Robo-Supervisor
Can the ECB harness the latest technology - and what would be the consequences?
The future of supervision
Elizabeth McCaul set out a high-tech vision of the future of supervision this week. Speaking at the ECB's Supervision Innovators Conference, McCaul - Supervisory Board member responsible for digital policy - described how ECB supervisors will benefit from cutting-edge technology.
Unprecedentedly rapid technological development, she said, is transforming every aspect of the economy. AI in particular is radically reshaping many activities, including banking, as banks become increasingly data-driven. Supervisors must keep pace, she argued, understanding the benefits and risks of new technology and equipping themselves with the latest tools to keep banks safe and sound.
Cyborg supervisor
Standing like Fritz Lang's robot at the centre of McCaul's vision was Pete. Pete is the supervisor of the future, where the People and Technology elements of the ECB technology strategy come together. The ECB, McCaul promised, will arm Pete with the latest data analytics, collaboration and communication technology. It will train him and support him to co-develop the tools of the future, working with supervisory technology champions across the eurozone and beyond.
Technologically tooled up, Pete will then use his augmented analytical powers to delve deeper into the risks facing modern banks. These will include both 'classical' credit, liquidity and operational risks as well as new emerging risks, such as those from climate change, environmental damage and cyber attack. Artifical intelligence, meanwhile, will simplify and automate workflows, keeping Pete from getting bogged down in routine tasks.
Will it work?
McCaul's vision of technologically-empowered supervision is an ambitious one. And the ECB has long committed to embracing new technology (most recently as one of the six pillars of SREP reform announced in May). Sceptics will of course note that the history of large public-sector IT projects is mixed. Meanwhile bankers who already like to grumble about poor quality assurance at the ECB (stories abound of decision letters that contain elementary errors like misspelling the name of the bank concerned) may question whether the ECB will really be able to boldly go into the future McCaul imagines.
Data hungry
A more serious concern, however, may be whether enhancing the ECB's analytical capabilities will only intensify its hunger for data. Banks already complain of the volume of data requests they receive from supervisors. These are costly to comply with and many, banks argue, bear little relation to the key risks involved in their businesses, but appear motived either by a dogmatic drive to treat all banks identically in the name of consistency, or by an almost academic desire to chart every corner of the banking system. This leads many bankers to question the proportionality and strategic focus of ECB supervision.
With advanced analytics and AI, the ECB could gain the ability to process vastly more data than at present. That could fuel its ambitions to dive ever deeper into the data lake in search of new patterns and correlations. Doing so could bring to light financial stability risks that had previously gone unnoticed, allowing future failures to be prevented. (Indeed, in her speech, McCaul mentioned the potential for AI in aviation to predict accidents before they happen as a goal worthy of supervisors.) But banks may worry that the urge to analyse will lead to ever more requests for data to feed the machine.
AI at Banks
As on previously occasions when she has discussed the ECB tech agenda, McCaul said little about bank's own use of AI. So far, we still have no comprehensive statement of ECB policy towards AI in banks. In August, however, the ECB took a small step in that direction, with the announcement that its forthcoming new update to its Internal Models Guide will include supervisory expectations on the use of machine learning in credit and other models.
Some technologists will dispute whether machine learning should really be counted as AI. But it seems highly likely that many machine learning techniques will fall within the definition of AI used in the EU AI Act. The AI Act designates creditworthiness and credit scoring systems as 'high-risk' AI applications, subject to requirements on risk management, data accuracy and human oversight. Banks' compliance with these requirements will not be supervised by the ECB, but by a mix of national financial supervisors and AI authorities. Good cooperation between these different agencies will be important to avoid inconsistent, or even contradictory, requirements. So it will be very itneresting to watch where the ECB's expectations come out when the new new Guide is complete.