Nordigen
Approve all creditworthy customers

Blog

What inspired us to build credit scoring models from transaction data

A short summary of case studies

FICO, the U.S. analytics software company, recently announced that bank account transactions add a lot of value in credit scoring. Previously, transaction data has been a luxury that only retail banks could afford, but today any lender is able to get this data via third-party banking APIs or directly from banks, as soon as Payment Services Directive 2 (PSD2) in Europe and Open Banking initiative in UK comes in to full force.

It feels obvious that transaction data can explain a lot about loan applicant's financial health, but in practice its biggest weakness is preparation — its possible to extract value from just account turnover and balances, but if you want to get higher uplifts, you'll want to categorise transactions and engineer features that capture effects not currently captured by your existing data sources.

This is where many data teams give up. All this categorisation and feature-engineering takes too much time.

What inspired us to keep researching the area of transaction-based behavioural scoring, despite it's steep learning curve, was the fact that there is real evidence of the positive impact it can have on lending decisions.

Below we've compiled a short list of research and case studies in the area of transaction-based behavioural scoring.

1. Transaction data is predictive

2. Transactions have proved to add value

3. Many lenders are already using transaction data in making credit decisions

To conclude, there are real-world examples that transaction data can add value to traditional credit scoring models. It's true that the field still seems to be in its early stages, but you can be sure that the easy access to transaction data means that more and more lending companies will be investing in this area, looking for a competitive edge.