Lend Academy Network Forum

Lending Club Discussion => Investors - LC => Topic started by: mrwhizzard on September 11, 2017, 02:19:15 PM

Title: LC Email: "The Next Generation Credit Model"
Post by: mrwhizzard on September 11, 2017, 02:19:15 PM
I just got the below email about an updated credit model. The one comment here that concerns me slightly is the shift to grades A and B loans. I'm a little wary of a credit model enhancement that makes moves more people up than down...

Quote
We have launched the most advanced and predictive credit model ever used on the LendingClub platform. It's the latest in LendingClub's long history of innovation on behalf of borrowers and investors.
 
The new model leverages machine learning and 10 years of LendingClub data to better assess and price credit risk. LendingClub has helped 1.5+ million borrowers since 2007; each borrower provides unique insight that we can use in our next decision.
 
What investors need to know about the new credit model:
 
•   It's 24% better at differentiating the likelihood of a borrower charging off than the fourth-generation model, and is more predictive than a borrower's FICO score alone.
 
•   It's built on a wealth of proprietary data, incorporates more data points for each borrower, uses best-in-class modeling techniques, and uses dozens of new custom attributes that are predictive in assessing risk.
 
•   We expect loan volume to shift toward higher quality grades (grades A and B) because some borrowers will qualify for lower interest rates under the new model, and other higher-risk borrowers, who may have received an offer previously, will be screened out.
 
•   We continue to see lower delinquency rates across grades and terms in the existing loan portfolio than in the second and third quarters of 2016.
 
We see this credit model as the latest innovation in our continuous enhancement cycle, and one that helps us continue to provide investors with solid risk-adjusted returns. See here (https://blog.lendingclub.com/lendingclubs-next-generation-credit-model/) for more on what makes the model unique.
 
As always, please reach out to us with any questions or concerns at (888) 596-3159. Monday-Friday from 7 a.m. through 5 p.m. PT, or investing@lendingclub.com.
 
Best regards,
The LendingClub Team
Title: Re: LC Email: "The Next Generation Credit Model"
Post by: Fred93 on September 11, 2017, 04:20:59 PM
...updated credit model.
Quote
We continue to see lower delinquency rates across grades and terms in the existing loan portfolio than in the second and third quarters of 2016.
This statement is misleading and self-serving.

Lets look at the numbers.  The only delinquency data that is available for 17Q2 is the delinquency at the 1-month point.  That's so early in the loan's life that its not worth trying to interpret.  For 17Q1 we have data at months 1,2,3,4.  Lets look at month 4 data.  That's the farthest into the loans we can go for 17Q1.  This is the fraction of loans delinquent at the point where the 4th payment should have been made.

17Q1    1.10%
16Q4    1.25%
16Q3    1.57%
16Q2    1.50%

So at first blush, you might say their statement is correct.  We have a recent vintage with delinquency lower than 16Q2 & 3.  What they don't say is that they're choosing really bad quarters to use for comparison.  Lets continue to look at this series (of delinquency at month 4).

16Q1    1.14%
15Q4    0.88%
15Q3    0.89%
15Q2    0.90%
15Q1    0.79%

Many of you investors know that the loans issued in 2015 performed very poorly for investors.  And yet, by this measure the most recent vintage (ie 17Q1) is even worse than any 2015 quarter.  Hmm.  Things are worse.  What to do...  Must be time to issue a press release saying things are getting better!

14Q4    0.78%
14Q3    0.84%
14Q2    0.79%
14Q1    0.67%
13Q4    0.65%
13Q3    0.84%
13Q2    0.79%
etc


Quote
We expect loan volume to shift toward higher quality grades (grades A and B) because some borrowers will qualify for lower interest rates under the new mode

Grade inflation.


Title: Re: LC Email: "The Next Generation Credit Model"
Post by: JohnnyP on September 11, 2017, 07:58:50 PM
That is pretty weak. All the talk about transparency... Very disappointing to see that kind of misleading comment.
Title: Re: LC Email: "The Next Generation Credit Model"
Post by: anabio on September 11, 2017, 10:14:38 PM
•   We expect loan volume to shift toward higher quality grades (grades A and B) because some borrowers will qualify for lower interest rates under the new model, and other higher-risk borrowers, who may have received an offer previously, will be screened out.

...Let me translate that for you...

Now that everybody and their brother is into this online lending bandwagon we can't compete anymore. We either lower our interest rates (which will turn off investors) or bump everybody into a higher grade credit risk...whether they deserve it or not...
Title: Re: LC Email: "The Next Generation Credit Model"
Post by: rubicon on September 12, 2017, 11:41:18 AM
it does seem that LC wins while investors lose if there's grade inflation.
Title: Re: LC Email: "The Next Generation Credit Model"
Post by: mrwhizzard on September 12, 2017, 11:52:42 AM
it does seem that LC wins while investors lose if there's grade inflation.

Well, in the short term anyway. I can't imagine that's a very sustainable practice...
Title: Re: LC Email: "The Next Generation Credit Model"
Post by: Rob L on September 13, 2017, 10:14:34 AM
I read the email and thought it was an April Fools joke; guess not since it's September.
Maybe this thing was supposed to encourage lenders to hang in there but it probably had the reverse effect.
Not a single positive comment on this thread nor was there anything in the email I found encouraging as a lender.
Title: Re: LC Email: "The Next Generation Credit Model"
Post by: Lovinglifestyle on September 13, 2017, 12:31:23 PM
I've been looking in vain for something encouragingly positive in the new loan profiles.
I hoped to find more that passed my filter, or, if not that, more in the A,B,C categories that subjectively appear safer.
I don't yet see any (subjective and/or emotional) changes that indicate better quality to me.
Title: Re: LC Email: "The Next Generation Credit Model"
Post by: bluto on September 13, 2017, 04:20:51 PM
If LC's model improves, that should usually mean less loans for us active filterers to find, either because they improved their model to fix a grading error that we exploit with filters, or because they updated their model to exclude some class of risky loans that weren't being identified previously. 
Title: Re: LC Email: "The Next Generation Credit Model"
Post by: Lovinglifestyle on September 13, 2017, 05:22:02 PM
If LC's model improves, that should usually mean less loans for us active filterers to find, either because they improved their model to fix a grading error that we exploit with filters, or because they updated their model to exclude some class of risky loans that weren't being identified previously. 

This makes sense to me.  Thanks.
The EFG loans do seem to be more uniformly designated.  Looked at individually, none seem to not belong in those classes as was occasionally the case.
So there's one positive--I can have more confidence that none of the loans in EFG grades are in my filter for a good reason, and I can save time by not checking them.
 
Title: Re: LC Email: "The Next Generation Credit Model"
Post by: Rob L on September 19, 2017, 10:44:13 AM
The email contained this table.
I'd prefer to see an expected range of the projected returns rather than a single number. How much risk am I taking?
How much better (due to improvements in economy, etc.) or worse (recession, etc.) performance can an investor reasonably expect?
If they can project returns to two decimal places seems they could take a shot at upside and downside risk using historic volatility.
Nah.

(https://i.imgur.com/XigKW6k.png)

[1] “Projected Annualized Net Credit Loss (w/ Prepayment)” also known as Expected Charge-Off Rate, is LendingClub’s projection of the aggregate dollar amount of loan principal charged-off, net of any amounts recovered and accounting for the impact of amounts prepaid, as an annualized percentage of the aggregate dollar amount of loan principal for all loans issued under the Prime Program after September 8, 2017. Projected Annualized Net Credit Loss (w/ Prepayment) is not a promise of future results and may not accurately reflect actual charge-off or prepayment rates. Actual charge-off and prepayment rates experienced by any individual portfolio may be impacted by, among other things, the size and diversity of the portfolio, the exposure to any single loan, borrower or group of loans or borrowers, as well as macroeconomic conditions.

[2] “Average Interest Rate” is the weighted average interest rate of the grade and maturity mix for issued loans for the period of February through May 2017.

[3] “Projected Return” is a measure of the estimated annualized return rate on invested principal (meaning for all funds then invested in Notes or loans) using an internal rate of return (IRR) methodology using a monthly term. Monthly cash flow projections are calculated as follows: the scheduled principal and interest payments based on the Interest Rate, minus the amount of such principal and interest payments lost due to the Expected Charge-Off Rate, minus Expected Fees. Monthly IRR figures are annualized by multiplying the monthly IRR figure by 12. Projected Returns are calculated based on grade and maturity mix for issued loans for the period of February through May 2017. Projected Return is not a promise of future results and may not accurately reflect actual returns. Actual returns experienced by any individual portfolio may be impacted by, among other things, the size and diversity of the portfolio, the exposure to any single Note or loan, borrower or group of Notes, loans or borrowers, as well as macroeconomic conditions. Individual results may vary and projections are subject to change. The information presented is not intended to be investment advice, guidance, or a guarantee of the performance of any Note or loan. Notes are offered by prospectus filed with the SEC and investors should review the risks and uncertainties described in the prospectus prior to investing. Actual results may vary.

Title: Re: LC Email: "The Next Generation Credit Model"
Post by: Skeptical on September 19, 2017, 01:12:10 PM
@Rob L

Let me take the heretical view that this new model and these new numbers are meaningless. In a couple of years, "The Next-Next Generation Credit Model" will lead us to the Promise Land. No thanks. You can only razzle-dazzle me once. This sheep has been fleeced by your previous model.
Title: Re: LC Email: "The Next Generation Credit Model"
Post by: Rob L on September 19, 2017, 06:08:01 PM
Hey, just to be sure you know I'm agreeing with you here.
I mean really, projections to two decimal places. So we're counting individual basis points now.
If this is LC's attempt to keep me interested it didn't. Think there's a pretty tough crowd around here right now that feels the same.
My guess is that these projections are 2% to 3% too high; and that's without a recession.

And by the way; a few months ago LC removed its sub-grade probability of default projections from its feeding time file releases.
Less transparency, not more. Wonder why? Embarrassment? Yeah right, these new projections are spot on.
Title: Re: LC Email: "The Next Generation Credit Model"
Post by: JohnnyP on September 19, 2017, 11:28:19 PM
How can they estimate an 8.19% return on combined E notes with a straight face? Or a 9.59% return for F/G notes? Returns for these notes have been in the toilet for a long time. There are people that are out there buying these notes based on LC expectations. Listen real close and you can hear the flushing sound - money down the shitter. Nobody can convince me that somehow it will improve from shitter to 8-9% in the short run. Not possible. Quit lying to people!

Title: Re: LC Email: "The Next Generation Credit Model"
Post by: AnilG on September 20, 2017, 12:29:29 AM
Interesting part I found in your table is discrepancy in Interest Rate between F/G Grade 36 month loans versus F/G Grade 60 month loans. For all other grades, interest rate is higher for 60 month versus 36 month except in the case of F/G grade. Why would borrowers take F/G 36 month loan @ 30.53% when they can take 60 month loan @ 29.89%. Supposedly 60 month loans will come with lower monthly payment commitment with option to pay off in 36 months without penalty?

Similar issue exist with Estimated Credit Loss for E/F/G grade. If E/F/G loans are riskier than A/B/C/D, they should have higher projected loss and higher interest rate to compensate for it.

I don't understand LC logic with the new credit model. There is something really wrong in their analysis and with new credit model as it defies the basic tenet between credit quality, risk and return. I think they are relying too much on ML and not enough on common sense. It will come back to bite them.

The email contained this table.
I'd prefer to see an expected range of the projected returns rather than a single number. How much risk am I taking?
How much better (due to improvements in economy, etc.) or worse (recession, etc.) performance can an investor reasonably expect?
If they can project returns to two decimal places seems they could take a shot at upside and downside risk using historic volatility.
Nah.

Title: Re: LC Email: "The Next Generation Credit Model"
Post by: SeanMCA on September 20, 2017, 01:08:21 AM
I guess this is why they just hired Anuj Nayar to be their new PR expert. It looks the investor community needs more convincing about how great everything is.
Title: Re: LC Email: "The Next Generation Credit Model"
Post by: nonattender on September 20, 2017, 04:29:02 AM
I guess this is why they just hired Anuj Nayar to be their new PR expert. It looks the investor community needs more convincing about how great everything is.

It'd be a great topic for Peter to cover on the blog - and ask some questions.  (Of a quant, not a PR guy.)
Title: Re: LC Email: "The Next Generation Credit Model"
Post by: rubicon on September 20, 2017, 10:21:46 AM
They expect lower projected annualized net credit loss on F/G loans on 36 months compared to 60 months: 17.95% vs 16.99%. My guess is that the prepayment for 60 months F/G loans is high enough that the credit losses is actually lower (compared to 36 months). Hence perversely they are able to offer lower interest rate on 60 months F/G loans as you still get higher net projected return with a term premium of app. 1% (9.84% vs 8.89%). I guess lots of 60 month F/G loans refinance into 36 months D/E loans after 1 - 2 years.

But yes it does seem heavily data driven.
Title: Re: LC Email: "The Next Generation Credit Model"
Post by: AnilG on September 20, 2017, 02:48:23 PM
Makes no sense. If the losses are lower, they shouldn't be F/G grade loans. The Credit Grades are supposed to be credit rating, on a risk-return continuum, of the loans being offered, 'A' lowest risk and 'G' highest risk. By claiming F/G/60 will have lower interest rate than F/G/36 because it has lower losses, the Credit Grade is no longer a rating for assessing risk and return of loans LC offers. Grades are turning into just a few "independent" buckets in which they distribute the loans with no relationship between them. Basically, they are undermining their own argument about selecting loans based on Credit Grade only. People using LC automated investing will be shafted.

Data-driven but knowledge-unaware, most probably performed by people who see data just as bunch of numbers and have no basic knowledge of credit modeling and aptitude for interpretation what data telling them.
 
They expect lower projected annualized net credit loss on F/G loans on 36 months compared to 60 months: 17.95% vs 16.99%. My guess is that the prepayment for 60 months F/G loans is high enough that the credit losses is actually lower (compared to 36 months). Hence perversely they are able to offer lower interest rate on 60 months F/G loans as you still get higher net projected return with a term premium of app. 1% (9.84% vs 8.89%). I guess lots of 60 month F/G loans refinance into 36 months D/E loans after 1 - 2 years.

But yes it does seem heavily data driven.
Title: Re: LC Email: "The Next Generation Credit Model"
Post by: Fred93 on September 20, 2017, 03:08:05 PM
Makes no sense.

Agreed. 

My guess is that the current model was data-fit (seems likely overfit) over some time interval, and then these estimates were produced by somebody else using some prognostication method (like maybe average over some different, shorter, historical interval).
Title: Re: LC Email: "The Next Generation Credit Model"
Post by: Data Junkie on September 20, 2017, 03:42:44 PM
"We expect loan volume to shift toward higher quality grades (grades A and B) because some borrowers will qualify for lower interest rates under the new model..."

Kind of reminds me of another quote by Will Rogers:  "When the Oakies left Oklahoma and moved to California, it raised the I.Q. of both states."
Title: Re: LC Email: "The Next Generation Credit Model"
Post by: lascott on September 20, 2017, 05:10:18 PM
Makes no sense. If the losses are lower, they shouldn't be F/G grade loans. The Credit Grades are supposed to be credit rating, on a risk-return continuum, of the loans being offered, 'A' lowest risk and 'G' highest risk. By claiming F/G/60 will have lower interest rate than F/G/36 because it has lower losses, the Credit Grade is no longer a rating for assessing risk and return of loans LC offers. Grades are turning into just a few "independent" buckets in which they distribute the loans with no relationship between them. Basically, they are undermining their own argument about selecting loans based on Credit Grade only. People using LC automated investing will be shafted.

Data-driven but knowledge-unaware, most probably performed by people who see data just as bunch of numbers and have no basic knowledge of credit modeling and aptitude for interpretation what data telling them.
 
They expect lower projected annualized net credit loss on F/G loans on 36 months compared to 60 months: 17.95% vs 16.99%. My guess is that the prepayment for 60 months F/G loans is high enough that the credit losses is actually lower (compared to 36 months). Hence perversely they are able to offer lower interest rate on 60 months F/G loans as you still get higher net projected return with a term premium of app. 1% (9.84% vs 8.89%). I guess lots of 60 month F/G loans refinance into 36 months D/E loans after 1 - 2 years.

But yes it does seem heavily data driven.
A college course I looked at recently being taken by a family member had the various types of 'premiums' that go into defining rates (eg. grade is a range of rates).  Maturity (time) Risk Premium is one of the parts of the formula. It seems they could have made two maturity of grades i.e. C36 and C60.   Perhaps they simplified it to make it easier to understand for the borrowers and investors.

Quote
Premiums Added to k* for Different Types of Debt

ShortTerm Treasury: only IP for ShortTerm inflation
LongTerm Treasury: IP for LongTerm inflation, MRP

ShortTerm corporate: ShortTerm IP, DRP, LP
LongTerm corporate: IP, DRP, MRP, LP

k = k* + IP + DRP + LP + MRP.

k   = Required rate of return on a debt security.
k*   = Real risk-free rate.
IP   = Inflation premium.
DRP   = Default risk premium.
LP   = Liquidity premium.
MRP   = Maturity risk premium.

Related page with short defns: The Five Components Of Interest Rates
http://www.investopedia.com/exam-guide/cfa-level-1/quantitative-methods/time-value-money-interest-rates.asp
Title: Re: LC Email: "The Next Generation Credit Model"
Post by: Rob L on September 20, 2017, 06:27:06 PM
+1 on all of the above. "Makes no sense".
If you were LC would you not be embarrassed by this?
Title: Re: LC Email: "The Next Generation Credit Model"
Post by: SLCPaladin on September 20, 2017, 07:35:05 PM
Is this LC's version of a partially inverted yield curve? It does make intuitive sense that interest rates should be higher on notes with longer maturity. The fact that this is not the case is perplexing.
Title: Re: LC Email: "The Next Generation Credit Model"
Post by: Rob L on September 20, 2017, 07:48:48 PM
Is this LC's version of a partially inverted yield curve? It does make intuitive sense that interest rates should be higher on notes with longer maturity. The fact that this is not the case is perplexing.

Okay inverted yield curve; now I'm worried a recession may be just around the corner.  ???
Oh wait; these aren't treasury bonds; never mind.
Title: Re: LC Email: "The Next Generation Credit Model"
Post by: nonattender on September 21, 2017, 08:44:39 AM
Is this LC's version of a partially inverted yield curve? It does make intuitive sense that interest rates should be higher on notes with longer maturity. The fact that this is not the case is perplexing.

That part of it could perhaps be explained by factoring prepayment rates into the pricing model.  No way to know for sure, until they talk - or until a few years from now.  (Probably best to talk.)  I nominate Anil as Chief Inquisitor - with the power to torture quants into telling the Truth - as Peter seems to be too busy writing for the Huffington Post, lately... :)
Title: Re: LC Email: "The Next Generation Credit Model"
Post by: JohnnyP on September 22, 2017, 01:17:36 AM
I do not think 36 and 60 month loans follow the same credit model. For example, a 60 month E3 will have fewer deliquencies the first year than a 36 month E3. This is because they tighten the requirements for 60 month loans.
Title: Re: LC Email: "The Next Generation Credit Model"
Post by: bdbrooks on September 25, 2017, 01:40:49 PM
I think you guys fail to understand the full scope of things. The reason for normal yield curves has to do with increasing uncertainty about inflation, economy, etc. as dates get farther out. 3 years vs 5 years really isn't much of a difference. Especially when you are only considering F/G borrowers paying ~30% apr (who cares if inflation goes from up half a percent).

The driving factor is credit loss, and here you have to switch perspectives from an investor to a borrower. If you are a borrower in the F/G grades you typically need the money more desperately and have far fewer alternatives. These borrowers are like travelers walking on the edge of a cliff. 60 month loans means smaller AND MORE MANAGEABLE payments for borrowers. It is like increasing the walkway for that traveler from 24 inches to the dropoff up to 36 inches to the dropoff. I have worked in unsecured consumer credit, and manageable payments is necessity on the lenders part. Yes there will still be people that have no intent on paying, but most people would like to pay. Though once they feel like they can't live and pay, they will stop paying: ACH Revoked, cease and desist, account issue (aka they closed their account or issued a stop payment). Maybe the lower interest rate is meant to encourage those risky borrowers to take the option that will be more manageable for them.

This concept is also what infuriates me about LC collection process, but that is too long of story.
Title: Re: LC Email: "The Next Generation Credit Model"
Post by: SeanMCA on September 25, 2017, 04:28:15 PM
Now that I think about it, someone I know just refi'd their student loan to an online student lender. They were given two choices from the lender, a long term loan and a short term loan. The long term loan had a lower interest rate (maybe 75-100 bps if I recall). The loan amount was the same and there were no prepayment penalties. I told them to take the longer one and pay according to the shorter one's schedule. He had to call up the lender in advance to make sure there wasn't a catch and there wasn't. 

Title: Re: LC Email: "The Next Generation Credit Model"
Post by: rawraw on September 26, 2017, 01:12:25 PM
Now that I think about it, someone I know just refi'd their student loan to an online student lender. They were given two choices from the lender, a long term loan and a short term loan. The long term loan had a lower interest rate (maybe 75-100 bps if I recall). The loan amount was the same and there were no prepayment penalties. I told them to take the longer one and pay according to the shorter one's schedule. He had to call up the lender in advance to make sure there wasn't a catch and there wasn't.
That's is really strange. I'm trying to understand what economic incentive could prompt that. Perhaps they had too few loans for a securitization and used the rare to have enough supply to deliver to the pool. Only thing I can come up with

Title: Re: LC Email: "The Next Generation Credit Model"
Post by: henrywilson on June 15, 2020, 03:37:21 AM
Nice information!