Home » Analyses » Bondora default rates and other statistics
          

Bondora default rates and other statistics

Disclaimer: I ran most of the calculations in one long session so can’t really guarantee that all data is entirely 100% accurate and free of any accidental mistakes with filtering or whatnot. Before making any investment decisions, double-check the results and run your own calculations as well.

Risk segment in Bondora's statistics

Source: The Risk segment of Bondora’s official statistics page on 10.02.2016.

This is essentially what covers the risks of investing with Bondora on their official statistics page. There is also some table about profitability and recovery below it, but those don’t really show anything about risks and they have no explanations around them and no-one who actually knows what they show, would really think that this is indeed what those tables mean.

In addition, the graph highlighted above, seems to be incorrect as well. The help text under the question mark says the following:

Expected Return measures potential returns for an invested principal within a particular Bondora Rating segment. It is calculated as the difference between the interest rate and the expected loss rate. Expected Return is based on historical inventory and relies on the country mix and loan maturity structure available for the previous month. Should the country mix or the maturity structure change or appear to be substantially different in your portfolio, your expected return may vary. Higher expected return generally also means higher volatility in potential returns.

In other words, it should be based on the loans issued in January of 2016 mostly. However, if we look at the expected return of loans, then since the introduction of Bondora Rating V2 in December 2015, the E(R) for HR group has been somewhere between 8-18% and nothing even reaching the 20% figure shown here.

In fact, let’s look at all the figures based on loans that were issued and not cancelled according to loan dataset.

Expected Return per Rating per year at Bondora

Graph 1: Expected Return (unweighted) for loans per Bondora Rating based on the year when loan application was started. For calculation, I used the formula of E(R) = I – EL%. Note that loans issued prior to 2015 were not priced according to Rating model and Ratings were calculated to already issued loans after the fact.

Not only is HR showing a lot higher number on Bondora’s statistics page than it should, the F Rating is also considerably higher. With the rest, I guess there could be a difference in the results if you weigh the values and/or look at all loans sent to market, instead of only those that received funding.

In general though, it is clear that in this situation, any investor worthy of the title, is forced to either do some necessary analyses on their own or allocate their funds to other investment opportunities.

In this post, let’s attempt to look at some of the numbers that one would perhaps want to and expect to see at a P2P-lending platform to help with assessing the risk and return of the opportunity.

The Risk – Default Rates

We are looking at a dataset as of 06.02.2016. and will start with general default rates per Rating based on the EUR amount of loans and EAD1 (exposure at default including only defaulted principal amount).

Bondora Default Rate based on EAD1

Table 1: Default rate per Rating based on defaulted principal amount. CurrentDefault is as of 06.02.2016 based on currently defaulted outstanding principal amount.

Default Rates per year and country

We can now also look at the same thing based on country and year when the loan was issued.

Bondora Default Rate based on year and country

Table 2: Default Rates based on amount of EUR issued per country and per year when loan was issued. CurrentDefault is based on the defaulted principal still outstanding as of 06.02.2016.

The Return

Interest rate per country and year

Now, while interesting and important information, the default rates alone don’t give us necessarily the whole picture, because we would also have to see the interest rates those loans were issued at.

Also, since most investors should be on average investing same or very similar amounts to each loan, then it might make more sense to look at the default rates based on counts of loans instead of amounts, so I’ve added this figure here as well.

Interest rates per year and country at Bondora

Table 3: Average interest rate (unweighted) for loans issued per year and country with Initial default based on SUM and on Count of loans.

While the actual interest rate when weighted, could be somewhat different from the unweighted result (relevant for when you compare it to default based on sum), it’s probably not too far from it. Especially for 2014 loans, where essentially most of the loans received very similar interest rates with up to a few percentage point differences on average.

Based on interest rate and default rate alone, you could make some simple calculations for rough return estimates already by assuming that defaulted loans don’t earn you interest and based on this you could probably easily identify which specific segments are performing best from these.

Although for a more accurate result, you would have to also consider the time when defaults happened and how much payments were made prior the default (you will end up with a slightly more optimistic result in case you use this count default rate, because this percentage has already been discounted by the principal proportion at default), and you would also have to account that some of the principal from performing loans will be earning interest for more than one year (while some of it is repaid also with every month). You may also want to account for some recovery, although I personally don’t since it’s relatively unpredictable and I prefer making a more conservative estimate, instead of a too optimistic one.

In other words, this sort of table is good for making quick at-a-glance conclusions about different segments of loans and judging the potential ballpark of a yearly return for those to judge which ones seem to be the best performers for your strategy (before or after taxes based on your situation).

To get a very accurate result from this, if you want the actual return figure, would take quite a lot of additional work though.

XIRR

A more accurate and mainstream method to calculate your return, would be to use XIRR calculation for your selected segments. However, while theoretically possible, it would require matching up several datasets and doing a whole lot of work before you actually can come up with some solution. The main issue would be with accounting for the cash flow coming back from the specific segment of loans.

In other words, it’s way too much work to do this in excel for me so I won’t be doing that today.

Recovery, Gross Profit and ROI

There is also the method similar to what Bondora is showing under the Portfolio Profitability section on their statistics page. I personally don’t think the Principal Overdue value used by Bondora is very meaningful for making conclusions about performance of the portfolio today and I’ve previously explained why in this post within the XIRR explanation section here.

Personally, I believe it makes much more sense to compare this figure with the defaulted principal amount instead. Why? Well, a defaulted loan loses 1/3 or even more of its value relatively instantly after defaulting, simply based on its status. In reality, it’s probably losing even more, since the liquidity is even lower than for other loans (not too many investors are out there buying defaulted loans I’d imagine compared to current or few days overdue loans). Also, the entire principal is considered by the contract and legally as overdue at the moment of default so by definition that principal is overdue.

Recovery is also taking significant amount of time and while we don’t know the exact recovery amount in the end, the outcome can’t really be expected to be over roughly 60% of the principal recovered (based on historical data, that seems pretty close to the ceiling) and that’ll usually take several years to achieve at least.

So after accounting for time value of money and the drop in value and liquidity of the defaulted loans, we can assume a considerable loss on those on average already instantly after the loan has defaulted.

While considering 100% of defaulted principal as a “loss” as of today, is overly pessimistic, this is somewhat leveled out by accounting for recovery so far and not deducting any part of simply overdue loans, out of which a decent amount will default later on.

Also, not considering defaulted loans as a “loss” almost at all as of today, is overly optimistic, as it inherently assumes a considerable recovery in the future, which for a freshly defaulted loan, is extremely optimistic, and which may or may not become true, while the default and the drop of value has already happened.

In other words, again, I personally prefer to be more pessimistic in these estimations than optimistic, since this leaves room to be positively surprised if additional recovery does come in and it somewhat protects me from unexpected losses.

Bondora funded, defaulted and recovered amounts per year

Table 4: Issued, defaulted, recovered amounts and interests received on loans based on the issuing year.

Based on this table, we can now calculate some other values such as:

  • Recovery% – percentage of defaulted principal that has been recovered to date (as of 06.02.2016).
  • Gross Profit – InterestAndPenaltiesPaid minus DefaultedPrincipal plus RecoveredPrincipal.
  • ROI – simple calculation of Gross Profit divided by FundedAmount. This is relevant because simply looking at an amount of money received on a bunch of loans, is relatively meaningless without the context. A €10 000 is a positive number, but it’s still a very bad investment if it is achieved on a €5 million investment over a year. Note that this is not a yearly figure, but total over the entire period. ROI is also totally different from XIRR and any other calculation methods where time element is also taken into account and thus should not be compared to these directly. Consider it as a very simplistic “put this much in, took this much out” type of calculation.
  • ROI Before Recovery – simple calculation of Gross Profit before RecoveredPrincipal divided by FundedAmount. Note that this probably still includes the interest and penalties received from recovery, if any. Note that this is not yearly figure, but total over the entire period.
Bondora ROI per year

Table 5: ROI, Principal Recovery and Gross Profit for Bondora loans based on the year when loan was issued.

Comparing the ROI Before Recovery and after at least some of it, we can see that it has historically had a relatively large effect on the end result and it’s already somewhat visible in 2014 year loans as well.

In short, while the results for more recent loans are not looking too good at the moment, it is reasonable to expect it to become at least somewhat better in the long run as the amount of new defaults starts slowing down, more interest payments will come in and additional recoveries will be secured.

Recovery, Gross Profit and ROI per country

Let’s look at the same numbers per different countries.

Bondora ROI per year per country

Table 6: ROI, Principal Recovery and ROI for Bondora loans per country based on the year when loan was issued.

From this table we can see a similar pattern, in regards to the effect of recovery, emerge. We can also relatively easily spot which segments are the best performing as of 06.02.2016. at least and which ones are worst.

Ideally you would also split these segments up into specific segments or compare in some other way the proportion of risk levels within these different segments to make sure that one is not significantly higher risk than another with a lot higher interest rates to compensate for it. As in this case the higher risk segment would almost always show a lower Gross Profit initially due to higher initial default rate, but the higher interest rate would then later on compensate for this (at least in regards to pre-tax returns).

In this case however, we can simply look at the average interest rate, which is also highest for EST for every year, with the exception of Spanish loans in 2015, where average interest rate is over 50% compared to 31% for EST. In other words, if there is no sudden unexpected large increase in defaults for Estonian loans so that it exceeds those of other countries significantly, Estonian loans will continue to show a lot better ROI and Gross Profit on average than other countries (prior to 2015 at minimum).

Another aspect to consider in this context would be to look at the average loan duration. In other words, with everything else being equal, the longer the duration, the larger the earned interest amount should be by now, since initial payments in long duration loans are majorly consisting of interests and the more interest can be earned by the end of the loan schedule as a whole.

In most cases Estonian loans are somewhere in the middle of the bunch compared to other countries and the interest amount relative to issued amount is significantly higher in close to every case, with the exception of 2015, where interest received for Spanish loans amount to 11.5% of funded amount compared to 10.8% for EST so far.

The following graph sums up where the Gross Profit is coming from on platform level up nicely.

Bondora Gross Profit per country and year

Graph 2: The amount of Gross Profit received from loans from different countries based on the year when the loans were issued.

ROI and Gross Profit after taxes

This simplistic calculation doesn’t tell us much about the actual annual return and the effect of time on return (for example, if 80% of loans from 2014 were issued in the end of the year, then the actual return would be higher than compared to issuing those loans early on in the year and if most of the interest had arrived early on in 2014 already, then also the return would be higher than when it is paid later on in the 2015 instead).

However, since it’s such a simplistic money in vs money out type of calculation, and we know the interest amounts received, we can very easily calculate the Gross Profit and the ROI after taxes by deducting the tax rate from interest earned.

So, considering that I’m an Estonian investor who has to pay 20% tax on earned interest (actually it was 21% for the earlier years, but let’s keep it simple), we can easily calculate the Gross Profit and ROI after taxes for same formula.

ROI per year and country after taxes

Table 7: Gross Profit and ROI per year and country after taxes.

I have to admit that the effect of taxes on this broad level seems larger than I had anticipated.

What you probably would want to look at when considering taxes though, is the split between risk levels because usually the higher the risk level, the bigger the effect of taxes since at least in Estonia we’re not allowed to deduct losses from taxable interest income. The higher the risk level, the bigger the portion of the return coming from interest payments that cover the defaulted loans.

ROI per Rating after taxes at Bondora

Table 8: ROI and Gross Profit after taxes per country and Rating.

Ideally you would also split them up by year for example and then see the performance more clearly, but since the table is also quite huge and the numbers in many segments are relatively small, then I’ll skip that part here.

So the Gross Profit and ROI after taxes has been better in general for the lower risk loans, compared to HR and others. It tends to even out a bit after the risk-based pricing was introduced in 2015 so the difference is not that huge anymore, but currently it’s still there. With the new V2 model where HR has a lower Expected Return, the difference will likely increase again.

Gross Profit and ROI based on Bondora’s statistics

Now, while I personally find this Principal Overdue concept a relatively impractical and in this context even misleading since contractually a defaulted loan is fully overdue, then I know that some people like this concept.

So here’s a small table on the Gross Profit and ROI after taxes based on the same Bondora’s Statistics page table as of 10.02.2016. Since the interest received is there, you can easily calculate these figures for yourself any time you want with your own tax rates.

ROI based on Rating

Table 9: ROI and Gross Profit based on Bondora’s statistics page and logic on 10.02.2016.

Unfortunately we can’t see this data split by country here so we can’t really look into why some figure is where it is today, but I guess you can use the tables provided previously in this post to make some assumptions and guesstimates on it anyway.

In conclusion

The statistics shown on Bondora’s official page don’t offer much practical value for making daily investment decisions since they don’t cover almost any aspect of risks nor allow any more granular look into the actual causes of why the outcome is as it is. Also, they don’t seem to be accurate in some cases for some reason, as seen from the image at the top of the post.

I also think it is pretty clear from these tables, that doing your own analyses has at least historically meant that, based on data, you can make a lot better quality decisions about your investments, understand better what’s going on at the platform as a whole and end up with a significantly better outcome than average.

While the new Bondora Rating V2 model is supposed to be a lot more accurate than its predecessors, and I’d expect it to be also, we will start getting data on the actual results perhaps in about 6-12 months on. I opted to wait for statistical evidence about the performance to come in when model V1 was introduced and stuck with my own analyses, and it seems to have paid off nicely, but in the end, it’s up to everyone to decide themselves how and where they will invest their money and what information they need to make those decisions.

What additional analyses or statistics do you use when making your investment decisions?

Subscribe to receive future updates

18 Comments

  1. Comment by dersandr:

    nice article. something what I was thinking too. but you confirmed with data. thank you for this big, hard work

    • Comment by Taavi:

      Thanks for the comment. Wanted to have a look at the data myself so thought I might as well share the results then. Perhaps someone finds it valuable or maybe even someone identifies some sort of mistake here so I can fix it and look at the correct results 🙂

  2. Comment by martins:

    Thanks Taavi for the analysis. Reading this is there any sense to invest in Bondora at all if they are cooking their stats & graphs?

    • Comment by Taavi:

      I’m not sure if you could call it cooking stats, but they do make sure to use calculation methodology that makes it look nicer. Where possible, an optimistic scenario seems to be used or negative data removed from stats. You probably remember that there were default rate stats in the newsletter per each country a while back…

      In several cases of course the stats are deliberately misleading since they have been notified about those cases, like the loan status thing, which clearly isn’t loan status. I mean, loan status is 60+ days overdue (which actually is 74+ days for over a month or even longer already, but no probs) and it is NOT part of my defaulted loans are overdue and the rest are current.

      It is extremely misleading to show this graph on stats page and show same info for investors’ personal stats page and claim that this is showing loan status of current and overdue. The overdue there is principal overdue (the one you see on dashboard and the profitability table), which has absolutely nothing to do with loan status in general. Claiming that the future payments according to the original schedules of my defaulted loans are “Current”, is indeed total BS.

      But apparently fixing translations is so hard that it takes several months to do it (or alternatively, no-one at Bondora understands that this is wrong, even though I’ve told them, or they simply don’t want to fix it).

      About the Expected Return graph highlighted above, I’m guessing they have changed the previous month part and looking at a longer period now instead. Possibly because most investors find it weird that you take on a higher risk loan and receive a lower return for it even pre-tax than for lower risk loans. And seems they forgot to notify investors about the change or fix the explanation texts apparently.

      It’s everyone’s own decision where they invest or don’t and it seems some people have even opted to use this new PM, so… As a disclaimer, I’m on full exit mode myself.

      Last straw was after they blocked my comments in the blog where I wanted to know how this XIRR and ER comparison actually would indicate that Bondora Rating performs better than expected.

      I then suggested through support that the EST A 13.6% XIRR (actual) is lower than the 13.7% Expected Return (target) in their “Bondora Rating has outperformed expectations: 98% of the portfolio above target return” blog post, thus being below the target return and suggested they fix it and let investors know about the mistake.

      I also suggested they consult with their credit scoring team if needed to see why this comparison of XIRR and ER doesn’t make any sense in this way and then make a comparison that actually is correct and also let investors know about this as to not mislead them by accident with the previous invalid analysis.

      In answer I received a soft type of legal threat in the lines of “We have forwarded your documentation and comments to our attorneys and should they have any additional questions they will contact you directly.”, as if attorneys would know anything about XIRR and ER comparison…

      They have also ignored some of my support e-mails like the one about why some of my Stage 4 loans were moved into Stage 1 for DCAs, so I’ve given up writing them and can’t really confirm whether this “Risk” graph has a bug or it’s been a change in underlying logic.

  3. Comment by martins:

    Thanks Taavi for comment. Blocking comments and legal treats looks so common approach today.

    • Comment by Taavi:

      Fortunately I haven’t experienced that in other platforms as of yet 🙂

      Omaraha used to delete comments they didn’t like in their forums though. Not sure if that still happens or not and as far as I know, none of my comments have been deleted there yet. Haven’t heard much about that in the recent years either anymore.

  4. Comment by martins:

    I am not from Estonia. Just curious, what is difference between Bondora and Omaraha? Is it hidden daughter company of Bondora?

    • Comment by Taavi:

      Totally different companies. Omaraha is more of a 1-man operation type of thing, although now they have more employees already.

      The business model and almost everything is different besides the fact that they both issue loans through the P2P-lending concept and both operate in Estonia (Omaraha is also testing waters in Slovakia).

  5. Comment by Jan:

    Thanks for very interesting statistics. I was just wondering whether you have analyzed timi ng of default. In my experience there is much bigger portion of loans defaulting without getting single payment, then there is another increase after some time, probably resulting from worsening economical condition of borrowers. The first category should be something Bondora team should address and work on its reduction, the other is simply result of many other factors. Thanks.

    • Comment by Jan:

      Just adding that I saw the other article on this, but it is a little bit old and thus not reflecting the current situation.

    • Comment by Taavi:

      I haven’t done any analyses on this specifically after this post:
      http://rahafoorum.ee/en/defaults-happen-bondora/

      However, from other analyses I’ve done, it does seems like there was a significant increase in 0 payment defaults, which mostly came from the introduction of non-Estonian markets and that was actually unexpectedly high proportion at some point.

      These stats are now messed up by other events as well, like this B secure issue where borrowers could take a grace period with only interest payments in the schedule and then didn’t default even after 6-7 months of zero payments.

      This issue was fixed at one point and then a relatively decent amount of defaults came in all at once so you can’t really make any serious conclusions about these stats for recent year or so.

      This thing kept the default rates artificially lower for a while and then brought them all in at once at a later time so the result will be skewed quite a bit probably.

    • Comment by Taavi:

      Redid an updated version of the default curves:
      http://rahafoorum.ee/en/bondora-default-curves/

  6. Ping from Bondora default curves per year and Rating - RahaFoorum:

    […] » Analyses » Bondora default curves per year and Rating « Bondora default rates and other statistics […]

  7. Ping from Is Bondora giving away your money? - RahaFoorum:

    […] At the same time, it might be somewhat more realistic for higher risk loans, at least by simply numbers, if we assume that higher risk loans recover at same pace as lower risk ones (which may be a possibility). […]

  8. Comment by Estonia-only Investor:

    so you are saying if I invested in Estonian-only loans I only made 1.9% return in 2015? My Bondora dashboard is showing 17% LOL

  9. Comment by Carlos:

    I also don’t get it (even after reading your suggested post).
    Are you saying the return from Estonian-only loans was 5.4% in 2014 and 1.9% in 2015 after recovery??

    • Comment by Taavi:

      a) 2014 means that the numbers include loans that were issued in 2014, not the performance of all loans during the year 2014.

      b) The return methodology used was used to make it comparable to what Bondora was highlighting on their site at the time. This is definitely not the best method to calculate return and for a more accurate result I’d suggest running a proper XIRR calculation.

      c) The figures you seem to be referring to are for Estonian investor’s after tax results at the time (20% tax rate, no deductions allowed).

      d) Please note that the article was written and the analyses were done early 2016 as of that period. Things have very likely changed by now by quite a bit. Especially for the relatively fresh segments, like 2015 was back then.

What do you think?