By Marion Leslie, head of Financial Information, SIX Group.
When you strip away the hysteria surrounding artificial intelligence (AI), what remains are numbers. Some of these tell a captivating story of relentless AI demand. For instance, computer chip developer Nvidia this year added $220bn to its market capitalisation in only an hour after reporting earnings – more than the entire value of Intel and Micron combined.
Yet, other numbers reveal a less certain picture of the scale of AI’s adoption – particularly in capital markets. According to the European Securities and Markets Authority’s (ESMA) ‘AI in EU Securities Markets’ report published in February, feedback received from central clearing counterparties and central securities depositories suggests the majority of these entities are in fact not widely using AI.
While AI models can, and will, deliver more efficiency in areas of capital markets like post-trade processing, the reality is the market is still some way off realising these benefits. In understanding why this is the case, the answer once again revolves around numbers – or, more specifically, data.
The devil is in the data
The emergence of AI tools can certainly help increase post-trade processing efficiency, but without a good understanding of the data being crunched and fed into it, as well as why this data is being used, firms will only scale any errors. This is doubly true within the complex capital markets space. The nature of our industry means there is a far more nuanced data challenge that needs to be addressed before an AI post-trade revolution takes place.
This is largely because the insights banks require to run comprehensive AI programmes are locked away across numerous databases and in different formats. Trying to marry this information together to predict settlement issues, such as trade fail rates, for instance, is incredibly challenging without first structuring and normalising this disparate information. Yet, only once this has been successfully achieved can banks even begin to start thinking about exploring where the future opportunities lie with AI.
Indeed, no matter what sector or business, AI requires a carefully developed data strategy – which means: identifying the necessary data, making it accessible, and understanding its provenance, ownership, governance, and entitlements. This presents firms with a significant hurdle to overcome – but not an insurmountable one.
To understand and embrace the opportunities on offer with AI, it is first important for businesses to understand the learning that needs to happen across the organisation. Like a stick of rock, attention to data must be evident right through the core of the business, from the boardroom to administration. That means C-Suite, sales, administration, technology, HR, risk and compliance, legal – they all have a critical role to play and all need to understand.
Vital decisions need to be made early. For example, what will the role of the human be? How much transparency, intervention and reporting will there be on insights, risks and opportunities? Of course, sustainability and ethics are central too. What is the company’s approach to data ethics, bias and exclusion? Here, the common sense that only a human can impart becomes a vital commodity.
For firms able to develop and execute an effective data strategy, several obvious use cases for AI are already emerging – particularly in post trade. Take, for instance, the transition to T+1 settlement in US securities, now just 8 months away. When it comes to settlement fails, counterparties’ historical settlement performance can be analysed, and AI tools utilised to predict the likelihood of future fails.
Fixed income may be a sensible place to start on this front. According to the latest data from ESMA, settlement fails in both corporate and government bonds have increased over the past 12 months, with even government bond fails sitting at around 4% as of December 2022.
One of the common use cases is with the new issuance of bonds. In the early lifetime of a bond, it is not uncommon for settlements to fail because liquidity is not yet sufficient. The shortening of the settlement timeframe for US securities next May will only serve to aggravate this issue. But if the industry can deploy AI in a way that links the settlement transaction to the new bond issuance, then financial institutions may be able to better predict and therefore mitigate the likelihood of settlement failure.
Nevertheless, firms that jump the gun on AI implementation are taking a considerable risk. A regulated financial institution cannot run its business based on information from unknown or non-expert sources. The implications for risk management and investor protection are enormous. While there certainly seems a place for AI in financial markets, firms must gain a deeper understanding of what has gone into the machine, before getting excited about what can come out of it.
Subscribe to our newsletter