had really low default rates, may not work with today's D & E grades... LC's smart folks did the same analysis and now those same borrowers would have gotten B & C ratings. ...
I share the same concern and a few more. I have developed filters on NS which reduce the loss rate on E notes to less than 4%. But the same filter has a loss rate of over 8% on F grade loans. To what can I attribute that?
This is what I meant when I pointed out it incorrectly rank ordered risk.  But very good points

Fred

  • Hero Member
  • *****
  • Posts: 1421
    • View Profile
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #18 on: July 27, 2014, 11:43:48 AM »
Their objective function is to get the overall grade to NAR relationship as advertised.

I am not sure what you mean by the above statement.  This requires LC to somehow fix their default rates.



IMO, LC goal is to minimize default rates across all grades, thus maximizing NAR.  This will provide incentive to lenders, thus increasing demand on LC loans.

trevor

  • Newbie
  • *
  • Posts: 39
    • View Profile
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #19 on: July 27, 2014, 12:34:01 PM »
My concern with all of the back-testing is that LC is changing their credit scoring all the time.  Their objective function is to get the overall grade to NAR relationship as advertised.  That great filter that looked at a subset of D & E grades and had really low default rates, may not work with today's D & E grades... LC's smart folks did the same analysis and now those same borrowers would have gotten B & C ratings.  We're not competing against other lenders... we're competing against LC's desire to have an efficient market where each loan is properly graded based on the likelihood of default.  The vast majority of us are bringing a dull knife to the proverbial gun fight.

The filter itself is still good, regardless if the underlying interest rates and defaults are different.

Very high income, loan reconciliation, 0 inquiries, etc. These filters have been shown to increase returns no matter where they are used. I have trouble thinking up a time in the future where the average 100k income loan WOULDN'T outperform the average 50k income loan. Or the average debt reconciliation outperforming other loan types. It just seems like common sense to me.


... That great filter that looked at a subset of D & E grades and had really low default rates, may not work with today's D & E grades... LC's smart folks did the same analysis and now those same borrowers would have gotten B & C ratings. ...
I share the same concern and a few more. I have developed filters on NS which reduce the loss rate on E notes to less than 4%. But the same filter has a loss rate of over 8% on F grade loans. To what can I attribute that?


Low number of loans that skews the returns, possibly? F and G loans aren't very common.












Rob L

  • Hero Member
  • *****
  • Posts: 2137
    • View Profile
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #20 on: July 27, 2014, 01:01:38 PM »
I was looking at the chart above and it was quite different than I remembered. Here's the current on the LC website.
LC is trying, but it doesn't look to me like risk taking has been sufficiently rewarded as I view the chart.



Personally I think LC's underwriting mechanism is its most valuable asset (other than perhaps its people).
Kind of a high wire act; balancing the needs of lenders and borrowers.
If they fail to do an extremely good job of that then eventually someone else will.








faeriering

  • Full Member
  • ***
  • Posts: 135
    • View Profile
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #21 on: July 27, 2014, 01:35:27 PM »
I would suggest one cavet to Rawraw & Anil's points above.  If you are only planning on holding the notes for 12 months, and then looking at selling them on Folio then a shorter loan history might be okay to review.  If you are looking to hold the notes to maturity their points are very valid.  Of course if you do sell them on folio and they have a high likelihood of default it might be harder to off load them.

jpildis

  • Full Member
  • ***
  • Posts: 158
    • View Profile
    • Email
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #22 on: July 27, 2014, 01:37:23 PM »
Their objective function is to get the overall grade to NAR relationship as advertised.

I am not sure what you mean by the above statement.  This requires LC to somehow fix their default rates.



IMO, LC goal is to minimize default rates across all grades, thus maximizing NAR.  This will provide incentive to lenders, thus increasing demand on LC loans.


The point is that it is in LC's advantage to pick those same attributes out of the history and improve the credit scoring on that subset, thereby lowering the interest rate.  Yes, the loan will "outperform" if you measure out-performance by default rate.  If, instead, you measure performance on total lender return, you will not necessarily see out-performance.

LC wants higher risk loans to have a higher net investor return... risk must be rewarded or, eventually, the capital will dry-up.  They are striving for an efficient market... anomalies will occur but they will be fighting everyday to eliminate them.

I over relied on filtering when I was big (>$100k) in Prosper... high income, repeat borrowers were gold. Past performance was not indicative of future results.  An expensive lesson was learned.

That all being said, if you have consumer underwriting experience and access to Masters-level statistics, you may be able to out perform the 'market' by some small but real factor.

trevor

  • Newbie
  • *
  • Posts: 39
    • View Profile
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #23 on: July 27, 2014, 01:44:16 PM »
Their objective function is to get the overall grade to NAR relationship as advertised.

I am not sure what you mean by the above statement.  This requires LC to somehow fix their default rates.



IMO, LC goal is to minimize default rates across all grades, thus maximizing NAR.  This will provide incentive to lenders, thus increasing demand on LC loans.


The point is that it is in LC's advantage to pick those same attributes out of the history and improve the credit scoring on that subset, thereby lowering the interest rate.  Yes, the loan will "outperform" if you measure out-performance by default rate.  If, instead, you measure performance on total lender return, you will not necessarily see out-performance.

LC wants higher risk loans to have a higher net investor return... risk must be rewarded or, eventually, the capital will dry-up.  They are striving for an efficient market... anomalies will occur but they will be fighting everyday to eliminate them.

I over relied on filtering when I was big (>$100k) in Prosper... high income, repeat borrowers were gold. Past performance was not indicative of future results.  An expensive lesson was learned.

That all being said, if you have consumer underwriting experience and access to Masters-level statistics, you may be able to out perform the 'market' by some small but real factor.


Out of curiosity....How exactly do you choose your loans now?

jpildis

  • Full Member
  • ***
  • Posts: 158
    • View Profile
    • Email
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #24 on: July 27, 2014, 01:55:32 PM »
Their objective function is to get the overall grade to NAR relationship as advertised.

I am not sure what you mean by the above statement.  This requires LC to somehow fix their default rates.



IMO, LC goal is to minimize default rates across all grades, thus maximizing NAR.  This will provide incentive to lenders, thus increasing demand on LC loans.


The point is that it is in LC's advantage to pick those same attributes out of the history and improve the credit scoring on that subset, thereby lowering the interest rate.  Yes, the loan will "outperform" if you measure out-performance by default rate.  If, instead, you measure performance on total lender return, you will not necessarily see out-performance.

LC wants higher risk loans to have a higher net investor return... risk must be rewarded or, eventually, the capital will dry-up.  They are striving for an efficient market... anomalies will occur but they will be fighting everyday to eliminate them.

I over relied on filtering when I was big (>$100k) in Prosper... high income, repeat borrowers were gold. Past performance was not indicative of future results.  An expensive lesson was learned.

That all being said, if you have consumer underwriting experience and access to Masters-level statistics, you may be able to out perform the 'market' by some small but real factor.


Out of curiosity....How exactly do you choose your loans now?

My LC strategy depends on fast purchasing of EFG loans and the greater-fool theory of Folio trading.

Lovinglifestyle

  • Hero Member
  • *****
  • Posts: 902
    • View Profile
    • Email
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #25 on: July 27, 2014, 02:50:35 PM »
I was looking at the chart above and it was quite different than I remembered. Here's the current on the LC website.
LC is trying, but it doesn't look to me like risk taking has been sufficiently rewarded as I view the chart.



Personally I think LC's underwriting mechanism is its most valuable asset (other than perhaps its people).
Kind of a high wire act; balancing the needs of lenders and borrowers.
If they fail to do an extremely good job of that then eventually someone else will.

This chart is using "adjusted" NAR.  You're right, risk doesn't appear very rewarded for the extra selection work it requires. 

Looking at this chart, I can see why my growing pile of problem notes is making me feel less successful.  Since I stopped selling them, I like that I can better see exactly where I am.  My NAR has dropped from being even at normal vs adjusted, to adjusted being 1.72% lower (at 16.19) than not-adjusted.  So I haven't been just imagining things!

I guess if all the NARs are about the same, the grade doesn't matter and that's a good thing?   

Rob L

  • Hero Member
  • *****
  • Posts: 2137
    • View Profile
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #26 on: July 27, 2014, 05:45:56 PM »
I guess if all the NARs are about the same, the grade doesn't matter and that's a good thing?   
Well... The grade tells us how risky LC thinks a loan is. I believe they really try to get this right.
If the macro-economy hits a rough patch NAR's about the same now probably won't be then.
All things considered it's gotta be better to be holding a portfolio of C's rather than E's if the NAR's are the same now.
If lenders aren't compensated for the risk, why go there? Why do we do that (and I plead guilty)?

AnilG

  • Hero Member
  • *****
  • Posts: 1123
    • View Profile
    • PeerCube
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #27 on: July 27, 2014, 06:25:56 PM »
If lenders aren't compensated for the risk, why go there? Why do we do that (and I plead guilty)?

Higher risk investments offer higher "expected" (not necessarily realized) return. There is no guarantee that just because you are taking higher risk you will "definitely" realize higher return. The mean return across hundreds of investors with similar risk may show higher with higher risk. But so is the spread (variance) in return.

P2P lenders seem to get fixated on "one" number without understanding the variance in that number. One lender return may be very different from another lender investing in similar risk loans.
---
Anil Gupta
PeerCube Thoughts blog https://www.peercube.com/blog
PeerCube https://www.peercube.com

Rob L

  • Hero Member
  • *****
  • Posts: 2137
    • View Profile
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #28 on: July 27, 2014, 09:39:29 PM »
P2P lenders seem to get fixated on "one" number without understanding the variance in that number. One lender return may be very different from another lender investing in similar risk loans.
True. I've tried to minimize that variance by diversifying in a large number of small higher risk notes.
That helps up to a point, but if the loans across which I'm diversifying have a high degree of underlying correlation I'm just fooling myself.
In the macro-economic sense they probably do. It's more a question of when.
Funny, but the title of the thread makes reference to "nickel steamroller"; a term coined in reference to this theme as I understand it.
Basically it works great; until it doesn't.

turing

  • Full Member
  • ***
  • Posts: 222
    • View Profile
    • PeerToPeerQuant.com
    • Email
Re: Have I found the Holy Grail of filters? (Nickelsteamroller)
« Reply #29 on: July 28, 2014, 02:29:28 PM »

My LC strategy depends on fast purchasing of EFG loans and the greater-fool theory of Folio trading.

Greater fool strategies have a bad history (Tulips in Holland, internet stocks in 1999, and tons of others).  Hopefully your timing is really good.