|
Post by soereng on Jul 21, 2016 20:51:23 GMT
Haven't checked the public reports, but the private loan data is now useless for certain core analytics. Not only some numbers are missing now (OutstandingPrincipal - OMG- how could they remove this one? ?) but also some columns changed their nature. E.g. interest rate is now in percentage points instead of percentage value. I also mailed support and cc:'ed Partel, I think that's a "typical Bondora action": Smart idea, but awful execution ,-/ cheers, Soeren
|
|
|
Post by soereng on Jul 1, 2016 22:03:21 GMT
When I have some time (not right now...) I will do some analysis on Bondora public data. For the moment I did it on my owns loans only (some 6-digit figure, I think it's relative representative in terms of statistic measurements).
Below is a comparison of expected and real loss, measured on EAD1
PRR = Percentage of real recovery [ SUM(Recovery) / SUM(Coalesce(EAD1) ] ERR = see below, Expected recovery rate DAD = Days since default occured
PRR and DAD are weighted by EAD1 and is relative to SUM(EAD1) of the respective rating grade. The universe of loans are the defaulted ones (AD=1) after I bought or issued them. Loans are anything from 1 to 4 years old since buying them:
Rating PRR ERR DAD 53.1 185.3 413 A 8.0 70.2 164 B 3.4 23.9 205 C 9.2 58.4 226 D 12.4 63.3 282 E 8.6 32.6 382 F 4.8 15.6 444 HR 14.3 39.7 519
As Bondora does not state how long it takes to recover to reach their assumed loss or recovery rate, I assume something between 2 and 4 years (in Germany it can take up to 6 years, but it's unlikely anything is recoverd after 2~3 years). PRL goes down in my calculation by every cent of payments done after default. The age in days since default is the DAD column.
So analysis now...Assume recovery will take 4 years until the loan remainder is written off - which of course is a personal guess from my side - we calculate: Assume that recovery payments until end of a 4 year period match the recovery payments from default day until today -> So we scale the Real recovery rate (PRR) by the remainding time from dafault day t until t+1440 => The result is ERR, my "Expected Recovery Rate" which is in % of EAD1. I know it's a rough guess, but let's check the results. For B e.g. we have 3.4% of real recovery with an expected recovery of rounded 25%. Default is on average here just 205 days old, so if recovery is the same for next 6/7 of total 4 years, the recovery increases by ~7 to 25%. So the expected loss in my calculation is 75%. For D it's ~65% recovery with ~35% loss expected.
This somehow matches the "Loss rate if default" on the primary market published or is even better, at least I would say that the LGD number seems not too bad. Unfortunately LGD and PD are not published on the investment list or loan data set any more (only in primary market), so it's currently not possible to calculate a proper comparison of target and actual loss and recovery rate.
The only issue is, that the above analysis is more a rough verification of the LGD parameter. The PD parameter however is at least as interesting as it has the same impact on what Bondora treats as "EL" - and the analysis still depends on the guessed time the recovery will take.
My gut feeling said before the analysis that "their" EL is far too low, however LGD seems to be okay, so I will take another approach to check the PD when I have some time on their full data set... Maybe the gut is not always as right as expected, but let's see later on ,-)
Regards, Soeren
PS: If you want to recalc the numbers on your portfolio: I have a MySQL database table with all columns from the investement reports. Select first all defaulted loans then apply the following SQL (some extra colums and different colum names are here used):
SELECT Rating , ROUND(SUM(EAD1), 2) AS EAD1 , ROUND(SUM(Recovery), 2) AS Recovery , ROUND(SUM(Recovery) / SUM(EAD1) * 100, 1) AS PctRealRecovery , ROUND(SUM(Recovery) / SUM(Coalesce(EAD1, PurchasePrice, InvestmentPrincipal)) * 100 * (1440 / (SUM(DATEDIFF(NOW(), Default_StartDate) * EAD1) / SUM(EAD1))), 1) AS pct_Excpt_Recovery , ROUND(SUM(DATEDIFF(NOW(), Default_StartDate) * EAD1) / SUM(EAD1)) AS DefaultAgeDays FROM tmp_defaults GROUP BY Rating ORDER BY Rating;
|
|
|
Post by soereng on May 3, 2016 21:25:35 GMT
"In the good ol' times"...this meant: How many days is the loan due. As at day 61 the loan turned into default, a strategy to avoid default loans in the portfolio would be to put them at day 57 on the 2nd market with reasonable discount. At least this was my strategy.
Bondora however did a significant change in how they determine defaults, e.g. they first apply overdue payments to outstanding principle, when that's zero, the apply it to outstanding interest. I haven't understood fully, when the CDD goes back to zero. The various other fields "InDebtXXXdays..." however don't get reset to zero once all is paid, so it's impossible to use them as alternatively. I also have various loans where CDD is >>100 days, e.g. because some restructuring whith still outstanding took place. Short story: The field is now useless, any other ideas are welcome. I've sent this request already to support twice, second time this Monday, but did not receive any answer then a questions why I need this ,-)
Cheers, Soeren
|
|
|
Post by soereng on May 2, 2016 19:40:02 GMT
Hi all,
until some weeks ago the "CurrentDebtDays" in the account as well as full data set was a perfect indicator to check wether a loan would be shortly before default. Bondora then changed the way how to determine the default event and how they would count those days - and voilĂ , the number is now useless.
Has anybody another good way how to determine if a loan will go default e.g. within then next 2, 3, ... n days?
Thanks, Soeren
|
|
|
Post by soereng on Mar 2, 2016 21:12:28 GMT
Just received an answer from support regarding priority of bids which they mentioned in their 2nd most recent newsletter, but did not explain:
Compared to other single bidding methods Portfolio Manager has priority access to loans. Current priority order for investing is: (1) anyone who takes the full loan and bids over API; (2) Portfolio Managers; (3) everything else.
This might be of interest for some of you.
Regards, Soeren
|
|
|
Post by soereng on Feb 8, 2016 19:30:39 GMT
A very comprehensive answer, thanks jevgenijs.
After reading my email again, I need to add one thing: The difference of course to a bond is, that the "bond" is virtually secured by loans. So in bancrupty case the guarantee is lost - but the loans aren't as jevgenijs pointed out. So it makes the loans somehow better then the bond.
However, as beeing involved with Trustbuddy for myself at the moment there is some huge operational risk involved with the administrator taking over the business. Depending on the reason for the bankcrupty the administrator (usually a lawyer) needs first to secure the bankcrup organisation - it takes some time until he understands that the major asset's of the clients are not the company itself, but the loan portfolio it manages ,-) However, this applies to really EVERY lending platform, so nothing very special for Twino then for any other platform.
Cheers, Soeren
|
|
|
Post by soereng on Feb 6, 2016 11:39:38 GMT
Hi nick,
your assumptions are right, although some extra details should be noted.
A loan has a counterparty risk aginst the original borrower. I.e. the borrower's ability to return principal and pay interest totally depends on the borrowers credit rating. BUT as loans are guaranteed by Twino (not Finabay!) the counterparty risk moves completely to Twino. With fixed interest rates (since I invest, I've never seen other rates then the 12.9%) this reflects in fact a bond against Twino at 12.9% with 1 month time to maturity. The baseline is: If one invests at 12.9% at Twino he/she should decide if this reflects the actual counterparty risk for an unsecured bond. Looking at current bond market, 12.9% reflects a high risk investment - one should be aware of this.
There is however an additional operational risk, what it means could be seen when looking at the Trustbuddy case: Even if the guarantee would in fact guarantee a 100% capital payback, a bancruptcy case could put an impediment on actually returning the "safe and guaranteed" money. If the company goes bancrupt operation is stopped until someone takes over, no payments are made at all. This cost for this specific risk is part of the 12.9%, so the counterparty risk alone is less then 12.9%.
The P2P market evolves currently towards the "payback guarantee" model - e.g. you can see this at Mintos where quite a number of providers there offer this type of loans now. But only time will show, if this business model will work as providers thought of. As you have mentioned, if the actual default rate of loans for a certain provider is higher then he initially calculated - which is always a tricky part for any new business without proper historical data to calibrate a default model against - he might move into bancruptcy after a while. So even the "guarantee" model bears some risks that - interestingly enough - a non-secured lending platform like Bondora does not have.
Regards, Soeren
|
|