AI: regulatory arbitrage on steroids?
Pretty much everyone has heard of Fintech by now, but a more focused approach to applying new IT technologies to banking is now the nerdier Regtech. Regtech aims at applying Fintech for regulatory and compliance purposes, simplifying a process that has caused headaches to bankers due to the exponential growth of the rulebook they had to follow, and which has also been a pain on the cost side given the number of extra compliance officers they had to hire in an era of lower revenues.
Indeed, the FT reports that:
Citigroup estimates that the biggest banks, including JPMorgan and HSBC, have doubled the number of people they employ to handle compliance and regulation. This now costs the banking industry $270bn a year and accounts for 10 per cent of operating costs. […]
Spanish bank BBVA recently estimated that, on average, financial institutions have 10 to 15 per cent of their staff dedicated to this area. This heavy investment has been necessary in response to the crackdown by regulators that followed the financial crisis. European and US banks have paid more than $150bn in litigation and conduct charges since 2011, Citi estimated.
What’s the solution? ‘Regulatory technology’:
New technologies mean that banks could make vast savings in compliance, according to Richard Lumb, head of financial services at Accenture, who estimated that “thousands of roles” in the banks’ internal policing could be replaced by automated systems.
Many of recent Regtech developments involve the use of artificial intelligence to simplify compliance issues that are very burdensome from a staff (and cost) perspective. As Deloitte outlines here (and see an interview on the Financial Revolutionist about applying machine and deep learning to investment strategies here):
The Institute of International Finance (IIF) highlights AI, among others, as it has a range of applications in regulatory compliance and reporting. It can be used in analysing complex trading relationships, trading schemes, patterns and communications between banks, exchanges and other market participants. AI can also be employed to monitor internal conduct and communication to clients, comparing it to quantitative metrics such as supervisory input. As AI relies on computer-based modelling, scenario analysis and forecasting, it can also help banks in stress testing and risk management.
But what I find particularly interesting is this bit:
Another field for AI in financial regulations is to simplify the regulations themselves: there are a multitude of different jurisdictions, products, institutional differences and enforcement mechanisms and it is hoped that AI systems are better in collecting and categorizing them according to rules.
Similar points in an Economist article published a few months ago about Watson, IBM’s AI product:
The next area is to provide clarity about rules. They are sorted by jurisdictions, institutional divisions, products and so forth, and then further broken down between rules and guidance. Watson is getting better at categorising the various regulations and matching them with the appropriate enforcement mechanisms. Its conclusions are vetted, giving it an education that should improve its effectiveness in the future. Promontory’s experts are expected to help Watson learn. A dozen rules are now being assimilated weekly. Thousands are still to go but it is hoped the process will speed up as the system evolves. Ultimately, IBM hopes speeches by influential figures, court verdicts and other such sources will be automatically uploaded into Watson’s cloud-based brain. They can play a role in determining what regulations matter, and how they will be enforced.
Below is a useful chart showing all current Regtech areas and start-ups (you can also find it here):
While the industry has not explicitly said it this way (and probably never will), it seems to me that we’re on our way to AI-driven regulatory arbitrage. Once those systems are ready, AI will be able to navigate through the thousands of regulatory pages and extract the most effective ‘regulatory optimisation strategy’ within and across borders.
If all AI systems used by financial institutions reach the same conclusion, this could lead to a build-up of imbalances and systemic risks that could eventually trigger a crisis, following a process similar to that which contributed to the latest financial crisis: Basel rules facilitated the accumulation of imbalances in the credit market towards real estate lending.
It of course remains to be determined whether AI systems reach the same conclusions in the end. But this is likely to happen, for the following reasons: 1. banks whose systems are less effective will progressively attempt to catch up with the competition, leading to harmonisation in the design of those systems and 2. if AI solutions are provided by third-party firms, harmonisation will occur from the start.
A glimpse of hope remains in that the optimal regulatory arbitrage strategy may be different for financial institutions with different business models (mortgage banks vs. universal banks for instance). But let’s not hold our breath: even in this case, imbalances would still occur and universal banks still account for most of the world’s banking assets by far.
For now, explicit regulatory ‘optimisation’ does not seem to be included in the chart above (although the ‘Government/Legislation’ category could well evolve into a more arbitrage-oriented segment). But how long before it does?