Monthly Archives: May 2014

Credit Scoring at Novum Bank: Data Mining defines success in high risk lending


Original article by Marcel Wiedenbrugge

Imagine you are active in the provisioning of (micro) credit and a customer wants to borrow temporarily a few hundred euros from you. How do you determine whether it makes financial sense to do business with this customer? For Joop Bruinzeel, Chief Credit Risk Officer (CCRO) at Novum Bank, this question is just another day’s issue. As a provider of micro-credit, Novum Bank daily provides relatively small amounts (from €100 to €600) to customers where traditional banks have no interest due to a high risk profile. Properly set up and tuned credit risk management is essential.
For assessing credit applications, Novum Bank recently started using STATISTICA, the analytical software solution from StatSoft (now a part of Dell). In this interview I speak with Joop Bruinzeel about micro-credit, the importance of credit scoring and the use of analytical software.
What are the activities of Novum Bank and what is your role?
Joop: “Novum Bank is a Malta-based bank with a full European banking license. We are a specialist in the field of payment: Consumer credit and prepaid debit cards in particular. Our cards division is offering white label programs in all European countries. Currently, we are strongly represented in Germany, where you can find our products at almost all petrol stations.
“Regarding credit, we focus primarily on high risk, short-term loans. At the moment we are active in The Netherlands, Poland and Spain. For the past 1.5 years I have been a member of the Executive Board in the position of CCRO. That means that I am responsible for credit risk, both in the area of consumer credit as well as in the field of risk analysis regarding the daily use of prepaid debit cards.”
Can you update us on the high risk short term loanmarket? What is your differentiator?
Joop: “We are in a special position because no other banks are providing this type of short-term loan. Traditional banks often consider the risk to be high and the reward (risk/margin/revenue ratio) too low. Additionally, traditional banks are dealing with a different cost structure. We have chosen to operate lean and mean and invest in technology. Our partnership with StatSoft is a logical continuation of our strategy. Currently, we are much better able to provide risk assessments, especially with regard to managing portfolios with a more complex risk profile. The combination of lean and mean and state-of-the-art technology ensures that we are able to achieve a significant competitive advantage, even despite challenging market conditions.”
What developments has credit risk management gone through at Novum Bank?
Joop: “Just eighteen months ago we were still dependent on human interpretation and gut feeling. The steep growth path that we have made and the expansion into several countries made a more professional approach mandatory. New markets and new ways of lending require a high technology approach. Furthermore, credit scoring plays a crucial role as StatSoft came into the picture.”
Can you explain the importance of credit scoring?
Joop: “Credit scoring is essential to us. The turning point between customers who will pay us back and customers who will not pay us back is very close. Expensive errors are easily being made.
“Statistically, we are in this market due to the law of large numbers. Modeling and testing of scoring models usually takes several months to complete before results appear. If the modeling process is not in control or not understood, you can find out—potentially six months later—that wrong choices that were made can be very costly. Credit scoring is, per default, important to us because we accept customers with complex risk profiles. Fluctuations, peaks and valleys, may increase sharply. Besides this, the scale of operation in several countries with different characteristics makes it necessary to manage credit scoring methods and techniques. The third point that I want to mention is that (near real-time) credit scoring allows (near real-time) customer feedback, whether a credit application is accepted or not.”
Why did you choose STATISTICA?
Joop: “Before I started working for Novum Bank, I immersed myself in modelling. I got back in touch with another company that was specializing in short-term loans and had (successfully) made use of STATISTICA. As the shareholders of Novum Bank required mature risk management, STATISTICA was perceived as a logical choice to go with.”
What functionality are you using and how does it work in practice?
Joop : “We use STATISTICA primarily to make a profit scoring model. This allows us to calculate the risk if we are dealing with a good or bad customer. We have cleaned and complemented the historical data of the past few years, after which we continue editing the data in STATISTICA for further processing and analysis.
“Four models are being used to determine which model works best. Although we use statistical logistic regression techniques, our main focus is the usage of data mining algorithms, such as Boosted Trees and Random Forrest. During data preparation the software clearly identifies the key parameters that affect the profit score model, which I think is a strong feature of STATISTICA. Once the best model is found, we apply the model on the historical data. The software clearly shows how much return we could have made on the portfolio if we had used the scoring model of STATISTICA.
What do you like about STATISTICA?
Joop: “The beauty of STATISTICA is that you can build decisioning models which you can test on older portfolios (also called backlog or backtesting). The workbench also offers the possibility to add your own insights to the models. That allows us to refine the models, so that we can achieve better results.”
Can you share with us your client acceptance process? 
Joop: “We are working with strict customer acceptance requirements, which are fully compliant with the rules set by the Dutch regulator AFM (Autoriteit Financiële Markt). The standard requirements in order to qualify for a loan are: At least 21 years of age, sharing a (recent) payslip, proving a steady income with sufficient funds, plus identification. In addition, there are twenty other parameters that we use to dynamically determine whether or not we accept anyone. With “dynamic,” I mean that, for example, we may or may not look at the age of the applicant, or whether someone is married or not. Based on historical data analysis, we know that these things can affect the likelihood that someone will pay back a loan (on time). We also look at less obvious things such as the point of time a loan is requested. STATISTICA can handle all of these parameters in the profit scoring model. This results in a final score, which serves as a basis to make a financially responsible decision whether we will grant a loan application.”
How are you dealing with fraud?
Joop: “So far we are doing this largely case by case. We check at least the standard documents for completeness, authenticity, and accuracy. We verify the payslips and check the age of the applicant against the identification number. In the future we will automate these kinds of control, as the verifying costs are relatively high.
“When you consider that we reject most of every hundred loan applications, then it will be clear that efficiency improvements and cost reductions have our ongoing attention.”
What was the implementation time of STATISTICA?
Joop: “A number of issues have played a role: First, I knew in advance very well what I wanted. Additionally the two-day training course was well conducted and focused purely on the functionality that we needed to calculate profit scores. We are experiencing a pleasant cooperation with StatSoft. All together we had a good working model in 2.5 months’ time, resulting in a significant improvement of the results in the test market.”
And the return on investment?
Joop: “So far we have applied the scoring model only to a part of the Spanish portfolio. We have invested in software, education, and time. Our return on investment was only four months. Taking it from another perspective, we are just at the beginning, as there are plenty of opportunities to refine the models.”
What are your future plans?
Joop : “I want to use STATISTICA not only for scoring, but also for its statistical functionality. There are plenty of opportunities to further improve and refine the models. We think about further modeling the portfolio, especially for marketing purposes. I expect that in the future even more data will be linked. For data analysis, STATISTICA will play a central role.
[This interview originally appeared in Dutch in “Credit Expo,” April 28, 2014.]