BNP Paribas Fortis Chair: research for a changing world

The world has been experiencing rapid economic change in recent years. As a result, policymakers need to act more frequently and more quickly. However making, evaluating and revising these decisions with half or no current insights is a hopeless task. Banks are at the source of economic activity as they directly process a large portion of economic transactions. Consequently, once fully anonymized, these transaction data could be used to make good estimates of, say, national growth and inflation rates before they are officially known. Professor Schoors refers to this exercise as "preplication," replicating national figures before they are published by national agencies. This chair was established to explore, within a strict compliance framework, how the social impact of insights from fully anonymized banking data can be positively amplified.


Three major challenges

Getting from anonymized individual transactions to socially relevant and actionable insights, requires a transparent approach. Machine learning techniques efficiently extract connections from large volumes of data, but there are three major challenges in applying them to support policy decisions. These challenges are the focus of the research sponsored by the chair.

The first challenge is that the flexibility of these techniques to capture highly complex relationships often comes at the expense of the interpretability of the results. This while for decisions it is just crucial to have a good understanding of why a prediction is made by the model to ensure that decisions are made on a correct basis.

The second challenge is that policy decisions prescribe interventions with a particular goal in mind (e.g., subsidies to help the disadvantaged). But machine learning is primarily focused on finding observational relationships, and these do not necessarily say much about the purpose of an intervention. For example, machine learning can predict that if there are lots of ice cream sales, this is a good indication that there will be lots of forest fires. But a policymaker who wants to fight forest fires by closing ice cream factories will, of course, be little effective with that.

The third challenge is that the data needed to create the insights that policymakers can use to better help populations cannot simply be shared. All of the chair's research is therefore done within a strict compliance framework, within BNP Paribas Fortis' secure systems, and only after approval of a strict legal and ethical process. And even after these steps, only strictly anonymized data is looked at from which all privacy-sensitive personal characteristics have been removed.


BNP Paribas Fortis




Prof. Koen Schoors