Bondora's analysis of DCAs

I was happy to see Bondora finally provide some reasoning behind their implementation by sharing with investors a calculator for the DCA costs and benefits.


However, it seems to have some calculation mistakes and a few questionable logic points in there.

Which kind of makes me worried that no proper analysis nor calculations were even made before implementing this new process. Which would be pretty weird for something that affects roughly half of all the loans on the platform.


Pärtel won't see this post, since he blocked me the last time I answered his comment here, but I'm sure there's plenty of eyes that can ask him to fix the points you find relevant.


1. Using single value for all countries for:

- Average principal balance at default is very much dependent on portfolio maturity. For EST it's actually lower, for ESP it's higher than the average used.

- Average collection fee should not be used to calculate per country results. Especially since FIN and SVK had 8% fee level announced in the previous deal. Average for EST and ESP is thus likely higher than 15% and for FIN is significantly lower than 15%.

- Cost of loss is definitely not the same in every country.

- Average number of late installments is also bound to be a lot different per country. EST being likely the highest by a big margin and ESP the lowest.


2. Deducting defaults fully

We can't make provisions for all future losses for the defaulted loans and assume that on the other side one intervention from DCAs will make the loans perform perfectly until maturity.

It looks like it should be treated similarly to the return calculation, where Bondora makes no provisions for future gains and thus is also not deducting losses any more than by the principal overdue amount at that point (roughly 2 mo. payments in this example).


Only once you know the stats for how many of the reperforming loans actually relapse, after how long from reperforming, and how many times, and you account for those in the calculations as well, you can start making estimates that go fully to the end of the loan period for both sides of the equation.


3. Incorrect use of average late installments

This figure is supposed to account for the fact that loans that go 7+ days overdue, do so on average 2.5 times.


The current formula used considers the 2.5 platform wide rate to mean that only the finally reperforming loans are in this status for 2.5 times while the defaulting loans are not.

Essentially, with 48% defaulting loans, it assumes that the loans go overdue 1.8 times on average instead of 2.5.


One could also find questionable the use of improvement rates gotten from an unknown sample of loans on a very short (1 month?) experiment, but there are other issues to fix here first.