[Fred93]

I buy notes on Folio, using the API.

When I place a buy order on Folio, the responses I get are a mix of NOTE_NOT_AVAILABLE and SUCCESS_PENDING_SETTLEMENT.

I'm going to be talking about just those orders that resulted in SUCCESS_PENDING_SETTLEMENT.  Those are the ones I think I bought.

A significant fraction of those never end up in my account.  I've been wondering why.

A few months ago I took a day's purchases, from the log written by my software, and checked them against my notes file to see how many actually arrived and how many did not.  I got a 50% failure rate!  Yipes.

I did this check again last month, and got a 38% failure rate.  38% of the time, the note does not show up.  (For the purposes of this little experiment, 38% and 50% are approximately the same number.)

I've written to LC a couple of times about this.  Most recently their response has been that the specific examples I sent them were all cancelled by payment.  I have quizzed them about the payment timeline, and how they decide which orders to cancel.  They say that the window of opportunity for payment after I place an order is only one day.  That would be 1 day out of about 30 days (because payments come about once every 30 days), so that would be 3.33% expected to fail.  But of course they only process payments on weekedays, so one of the 5 weekdays absorbs the hazard of the two weekend days, so maybe that would legitimately get them to 4.66%

Long way from 50%.

So I'd like to hear the experience of other LC folio buyers.  I'd like to know if I'm alone in this.

What fraction of successful orders fail to arrive in your account?

Do many of you see a large fraction of buy orders cancelled, ie say SUCCESS when you place the order, but it never makes it to your account?

LC is asking me questions aimed at a theory that my selection process accurately predicts who will pay on the day I order.  I've looked over what I'm doing, and I just don't think that's possible.

PS: Jheizer made a brief comment in another thread that he had seen "like 50%" failure.  I can send that to LC, but I'd prefer to send comments from several different people, showing that many people have similar experiences.  That will allow us to get past their theories that this is caused by some strange thing I'm doing.

[Dave101]

Since I started tracking I've bought 519 notes, 175 of which are not presently in my account. A few are in pending and a few are sold, but I doubt those add up to 175. Still not close to 50%.

I'll re-tool my process so I can get an exact number and check back in in a couple of days.

Incidentally... any idea why it takes notes 4-6 minutes to be visible in the API listing, after which time several are not available? It seems to coincide with how often the downloadable CSV file is updated. Screen scraping doesn't seem to be much better. I know notes sold are available for purchase instantly, since I've tested it across accounts.

[TravelingPennies]

I found a simpler way to explain what the cancellation rate should be...

Payments are processed 5 days every week, and there are 52 weeks in a year.  5x52 = 260 payment processing days.  A loan generates a payment 12 times a year.   That's 12 payments in 260 days, or a note should see a payment about 4.6% of days. 

Therefore, note purchasers should see only about 4.6% of their note purchases cancelled for payment.

[TravelingPennies]

Getting a bit more serious, I decided not to rely on the stats I previously got by manually counting one day here, a half a day there, etc.  I wrote a program to process my log files for the past 6 months, and produce complete stats...



If the hazard is truly one day, as LC says, then the failure rate should be 12/(5x52) = 4.62%

The last couple of months look to be about 5x this number.  Months before Nov 17 have lower failure rates.  Don't know why.  I've asked LC whether they made a change in this area in Oct '17.

[AnilG]

Your analysis assumes loan payments and issuance are processed equally every day and not batched. I seriously doubt that is the case. Also, typically loans don't stay in payment processing for only one day. From payment due date to when payment is processed, completed and the payment processing flag is removed could be as much as 5 days.

from: Fred93 on January 12, 2018, 04:40:27 AM

[TravelingPennies]

from: AnilG on January 12, 2018, 05:24:28 AM

[apc3161]

I too had a similar problem with canceled payments, severe problem. I also talked to LC and we decided it was probably secondary to payments being too close to the sale date, such that payments automatically cancelled the sale. What I did to avoid this is stop purchasing notes (on-time notes) with payment date > 25 days (i.e. they have an upcoming payment). I did this recently. I'll see if this helps improve the situation.

[TravelingPennies]

I'm thinkin I should add the "days since last payment" field to my log file.  That way I will be able to see whether this correlates to the notes that "disappear" during settlement.

[Rob L]

I want to add another voice and say I feel your pain so to speak. The comedy of Laurel and Hardy, Abbot and Costello, etc. are sadly in play.
The best of luck working this out. If the API worked I'd consider dabbling my toe in the water and maybe others would too.
When things are not a priority they can become an embarrassment; especially for the "tech" in "fintech".

[hdsouza]

I used to purchase notes with nsr investing earlier and they used to get about 1-2 notes/day. At that time I never really looks at the failures, although I did not see a few failures.
 
After recently automating  the process I purchased a bunch of Notes over the last few days and I noticed Like Fred that about HALF the notes were cancelled. This is after I received the SUCCESS_PENDING_SETTLEMENT.

[TravelingPennies]

I think I have figured out the order processing timeline, and why so many buy orders are cancelled. 

In the past I'd been looking at historical data for months gone by, and LC gives us no way at all to track down what happened for notes we bought in prior months but did not settle, so I couldn't look at note by note details.  Generally, you can't look up the payment history on the note unless you own it, or it is offered for sale on Folio.  One exception is notes that you've tried to buy during the CURRENT month.  The "my account" page on Folio has links to the note detail for all these notes.  However, if you let time slip into the next month, all these links disappear.  This page provides an opportunity to look at the payment details of notes for which a buy order has been cancelled!  You've just got to do it before the links disappear!  That's what I did tonite.

Tonite I looked at notes I bought this month, looking up the status of the purchase transaction, and the note payment dates on the note detail pages for a considerable number of notes.  Lots of notes.  My fingers are very tired.

I now believe I know how the purchase timeline works.  Its pretty wonky.  We all know that lots of sell orders are cancelled for note payment, and I always figured this was done to protect the buyer.  Not so.   The sell orders are cancelled way too late to help buyers.  Sale offers are NOT cancelled while payments are processing.  As a result of this, the buyer has quite a hazard.   

Loan payment processing timeline...

Lets reference dates to the note "due date".  I'll call this 'D'.  This is when the payment processing window begins.  No payment has been made yet.  This is simply the day when a process starts.  On that day, or the closest business day, LC sends a request to the bank, via the ACH system, to grab some money from borrower's account.  A few days later, if the money has arrived, LC considers the payment "settled".  This is the "settlement date".  Lets call it 'S'.  (Due date and settlement date are shown on the note detail page.) 

Settlement date S can be = D+3, D+4, D+5 or D+6.  This occurs because LC only works on weekdays.  So if no weekend or holiday intervenes, settlement is D+3.  If a holiday is in there, then it's D+4.  If a weekend intervenes, it's D+5.  If you have both a weekend and a holiday, then settlement is D+6. 

So payment processing takes 3,4,5, or 6 days.  ...but there's more!

Now, the LC/folio note order processing timeline...

When I place a successful buy order on day 'B', LC does not process that the same day.  They process it the next day.  You can see this on your "my account" page.  The "order date" is always 1 day later than the day you placed the order.  Recent email from LC claimed the settlement process runs every day at 3PM, but this is misleading.  That may be the time that the software runs, but the time cutoff it uses is midnight.  Buy orders up to but not including midnight on 1/1/18 are logged by LC as ordered on 1/2/18.  In other words "order date" shown by LC is always B+1.

Now, when is a buy order cancelled?

A buy order is cancelled if B is in the range [D-1 thru S] inclusive.  In other words, the hazard window is one day more than settlement.

As an example, suppose a note had a due date on the 1st of a month.  Suppose there's a weekend and a holiday in there, so settlement is the 7th of the month.  Now you come along on the 7th and see this note listed for sale, and you buy it, and get a SUCCESS_PENDING_SETTLEMENT response.  Cool.  You think you've bought a note.  You haven't.  That order is gonna get cancelled for sure.

This is astonishing.  Its goofy that they know that they're gonna cancel the order if you buy, and they have known this for SEVERAL DAYS before you buy, but they let you buy, and then cancel.  Extra processing for them.  Extra processing for you.  Things are harder to track.  The right thing to do would have been to remove the sell order from folio once the window began, so that buyers didn't waste their time.  That's not what LC does. 

I've had orders where I bought on D-1 and the order was cancelled after I bought.  (For the record note# 126688481.  I bought 1/1/18.  LC order date 1/2/18.  Due 1/2/18.  Settle 1/5/18.)

I've had orders where I bought on D+6 and the order was cancelled after I bought.  (For the record note# 119577848.  I bought on 1/2/18.  LC order date 1/3/18.  Due 12/27/17.  Settle 1/2/18.)

So the bottom line is that for each payment, ie every month, each note has a window of 4,5,6, or 7 days in which if you order in that window, the order will be cancelled.  Most of the time its 4 or 6 days.  The other two possibilities require holidays, and there aren't very many of those.  There are 4 days of each non-holiday week which have 4 day windows, and three days that have 6 day windows.  From this we can calculate that the average window is 4.857 days.  (Not considering holidays, which would make it a little bit bigger.)

Payments come 12 times per year.  Each payment has a 4.857 day hazard window.  That's 12x4.857 hazard days per year.  (The occasional extra payment will bump this up a little, as will holidays.  Missed payments will bump it down a little.)

D can occur on any day of the year.  Yea, I know LC only processes on weekdays, but we've already handled that with all that stuff about settlement sometimes being D+3 and sometimes D+5.  So there are 365 days to consider, and ...

Probability that a note order is cancelled = 12x4.857 / 365 = 18.46%

Now you may recall that I've had some periods when I calculated much higher failure rates for my own account.  Keep in mind that this calculation is an average.   Different months have different numbers of holidays and different alignment to weekends.  Also the luck of the draw will cause some variation.  Finally some of the much higher numbers I calculated a few days ago were from manual calculations which because of the difficulty of calculating by hand were for single days, which of course will vary much more than months.  There is probably also some day-of-week bias when loans originate.  This number above is just an average calculation all other things being equal.

Here (again) are the calcs I did for the last 6 months on all buy orders in my account.

November and December have a couple of holidays each, so they're expected to be a little higher.  These numbers look like they sorta agree with the theory.

The marketplace is simply designed so that about 1 out of 5 buy transactions will fail.

[TravelingPennies]

from: Fred93 on January 14, 2018, 08:21:29 AM

[TravelingPennies]

from: Dave101 on January 14, 2018, 10:53:29 AM

[TravelingPennies]

from: Fred93 on January 14, 2018, 08:49:24 PM

[jheizer]

I just tried to do a really rough guess on mine to see how it actually compared to my "I was annoyed too 50% guess".  I assumed for every time I logged out that a purchase ran that I only bought a single note.  I do see some had 2-3, but also some errors.  So if they do cancel each other out to average non per file, then about 20% of mine failed.  That's for the time period started about 2 weeks after the May mess in 2016 through October of that year.