|
|
Post by rahafoorum on May 8, 2020 19:27:10 GMT
Bondora shares some "interesting" stats about recovery rates, but there are actually good ones as well. I put together a quick guide on which of the figures show what exactly. Perhaps someone will find this helpful.
|
|
|
|
Post by badlord on May 12, 2020 18:41:03 GMT
Bondora shares some "interesting" stats about recovery rates, but there are actually good ones as well. I put together a quick guide on which of the figures show what exactly. Perhaps someone will find this helpful.
rahafoorum.ee/en/bondora-stats-crash-course-recovery-rate/Disclaimers: My investments on Bondora during 2010-2016 ended with a total annualized return above 28%. I worked at Bondora nearly 2 years and quit at the end of 2015, when it became clear that the company had chosen a path I did not want to participate in. May I ask what you were not comfortable with?
|
|
fric
Member of DD Central
Posts: 199 
Likes: 79
|
Post by fric on May 13, 2020 5:53:53 GMT
Bondora shares some "interesting" stats about recovery rates, but there are actually good ones as well. I put together a quick guide on which of the figures show what exactly. Perhaps someone will find this helpful.
rahafoorum.ee/en/bondora-stats-crash-course-recovery-rate/Disclaimers: My investments on Bondora during 2010-2016 ended with a total annualized return above 28%. I worked at Bondora nearly 2 years and quit at the end of 2015, when it became clear that the company had chosen a path I did not want to participate in. May I ask what you were not comfortable with? My guess is the risky expansion abroad?
|
|
|
|
Post by rahafoorum on May 14, 2020 19:14:42 GMT
Bondora shares some "interesting" stats about recovery rates, but there are actually good ones as well. I put together a quick guide on which of the figures show what exactly. Perhaps someone will find this helpful.
rahafoorum.ee/en/bondora-stats-crash-course-recovery-rate/Disclaimers: My investments on Bondora during 2010-2016 ended with a total annualized return above 28%. I worked at Bondora nearly 2 years and quit at the end of 2015, when it became clear that the company had chosen a path I did not want to participate in. May I ask what you were not comfortable with? It's a long list, but I guess you could sum it up with the following. At the time I quit, Bondora was adamant that the reason the company can't grow, is that investors are picky and choose which loans they want to invest into. Not that the non-EST loans are actually mispriced and way worse than junk in most cases.
One approach was to "beautify" the stats and reports. Other, that finally solved this problem (for a while at least), was Go & Grow. This product has no problem with investors choosing anything and there's no data to "beautify" either. They don't publish practically anything about its portfolio (you can still get most of it though, if you really want to) and Bondora singlehandedly decided what gets funded with that money.
|
|
|
|
Post by badlord on May 14, 2020 21:11:48 GMT
May I ask what you were not comfortable with? It's a long list, but I guess you could sum it up with the following. At the time I quit, Bondora was adamant that the reason the company can't grow, is that investors are picky and choose which loans they want to invest into. Not that the non-EST loans are actually mispriced and way worse than junk in most cases.
One approach was to "beautify" the stats and reports. Other, that finally solved this problem (for a while at least), was Go & Grow. This product has no problem with investors choosing anything and there's no data to "beautify" either. They don't publish practically anything about its portfolio (you can still get most of it though, if you really want to) and Bondora singlehandedly decided what gets funded with that money. Clear story. Thank you for sharing. Yep, I noticed the questionable quality of non-EST loans when I analysed the default probabilities of Bondora loans. I also came to a conclusion that Bondora made another choice: instead of optimizing for avoiding defaults (not giving loans to those who are unlikely to comfortably pay them back) they chose to optimize for better recoveries. I tried to discuss it with their support, however heard back that they didn't comment on that and that each investor could draw whatever conclusions they saw fit from data analysis.
|
|
|
|
Post by rahafoorum on May 15, 2020 8:49:55 GMT
I haven't seen any proof of better recovery yet. They opted to eliminate costs associated with recovery by changing the terms from Bondora covering the costs to pushing those costs to investors. They initially said it's a 15% cost, but then silently increased it to 35% quite soon afterwards and also added these costs to old loans that had already defaulted long time ago and were already in the hands of bailiffs in many cases. This has saved/earned Bondora millions over the years, which otherwise would've been their cost. Now it's investors' cost, so Bondora doesn't really have to worry too much about bad loan quality.
If you look at the cumulative recovery statistics, you can't really see any improvements. Even if it has occurred, it's not enough to cover the additional 35% cost.
Haven't bothered to look lately, but several years ago the recovery was mainly carried by Estonian loans with low recoveries elsewhere. Especially in Spain, where a few loans recovered a lot and the rest had recovered almost none at all. Perhaps the picture has changed by now, but somehow I doubt it. Otherwise we'd see it in the overall stats as well.
Note also that this means that investors who invest into Estonian loans, pay the majority for the recovery efforts of non-EST loans that don't really recover.
There are some graphs here on this from 2016. Text is Estonian, but the graphs show the proportion of defaulted loans that have recovered a certain proportion. For example this shows that over 75% of ESP loans that defaulted in 2014, had no recoveries whatsoever by July 2016. Worse even than Slovakian loans which were an utter disaster.
|
|
|
|
Post by badlord on May 15, 2020 17:52:57 GMT
I haven't seen any proof of better recovery yet. They opted to eliminate costs associated with recovery by changing the terms from Bondora covering the costs to pushing those costs to investors. They initially said it's a 15% cost, but then silently increased it to 35% quite soon afterwards and also added these costs to old loans that had already defaulted long time ago and were already in the hands of bailiffs in many cases. This has saved/earned Bondora millions over the years, which otherwise would've been their cost. Now it's investors' cost, so Bondora doesn't really have to worry too much about bad loan quality.
If you look at the cumulative recovery statistics, you can't really see any improvements. Even if it has occurred, it's not enough to cover the additional 35% cost.
Haven't bothered to look lately, but several years ago the recovery was mainly carried by Estonian loans with low recoveries elsewhere. Especially in Spain, where a few loans recovered a lot and the rest had recovered almost none at all. Perhaps the picture has changed by now, but somehow I doubt it. Otherwise we'd see it in the overall stats as well.
Note also that this means that investors who invest into Estonian loans, pay the majority for the recovery efforts of non-EST loans that don't really recover.
There are some graphs here on this from 2016. Text is Estonian, but the graphs show the proportion of defaulted loans that have recovered a certain proportion. For example this shows that over 75% of ESP loans that defaulted in 2014, had no recoveries whatsoever by July 2016. Worse even than Slovakian loans which were an utter disaster.
Very useful info. Many thanks for sharing. I didn't analyse their recoveries based on their dataset (perhaps a good idea for me to update my Pandas Python script I created to analyse default probabilities to do that). I was just trying to reconcile the default probabilities I discovered when analysing their dataset with Bondora's statements on expected losses: support.bondora.com/hc/en-us/articles/212798989-Risk-scoring . When I saw annual default probabilities in excess of 8% for AA Estonian loans and at the same time claims of maximum expected losses for them of 2%, the only explanation was a recovery rate of around 75%. When I saw that their default probabilities where worsening (I had discussions with Bondora support about it in 2018), while the statements of expected losses being claimed to remain the same, I judged that they must be improving recoveries.
|
|
|
|
Post by rahafoorum on May 16, 2020 8:06:34 GMT
Initially they used a fixed rate for expected recovery (1 - LGD figure essentially), which was relatively low for non-EST. Then at some point they increased it supposedly based on the actual recovery rates thus far. So in that sense they have increased expected recovery rates that they used in pricing model. You can find these figures from dataset or unweighted average here in a table:
This increase is just not visible in actual recovery rates as far as I have seen. Now they have removed this info from the dataset for some reason. Newer loans have a 0 in that column since mid-2019. Either they stopped trying to estimate recovery, started using some other figure, the dataset broke or simply decided to not publish this info anymore. Who knows.
|
|
|
|
Post by badlord on May 16, 2020 12:39:40 GMT
Bondora shares some "interesting" stats about recovery rates, but there are actually good ones as well. I put together a quick guide on which of the figures show what exactly. Perhaps someone will find this helpful.
Excellent article. This "creative recovery rate" is probably one of the main reasons the expected profit as advertised by Bindora stats on an individual's portfolio looks great initially, but then starts to plummet towards 1%-2% as time passes by if one stops investing in new loans. Many thanks for debunking Bondora's recovery rates! I find the approach chosen to be most misleading. That would be a worthwhile exercise to undertake -- calculate a posteriori recovery rates based on defaulted loans in the dataset.
|
|
|
|
Post by rahafoorum on May 16, 2020 14:46:49 GMT
Excellent article. This "creative recovery rate" is probably one of the main reasons the expected profit as advertised by Bindora stats on an individual's portfolio looks great initially, but then starts to plummet towards 1%-2% as time passes by if one stops investing in new loans. Many thanks for debunking Bondora's recovery rates! I find the approach chosen to be most misleading. It is the same logic, but their creative recovery rate is not to blame for that high XIRR value. Their XIRR also deducts only the principal that is overdue according to the original schedule. So, a 5-year loan that makes no payments whatsoever, is still considered more than 50% current in the calculation after 2.5 years of no payments.
You can see that calculation in action very nicely if you open up their blog articles where they show return of different segments of loans and then compare the returns over time. I put some of those on a graph here, so it's easy to see (haven't bothered to update it lately).
For an individual investor, the best XIRR is achieved, if you buy some relatively freshly defaulted loans at high discounts. Bondora will consider most of that defaulted principal essentially as capital gains and increase your return for years. Not sure you'd want to do that with your money though and I doubt Bondora themselves will buy your loans at the value they use in XIRR calculation either.
|
|
fric
Member of DD Central
Posts: 199
Likes: 79
|
Post by fric on May 18, 2020 5:50:37 GMT
Cross border recoveries are very expensive, time consuming and often not worth it really in the end. I have heard multiple cases of people who have taken out payday loans in the UK (5-20k) before moving back to their home countries. Its very hard to produce a criminal case of fraud, and otherwise its just a civil case which will take forever and any lawyer and other fees and expenses will just eat up all the principal. Sure, if you owe somebody hundreds of thousands than its another story.
That's why Estonian loans always have better recoveries, since its a lot easier for Bondora to actually go through with recoveries.
|
|
|
|
Post by badlord on May 21, 2020 15:47:11 GMT
Excellent article. This "creative recovery rate" is probably one of the main reasons the expected profit as advertised by Bindora stats on an individual's portfolio looks great initially, but then starts to plummet towards 1%-2% as time passes by if one stops investing in new loans. Many thanks for debunking Bondora's recovery rates! I find the approach chosen to be most misleading. It is the same logic, but their creative recovery rate is not to blame for that high XIRR value. Their XIRR also deducts only the principal that is overdue according to the original schedule. So, a 5-year loan that makes no payments whatsoever, is still considered more than 50% current in the calculation after 2.5 years of no payments.
You can see that calculation in action very nicely if you open up their blog articles where they show return of different segments of loans and then compare the returns over time. I put some of those on a graph here, so it's easy to see (haven't bothered to update it lately).
For an individual investor, the best XIRR is achieved, if you buy some relatively freshly defaulted loans at high discounts. Bondora will consider most of that defaulted principal essentially as capital gains and increase your return for years. Not sure you'd want to do that with your money though and I doubt Bondora themselves will buy your loans at the value they use in XIRR calculation either.
I am thinking of calculating the actual recovery rates from Bondora's dataset using my Python script and the pandas library. I went through the description of fields in the public dataset. The following 5 fields look relevant. Would the following be a correct way of going about it going through loans that defaulted? PrincipalRecovery + InterestRecovery - PrincipalDebtServicingCost - InterestAndPenaltyDebtServicingCost
------------------------------------------------------------------------------------------------------- x 100%
PlannedPrincipalPostDefault
I put the same question to Bondora support, however they made a statement that they couldn't advise me on that and referred my to their own reports.
|
|
|
|
Post by rahafoorum on May 22, 2020 7:32:10 GMT
I am thinking of calculating the actual recovery rates from Bondora's dataset using my Python and script and the pandas library. I went through the description of fields in the public dataset. The following 5 fields look relevant. Would the following be a correct way of going about it going through loans that defaulted? PrincipalRecovery + InterestRecovery - PrincipalDebtServicingCost - InterestAndPenaltyDebtServicingCost
------------------------------------------------------------------------------------------------------- x 100%
PlannedPrincipalPostDefault
I put the same question to Bondora support, however they made a statement that they couldn't advise me on that and referred my to their own reports. I believe the debt servicing cost is a separate figure that's already excluded from recovery figures. So principalrecovery and interestrecovery are already the numbers that reached investors after the fees.
You can do a quick double check if you find a loan with PrincipalDebtServicingCost and then add PrincipalBalance + PrincipalRecovery + PrincipalDebtServicingCost + PrincipalWriteOffs and compare it to EAD1. If those numbers are equal (which they seem to be), then debt servicing cost is a separate figure.
If you want to be certain that you have the correct figures, you could round up loans that defaulted in a certain month and compare the recovery amount with the figures in Cumulative recovery rate graph on their statistics page. It has percentage and sum both available, so you should get a figure that's somewhat close to that.
|
|
fric
Member of DD Central
Posts: 199
Likes: 79
|
Post by fric on May 22, 2020 8:33:41 GMT
If you look at loan level you usually see that they are multiple numbers. For example 1 eur was recovered at a specific date for your loan, 40 cents principal and 60 cents interest. Usually all the interest recovery is written off as debt servicing cost.
|
|
|
|
Post by rahafoorum on May 22, 2020 10:49:57 GMT
If you look at loan level you usually see that they are multiple numbers. For example 1 eur was recovered at a specific date for your loan, 40 cents principal and 60 cents interest. Usually all the interest recovery is written off as debt servicing cost. We're talking about dataset here not the UI.
|
|
