The Bank of England is experimenting with automation and machine learning to reduce the amount of human effort required in the interpretation, submission and analysis of governance data, according to Peter Thomas, senior manager of regulatory technology at the Bank of England.
Speaking at the Advanced Analytics & Artificial Intelligence Summit in London this week, Thomas said that the Bank of England is committed to making its rule book machine readable within the next three to five years. “We want to make it more accessible and make sure that at least part of it can be accessed through automation. One of the obvious areas for this is regulatory reporting. Currently a huge amount of human effort is spent interpreting our rules on what we need, where we need it and how we measure it.
“We’re working with the Financial Conduct Authority (FCA) and external firms in creating a pilot to see whether we can actually get end-to-end regulatory reporting, building off that idea of a machine-readable rulebook, and we’re working towards automating the entire reporting chain. That’s not easy, it’s a long-term project and we’re talking about a 10-year timeframe.”
Thomas added that the Bank of England is conscious that it pulls in large amounts of unstructured data and information.
“People tend to process this information manually themselves and people aren’t particularly good at it [when compared to machines]. We’re looking at teaching machines – in what would be called basic machine learning – to do the simple stuff and ingest information, categorise it, file it and extract insights from it.”
The reporting process could be further streamlined by a switch of models, argued Thomas. “It boils down to a pull versus a push model. We currently operate a push model, where any firms in the UK will submit financial and conduct returns via Gabriel and the FCA. You ‘push’ them in and then we process them.
“The idea of a ‘pull’ involves us sending out a query and potentially interacting with the bank’s systems directly. Now, there are loads of very interesting governance and ethical questions around how you do that and what you do with that, and I completely understand firms’ reluctance around giving us access to potentially anything and everything.”
Thomas said that while the regulator can already ask firms for specific sets of data, most of the complaints it receives focus on how onerous its ad hoc requests can be to comply with. The Bank of England has investigated using an application programming interface (API) through which firms can upload their data. “It’s much more granular. Quite frankly, we don’t want you to send us and give us access to all of your data, because we can’t process it. It’s terabytes and terabytes produced on a daily basis.
“But finding that level of granularity with the data a bank can upload and say, ‘we’re happy for you to query that’ which we will then query as and when required, rather than requiring fixed returns on a monthly or quarterly basis. This is not going to happen overnight. Like I said, this is a 10-year project.”
Thomas said that the Bank of England has been in discussion with eight major banks on the matter, and the consensus has been positive. “They’re really interested in cutting the cost of compliance, making things less manual, and doing things in a quicker, more efficient manner. They recognise that we are going to ask for the data at some point and are at the same time concerned about making sure the numbers aren’t uploaded until they’ve been signed off at all governance stages.
“That’s what leads to this point about an API. One of the ways the eight banks were comfortable with the idea was that the data would not be released until they sign it off, after which it can be queried.”
Thomas added that there remain data autonomy challenges and data model challenges around creating something that can interact with legacy systems. “It’s a huge challenge, but it’s the kind of conversation we’ve been having: not doing away with the governance processes in place, but how to make that reporting part more efficient.”