How to decide if the DCA deal is good?

IN SHORT:

• The percentage of actual 7+ days overdue CASES that end up defaulting, is roughly around 23%. This is the number you actually should be looking at when figuring out the costs.
• With current deal structure, risk averse investors who prefer lower risk / higher quality loans, are practically guaranteed to lose money with this deal.
• More mature loans, just as any other better quality segment, have a lot lower percentage of first case 7 day overdue loans actually ending up defaulting. In other words, implementing this new process to all old loans is almost definitely simply an additional cost for investors.

After Bondora‘s announcement to send all 7+ days overdue loans to DCAs for “a more aggressive collection strategy”, I made an attempt to run some numbers to find out what the possible outcome of this move would be for us investors, as we are the ones who will ultimately cover the costs associated with the DCAs.

Now, that analysis was something I put together in a day with very limited information available at the time and me thinking of this topic for the first time, so it’s definitely not the perfect be-all and end-all final word on the topic.

This post will be a bit deeper look into the possible effects of this deal and some thoughts on how to decide whether it’s a good one or not.

Some definitions

In the previous post I was using recovery in two different meanings interchangeably. To avoid confusion, I’ll define some terms here and try to use them in the defined context only. Hopefully this will make it easier to read and understand:

• recovery – repayments received from a loan after it has defaulted
• reperformance – a loan that makes a payment to become current or reduce overdue days after becoming 7+ days overdue
• relapse – a loan that becomes 7+ days overdue again after reperforming
• default – a loan that has defaulted (60+ days previously, 74+ days nowadays)

1. How to determine if DCA will be profitable?

The first thing you might want to ask when deciding whether this sort of deal should be implemented or not, is considering the likelihood of it being profitable for the investors, since they’re the ones paying for it.

Ideally, this would entail a proper controlled experiment over a long enough period of time, with a randomly assigned control group that doesn’t get sent to DCAs for comparison and with statistically significant sample sizes in all countries (and all DCAs if there are more than 1 per country) and preferably from lower and higher risk segments as well.

Getting it for all Ratings would probably turn out to be very difficult, considering the low number of loans in lower Ratings, but grouping them even in the fashion that Bondora has done previously as sub-prime and near-prime should somewhat suffice.

The period of time should be long enough to:

• make sure any possible effects aren’t simply seasonal, since some seasonality effects have been noted previously with Bondora’s overdue loan performance and
• to see how long the loans are reperforming and how likely and how often a relapse occurs.

You might not want to take this experiment to the extreme, since this is a relatively changing environment and we’re not looking to publish the results in a peer-reviewed scientific journal.

But, considering that the change is going to affect roughly half of all the loans on the platform going forward, it should be serious enough to give a clear indication of what the possible results would be with good enough confidence. Especially, since it’s the investors’ funds it will take to implement this.

Of course, if Bondora were to be so certain of a positive outcome that they decided not to run this sort of test, a solution where they will cover the costs if the outcome is not a desirable one, could be used to move forward with this idea. Of course, this requires a specific outcome as the benchmark and a way to measure for it, but we’ll get to this in a paragraph later on.

Unfortunately we don’t know the likelihood of it being profitable, since Bondora hasn’t published such analyses.

Bondora has mentioned some experiment that was done in January 2016 and I quote: “Certainly the period of testing was not long enough to make absolute conclusions…”.

Not only was it too short to make absolute conclusions, it was also too short to know anything about relapses. If the only effect of the DCAs is that defaults will on average occur with some months delay, it will be a lot cheaper to simply let them default and pay for the recovery of defaulted loans only in the first place.

If no proper experiment was done, then I’d expect to see some sort of extensive reasoning behind this decision shared with investors soon hopefully.

The most prominent piece of data shared with us is that 58% of loans going 7+ days overdue end up defaulting and 42% will reperform. The new DCA process is supposedly going to improve this.

However, these figures don’t tell the whole story. We’ll look into this in the next section.

2. Reperforming more than once

A reperforming loan may relapse and then also reperform again and so on for several times. If you have invested for a while and have been looking at your loans or loans from Secondary Market, you have probably seen many such loans yourself.

My previous post did not account for this fact because we only have one date for 7 days overdue available in the loan dataset, making it close to impossible to determine how many times it has actually happened. So we could only look at the percentages based on the first occurrences of 7 days overdue.

From what I understand, the 58% defaulting and 42% reperforming figures provided by Bondora are also based on first time 7+ day overdue figures.

Unfortunately, running calculations on purely these figures, is misleading, because if a loan defaults after reperforming twice, there are already 2 occasions where loan reperformed and cost for investor occurred and 1 occasion of default. These costs of 2 additional reperformances for defaulted loans are entirely unaccounted for with the previous calculations.

In other words, the actual proportion of reperformances from 7+ days overdue should be higher than the one previously indicated.

Thankfully, Bondora has provided us with a figure that a loan that has been overdue 7+ days, has been this way on average for 2.5 times. While it’s still not a perfect figure to help us with a more accurate calculation, we are able to use it to come up with some estimates with relatively decent accuracy.

In loan dataset we have the date of first time the loan was 7 days overdue and we have the date when a loan defaulted. Based on this, we can calculate the average time between those two events to get the average time it took for a defaulted loan to default.

Based on this, we can then calculate the amount of times on average a loan was possibly overdue for 7+ days before defaulting (dividing the average time by 60 days, the time it takes to default).

While we may overestimate the number a bit here with this method, it doesn’t really matter for the end result. We know the average should be in total 2.5, so we’ll even it up when calculating the average number of times for reperforming loans.

(I won’t bother you with the exact calculation logic in the post here, but if you’re interested, you can grab the DCA profitability calculator 2.0 with the formulas below and ask about it in comments.)

The results I ended up with, were following:

• Defaulted loans on average have reached 7+ days overdue 2.3 times before defaulting.
• Reperforming loans on average have reached 7+ days overdue 2.7 times.

Note that these figures are not a fixed ratio and only matter for this specific segment of loans and at this specific point in time. These ratios will change over time as loans mature or fresh loans are added to the portfolio.

These figures are not 100% accurate, since the data that was used to provide us with this 2.5 figure, is likely somewhat different from the data I used, but it should be close enough and will suffice to make the point.

Based on this new info, we can calculate the proportion of actual occasions of 7+ days overdue that lead to default vs those that have reperformed so far without the more aggressive DCA process:

• 23% cases of 7+ days overdue, have ended up in default
• 77% of cases have reperformed before adding more aggressive DCA process

Why did we do this? Simply because the fee is paid on every occasion of 7+ days overdue reperformance, not only the first or last case.

In other words, this will be the actual measure to use to determine how many additional cases of 7+ days overdue loans would have to reperform with the new DCA process to cover the related costs.

If we now put these figures into the DCA Profitability Calculator 2.0, we will see that the reperforming rate would have to increase by an additional 14% to 91% of the 7+ overdue cases reperforming to reach breakeven point.

This means that for a breakeven, roughly a whopping 58% of defaults should be additionally prevented from occurring by the DCAs.

Note: My method of calculation is relatively conservative. You could attempt to calculate some sort of gain based on preventing defaults, but since we don’t know any stats on relapses after DCA intervention, there’s no meaningful way to understand how much defaults are actually prevented, instead of simply delayed with additional costs. Thus it makes no sense to assume a gain in the future and ignore any additional costs in the future.

All of this is without accounting for the fact that if DCA is actually successful in making these large numbers of loans reperform, this means that some of them will relapse and default eventually.

This in turn will increase the number (and proportion) of reperformances in the process without necessarily reducing default rate as much in the end.

Grab the DCA Profitability Calculator 2.0

(You’ll be added to an inactive list for the delivery system to work and then deleted. If you want to receive further updates from the blog, subscribe from the end of the post.)

NB! While we don’t know the number of times a loan goes 7+ days overdue before defaulting per country, it is pretty safe to assume that Estonia is leading here because of the simple fact that the loans issued here have had the longest time to mature and go into this state more times.

EST, the country with lowest proportion defaulting in the figures before, is likely the one where the percentage of actual cases ending in default will reduce this figure by most as well.

3. Subsidizing lower quality loans

We may assume for the sake of the argument that my calculations were incorrect or the 58% figure provided by Bondora was actually too low. From the loan dataset we seem to be getting even lower figures, but let’s assume this anyway…

Once we have the data from an experiment such as the one in section 1, we can start thinking of the structure of the DCA deal we can use to bring this intervention to profitability.

The deal currently used, is a flat fee on all 7+ day overdue cases, irregardless of risk level or other criteria. I assume the prices are somewhat different per country due to different DCAs doing the work or whatever reasons, but we have been served this deal as a flat fee deal with 15% success fee on average per every payment during the process of making the loans reperform.

As we saw in the previous post, this is actually leading to a situation where the lower risk loans are essentially being used to subsidize the reperformance efforts of higher risk loans, because the proportion of loans that default are actually lower with lower risk/better quality loans.

In fact, what we also saw, was that Estonia had a significantly lower proportion of loans defaulting on average compared to other countries, but at least judging by the DCA fees table, it seems that Estonian DCAs collect one of the highest fees of the bunch. FIN and SVK were said to be at 8%, so EST and/or ESP have to be considerably higher than 15%.

This table is also based on the first 7 days overdue case only, not all the cases, like we looked in section 2.

This sort of flat structure is definitely not the one to use, if you take a look at the data at hand, because it will make it impossible to be profitable for lower risk/higher quality segments and would thus simply be an additional expense with no added benefit for already lower expected return segments.

This will essentially be using more risk-averse investors’ funds for subsidizing the recovery for investors who have decided to take more risk by investing into lower quality/higher risk loans.

A granular pricing depending on the proportions could be a possible solution here, but for one issue that we’ll look at in the next section.

4. Measuring effectiveness of DCA process

In a normal course of action, you would have your first measurement of effectiveness in the 1. section prior to implementing a new process such as this. Only in case of a clearly positive result, once you can be relatively confident of a profitable outcome, you will start implementing it.

Once you have implemented it though, it will become significantly more complex to actually measure its efficacy, which makes it even more important to have a good positive result from a proper experiment beforehand available.

I’m not an expert on how one can measure the success or failure of this process after it has been implemented in the fashion that it has. Currently the only reliable way I can think of, is creating occasional controlled experiments where some loans randomly aren’t forwarded to DCAs to measure if and what the difference with DCA process is. But maybe there are some other ways as well?

One thing I know that people would be tempted to do, is compare the proportions of defaulted vs reperformed loans with the results from the past.

This would be wrong for a multitude of reasons, most of which are outside the scope of this post, but let’s look at some simple examples that can be summed up by a phrase “correlation does not imply causation”.

What this essentially means, is that simply because two things occur simultaneously or one after the other, it doesn’t mean that one caused the other to happen.

In other words, if you were to look at the proportions of defaults in 6 months’ time and saw that the proportions are lower than they were before, one would be tempted to say that the DCA process has worked. Not so fast though.

Remember the result that better quality loans (lower risk vs higher risk or EST vs ESP etc) have a higher proportion of loans reperforming without additional intervention from DCAs?

Well, Bondora introduced a new Bondora Rating V2.0 in December 2015, which, among other things, “takes into account more parameters to estimate the probability of default of the specific customer even better”. In addition, 2014 was one of the worst quality borrower years in Bondora portfolio, outside of EST at least.

Read: the quality of the borrowers is bound to become better, especially in the lower quality markets of Spain and Finland, which previously were based on very low amount of data, and risk estimates weren’t exactly accurate.

As a result, given that Bondora Rating V2.0 actually delivers what it was supposed to, we would see a lower proportion of defaults and higher proportion of reperforming loans irrespective of any more aggressive DCA process. This would make the DCA process even more likely to be simply an additional loss to investors.

Of course, since the DCA process was added to the mix today, it won’t exactly be possible to separate the two effects. If you then compare it to the past, you could falsely conclude that it has been successful.

Even if it has actually been worse than initially thought because loan quality has increased and the proportion of loans that would have reperformed without DCA has increased and thus additional cost has been added instead.

It would also work the other way around, if for example the loan quality were to deteriorate for some reason (let’s say a macroeconomic event), it would be impossible to detect the efficiency of DCAs and one would be led to a conclusion that they’re not efficient. Even though they could have been preventing an even larger downturn.

In essence, the current model is putting the investor into a double bind – if loan quality improves, investors pay more for unnecessary DCA actions and lose the improvement from this, possibly making better loan quality a negative outcome for investors.

If loan quality deteriorates, the unnecessary DCA costs will become lower, but investors will lose due to drop in loan quality.

5. Sending existing loans to DCAs

Another aspect of doing such a change is the thought of whether it should be implemented going forward only, or if this should also be applied to loans issued in the past. Thus changing the outcome from these loans and additionally making the considerations the investors made when making these investments, partly or fully irrelevant and affecting the profitability of these historical transactions (especially any Secondary Market purchases).

You can probably guess what I think would be the correct way to proceed here, but I’ll mention anyway that Omaraha did a change of a somewhat similar scale in 2014 by introducing a forced selloff of all 90+ overdue loans at a price of 60%-80% of outstanding principal value.

The process was announced roughly 3 weeks before going into effect, all investors had to manually agree to the new terms to continue investing after the new process took effect and all of the loans issued prior to this process continued with the old rules with none of them being sold automatically with the new rules.

I personally didn’t initially like the new process, but I did appreciate the option to opt out.

To support my opinion, let’s look at some data:

For example, the graph shows that if a Finnish loan goes 7 days overdue after 1 month of origination, then roughly 90% of those have ended up defaulting afterwards, if however it goes 7 days overdue for the first time on month 3, the likelihood of default happening has been around 50%.

Just in case the previous graph had too much going on and some of you missed it, here it is again with all countries together on a single line:

These graphs only show the proportions based on first time 7 days overdue cases. My guesstimate based on when most defaults happen, would be that looking at all instances of 7 days overdue, would make this graph decline even faster.

Any investor with a bit more experience would know that the quality of paying loans increases with maturity, so this is not too big of a surprise here. Especially since this pattern has been visible in every segment where loans are of higher quality.

So, you know my opinion, but based on the maturity of your portfolio, do you find it reasonable to implement this new DCA process on already existing loans?

How reasonable do you find sending EST loans, lower risk loans and mature loans to DCAs?

P.S. This also affects the results in section 4 in same way as increase in loan quality, since this essentially is increase in loan quality because lower quality loans have already defaulted.

Disclaimer: Please be advised that all of the calculations for the DCA profitability are based on all reperforming loans paying for DCA fees. In reality, it is possible that some percentage of these loans will make payments directly to Bondora and thus sidestepping the DCA fees. Unfortunately we can’t account for this, because we have no figures on it. These could be obtained as well through an experiment such as the one described in section 1 of this post though.