New (Especially Quality) Notes Really Trailing Off?

[avid investor]

Have had to lower the APRs of the loans I am funding to get anything that passes through my (admittedly stringent) filters.  Sideline cash was piling up.  But noticing that each feeding time only sees about 100+ new loans being dropped.  Frequently, 0 to 1 of those make it through my filters.  Anybody else having trouble staying invested at good return rates?  I don't do any A's or lower interest B's.

[Kombinator]

Yep most definitely last few days there has been a drop off in decent notes, but this happens in waves periodically, so nothing alarming yet.

[Lovinglifestyle]

I do mostly Es.  Yes to the same problems as above.  I noticed my last E1 at 18.99 was  on Oct. 23, and my first E1 at 18.54 on Nov. 2.  Don't know how the other E APRs compare, but they all seem lower.  That's good news for borrowers.  Now they can always make their payments on time, and I love when that happens!

[avid investor]

Yep most definitely last few days there has been a drop off in decent notes, but this happens in waves periodically, so nothing alarming yet.

Agreed that it happens in waves.  We've been LC investors for about 4 years, and using the API for over 1 year.  What I have noticed over the past 1 year is our sideline cash (un-invested cash or cash In Funding/In Review notes) has grown from around 2% to recently up to 3.7% of the overall account balances, which is approaching the intolerable.  I loosened the filters temporarily to get more "B"-grade notes through and get the cash down, but I wouldn't want to do that for long.

[rawraw]

By quality, I think you may mean mispriced.   If that's the case, one should never get used to relying on mispricing to last forever.  What you label as good may not be a realistic return.

[avid investor]

By quality, I think you may mean mispriced.   If that's the case, one should never get used to relying on mispricing to last forever.  What you label as good may not be a realistic return.

Actually, no I don't.  The number of what I would call "junk" applications - those that will be rejected at underwriting appears to be up.  We believe that we have the filters on all of our accounts set realistically.  We're not expecting "never a late, never a problem" borrowers at C, D, E rates.  We have been able to stay fully invested, however with fairly stringent requirements.  Are you suggesting that LC is improving their scoring algorithms?

[Lovinglifestyle]

By quality, I think you may mean mispriced.   If that's the case, one should never get used to relying on mispricing to last forever.  What you label as good may not be a realistic return.

In the PeerCube "Loans With Similar Risk Profile" box on the Loan ID page, do you consider some of those to be mis-priced as well?
This E3 is compared to 3 Bs and 2 Ds:   https://www.peercube.com/comment?loanid=35053672
Does that mean the Bs are priced too low, or the E3 too high?  The way my thinking goes is if a B is going to have payment problems anyway I might as well get the most return of and on my investment as is available in the risk range. 

Right now I'm undecided and waiting for the next wave of change instead of dropping back to BCDs.

Thanks for the discussion!  The subject is timely for me.

[rawraw]

By quality, I think you may mean mispriced.   If that's the case, one should never get used to relying on mispricing to last forever.  What you label as good may not be a realistic return.

Actually, no I don't.  The number of what I would call "junk" applications - those that will be rejected at underwriting appears to be up.  We believe that we have the filters on all of our accounts set realistically.  We're not expecting "never a late, never a problem" borrowers at C, D, E rates.  We have been able to stay fully invested, however with fairly stringent requirements.  Are you suggesting that LC is improving their scoring algorithms?

I've only worked with a few lenders who used credit score cards extensively.  What I do know from those lenders as they are constantly revised, as predictive powers of the variables change and more data is acquired.  I think it is foolish to assume the scoring remains static. 

And I'm not familiar with the first point.  Are you talking about loan rejection rates are increasing for LC? 

In the PeerCube "Loans With Similar Risk Profile" box on the Loan ID page, do you consider some of those to be mis-priced as well?
This E3 is compared to 3 Bs and 2 Ds:   https://www.peercube.com/comment?loanid=35053672
Does that mean the Bs are priced too low, or the E3 too high?  The way my thinking goes is if a B is going to have payment problems anyway I might as well get the most return of and on my investment as is available in the risk range.

Right now I'm undecided and waiting for the next wave of change instead of dropping back to BCDs.

Thanks for the discussion!  The subject is timely for me.

Anil is a smart guy and if I ever get the privilege of managing a P2P portfolio for a financial institution, he is one of three people on this forum I'd reach out to for their services.  That said, his model is just a model and I'm not very familiar with how it works.  To answer your question, you really need to have the probability of default estimate produced by the model.  You then can use this to determine whether something is priced too high or too low.

[Lovinglifestyle]

"To answer your question, you really need to have the probability of default estimate produced by the model.  You then can use this to determine whether something is priced too high or too low. --rawraw"

Thanks for reminding me.  It's easy to click on the individual loans in the PC Similar Risk box and then note the default rates, although I'm not sure if they were model produced.  I haven't paid enough attention to that.  I was letting my eyes skip over the default numbers.
Thank you.

[AnilG]

As I have never discussed Loans with Similar Risk Profile feature, I just wanted to put more color on how PeerCube determines the 'Loans with Similar Risk Profile'. PeerCube uses a clustering algorithm to determine what loans are similar to each other. The Loans with Similar Risk Profile box displays the five closest loans similar to the loan being viewed. There are multiple ways to use this for making lending decision.

• If your filter selected a loan, by reviewing Loans with Similar Risk Profile, you will be able to select similar loans that may not have met some of your filter criteria but are still very similar to the loan selected by filter according to the algorithm. It may just have missed a few selection criteria of your filter. This may works well when your filter is very tight and doesn't find enough loans on its own in lieu of cash buildup. The main drawback is that you will be building a portfolio of very concentrated loans that are similar to each other so they may move in tandem.
• If you prefer a very diversified portfolio, you can use the Loans with Similar Risk Profile to avoid investing in loans similar to one you are viewing.
• If you are after higher return, select the loan with higher interest rate among the Loans with Similar Risk Profile and the one you are viewing. For example, if you are viewing a C2 Grade loan and Loans with Similar Risk Profile displays a E4 Grade loan, you may consider selecting E4 loan that will give you higher potential return for similar risk profile.

In the past, a few PeerCube users mentioned that they leverage Loans with Similar Risk Profile feature by using a combination of method 1 and 3.

It's easy to click on the individual loans in the PC Similar Risk box and then note the default rates, although I'm not sure if they were model produced.  I haven't paid enough

The expected default rate on loan details page is provided by Lending Club and not generated by any model used by PeerCube. I typically ignore this number as you can be certain it rises with credit grade and interest rate. PeerCube internally uses this number a few times but none of those usage are directly visible to users.

[rawraw]

Is the cluster analysis done on just the credit variables or does it factor in past performance of those combination of credit variables?  If I recall, cluster analysis is often used to find similar groups of notes but generally a firm then has a cluster specific scorecard to evaluate the credit of notes within that cluster.   But I'm not very familiar with cluster analysis, so appreciate any insight on how it works

[Lovinglifestyle]

As I have never discussed Loans with Similar Risk Profile feature, I just wanted to put more color on how PeerCube determines the 'Loans with Similar Risk Profile'. PeerCube uses a clustering algorithm to determine what loans are similar to each other. The Loans with Similar Risk Profile box displays the five closest loans similar to the loan being viewed. There are multiple ways to use this for making lending decision.

• If your filter selected a loan, by reviewing Loans with Similar Risk Profile, you will be able to select similar loans that may not have met some of your filter criteria but are still very similar to the loan selected by filter according to the algorithm. It may just have missed a few selection criteria of your filter. This may works well when your filter is very tight and doesn't find enough loans on its own in lieu of cash buildup. The main drawback is that you will be building a portfolio of very concentrated loans that are similar to each other so they may move in tandem.
• If you prefer a very diversified portfolio, you can use the Loans with Similar Risk Profile to avoid investing in loans similar to one you are viewing.
• If you are after higher return, select the loan with higher interest rate among the Loans with Similar Risk Profile and the one you are viewing. For example, if you are viewing a C2 Grade loan and Loans with Similar Risk Profile displays a E4 Grade loan, you may consider selecting E4 loan that will give you higher potential return for similar risk profile.

In the past, a few PeerCube users mentioned that they leverage Loans with Similar Risk Profile feature by using a combination of method 1 and 3.

It's easy to click on the individual loans in the PC Similar Risk box and then note the default rates, although I'm not sure if they were model produced.  I haven't paid enough

The expected default rate on loan details page is provided by Lending Club and not generated by any model used by PeerCube. I typically ignore this number as you can be certain it rises with credit grade and interest rate. PeerCube internally uses this number a few times but none of those usage are directly visible to users.

Thank you very much for the additional information.  I like the #1 and #3 explanations and the bolded LC default model comment about ignoring it.  Thanks for explaining the one provided is not internally generated. 

Off the subject somewhat--I find risk index numbers of 130 and under are usually loans I've decided I want from the filtered ones LC gave me.  However, for a Major Purchase I'm not quite as confident that the added debt is taken into account (whether loan purpose is part of the scoring model).
 
I am also interested in this question from rawraw: "Is the cluster analysis done on just the credit variables or does it factor in past performance of those combination of credit variables?"

Thank you for your ongoing participation and help on this forum.

[RaymondG]

Have had to lower the APRs of the loans I am funding to get anything that passes through my (admittedly stringent) filters.  Sideline cash was piling up.  But noticing that each feeding time only sees about 100+ new loans being dropped.  Frequently, 0 to 1 of those make it through my filters.  Anybody else having trouble staying invested at good return rates?  I don't do any A's or lower interest B's.

This have been noticed since last week of September. I moved up grade of loan candidate by 2 sub-grades, from C5 to C3 and increased note size. Not much helpful. It became better in past 2 weeks and my cash balance is stable now. But, there are much less both 0-inquery and higher quality loans found by filter. Overall, quality of D and E loans are worse in past one and half month.

[AnilG]

The clustering algorithm takes into consideration the probability of default from past loans as weight for distance estimates. This enables the important variables to have more influence when doing similarity analysis. And, this approximation avoids two step analysis of first identifying cluster and then evaluating credit quality of cluster. The speed is more important as calculations are performed in real-time.

Is the cluster analysis done on just the credit variables or does it factor in past performance of those combination of credit variables?  If I recall, cluster analysis is often used to find similar groups of notes but generally a firm then has a cluster specific scorecard to evaluate the credit of notes within that cluster.   But I'm not very familiar with cluster analysis, so appreciate any insight on how it works

[avid investor]

Yep most definitely last few days there has been a drop off in decent notes, but this happens in waves periodically, so nothing alarming yet.

Resurrecting this thread as the trend has continued, and I've noticed some interesting variants.

First, it's been two full weeks with few notes added at each feeding time.  Typically, the number is in the 80 - 100 category.  I've also noticed that random checks at other times are seeing more notes dropping (very few, but some) that weren't there from the last "major" drop.

If loan demand hasn't waned, then perhaps a higher percentage are being diverted to the wholesale market?