Author Topic: Why Does Lending Club Set Interest Rates?  (Read 8227 times)

Shylock

  • Newbie
  • *
  • Posts: 11
    • View Profile
Why Does Lending Club Set Interest Rates?
« on: April 25, 2016, 02:37:02 PM »
If Lending Club is merely creating a marketplace for borrowers and investors, why does LC set the price? Why not let them determine it independently? The external credit models are fundamentally about finding loans with the highest expected return relative to their risk.

The only reason I can see is if LC is using information that is not disclosed to influence its rate setting - information that cannot be disclosed to prospective investors. I don't know if this is the case or not.

I'm imagining a Dutch auction - let's say a borrower signs up for a loan at a maximum 25% interest rate. Investors keep bidding against each other until the loan is not only fully funded, but there are no investors who are willing to bid lower. External credit models are already implicitly doing this, except there's only one price at which they decide to bid or not, and not multiple rounds.

One pitfall of this model is that the buyer wouldn't know upfront what his interest rate would be. This could be mitigated by either being given an estimate beforehand, or being able to opt out of taking the loan if the interest rate isn't to his liking. (And if he opts out, he loses say, $10 or 0.1% of the value of the loan).

Booleans

  • Jr. Member
  • **
  • Posts: 74
    • View Profile
    • Email
Re: Why Does Lending Club Set Interest Rates?
« Reply #1 on: April 25, 2016, 03:30:04 PM »
Because humans are terrible at estimating risk and will end up accepting rates that are too low. Especially with the current glut of capital and ~0% interest rates. Once the defaults start coming in LC will have to deal with bad press and angry investors.

LC would lose a lot of its appeal to me if I was just given loan data and had to analyze each loan to determine what I'd be willing to pay for it. And then afterwards it's likely that I'll just be underbid anyway, wasting my time.

So let's just remove the uncertainty for both buyers and sellers and let LC price the loans.

rawraw

  • Hero Member
  • *****
  • Posts: 2759
    • View Profile
Why Does Lending Club Set Interest Rates?
« Reply #2 on: April 25, 2016, 04:10:25 PM »
Although it's a young Industry  history helps. Read about the original prosper

Sent from my SAMSUNG-SM-G935A using Tapatalk


AnilG

  • Hero Member
  • *****
  • Posts: 1090
    • View Profile
    • PeerCube
Re: Why Does Lending Club Set Interest Rates?
« Reply #3 on: April 25, 2016, 07:30:41 PM »
I agree with rawraw and Booleans. Prosper originally started out with investors setting the interest rate and failed miserably. Investors are not very good at judging risk. I see this play out on LC secondary market quite often.
---
Anil Gupta
PeerCube Thoughts blog https://www.peercube.com/blog
PeerCube https://www.peercube.com

Fred

  • Hero Member
  • *****
  • Posts: 1421
    • View Profile
Re: Why Does Lending Club Set Interest Rates?
« Reply #4 on: April 26, 2016, 02:34:38 AM »
I see this play out on LC secondary market quite often.

Yes, in Folio people are setting the note's YTM through the premium or discount.

Please come to Folio if you want to set your own "rates."

fliphusker

  • Sr. Member
  • ****
  • Posts: 463
    • View Profile
    • Email
Re: Why Does Lending Club Set Interest Rates?
« Reply #5 on: April 26, 2016, 05:34:12 AM »
I see this play out on LC secondary market quite often.

Yes, in Folio people are setting the note's YTM through the premium or discount.

Please come to Folio if you want to set your own "rates."

I would love to see how many notes are sold on folio daily, and what percent are sold from -2 to +5.  I will bet it is under 2%. 

I think buyers on folio who are willing to take risks, can make a killing there.

jz451

  • Full Member
  • ***
  • Posts: 118
    • View Profile
Re: Why Does Lending Club Set Interest Rates?
« Reply #6 on: April 26, 2016, 12:34:14 PM »

I would love to see how many notes are sold on folio daily, and what percent are sold from -2 to +5.  I will bet it is under 2%. 

I think buyers on folio who are willing to take risks, can make a killing there.

That is what my strategy is. Starting last month I'm testing only using notes from folio so hopefully I beat most of the default risk and earn +10%-15%.

Shylock

  • Newbie
  • *
  • Posts: 11
    • View Profile
Re: Why Does Lending Club Set Interest Rates?
« Reply #7 on: April 26, 2016, 03:17:52 PM »
Because humans are terrible at estimating risk and will end up accepting rates that are too low. Especially with the current glut of capital and ~0% interest rates. Once the defaults start coming in LC will have to deal with bad press and angry investors.

LC would lose a lot of its appeal to me if I was just given loan data and had to analyze each loan to determine what I'd be willing to pay for it. And then afterwards it's likely that I'll just be underbid anyway, wasting my time.

So let's just remove the uncertainty for both buyers and sellers and let LC price the loans.

That's all true, and my comments were more aimed at the onset of algorithmic trading. I.e. if you have LC's algorithm setting rates, versus a panoply of third parties, then it's just one algorithm versus another. Plus, hiring smart people to design a credit rating system isn't free. When you're using say, Lending Robot, you're effectively paying for two algorithms, or two credit rating bureaus - LC and LR. I'm imagining a system where LC still procures the loans, but doesn't automatically offer loan grades. Loan grades would be provided by the third parties of your choice (including possibly LC), at additional cost.

It would be interesting to see how much stock the third parties *currently* put in the loan grades that LC issues. I.e., if all the information can be extracted from the borrower info, there's no sense in using someone else's number crunching of that same data - unless, again, LC uses data to set its loan grades that it doesn't release to investors.

As for people underestimating risk, the economist in me likes to think, well, that's they're problem. Or that you if people are accepting rates that are too low, you need more loans.

Off-topic, but my hunch is that if the algorithmic pickers got into the secondary markets (and actual picking, not just filtering and buying), the P2P lending market could become a whole lot more liquid.
« Last Edit: April 26, 2016, 03:21:35 PM by Shylock »

dompazz

  • Full Member
  • ***
  • Posts: 224
    • View Profile
Re: Why Does Lending Club Set Interest Rates?
« Reply #8 on: April 26, 2016, 03:44:37 PM »
That's all true, and my comments were more aimed at the onset of algorithmic trading. I.e. if you have LC's algorithm setting rates, versus a panoply of third parties, then it's just one algorithm versus another. Plus, hiring smart people to design a credit rating system isn't free. When you're using say, Lending Robot, you're effectively paying for two algorithms, or two credit rating bureaus - LC and LR. I'm imagining a system where LC still procures the loans, but doesn't automatically offer loan grades. Loan grades would be provided by the third parties of your choice (including possibly LC), at additional cost.
Interesting idea, but I'm not sure it works economically for LC.  Most of what you pay them for is client acquisition, servicing, and general infrastructure.  The grading is probably not a huge ongoing cost of the system.

It would be interesting to see how much stock the third parties *currently* put in the loan grades that LC issues. I.e., if all the information can be extracted from the borrower info, there's no sense in using someone else's number crunching of that same data - unless, again, LC uses data to set its loan grades that it doesn't release to investors.
I grade loans through the API.  LC Sub-grade figures heavily in there.  It is highly significant.  I do find adding other variables helps create better fits.  But the standard to which I grade any model is a comparison against a sub-grade only model.

Off-topic, but my hunch is that if the algorithmic pickers got into the secondary markets (and actual picking, not just filtering and buying), the P2P lending market could become a whole lot more liquid.
Yup.  Problem is that you don't have up-to-date underwriting for the borrower.  You have to depend on payment history and FICO score.  Not a lot to work with.


Rob L

  • Hero Member
  • *****
  • Posts: 2043
    • View Profile
Re: Why Does Lending Club Set Interest Rates?
« Reply #9 on: April 26, 2016, 04:49:42 PM »
There is no reason to think LC provides us all the data they use in their underwriting model. Quite the contrary. To protect the identity of the borrower LC purposefully restricts the data provided. Over the years they have even removed data (i.e. employer, etc.) that had been previously been provided. On the other hand data mining of an individual borrower's social media accounts may possibly be used by LC to enhance the accuracy of their underwriting. Who knows what other data sources they may exploit. In short they have a huge informational advantage.

Meanwhile, refer to this old thread. You should find it interesting reading:

http://www.lendacademy.com/forum/index.php?topic=1953.0

PhilGD

  • Full Member
  • ***
  • Posts: 150
    • View Profile
    • Email
Re: Why Does Lending Club Set Interest Rates?
« Reply #10 on: April 27, 2016, 08:52:31 AM »
Quote
It would be interesting to see how much stock the third parties *currently* put in the loan grades that LC issues. I.e., if all the information can be extracted from the borrower info, there's no sense in using someone else's number crunching of that same data - unless, again, LC uses data to set its loan grades that it doesn't release to investors.
I grade loans through the API.  LC Sub-grade figures heavily in there.  It is highly significant.  I do find adding other variables helps create better fits.  But the standard to which I grade any model is a comparison against a sub-grade only model.

Subgrade is highly significant in back testing, sure. But I don't use the LC subgrade as an input in my models.

I think that to actually run regressions off of LC's subgrade would be a bad idea, because LC updates its models so frequently that you can't be sure if a loan that LC assessed as D4 in 2014 would also be graded D4 in 2016.

dompazz

  • Full Member
  • ***
  • Posts: 224
    • View Profile
Re: Why Does Lending Club Set Interest Rates?
« Reply #11 on: April 27, 2016, 10:46:27 AM »
Subgrade is highly significant in back testing, sure. But I don't use the LC subgrade as an input in my models.

I think that to actually run regressions off of LC's subgrade would be a bad idea, because LC updates its models so frequently that you can't be sure if a loan that LC assessed as D4 in 2014 would also be graded D4 in 2016.
I don't disagree in principal and this is something I struggled with.  Do we actually know how frequently LC updates their model?  Are those updates major or are they minor/marginal?  I don't know the answers to either of those.

I get effects that are monotonically increasing up the sub-grades.  I have to assume that is LC's goal.  So if they have put a lot of work into grading loans, seem to have done a decent job at it, and have some information which I do not have, then it seems illogical to throw that out. 

Might the effects be off because of unobserved changes, yes.  Are there lots of other unobserved changing variables that effect default rates, yes.  I don't think using sub-grade as a first proxy of default is necessarily bad.

It is also a good first champion to throw each challenger up against.  That is, can a new model better predict default than simple historic default rates by sub-grade?  If so, it's worth a further look.

thezfunk

  • Sr. Member
  • ****
  • Posts: 316
    • View Profile
Re: Why Does Lending Club Set Interest Rates?
« Reply #12 on: April 27, 2016, 11:47:27 AM »

I would love to see how many notes are sold on folio daily, and what percent are sold from -2 to +5.  I will bet it is under 2%. 

I think buyers on folio who are willing to take risks, can make a killing there.

That is what my strategy is. Starting last month I'm testing only using notes from folio so hopefully I beat most of the default risk and earn +10%-15%.

After increasing defaults on newly issued loans, I have moved exclusively to folio for all my buying now that LendingRobot has automated investing working on Folio.

PhilGD

  • Full Member
  • ***
  • Posts: 150
    • View Profile
    • Email
Re: Why Does Lending Club Set Interest Rates?
« Reply #13 on: April 27, 2016, 01:31:03 PM »
Subgrade is highly significant in back testing, sure. But I don't use the LC subgrade as an input in my models.

I think that to actually run regressions off of LC's subgrade would be a bad idea, because LC updates its models so frequently that you can't be sure if a loan that LC assessed as D4 in 2014 would also be graded D4 in 2016.
I don't disagree in principal and this is something I struggled with.  Do we actually know how frequently LC updates their model?  Are those updates major or are they minor/marginal?  I don't know the answers to either of those.

I get effects that are monotonically increasing up the sub-grades.  I have to assume that is LC's goal.  So if they have put a lot of work into grading loans, seem to have done a decent job at it, and have some information which I do not have, then it seems illogical to throw that out. 

Might the effects be off because of unobserved changes, yes.  Are there lots of other unobserved changing variables that effect default rates, yes.  I don't think using sub-grade as a first proxy of default is necessarily bad.

It is also a good first champion to throw each challenger up against.  That is, can a new model better predict default than simple historic default rates by sub-grade?  If so, it's worth a further look.

To your first point, I remember reading an article where Renaud Laplanche stated that LC rebuilds their credit model every 6 months (I don't have a link unfortunately). So If you're backtesting and running regressions on seasoned loans, then you're looking at subgrades assigned by a model that was used 2 or 3 generations ago. And the fact that he said the model gets "rebuilt" suggests to me that the changes are significant.

Also, lets hypothetically say that LC's model considers DTI when calculating subgrade. DTI is a variable that I will definitely be including in my model, too. This means that if I were to include subgrade, DTI, and other variables, and then LC's model in one generation heavily factored DTI, but in the next generation the influence of DTI on subgrade was much smaller, then my model could see a heavy reduction in accuracy.

A big assumption in regression analysis is that the influence of predictor variables is basically constant over time. So to include a variable like subgrade violates that assumption in my opinion.
« Last Edit: April 27, 2016, 01:33:20 PM by PhilGD »

rawraw

  • Hero Member
  • *****
  • Posts: 2759
    • View Profile
Why Does Lending Club Set Interest Rates?
« Reply #14 on: April 27, 2016, 02:36:51 PM »
Doesn't sub grade introduce serial correlation as well?

LC is limited in the data they can use for compliance risks. For example, lending based on zip code could be risky for LC and last I saw their models didn't use it. If you used it, you may be able to improve their model.

Sent from my SAMSUNG-SM-G935A using Tapatalk