Speed of Funding

[profwolfe]

Anyone know of a way to gauge how fast notes are getting purchased each day?  Some type of measure for the speed of (note) funding?  We're trying to cook up some measure of adverse selection and thought if a large chunk of the loans were purchased quickly on a given day - it might be indicative of more severe adverse selection issues for (the less speedy) investors. 

I think API users can grab listD and then compare to the note's issue date/time but I haven't found anyone tracking that besides Monja (https://www.monjaco.com/blog/3-types-of-loans-algorithmic-investors-buy/). 

Thoughts?

https://sites.google.com/site/swimwolfe/home/research

[Fred93]

What your suggesting is a measure of "popularity" rather than being a measure of "goodness".  There's no evidence that popular notes, ie those for which there is great demand, so get bought up quickly are the best for investors.

[profwolfe]

Fred -

I totally agree that some of the reason notes disappear quickly is due to popularity.  Intuitively, there's no reason to hurry and fund a bad (higher default given the yield) loan.  So when investors rush to a certain type of loan, I can think of three reasons:

a) They are herding toward what's popular - if that's the case, there will no difference in performance (default-adjusted yield) between those notes and the slowly funded notes - there is definitely some evidence of investors herding based on friend networks - (https://pubsonline.informs.org/doi/abs/10.1287/mnsc.1120.1560)
b) They are competing because notes are scarce and capital is plentiful - but then there shouldn't be any notes that have long funding times which we do observe
c) Investors know something others don't - which means controlling for all the other characteristics (which is tricky but doable) they have lower default-adjusted-yield

It's hard to know that (c) does/doesn't happen without a "funding time" measure.  That's what we academics like to do - make sure we know what "we know". 

[AnilG]

I published a 5-part series of blog posts in 2014 on similar topic. My conclusion: it is a losing proposition.

Mad Rush at Lending Club Loan Release Time: Part I https://www.peercube.com/blog/post/mad-rush-at-lending-club-loan-release-time-part-i

Mad Rush at Loan Release Time: Part II - Loan Performance with Time to Fund https://www.peercube.com/blog/post/mad-rush-at-loan-release-time-part-ii---loan-performance-with-time-to-fund

Mad Rush at Lending Club Loan Release Time: Part III - Delinquency Rate and FICO Score Change https://www.peercube.com/blog/post/mad-rush-at-lending-club-loan-release-time-part-iii---delinquency-rate-and-fico-score-change

Mad Rush at Lending Club Loan Release Time: Part IV - Interest Rate with Time to Fund https://www.peercube.com/blog/post/mad-rush-at-lending-club-loan-release-time-part-iv---interest-rate-with-time-to-fund

Mad Rush at Lending Club Loan Release Time: Part V - Loan Term with Time to Fund https://www.peercube.com/blog/post/mad-rush-at-lending-club-loan-release-time-part-v---loan-term-with-time-to-fund

[profwolfe]

Anil -

Really interesting work and well executed.  It looks like you used the API to grab the listing time variable (listD) and then used the disappearance from an API call as the measure of funding time - does that sound right?

A couple thoughts
1) The timing of these loans is right around the period when the platforms were splitting institutional investors and retail investors into their own markets.  It would be interesting to see if now that retail investors have their own dedicated market if this is still the case.  Even more interesting would be to look at the institutional (whole) loans to see if they have a similar issue.   

2)  Because you're comparing loans that are 9-12 months old, its really important to use statistical techniques like a hazard model to look at these types of questions.  Implicitly they help deal with a truncated sample such as this.  It would be different if we were to repeat the exercise on that sample now since all the loans would have reached full term - we could use some simpler statistical techniques like logit/probit or OLS.  It's also important to control for multiple characteristics simultaneously to really get down to this question.  Is default more likely given credit grade, term, dti, etc... for a particular loan.  Sometimes single sorts can be a bit misleading. 

If you're ever interested in swinging around to look at this again I'd be glad to collaborate with you on it...

[mikedev10]

isn't the greater problem now that there are barely any loans out there?  it doesn't feel to me that i'm not fast enough to get into those "juicy" loans that come out at 4 times a day - because checking 1 second after the release, there are barely any loans to choose from.

per this https://www.lendingclub.com/investing/investor-education/how-loans-are-allocated-to-investors the primary market api seems to be kind of a "leftovers" api.  i only started playing around with this stuff the last week of december, but it seemed like each release period just 4-10 loans leaked out.

looking at https://www.peercube.com/histperf/available_loans/lc this seems to be the case as well - the last 4 release periods here had 1, 4, 12, and 18 new loans released respectively.  so there's no great race/effort to pick the best of the 1000 loans issued in the past 24 hours; there's just a lazy stroll to review the 35 that became available.

[profwolfe]

Mikedev10 -

I guess I didn't realize how drastic the falloff has been.  You're right though, the retail market does continue to shrink - I put together some quick charts from the SEC data https://sites.google.com/site/swimwolfe/home/news-noteworthy/crowdfunding-active-research/falloffinretailnoteavailability. I would think this puts more emphasis on the secondary market investment opportunities right?

What we've been thinking about lately is how quickly the whole loan market loans get snapped up.  Occasionally those loans go unfunded and we were thinking a little about how that speed of funding in the institutional market might be an indicator of adverse selection issues for 1) slow institutions 2) retail investors on those rollover loans.  I know most of those rollover loans went away after the management change in 2016 but they still happen from time to time. 

[Fred93]

I would think this puts more emphasis on the secondary market investment opportunities right?

No, because the secondary market is so tiny that the "opportunities" there are incredibly small relative to the primary market. 

My experience is that I am able to invest 1000x more $ per day on the primary market than on the secondary market.

Individuals results will vary, depending on their investing criteria, pricing policy, etc.

So while the primary market volume has decreased, secondary market volume never existed.

[mikedev10]

Mikedev10 -

I guess I didn't realize how drastic the falloff has been.  You're right though, the retail market does continue to shrink - I put together some quick charts from the SEC data https://sites.google.com/site/swimwolfe/home/news-noteworthy/crowdfunding-active-research/falloffinretailnoteavailability. I would think this puts more emphasis on the secondary market investment opportunities right?

What we've been thinking about lately is how quickly the whole loan market loans get snapped up.  Occasionally those loans go unfunded and we were thinking a little about how that speed of funding in the institutional market might be an indicator of adverse selection issues for 1) slow institutions 2) retail investors on those rollover loans.  I know most of those rollover loans went away after the management change in 2016 but they still happen from time to time.

i'm curious why your chart to current day does not appear even lower - what does it look like for the last 3 weeks?  this chart (and maybe just because of the size and render etc) makes it look to me like there's still hundreds a day vs. the tens a day.

it would be nice to have access to the above mentioned "allocation 1" although i'm not sure my algorithm could compete with lending club's own internal simple based rules process.  my end to end process takes about 60 seconds, i could make it faster, although never as fast as an internal process with simple rules.  having hundreds of retail loans plop out a day would be nice though.

i asked about access to loans earlier or through a different way - it took some pestering as no one was getting back to me, but it sounded like you had to do whole loans then and the starting point for access to that was 50k.  i think i have quite a good loan picking methodology but i'm not on board with buying whole loans with my 50k, i'd be up for OPM and a few million to play with though :p

I would think this puts more emphasis on the secondary market investment opportunities right?

No, because the secondary market is so tiny that the "opportunities" there are incredibly small relative to the primary market. 

My experience is that I am able to invest 1000x more $ per day on the primary market than on the secondary market.

Individuals results will vary, depending on their investing criteria, pricing policy, etc.

So while the primary market volume has decreased, secondary market volume never existed.

i have been tweaking my model and fear the impact of the earlier hungry day where i had money in my account and it bought something like 200 notes in 24 hours... cuz they probably weren't top shelf picks...  in any case my current buy rate is 2-10 a day.