Studies of Life

Learning by doing.

Filtering Bondora Loans for Fewer Defaults

22 August 2014 by Jim

As some of you may know, Bondora is from Estonia, and Estonia has a wonderfully digital approach to almost everything. When you file your taxes in Estonia, you do so online, and within a few days you get money back if the government owes you, for example. Other nice features include the possibility to digitally sign documents in a legally binding way, and precise control and monitoring of how and when the government uses your personal data.


So it’s not surprising that the main P2P lender from Estonia is ambitious. As you may know, Bondora wants to become a pan-European lending service, catering to the biggest economy in the world with its 500 million potential customers. In order to allow such a lending system to function on such a huge market, where the national requirements for every country may be different, you need a really good system, and that’s what Bondora is working on.

For those interested in the actual loan data: you can download a CSV (currently about 6 MB in size) of all the loans on Bondora to analyse yourself. I’ve done this, and I’ve used Excel’s pivot table feature to analyse the dataset. If you want to know how pivot tables in Excel are used, check out this video on Youtube. The same features are also available in Google Spreadsheets, in case you don’t have Excel. Basically, pivot tables allow you to compare different parameters to others by grouping the data from a huge table like Bondora’s CSV file into different categories that you can customise.


So what are the results of my analysis?


First, I’ve only been looking at Estonian and Finnish loans in my analysis, because Spanish and Slovakian loans seem to be 1) less numerous, so the analysis isn’t as meaningful right now and 2) somewhat more risky, so I want to restrict my investing to the two countries for now. The following analyses are therefore applicable to loans from Finland and Estonia only.


Next, what do we look for in our analysis? The main problem with P2P investing are the default rates. If borrowers stop paying their debt, we don’t earn any return, so we’d like to reduce the amount of defaults as much as possible. For individual borrowers, it’s impossible to say with any kind of certainty whether or not they will default at some point or not. So we look at probability of default by searching for groups of borrowers that have default rates below average.


The parameter used to indicate default was the ‘In Debt 60 days’ column, which indicates whether or not (‘1’ or ‘0’) the loan has, at some point, been 60 days overdue, no matter what happens after the default. In many cases, defaulted loans are eventually paid back with a delay, but we want to know how to avoid borrowers with a higher probability of default altogether.




The first things I looked at was the home ownership. The categories ‘tenant, unfurnished’ and ‘mortgage’ had the lowest percentage of defaulted loans. The mortgage makes intuitive sense for me, because when a borrower has a mortgage, he 1. is used to paying back a loan and 2. has something to lose if he doesn’t consistently repay his debts, so his overall discipline might be higher. The ‘tenant, unfurnished’ category could make sense in so far as people who live in apartments that are unfurnished probably have some cash on hand to spend on new furniture, so they might be overall more financially healthy than people who do not have any cash for such expenses. These explanations of why these two categories seem to be better are pure conjecture, I do not have any evidence for their validity. I only have evidence for the lower default rate, but not for my explanation of why this is. In the end, it doesn’t matter if we can explain it, as long as we can take advantage of it.


Home Ownership Stats


Same thing as before, the numbers in the leftmost column represent a code for different occupational areas that can be looked up on Bondora’s website. The low default rates highlighted here belong to the categories: Utilities, Real Estate, Research and Administrative.
Occupation Area Stats

Unfortunately, these two first aspects we’ve looked at cannot be easily filtered for when looking for new loans to invest in. You can select one type of home ownership to search for, and one occupational area, but you can’t select a group of values that are OK for your search. So to filter based on the results for occupational area, we’d have to execute 4 searches. Other parameters are however easier to use when searching for loans. Here they are:
3) AGE
As you can see on this chart, the default rate goes down the older the borrower gets. And there’s a quite visible drop at age 26, so borrowers younger than 26 have a higher probability of defaulting. It seems intuitively correct that as people get older, they take fewer risks and may take on a loan only if they are sure to be able to repay it, whereas younger borrowers may be less cautious and therefore more risky overall.
Age Statistics

The next parameter is the credit rating on Bondora (500 to 1000). we can see in the chart below that the default rate is very high for 500 and then quickly drops. Personally, I focus on loans with a credit rating of 800 or higher.

According to the chart below, loans with a longer duration seem less likely to default. Does that make intuitive sense? I don’t think so. I think this chart is not reliable as a source of information because Bondora is not that old yet, so the amount of loans that are repaid over 60 months / 5 years is still very low. Therefore, I will not use the information in this chart for any practical investment purposes myself.
Duration Stats

Here we see that it’s better to focus on fully employed borrowers than partially employed ones, and entrepreneurs and retirees seem to be an even better bet: entrepreneurs are on average more financially robust than employees, and retirees are generally older, so they take fewer risks than younger people would.

Your financial wealth depends on where you live. But Bondora operates in Europe, and the living standards, although quite diverse overall, are not as different as those of Switzerland and Nigeria. Even though we are looking at income data from Estonia and Finland together, we can clearly see that there’s a somewhat smooth slope in the chart: the more people earn, the easier it is for them to pay back their loans and not default. So who should we exclude from our investments? In order to still find enough opportunities, I focus on borrowers with 900 euros in income or more. This cuts out most of the low-earners.

Income Stats
To make this a bit clearer, below is the average default rate of people earning less than 900 compared to those earning more. Quite clear, isn’t it?
Income Stats 2

Finally, the amount funded. As you can see in the chart, the only real outlier seems to be loans for 0-1000 euros, which default in almost 30% of cases. Why? 1000 euros is not a lot of money, so if people need to borrow 1000 euros, this means they don’t have any cash or savings (and probably no job either, since a monthly net salary is about 800 euros in Estonia and 2300 euros in Finland), so they are not very healthy financially. It’s best to not invest in those kinds of loans.

The ideal loan to invest in, therefore, would be a loan to a borrower older than 26, with a credit rating at or above 800, who is fully employed, earns more than 900 euros a month and wants to borrow more than 1000 euros.
This is what I now use to filter the investments on Bondora before I invest. You can now easily save searches when looking for new investments, and you can invest in bulk to more quickly invest your money on Bondora.
What do you think about these criteria? If you have any other interesting conclusions, or a better explanation for the results I’m seeing here, let me know in the comments!

7 comments | Categories: Investing | Tags: , , , , , , , ,

Comments (7)

  1. Pingback: Tipps für Bondora | geldexperimente

  2. 1.Do you mean 900€ Income before or AFTER calculating all other obligations,cause some have 3000€ income, but only 400 left for a living after deducting debts and living expenses?
    I mean i can read anywhere Estonian borrowers are the saftest,however my first default was Estonian with 1k income, but only 200 left for a living after deducting debts and other expenses.The income gap seems to be to me in that case.
    2.Interesting would be filtering by the new features at Bondora ,fully verified,income verified,income not verfied.
    3.Divorced women above 28 seem to perform better.Singles are the worst.What about married,what about married with child(ren)?
    4.Avoiding small debts is new to me but sounds logical,so that will deffinetly a point i will look for in the future.
    5.I would give my data away to you for further investigation,i think this filtering is a very profitable thing if done correct and extensive.
    6.Job would be interesting , I read constructors and other not so high educated jobs (fisherman,forest work) do significantly worse that social/mediacal workers.Education surely belongs also to this area.
    7.Do they already have debts ?
    (how many different debts at how many banks,whole sum ,mortgage,car loans).A 24 year old male ,living with parents(amazingly a higher than normal risk)having 12 different ,even small,consumer credits is a huge favorite for default.

  3. Here are a few answers:
    1. I use Bondora’s dataset for the income level, and I suppose it is 900€ TOTAL income, i.e. before other obligations. Estonian loans are overall safer than other countries, but that does of course not mean that every Estonian loan will perform as expected. Diversification is key here.
    2. You’re right. I’ll add a new post soon where I use the new dataset.
    3. Good point, too. I’ll have a look if this is included in the dataset Bondora offers.
    4. Small debts are an indication that the borrower is very bad at managing money. If he needs to borrow 1000€, that means he has NO SAVINGS whatsoever. And that’s not a good sign.
    5. Look up “pivot tables” in Excel, that’s how I did the analysis. It’s quite easy and fun to experiment with this.
    6. Jobs do have an impact, but it’s not a good overall criterion because if you limit yourself to just certain types of jobs, the pool of loans you can invest in shrinks dramatically. You need to use filter criteria that do not eliminate too many loans if you want to invest a lot of money.
    7. I suppose they often do (the purpose of a Bondora loan is often loan consolidation) but this should already be included in the disposable income figure. A 24 year old male living with parents could be a higher risk than normal because he’s not yet used to being independent and paying his bills, and his situation may change radically very soon (moving to a different city or country, getting a new job, marrying someone, etc.) whereas older borrowers who are married and living in their own home have much more to lose and a more stable situation.

  4. @Point 6 :
    I agree we should not exclude certain jobs.
    All criteria, including Job, have an impact,and so in my oppinion we should keep at least an eye on every single criteria including the job.
    As is dont want to exclude a certain criteria (same applies for council house vs mortgage),i have a basic investment of lets say 50€ per loan,but add 1 degree (=10€) for every positive, and deduct one degree for every negative criteria.The higher the risk seems,the lower the loan and vice versa.
    I dont give flat loans,its higher the better the customer fits my criteria.
    A constructor living in a council house with 4 existing consumer-loans will get a max of 20€ from me if at all(depends on age and income),wheras an entrepreneur with mortgage and zero loans exept property will get 100€, or even more if hes between 40 and 60 years old ,maried and 2 children.
    @Point 7: New for me but interesting ,these young people where never trained for independance,and so they have to learn the hard way,but not with a loan from my pocket in the future.First i thought living with parents could be a good criteria, cause the parents might pay the debt,but thats seems not correct.

    In review,if i would have had this knowlegde about default-criteria 1 year ago,i would have a minimum of 50% less defaults.

    • It’s a good idea to increase the amount you want to lend for every positive criterion. However, this is difficult to do if you invest a lot of money. With the size of my current portfolio at Bondora, I do not have the time to manually invest and I use the portfolio managers to do it automatically, so I cannot adjust the loan amount individually. But maybe your strategy could be automated IF Bondora decides to publish an API for their loans one day. This was something that people talked about in the forum and maybe they will allow for that one day. Maybe they already do for institutional investors.

      Personally, I don’t want to spend my entire day managing my portfolio, so if I need to spend an hour each day to increase my overall return by 1% per year, that’s not really worth it for me. In general, I am looking for information that can help me increase the return with AS LITTLE WORK as possible.

  5. Nice analysis. Good tip about the smaller value loans – I might avoid these in future! I did a bit more analysis here:

    This article relates to the Portfolio Manager factors, I’ve also done articles relating to other factors (e.g. age, employment type).

    My results seem pretty similar to yours, except I used the AD field = 1 for bad loans, and I only did research for Estonian loans.

  6. Pingback: Bondora high yield peer to peer lending reviewed | Investment Hacking

Leave a Reply