Can we predict who is more likely to repay a Balloon loan?
Throughout 2018 we collected over 500 data points on each entrepreneur we supported. This data is collected to measure impact but also to make us smarter in answering key questions. One of these key questions is: what (if anything) predicts which entrepreneurs are more or less likely to repay the loans that Balloon gives them?
We recently started a project to comprehensively analyse this question and have some preliminary findings we wanted to share. With that in mind, the findings presented here are not conclusive, they are just what we’ve learnt so far.
What we did
Our starting point was to review the loan repayment records of all of our entrepreneurs who had completed one of our programmes in the first half of 2018. We then grouped these entrepreneurs into two groups, those who were repaying their loans on time and those that were not (including those who defaulted and those who were late in their payments). Then we used statistical methods (something called stepwise logistic regression) to tell us whether any programme related data predicted whether an entrepreneur ended up in the ‘good repayments’ or ‘poor repayments’ group.
Before jumping into the findings, let’s look at what predictors we considered. There are two important bits we looked at in this analysis. The first, is data on the entrepreneur’s characteristics. Our work involves providing long term support to entrepreneurs. Therefore, they get to know them very well. To leverage this knowledge, we ask our staff to provide character references of the entrepreneurs they are working with. Amongst other things, this captures how passionate entrepreneurs are about their business, how opportunistic they are, their knowledge of finance, etc.
The second chunk of data comes from our loan application and appraisal process. This includes three parts: preparing an investment proposal document; a visit to their business; and an investment meeting (a 30-minute presentation and Q&A).
An entrepreneur makes their pitch below:
At every step, our loan officers are gathering evidence to score the entrepreneur on Balloon’s 12 evaluation criteria. These are criteria that we have learnt are important for business success. Examples include: proof there is demand for the entrepreneur’s product/service; the entrepreneur’s experience in running similar businesses; and the financial viability of the business.
So, in summary, we looked at whether the character references and loan officer ratings of entrepreneurs predicted loan repayment performance. What did we find out?
What we learnt
The character reference ratings were important predictors of whether someone repaid or not. Interestingly, amongst all of them, the best single predictor was ‘appetite to take risks’.
For example, if an entrepreneur was rated a 4 (out of 5) instead of 3, on their appetite to take risks (i.e. they are riskier), they were five times less likely to repay their loan. This makes sense, people who are riskier might be more open to not repaying a loan (when they have the money to do so) but also might engage in overly risky behaviour which causes them to lose the money they would’ve used to repay their loan.
For loan officer ratings, there were similarly positive findings. Here, the strongest individual rating of the 12 pitch criteria was: ‘The entrepreneur can repay the loan within twelve months, taking into account costs of the business and any salary or personal expenses’. Once again, a one-point difference on this rating was associated with an increased likelihood of repaying the loan of over four times.
This is promising because it means that our loan officer assessment of the likelihood of loan repayment is actually accurate.
Findings like these are exciting. They mean that we can become smarter in our decision making, using data and evidence to maximise impact. For example, we might in the future choose not to give loans to people who are rated as riskier.
To make those kinds of decisions we have to collect a lot more data to be absolutely sure that what we are seeing are true patterns in the data. We also have to check that ratings on these variables are not biased against any demographic as this might then make us unconsciously biased in how we allocate loans.
For now, we are doing two things. First, we are finishing off the rest of the preliminary analysis to see what else we can find. Second, we are laying the groundwork to collect even more data to strengthen our conclusions. With this, we can begin to make practical improvements across our programmes to maximise the impact we create.
This piece was written by Nicholas Andreou. Nicholas leads the Insight & Impact function at Balloon Ventures. He holds a PhD from the University of Nottingham and has previously held research positions at Harvard University and the World Health Organization as a student.