Important! If you find some mistake in the analysis or logic here, let me know in the comments below so I can deal with it. CONSTRUCTIVE criticism and feedback is welcome and encouraged.
- The logic used by Bondora to claim that Bondora Rating has outperformed expectations, is invalid and misleading.
- The compared XIRR and Expected Return in Bondora Rating are totally different constructs and will practically never be equal. With XIRR being higher of the two, especially for such fresh loans as 2015.
- Bondora Rating model V2 expects higher losses for the vast majority of the loans issued in 2015 compared to model V1. Including a negative expected return for roughly 15% of them. In other words, it seems that either V1 wasn’t very accurate (specifically outside of Estonia) or the update of V2 made the model less accurate.
Why write this post?
In the beginning of December 2015 I published an analysis of Bondora Rating on loans issued in 2014.
Before publishing it, there were 4 people, who deal with statistics/data analysis on a more or less professional level and are familiar with Bondora’s datasets, who reviewed the analysis. They all confirmed that the logic seems legit and no additional mistakes were found in the analysis after their feedback.
Before publishing it, I also asked feedback from Bondora about the calculations and numbers to make sure I didn’t accidentally make any serious mistakes in the analysis. The feedback I received, was a 12-point e-mail with several sub-points.
The relevant feedback was taken into account and the required modifications made. Irrelevant parts were set aside. Additional clarification was requested on some of the more vague points, where it was not entirely clear what Bondora meant by those. Bondora refused to provide further comments.
Because some of the feedback was already taken into account, some of it was invalid and some of it unclear, I couldn’t account for the first thing that Bondora requested in their feedback:
We would kindly ask you to share our statement prominently and above the rest of your analysis in full, along with all supporting documents in their original form, as otherwise we consider you intentionally misleading your readers for personal financial gain.
I also wrote up a thorough reply for all of those 12 points, which I haven’t published and I wasn’t really sure whether I should either. The feedback was sent to me privately and was probably meant to persuade me not to publish the said analyses.
However, on the 6th of January 2016, Bondora has published a blog post titled “Bondora Rating has outperformed expectations: 98% of the portfolio above target return“, which is based on the 11th point from the 12 point feedback.
This point is unfortunately invalid, inaccurate and (hopefully not purposefully?) misleading information.
In other words, this point of the feedback has been publicly shared and promoted to all the investors, so I think it deserves some public feedback as well.
What’s wrong with the XIRR table?
Since some of this information is not mentioned in the latest blog post, I will start with the feedback that was given me, with the image that I received in the e-mail.
The table in the blog post has been somewhat updated, but let’s look at this one first, because this is the one Bondora requested that I should be publishing as the most prominent part of my analysis on Bondora Rating’s performance on 2014 loans.
11. Due to the number of limitations set out above we cannot agree with the calculations for expected/realized returns pre nor post-tax. The enclosed table provides actual returns for across all countries for 2014 and 2015 as of 3/12/2015. 2015 results are also split between grades (full data enclosed as a separate file).
a. All country-grade segments, other than 3, have produced returns in line or above the expected return. The three grades in question make up less than 2% of the inventory available on Bondora. With the release of our new scorecards we expect that these inefficiencies have been solved.
E-mail also included an excel with the data used for this table. You can download it here, should you want to.
So there you have it. My analysis obviously MUST have been incorrect, because that table clearly has a bunch of green cells…
I won’t stop at all the things wrong with this picture to save you time, but let’s look at two:
a) My entire analysis was based on loans issued in 2014 only. This table has main focus on 2015 loans for some reason instead. How would that say anything about my analysis or the sample I analyzed? Short answer would be that it doesn’t.
I suppose we could compare it with the other analysis I did on first 6 months of Bondora Rating in 2015. Easiest and fastest comparison would be Spanish E Rating, which supposedly is accurate here, but in my analysis had Target of 16.3%, default rate of 37.9% (34.11% expected loss after the expected 10% recovery that was used in the Rating algorithm) and an interest rate of 31.4%. One would ask, how is it possible to have something close to 16% return on such figures if the calculation is Interest rate – Expected Loss, and we’ll take a look at this below.
One COULD also ask, how can Bondora claim that the return of 15.9% is accurate while the target is 16.2% and these loans haven’t even reached a one full year of maturity yet (in other words, most of the expected defaults have not yet happened). Yet, the cell is green and the 11% proportion of loans is somehow considered as on the mark (the actual proportion of ESP E for 2015 loans at that point would be 7.7%, but never mind that).
On the updated table from Bondora’s announcement, you can see that the 3.8% proportion of Estonian A loans is also below expectations with 13.6% vs 13.7% Target, but is still somehow shown as green.
There is the summed up part on 2014 at the very top. It doesn’t say anything about Bondora Rating nor my analysis though, so no idea why it’s there. It doesn’t even have the E(R) according to Bondora Rating based on their own estimated EL%. It should, since we’re talking about the accuracy of the model.
I’ll help complete it with the EL% that was there before the Bondora Rating model update.
You see? It’s all positive here, we could add 4 more green cells to support Bondora’s argument. Of course, this table would raise the question in investors: Are the Ratings accurate and can people really expect to lose money on a large portion of 2014 loans or are the Ratings inaccurate by being too pessimistic for those loans?
Since we already started, let’s also look at same thing with the more accurate model V2 EL% values.
More green cells added to support the claim. The case grows stronger with every table.
Actually there’s nothing here that could be used to claim anything about the accuracy of the Ratings, but more interesting though is that Ratings were supposedly accurate everywhere and investors are expected to earn positive returns all over the place, but yet, Expected Loss has increased about 6.5% for FIN and a little bit for ESP as well on average.
In other words, the updated Bondora Rating model is assuming an even bigger loss for Spanish loans than the previous one and a drop in returns for Finnish loans as well (of course, this assumption is not accurate because it was done on historical loans where all data was not available etc etc etc).
While mostly these tables are useless for determining Bondora Rating’s accuracy, we can use this XIRR info in a later example to prove it.
Why XIRR and Expected Return should NEVER be compared directly?
b) Now that the easy one is out of the way, let’s proceed with a bit more complex issue: my analysis was done to look at the accuracy and performance of Bondora Rating by converting actual default rate into a loss we could expect based on Bondora Rating figures and seeing what the end game for those loans could be based on Bondora Rating calculations logic.
Yet, Bondora has instead provided a XIRR value, compared it to Expected Return on a different sample of loans, and is saying that this is the proof of Bondora Rating’s accuracy and the invalidity of my analyses.
Moreover, Bondora has used the same logic in their new blog post and claimed that according to this:
“98.1% of the loans originated in 2015 have delivered net returns above or in-line with expectations. Only three grades performed below targets (Spanish B, C and D grades) however in total these made up 1.9% of the originations in 2015.”
You can find the table here:
Dear reader, I will forgive you if you don’t straight away understand why you can’t directly compare Bondora’s XIRR and the Expected Return from Bondora Rating. You don’t have to understand it, it’s not your job and in fact, you should use your own calculation instead of Bondora’s XIRR for returns anyway to make it comparable with other platforms you invest on (which is something Bondora has not done in their site comparison).
Since Bondora claims in their announcement and sent this to me as a proof of Bondora Rating’s accuracy, there’s two possibilities here:
- Bondora doesn’t understand the difference between the two and that those can’t be compared directly. Which should make you worry about what’s going on with your investments, which numbers on the site can you actually trust and which you can’t and perhaps about the competency of some of the people working on this area… OR
- Bondora does understand the difference, but they sent this to me (and now to all of the investors publicly as well) on purpose, hoping to confuse and mislead their investors, my readers and perhaps myself as well. Although, they should know that I’d understand the difference, since I’ve said it a few times in the past, but who listens anyway 😉
Either one of those I find a lot more difficult to forgive though.
Difference of XIRR and Expected Return
So what’s the difference between Bondora’s XIRR and the Expected Return in Bondora Rating? (besides the fact that they’re both totally different concepts to begin with)
- XIRR is looking at current situation, Expected Return shows the result by the end of the period. In other words, if you look at a loan at point A, B, C and D, where A is the time when loan is issued and D is when the loan has ended and all the expected payments/recovery has taken place, Expected Return would always show the result as if the loan is at D, XIRR will show a result based on the point where you look at it. There’s some sort of XIRR in point A, a different one in B and another one in C and yet a different one in D.
- As a result of the previous point, XIRR is dependent on how mature the loans in the sample are with XIRR values differing on exactly equally performing identical loans depending on where they are in their maturity. In other words, if you have a 100 12-month old loans and add another 100 fresh loans identical to the ones already in the portfolio, even though the return is exactly the same, your XIRR will change after the fresh loans reach first payment dates (if default rate is not 0 or extremely high, then XIRR will increase initially in this instance).
This is also the reason why Bondora’s XIRR is usually highest for the loans when they are fresh and just have reached their first payments, compared to the XIRR of same loans later on after 12+ payment dates have passed. It will take a while to start adjusting and closing in towards the actual Expected Return.
Expected Return will not be affected by adding more identical loans to the pool since it’s only looking at the end result for these loans and the point in maturity has no effect on this.
- Expected Return includes accrued interest at the moment of default as part of the defaulted amount and deduces this from return. Bondora’s XIRR only deducts missed principal payments and ignores any unpaid and accrued interest. This means that if a loan behaves exactly like Bondora Rating expects, the values will be different at any given point in time, even in the end. More precisely, XIRR will be higher than Expected Return at almost any point in time. Perhaps with the exception of in the end of a schedule for defaulted loans where loan has a relatively short duration, but recovery takes a lot longer time to come in. (For example the XIRR for a 6-month loan at month 7 that defaulted with first payment and is recovering slowly, but is expected to recover a considerable amount.)
There are probably some more significant differences, but I think these three will suffice. Let’s look at some examples instead to illustrate this so it would be easier to understand this complex mumbo-jumbo.
Mechanics of annuity payment schedule
To understand the XIRR calculation, we first have to understand what annuity payment schedule looks like. For this, I have taken a loan from Bondora that’s as much close to the average loan on the platform as of 07.01.2016. as I could find:
- Loan number: 399328
- Interest: 28.40% (average 28.68%)
- Duration: 48 months (average 47)
I have mapped this loan schedule to show you the proportion of principal and interest payments within each month’s payments in the image below.
As you can see, the payments in the first part of the loan schedule consist mostly of interest payments and smaller portion of principal payment. The largest part of principal payments are in the other half of the loan schedule with principal payments exceeding 50% of the monthly payment since month 20.
The longer the loan duration, the farther away this point is where principal proportion in the monthly payment exceeds 50% and the smaller the principal proportion in the first payments of the loan. The shorter the loan, the earlier it is.
Bondora’s XIRR calculation
You probably also know that Bondora deducts from XIRR and other calculations, ONLY the principal portion that is overdue according to the original schedule. Even in the case where the loan has defaulted and turned fully collectible.
In other words, if you have 2 identical loans issued at the same time and one of them defaults with no payments and another is paying properly, the dark green part on the image below would be your earned interest and the red portion will be considered as “loss” in XIRR calculation on month 3.
Now, a more attentive reader might be saying at this point something like: “Hey! Wait a minute now! It takes around 36 months until Bondora’s XIRR starts showing that you’re actually losing money?”
Granted, you probably won’t know that 3 year point automatically by looking at the graph, but don’t worry, you can download the file here. Besides that though, yes, you’d be correct on this assumption.
Someone else might be asking “How come?”. I’m glad you asked. It’s actually quite simple. You see, if the principal payments from defaulted loans are only deducted at the same pace as the interest payments added from paying loans, there’s only two ways for the deducted principal to grow larger than the received interests:
- The default rate is so insanely high that even at 30% principal vs 70% interest payments, the principal overdue increases faster than interest received. This is unlikely in most cases where you have some diversification and you didn’t invest all your money into some weird loans from 2014. To put this into perspective, for the loan in current example at the interest rate of 28.4% and loan duration of 48 months, the default rate would have to be somewhere above 85% to be able to see that you’re losing money already with first payment. This would have to increase even further for higher interest rate loans. In other words, it will never happen with first payment, however, with 70% default rate, you may see it after 6 months’ time.
- The loans mature and pass the point where income from interest is smaller than the overdue principal in each payment. There is still an abundance of previously received additional interest from the first 20 months though. That’s why you’ll see it about 16 months later than the moment where the 50% point is passed.
Of course, in an actual portfolio it would take longer still, because repayments are reinvested and another one of these “36-month” cycles are added to the portfolio every time. Additionally, any recoveries for a loan will first be used to pay the first overdue principal amount, meaning that with a very nice recovery of 30% of the overdue principal at month 20, you could essentially see €0 overdue principal and all previously earned interest included as pure profit would show you a decent XIRR. Even if there is no further recovery whatsoever and you’re still losing money in the end.
This is not to say that the XIRR calculation is wrong per se, there is no one correct or incorrect way to do it as far as I know (although this method is usually used by companies who also deduct the accrued interest as far as I know). It’s just extremely optimistic in this format for not fully mature portfolios and it is quite useless when considering the performance of your portfolio or the effectiveness of some strategy you recently implemented.
In fact, it might even mislead you into thinking that you’re some sort of investment genius, because the fastest way to increase your XIRR, would be to buy hugely discounted relatively freshly defaulted loans. The bigger the discount and the more recent the default, the higher the jump in XIRR. Of course, you would have to continually do that with increasing amounts of funds, because otherwise you’d see yourself losing money within a few years’ time if there’s no miracle recoveries.
Additionally, this XIRR method tells you nothing whatsoever about the accuracy or inaccuracy of Bondora Rating’s model. Don’t trust me on this though, let’s look at some examples instead.
XIRR vs Expected Return Examples
Example 1 – positive scenario
Let’s now put this into context and look at a positive scenario. Positive is always good, right?
We invest at the same time into the same 100 identical loans described above. This time, all of them are ideal payers so default rate and thus EL% = 0%.
E(R) = I – EL%, so:
E(R) = 28.4%
What’s the XIRR? Also 28.4% the entire time?
Wait, not the same? God damn… Maybe we looked at a too optimistic scenario. I mean, no-one pays that well, there are always some bad apples.
(As a side note, this is what happens most likely due to rounding and some other issues that those schedules might have due to also having to add Bondora’s maintenance fee into those etc so not all monthly payments are equal. Theoretically, with an expected 0% default rate, Bondora’s XIRR calculation could equal to E(R), I think. Those cases don’t exist in reality though.)
Example 2 – negative scenario
So let’s assume that we made an investment into another 100 identical loans simultaneously that match the loan 399328 described above.
In this case 50% of the loans defaulted with first payment and will never make any payments whatsoever and the other 50% is paying properly till the end.
Since we know that the recovery rate is 0%, then EL% = 50% + accrued interest at the moment of default = 52.45%. As a result:
E(R) = 28.4% – 52.45% = -24.05%
Now let’s look at how the XIRR would look like for the same scenario at different points in time:
I’m not sure if it’s clear at first glance, but if you follow the line carefully, you may notice a small change in the XIRR results at different points in time. The more attentive reader might even notice that during each of the following 12-month periods the speed of that change is increased as the incoming interest amount with each new payment is reduced and the principal amount that’s counted as overdue is increasing simultaneously.
In other words, the process that’s keeping the returns positive in the beginning, is changing its direction and begins accounting for the losses with increased speed in the end.
It is even possible that some of the more attentive readers at this point have already attempted and failed to find any point in time where E(R) would be equal with XIRR or even close to it. Mind you, it’s not a major difference, you could after all reach the E(R) of -24.05% by multiplying the final XIRR with 3…
Example 3 – with recovery
Let’s assume another scenario with some recovery as well to see the calculations in action for a more realistic case.
This time, we have the same situation with 50% defaulting, but we expect a relatively optimistic recovery of 30% (it is a Finnish loan after all, where 10% was used in the Rating).
We will assume that this recovery is coming due to a new payment agreement on month 13 and all of the 30% is equally paid over a 24-month period after which the borrower stops all further payments.
What this 30% recovery means in EL% context, is that the actual recovery amount in EUR is bigger than this 30%, because this amount is discounted by the time value of money. In other words, a €100 in future is worth less than a €100 today, so to receive this 30%, the EUR amount should be higher than 30% of the initial Exposure_at_Default.
So let’s assume that the 30% we use in EL% is the result after discounting.
To calculate E(R), we first need to account for the fees paid to collection agency since those are taken from each payment equally and can’t be accounted for within recovery. For FIN loans it’s 8% according to Bondora’s blog:
So with 30% recovery from 52.54%, it means that EL% is lowered to 38.04% when we account for recovery.
The Expected Return will be:
E(R) = 28.4% – 38.04% = -9.64%
In the XIRR, we won’t use a discounted value for simplicity. In other words, the payments in XIRR are smaller than they should be to actually achieve this Expected Return.
We are also showing less recovery, because we don’t assume the accrued interest as part of the loss, but only consider 30% recovery on the 50% default rate as the XIRR does.
In short, our XIRR should be lower after the recovery kicks in than it is in reality in such cases where Bondora Rating expects a 30% recovery. It should be accurate enough for our purpose of looking at the interaction of these calculations.
As clearly visible, you can see the XIRR go up after the recovery kicks in, but later the principal amount going overdue is increasing faster than recovery and at month 36 (the last month of recovery) XIRR is already lower than at month 24. In month 48, you see a small negative return on the investment.
However, you may also notice that Bondora’s XIRR is at any point in time significantly higher than E(R) and will never drop to its level. Also keep in mind that this is in a scenario where the actual XIRR would be even somewhat higher due to the discounting and higher recovered amounts, which we didn’t account for here.
XIRR will never equal Expected Return
This is also the main reason why Bondora could show a 0% return for Slovakian 2014 loans while any person with half a brain knows that it’s hemorrhaging money from every direction with default rate of 80.73% and average interest rate of 28.28%. Even with this, Bondora’s XIRR is still considering €576k (77%) of the €742k as performing normally in that 0% XIRR. Clearly, it’s smooth sailing from here on. (While Bondora’s own Expected Return assumes a loss of over 30%)
This is also the reason why Bondora can show a XIRR of 3.36% for Spanish 2014 loans and 14.70% for Finnish 2014 loans (within the month’s time, FIN has had some recovery come in and Spanish has dropped a bit).
As you can see when comparing the table provided to me and the ones shown in the 6th January blog post, even after less than a month, the extremely high default rate in Slovakia has already had its effect and the XIRR in Bondora’s new analysis is a negative -1.5% and I can pretty safely say that it’ll get a lot worse before they reach the end of those loan schedules. Unless there’s some real miracle worker at one of those DCAs, then it’ll take longer to drop more.
AND most of all, this is the reason why I couldn’t publish this table that Bondora provided as the most prominent (or even least prominent) part of my analyses and why I didn’t change anything in my analysis based on this feedback. Because you see, by doing this, I would have misled my readers and it is invalid to compare XIRR with Expected Return and expect to make some conclusions based on this.
By the way, since I did the analysis on 2014. loans, Bondora has quietly fixed an issue where loans on grace period wouldn’t default even after 6+ overdue interest payments. A quick check on the 2014. loan sample shows an additional 185 defaulted loans between 15th-18th of December with Exposure_at_Default amount of €530k total.
In other words, the Exposure_at_Default amount has increased from €4.82 million by more than 10% within less than a month, proving my assumption that no new defaults would happen, clearly inaccurate and actual results worse than the optimistic results in my analysis would suggest.
On a more positive note, this has not much effect on XIRR, account value or your profit on the dashboard for a while, since there are no principal payments during Grace period. As a result, there’s no principal overdue to discount.
This is not to say that the XIRR calculation used by Bondora itself is incorrect. You can calculate it in whichever way you want and most P2p-lending platforms do, with many using a totally different logic. It’s just not very useful because of the huge delay in reaction to actual events and you can definitely not use it to claim anything about the Bondora Rating performance.
Edit 09.01.2016: While writing this post, I was focused on the analysis done by Bondora. In reality I realized though, that XIRR can be only a little bit above of or lower than Expected Return. In such situations though, it is almost certain that the loans are performing way under expected performance. Especially, if the XIRR is close to Expected Return at such early stage of the loan maturity, it definitely means a lower than expected performance on the loans.
Bondora Rating accuracy
Now that we’ve shown that the method used by Bondora, cannot be meaningfully used to claim anything about the accuracy or performance of Bondora Rating and the Expected Return of those loans, let’s look at a way that perhaps can.
As we already saw previously, Bondora is using the following formula to calculate Expected Return:
E(R) = I – EL%
To claim that actual Expected Return is above the Expected Return (Target in the blog post table, although this information was omitted from the blog, you can see it in the feedback given to me earlier), we can today use two different methods:
- The one I used in the analysis I did on 2014. loans where I looked at current default rate levels and translated this into EL% to calculate the Expected Return. This would be relatively meaningless on such a fresh batch of loans as 2015. though, since only a very tiny portion of all defaults have yet happened on those loans.
- We could compare the Expected Loss of Bondora Rating V1 with the EL of Bondora Rating of V2 for the same loans. Since the EL% is only affected by probability of default, exposure at default and loss given default, which are all measures of default rate and recovery and the interest rates for the already issued 2015. loans aren’t changed, then we can use it easily enough.
Mind you, Bondora is claiming that the previous model was accurate in 98% of the cases AND the new one is even more accurate. So let’s rely on it being more accurate and thus consider that any differences from previous one in regards to EL, mean that new model has given a more accurate figure for this.
We will not look at the actual returns, because for this we would have to wait until the end of the loan period for those loans. What we will do, is simply comparing the estimations of loss for the same loans based on two different Bondora Rating models.
In a case where previous model was accurate in 98% of the cases, you wouldn’t expect to see too much difference in the EL% for loans that were issued in 2015 and have both the Bondora Rating V2 and V1 estimates. The other elements that were updated within the model, such as country risk, risk-free rate of return and market rates of return should only affect the interest rate, not EL%.
You can download the analysis file from here.
Changes to Ratings
First, let’s look at the Bondora Rating level, how many loans have received same or different Rating with this new more accurate scoring model.
In the first row is the total number of loans in each Rating based on model V1. In the last column, are the number of loans within each Rating based on V2.
Bold cells within the table show loans where Rating stayed the same. All the numbers below the bold cells are loans where model V2 considered the loans as higher risk than V1, and all the numbers above the bold cells show the number of loans that the new model considers as a lower risk Rating compared to V1.
In total, more than half of the loans issued in 2015, are considered as higher risk Rating by the new model, compared to the model that supposedly was accurate in 98% of the cases. And 15% are considered as lower risk.
Just to illustrate the scale of difference between the opinions of those two models:
- V1 thinks 43% of Spanish loans are HR. V2 thinks 93% are HR.
- V1 judged (and priced accordingly) 5.5% of Finnish loans as HR. The more accurate V2 model puts 51.5% of these same loans into HR group. That’s almost a 10x difference.
And I don’t think we really even need any percentages to see where most of the Spanish and Finnish loans are according to the updated, more accurate model. I don’t think any of the more statistically savvy investors and a lot of the others disagree with the new model on the direction of this assumption.
Changes in Expected Loss
But let’s take a closer look into the changes, because we should look at what the E(R) is now in the light of these new and more accurate EL% figures and if we can really claim that Bondora Rating V1 has been accurate in 98% of the cases about the returns and the loans are expected to outperform the previous expectations.
Mind you, in order to outperform the previous expectations, the V1 model should have been overly optimistic on these loans, meaning that the new model probably should have considered these loans as less risky.
In the following table, the Rating on the left scale is based on model V1.
So wait a minute. According to the more accurate new Bondora Rating model V2, at current prices that the model V1 set, we are to expect a negative return on about 16% of the portfolio and a significantly lower return than V1 expected, on about additional 34% of the total loans issued in 2015 before model V2 was introduced (Finnish C-F).
In total, the new Rating expects the return to be somewhat lower for close to 90% of the loans (issued from 1st January until 3rd December 2015) compared to what was expected by model V1.
This could be partly because the unexpected loss figures have changed and Expected Loss is now assumed more accurately. It wouldn’t help with the accuracy of the previous model though.
P.S. This tells us nothing about the actual returns. Just the assumptions of the two models and the relative accuracy of previous model compared to new one. For actual returns, we need to wait for a while to see any meaningful data.
Returns are outperforming?
By now you already know this formula inside and out:
E(R) = I – EL%
This means, that if EL% is increased, and Interest rate stays the same (like in the case of our 2015. loans), then Expected Return for those loans is lower than it was before.
On average, the new model thinks that for those same 2015 loans, the Expected Loss is:
- 0.4% on average higher for EST loans (it’s relatively evened out because it’s lower for higher risk loans and higher for lower risk loans)
- 19.1% higher on average (meaning 19.1% lower Expected Return on the 18.2% Expected Return figure) for Spanish loans
- 11.3% higher on average (meaning 11.3% lower Expected Return out of the 15.5% Expected Return based on V1 model) for Finnish loans.
The method used by Bondora to measure the performance of Bondora Rating is invalid and should not be used in this fashion EVER. I assume Bondora will promptly fix this mistake and let’s investors know about the unfortunate misleading information. Although, if I understood the replies in the comments correctly,then Bondora didn’t apparently claim anything about the accuracy or performance of Bondora Rating…
Judging by the EL% estimations of Bondora Rating model V1 and model V2 on the exact same loans at the moment those were issued, we are left with understanding that the EL% figures are vastly different from model to model. In fact, the average absolute change in EL% compared to EL% in model V1, is a whopping 97% (for example a 10% EL% according to V1 would be 19.7% or 0.3% on average according to V2).
We have also seen that in the majority of the cases, the EL% is (often considerably) higher according to model V2.
This puts us into a bit of conundrum, since Bondora has led us to believe that the following statements are true:
- Bondora Rating V1 has accurately predicted return in 98% of the cases and returns are outperforming expectations.
- Bondora Rating V2 is even more accurate than its predecessor V1.
Now, if the first statement is true, then the fact that according to the EL_V2 on the exact same loans, we are expected to lose money or earn less than Expected by V1 model in about 90% of the cases, should mean that model V2 has been made less accurate than model V1 in at least several cases. In other words, we are assuming that these loans should be performing even worse than initially expected, while the actual performance supposedly is at par or better than expected for 98% of the cases.
If the second statement is true, it would mean that the first statement can’t really be true…or can it? (it can, if you compare apples with asteroids obviously, as seen with the XIRR and Expected Return comparison)
Be as it may, I’ll leave you to judge it.
Which claim do you think is (more) correct?
There’s also a third option for the skeptics out there that neither of them are correct. We’ll start to get some idea about the second one in 6-12 months’ time though.
P.S. We could also look into the other claim from Bondora’s announcement of “Net yields on Bondora are highest in the marketplace lending industry”, which, judging by the pop-up I’ve seen, they seem to be comparing returns with Lending Club and other platforms who use different logic to calculate those returns.
For example, Lending Club discounts fully all loans that are charged off, meaning loans that are overdue between 120-150 days max. I’m pretty sure Bondora’s XIRR would show somewhat different results if we did the calculation using similar logic and tried to make those two numbers comparable.
But that’s an entirely different topic on its own…