Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.


Messages - hoggy1

Pages: [1] 2 3 ... 27
1
As I understand it, a Delinquency is 30 or more days late and I don't count one in the last two years against the borrower. By contrast a major derogatory is more than 120 days late and I don't buy notes with any.

2
Introductions / Re: Hi from Orca Money, a UK aggregator
« on: December 07, 2015, 02:03:57 PM »
How do we get to your beta site. You haven't given us anything to comment on?

3
Introductions / Re: Hello from Angie
« on: December 01, 2015, 05:17:51 PM »
Law school? We often need legal help here. What do you know about security law?

4
Introductions / Re: Hello from Angie
« on: November 26, 2015, 04:19:09 PM »
Welcome aboard. We need more female voices to keep the conversation civil.

5
General P2P Lending Discussion / Anyone know where Core went?
« on: November 25, 2015, 12:25:54 PM »
Hasn't posted since March 6th?

6
Investors - LC / Re: How could LC update their website so casually?
« on: November 25, 2015, 06:44:37 AM »
Everyone will be thankful I'm speechless.

7
Thanks for the excellent post. ANAR has some problems for sure but it has some value.
....
 I keep an eye on it over the short term and that has been informative. Over the past 6-8 weeks mine dropped 50 basis points and my portfolio age is about 17 months. It was my first indication that I was having a bad month and it turned out I did.


Well, I will at least conceede that: relative changes over time are probably (I'm not promising anything) in the right direction.

... If all loans were either 36 month or 60 month that would be fine. While we may have only one set of loan status migration percentages (over 9 months) for both 36 and 60 month loans (maybe they are roughly the same)...  Said another way a 10% ANAR portfolio with all 36 month notes and an average age of 15 months is the same dot on the graph as a similar portfolio with all 60 month notes (or any mixture of 36 and 60 month loan lengths). In this way the graph is apples to every other fruit.


You are correct; the mix of 36 and 60 month notes is another complication I considered but did not delve deeply into (there are just too many). As for the different discounts for 36 vs 60, we can be certain they can not be the same or the default curves for the first 36 months would have to be the same for both and they are not. You can see the cumulative monthly default for 36 month loans on the last line of the 36m amortization schedule in reply #121 above. It is 15.18% as discussed in the text. The cumulative default rate for 60m notes at month 36 (a hidden row) is 18.34%. So the "borrowers" themselves are different and behave differently.

an adjustment for current notes would definitely be very different.

I am so pleased someone got this point. Our personal data point does NOT include any discount for current note, BUT the population points for all other accounts are corrected. That is why everyone reports their performance not just above average, but 1-4 SDs above average.

If you take a look at the comparative spreadsheet by lascott it's clear that P2PQ performance is heavily influenced by the percentage of 36 and 60 month loans owned.

I forgot to mention one important point. The data I used and illustrated is precisely the same data that P2PQ used; in fact Turing of P2PQ was the one who told me how to download the data

Throw in folio trading and we have fruit salad.

Nicely said but I am at least 2SDs above the salad. The numbers had to be nonsense.

As an improvement LC could possibly compute a WALL (weighted average loan length)  factor and apply it as a current note adjustment to compute a new AANAR. This would not address rawraw's valid point regarding the differences in portfolio riskiness but it would at least make the graph apples to apples and would be a pretty easy fix. We already have a button to display similarly risky portfolios (WAIR).  This is no substitute for IRR but a nice improvement with no chance of being implemented. :)

LC knows how to properly adjust each users ANAR; they just don't do it. If we could input a discount rate for current notes, even we could compute the correct numbers to use. You would have to do what I did for E grade loans for all the other grades. Then you would make two tables that would work for everyone. The tables would be age x grade (2 of them, one for 36 and one for 60) containing the discount rates in the rightmost column of my amortization tables. I described someplace in the text how to interpolate between the two based on you relative percentage holding of each. But you could go further and created the weighted average for each loan grade based on your holdings.

Anyway, thanks for the thoughtful response.

8
If you are trying to adjust results, just build an allowance like a bank will.

There are two pools.  Individually evaluated for impairment (late notes) and collectively evaluated (current notes).

The way to build this reserve for LC is easy.  For collectively evaluated, take the average charge off of each grade over 1 year and apply it to the balance of current notes.  Then for individually evaluated, apply LC's loss estimates (possibly modified by a fraction of avg charge off for the grade to total charge offs to adjust for grade levels).  This total balance would be your expected losses in 1 year and 'fair value' your portfolio unless the economy changes.

And of course ANAR is going to be more inflated the more risky your portfolio is.  There is higher embedded losses in the current portfolio that are not being accounted for.

I didn't know until last week that the charge off data was available to us. That's what made this investigation possible. The adjustment you suggest woud certainly work. But I am and have been happy with my ability to understand my returns. I just never had the time to figure out how and why LCs repoted ANAR was so much higher than what i knew the returns to be.

My post was a public service which took quite some effort as a caution to newer members or some who struggle with the math and statics and might mistakenly depend on LC data which provides a false picture for the reasons stated and several others I haven't even gotten to yet. The "What does yours look like" thread is the 3rd most often viewed and the 2nd most replied to thread of any threads begun this year, so I assumed there was a large mislead audience to worry about.


9
I also understand the difference. I am not trying to show their equivalence. I am just try to demonstrate that what LC reports on you screen does not in any way, shape or form represent anything meaningful about what you are earning. If you sell on Folio as I do the numbers are even less meaningful. One would have thought once you included what LC reports as my combined results (including trades) might have had some relevance. I haven't gotten into that yet, but you won't have to look hard to see that that is not the case either.

10
Investors - LC / What does yours look like: This can't be right
« on: November 21, 2015, 03:40:51 PM »
The correct adjustment

My thanks to Anil for bringing this file to my attention.
https://www.dropbox.com/s/0l5tijquoma4k3y/Cumulative%20CO%20Rates%20Projected%20Loss%20Curves%20v15_OCT2015.xlsx?dl=0

This is LC loss data which shows for each quarter and every loan grade the monthly losses (hidden lines) and cumulative monthly losses (select particular loan grades or combinations from pivotTable in upper left corner). For a particular loan grade the raw data is hard to work with because for loans that have fully matured (issued more than 3-5 years ago) there are few loans in any particular grade/issue_quarter so the statistics are poor. But included on the sheets “36M Future…” and “60M Future …” do what I believe after filtering and averaging the raw data, likely reflect the expected monthly charge offs for each grade. (Below)



I chose E only grade loans because my portfolio most closely resembles E notes, but the method I will employ will work for any grade or combination of loan grades.



So let’s look at recent E grade performance. My account has a weighted average interest rate of 18.56%, made up predominantly of E grade notes which average of 20.12%. So the analysis that follows uses E grade data only. You can see below LC data for the last 10 quarters (30 months).
https://www.lendingclub.com/info/demand-and-credit-profile.action



Whenever I have a tough question, I run the amortization schedule. Let’s look at the amortization schedule for E loans subject to the monthly losses above.


PLEASE NOTE HIDDEN ROWS AT HORIZONTAL LINE
There are two amortizations at two different interest rate, 20.12% and 9.96% (Orange) from the LC table above.  Columns labeled “Multiplier” are the fixed multiplier (1 + decimal monthly interest) that will when applied to the prior monthly account value give you the current value (remember E grade loans). In the second “Value” column we see the amortization of $100 for 36 months at the reported ANAR rate of 9.96%. In the first “Values” column is the amortization at the 20.12% average note rate BUT where the multiplier is adjusted monthly for the charged off loans extracted from the LC charge off file (labeled “Prin Loss”). When adjusted, both these should result in the same final account value after 36 months if LCs figures are correct and you can see that they do (green). But it took more than just that adjustment to get this result which confused me for some time. You actually lose twice. You lose the interest income as soon as the loan stops performing, and then you lose the principal 4-6 months later. Before this realization I could not get the final values to match. When you add together the income loss (advanced 4 month) to the principal loss as it occurs and adjust the multiplier by subtracting both, you can see the values match almost exactly.  This set of LC data is at least internally consistent.

So how much would we have to discount our CURRENT notes to get the right correction to our individual data point? That turns out to be easy. The column “Accumulated Loss” shows the running sum of monthly defaults. You can see after 36 months, a total of 15.18% of notes will have defaulted. So we need to discount our CURRENT notes by the fraction of defaults already encounter divided the total defaults to be expected. The second to last column contains this fraction. Clearly, if we had a brand new account with all newly purchased notes, the adjustment would be 10.16% (green, difference column). The last column show the discount that should be applied for each month of average age[ (1-.03)*10.16 = 9.87 ]. Notice current notes in my 13 month average age account should be discounted is 3.81 percent.

That is not quite the end of the story. My account (and most others in the population) is composed of 36 and 60 month loans. So we need to look at the 60 month amortization as well.


The blue dots above are the raw data. I told you it was all over the place. The red line is what LC says the “Future” monthly defaults should look like based on this historical record. The dashed line is a 5th order polynomial fit using least squares to the raw data. As you can see it is a reasonably close approximation to what LC says we should expect. However neither collection of default projections will cause the final values after 60 months of amortization to equal LCs reported ANAR (Es for the last 30 months). The losses had to be higher. The green line represents a corrected default rate 28% higher than reported by LC that does properly account for the difference between average note rate of 20.12% and ANAR of 9.96%. The corrected amortization is below.


PLEASE NOTE HIDDEN ROWS AT HORIZONTAL LINE

The blue line in the chart above is LCs reported default rates, and the red the adjusted rates I had to use to make the amortization of 60 month notes reconcile. So on the surface of it LCs data for 60 month notes is internally inconsistent. Their reported default rate could not adequately account for the difference between the average interest rate on notes and the reported adjusted NAR. There are some possible explanations. The defaults and recoveries are reported in the quarter the notes is issued, not in the quarter when the note actually defaults or what might be a substantial time later when a recovery might be made.

So let me return to the subject of the proper amount to discount our CURRENT notes. My account is made up of 85% 60 month notes and 15% 36 month notes. For 13 month old average age account, 60m notes should be discounted 6.1% and 36m notes 3.81. My composite CURRENT notes should then be discounted by 5.76% (0.85*6.1 + 0.15* 3.81). Remember according to LC, my ANAR was 15.11%. If I add a discount (if I could) to my current notes of 5.76% my real ANAR would be properly reported as 9.35%. That is a 0.38% difference from what my XIRR properly reports at 8.98%.

There is more to say, but I’ll wait for my punishment from the better mathematicians and statisticians on the board for not having a brain in my head. But in the interim, please don't put any confidence in, or make any important decisions based on what LC says you are earning.


11
Investors - LC / What does yours look like: This can't be right
« on: November 21, 2015, 03:40:05 PM »
The lending Club Adjustment


Below is my plot of the raw data. The Orange dots are the mean, the darker blue dots are one SD above the mean, and the black line is 3 SD’s above the mean. The big red dot is me, 4 standard deviations above the mean. Each one month period has more than 1000 data points.
http://i1027.photobucket.com/albums/y335/hoggy13/raw%20UnAdj_zpsvc3jsvpx.png

http://i1027.photobucket.com/albums/y335/hoggy13/raw%20Adj_zps9esk3glg.png

If you download the two files and flip between them (or just look closely) you will see the young population falls 4+% with LCs adjustment. But my adjustment was less than 1%, and 3% of my notes are troubled. Why do we see such a large adjustment in the population and small one in my personal adjustment? Surely not for a few troubled loans discounted at LC’s stated default rates? In fact 3 month old loans actually have no real charge offs yet. That is why LC uses a 9 month sliding window to measure charge off rates (See description below).

http://i1027.photobucket.com/albums/y335/hoggy13/LC%20loss%20adjustment_zps7vallmk9.png

There is a sensible reason. Because it is not just troubled loans (IGP, 16-30, 31-120, default) that will ultimately charge off as suggested by LC table above. Some fraction of CURRENT loans will also end in default and LC’s adjustment to the population points comprehends this. Now note that we cannot add this effect to our custom adjustments (not that LC “understanding your returns” page uses them anyway) because LC won’t allow us to enter a discount rate for current loans. The current loan discount is fixed at zero. (see below)



It is not true that 0% of current loans will eventually charge off. Exactly how many will depends on the grade makeup of each portfolio and decline over time as the portfolio matures.  Whether LC actually takes this in account or not I don’t know but they certainly have the data that would allow this type of correction.
LC stated charge off rates from each troubled category cannot explain what we know from LCs own data the ultimate default rate will be (again differences by note grade), but we can get an idea.

12
Investors - LC / What does yours look like: This can't be right
« on: November 21, 2015, 03:39:21 PM »
MY DATA

Here is my data as of a few days ago when I began this study.



Notice my ANAR of 14.32% The chart is below.


However my ANAR of 14.32% is changed to 15.11%, and the plot shows it there. What is the discrepancy?
I have custom adjustments that discount all problem notes 100% and that results in the 14.32% reflected on the main page. But the adjustment made on the "understanding your returns" page is using the default LC discounts. It’s a small thing but caught my attention and started this investigation. Notice this error makes my return look better.

If I’m earning 14.32% since inception why does my XIRR (in excel) report I’m earning 8.98%, a 5.34% difference? I'll get to that answer shortly. For now just notice again LC plotted results makes my account look better against the population.

When I plug these numbers (either one) into PeerToPeerQuant’s percentile calculator here is what was reported: 99%, 99%, 99%.  Below is Scott’s latest version of his spread sheet.


My account would be the best performing account, even better than jonpildis. Thanks to help from RaymondG and Turing I was able scrape the raw data (36000+ points) from the performance chart. Armed with the raw data, even P2PQ underrated me: the number should have read 99%, 100%, 100%. Yeah, my account is the absolute highest returning account in a one month window around 12.5 months.

Below are the means and SD from the adjusted population.




As Scott added each of the members, I was certain something was wrong. Only 2 of 24 accounts are below 50% of similarly aged accounts. Now I have no problem believing member of this board might outperform the average LC investor, but almost everyone outperformed by 2 or more standard deviations. My data point is 4SDs above the mean. Too Lake Wobegon for me.
I thank Scott for sharing the data with me. Here is your performance in SDs:


What I will demonstrate is that LC makes different adjustments to the population data than they make to your individual data point.

13
Investors - LC / What does yours look like: This can't be right
« on: November 21, 2015, 03:37:14 PM »
I have already tried once to explain why the data being shared in this thread is suspect (reply #72, page 5, Aug 31). I am certain what LC reports for my account is totally nonsensical. But the post is so popular, I thought I must be missing something. I re-read the entire thread; twice; three times. Certainly these highly respected members are not just posting to boast. Scott's graph is beautiful, but that alone can't explain the interest. Surely members think there is valuable information that can help them. I hope you are right and that someone can explain it to me, but I've been studying now for several days and have concluded there are more problems than I originally thought.

I apologize in advance for the length of this discussion. I will break it up into multiple post following this index of deficiencies:

1.   My data
2.   The Lending Club adjustment
3.   The Correct adjustment?


Skip the next 3 post if you are not interested.

14
Investors - LC / Re: Haven't seen this before
« on: November 19, 2015, 02:47:10 PM »
Thanks Edward,

I missed that. I tried searching the forum for joint but got nothing. I must typed it incorrectly because it works now.

Steve

15
Investors - LC / Haven't seen this before
« on: November 19, 2015, 12:42:51 PM »
https://www.lendingclub.com/browse/loanDetail.action?loan_id=65424600&previous=browse

Look at this loan. It lists "Individual Gross Income" and "Joint Gross Income" together with "Joint Debt to Income".

New?

Pages: [1] 2 3 ... 27