Tauri: I am happy to announce the very first English written guest post in Rahapuu’s blog (by the way, Rahapuu = MoneyTree). First of all I would like to thank David for such a great piece of overdue statistics he has made. I was impressed by some of the results that are covered in this analysis below. Hopefully all of you, secondary market traders, are looking this information that David is sharing with us. Any ideas and thoughts are welcomed in the comment section below. Happy reading!
Introducing a better way to evaluate the overdue portion of your portfolio
It always happens this time of the month. I eye my Bondora portfolio and feel that sweaty feeling of impending doom fill my belly when I see the size of the orange portion of the investment pie chart.
Panic soon turns to general outrage. I mentally shake my fist at the screen in front of me and feel like shouting “Time to pay up, you deadbeats! I’m not running a charity here!” But then something surprising happens near the end of the month: people start sending in their payments, and the big orange wedge of doom retreats, allowing me to rest again, at least for a few more weeks until the cycle repeats itself.
Experienced investors have learned over time that late payments are a regular and acceptable part of p2p-lending and that the majority of borrowers eventually cough up their money, despite being a few days or even weeks late. Some investors even attempt to capitalize on the panicky risk avoidance of those investors who sell late loans at juicy reduced rates, gambling that history will repeat itself and that there will be a nice pay off. But how can we predict which overdue loans will really turn into defaults and how can we use this knowledge to better evaluate our portfolios or even make an extra coin or two? Lucky for us all, work has been dead as usual around here during the summer, so I made it my summer project to look for answers in the Bondora dataset.
My gut feeling regarding overdue loans has up until now been a belief that “more days overdue means higher risk for default.” But exactly how much more? And is there any interesting trends in the data that might be useful?
Using the Bondora dataset, I filtered out all but Estonian loans from the time span of 2011-2013, reasoning that most of the loans destined for default will have already done so by now. Here’s some general stats for the 3 153 remaining loans**:
51.6% (1627) of loans were 1 day late at least one point in time
25% (795) of loans were 14 days late at least one point in time
16.8% (530) of loans were 30 days late at least one point in time
12.6% (397) of loans were 60 days late at least one point in time
12.0% (379) of loans were listed as having actually defaulted
** includes all credit grades and groups, ABC 600 – ABC 1000
Here’s another way of looking at that same data:
23.3% of loans that were 1 day late at least once ended up defaulting
47.7% of loans that were 14 days late at least once ended up defaulting
71.5% of loans that were 30 days late at least once ended up defaulting
95.5% of loans that were 60 days late at least once ended up defaulting
Charting this data reveals an interesting trend:
Whoa! That is a very pretty linear relationship we have going on there. This is convenient because we can use known variables (overdue days in this case) to factor for the unknowns (ie, average default rate). For example, let’s say you have a loan that is overdue by 11 days and want to know the risk for default. Eyeing the above graph and using some rough extrapolation allows for the following estimation:
So the average default rate for such a loan would be around 40-45%. You could use the same method to estimate the average default risk for any overdue loan. Curious as to how this relationship would hold up for the different credit grades, I did a breakdown of the data for grades 1000-600. Here’s a plot of what I found:
They say that a picture is worth a thousand words, so I’ll keep my comments brief and let you draw your own conclusions for the most part. Worth pointing out is that, around the 14 day overdue mark, the average default rate becomes approximately 50%. This is regardless of credit grade. And our beloved grade 1000 category, that had been performing so well compared to the other grades, shoots up to highest default risk around the 30 day mark. Why? I’ll leave the guesswork to you. The most valuable piece of information to bring from this, in my opinion, is that all loan groups generally behave in the same linear pattern, with each passing overdue day decreasing the chance that the borrower will start paying again (before the bailiff knocks on their door, that is). Nothing shocking I know, but once again, having approximate numbers is always better than just a gut feeling.
But what about the other markets? How do these compare to Estonia? Even if they are still young and have the loans have not yet matured enough to see the full default rating, we now have enough data to see some patterns emerging, at least for Finland and Spain**. Slovakia will have to wait for now.
** 348 loans for Finland and 253 for Spain, after filtering out for 60 days from last payment
The first thing that you will notice is that there is quite a difference between countries when it comes to the overall default rate, including after one day overdue. Another thing that sticks out for me is how well Finnish loans are performing at the 1-14 day overdue mark compared to other markets, something that Bondora has mentioned on the forums. Spanish loans, on the other hand, have a higher total default rate and are more prone to default when overdue early on. But look at how the difference in default rate between markets seems to decrease at the 14 day mark. What’s more, at the 30 day mark this difference has all but disappeared. This means that a loan has essentially the same default risk ( ie, around 70%) after 30 days overdue, regardless of its market origin.
So how can you put all this information to use? Where as before we had to rely on our gut feeling telling us that “more overdue is worse”, we can now better estimate the actual value of the overdue portion our portfolios. For example, knowing that the average default rate of 30-day overdue loan is approximately 70%, you can substitute this value into your net return calculations to get a better indication of the value of those loans.
((1 – seasoned default rate ) * (1 + average interest rate) + (seasoned default rate * expected recovery rate))-1)
((1 – 0.70 ) * (1 + average interest rate) + (0.70 * expected recovery rate))-1)
The wildcard in the above calculation is of course the expected recovery rate. How to estimate that is an entire different discussion and beyond the scope of this article. Which is fancy way of saying “I don’t know yet.” Bondora has some estimates that can be used in a pinch, until something better comes along.
Another potential use for this information is in evaluating overdue loans on the secondary market to fish for potential bargains. Late loans still have a value due to the chance of becoming current again and the potential for recovery, thus some nice deals can be readily found for the investor willing to play the numbers game.
It perhaps goes without saying that all of this information is based upon past performance, which may or may not hold true for future loans. Never the less, we at least now have more than a gut feeling to guide us in our decision making based upon overdue loans.
Oh, if you should see any of my deadbeat overdue borrowers, tell them I want my money!
About the author:
David James is a happy amateur statistician and p2p investor, managing portfolios in Bondora and several other platforms. He is also self-subscribed “Bogglehead” and minimalist.