Germany’s financial authority BaFin published supervisory principles for the use of algorithms in decision-making processes by financial institutions. These principles are intended to promote the responsible use of big data and artificial intelligence (BDAI) and facilitate control of the associated risks.
Any regulatory questions in relation to BDAI are beset by a fundamental conundrum: how to draw the difficult distinction between BDAI processes and processes driven by conventional statistics. From a risk standpoint, there are three characteristics of particular relevance to modern BDAI methods:
- First, the algorithms used are frequently much more complex than conventional statistical processes. This renders them opaque.
- Second, recalibration cycles are getting shorter and shorter. This is due to the combination of algorithms that are constantly learning with the fact that new data becomes available on an almost daily basis. As a result, the boundaries between calibration and validation are increasingly blurred.
- Third, the use of BDAI methods is leading to an increase in automation. This makes it ever-easier to scale processes, with the impact of the individual algorithm being amplified.
The principles thus apply primarily to those algorithms which exhibit these three traits.
A two-phase approach
In order to formulate the principles as precisely as possible, the algorithm-based decision-making process was broken down into two very simplified phases: development and application. The development phase examines how the algorithm is selected, calibrated and validated. For instance, principles for the development phase relate to the relevant data strategy, as well as documentation to ensure clarity for both internal and external parties.
In the application phase, the results of the algorithm must be interpreted and included in decision-making processes. This can either be done automatically or by involving experts. A functioning mechanism comprising elements such as sufficient checks and feedback loops for the development phase must be established in all cases. Aside from these two phases, there are overarching principles such as the necessity for a clear responsibility structure and adequate risk and outsourcing management.
Also on the radar, BaFin and Bundesbank aim to publish a discussion paper in July specifically on the use of machine learning that will focus on which models must be granted regulatory approval in the context of solvency supervision or which are subject to prudential review. BaFinJournal will keep you updated.