Although asset managers leverage artificial intelligence (AI) in their investment processes, risk management and compliance, it is not as pervasive across the trading cycle, according to the European Security and Markets Authority’s (ESMA) new paper – Artificial Intelligence in EU Securities Markets.
The benefits have been well documented. AI models allow traders, brokers, and financial institutions to optimise trade execution and post-trade processes, reducing the market impact of large orders and minimising settlement failure.
For now, firms are at different stages of their use in the trade cycle which ESMA breaks down into three parts – pre-trade analysis, trade execution and post-trading.
The paper notes AIÂ application in trade execution holds the most promise. “Accurately estimating market impact has become particularly important for investment banks and other brokers operating low-margin businesses,” it said.
One of the biggest challenges is the lack of data which makes it difficult to model, especially for less liquid securities.
The result is that some brokers and large buy-aide firms such as pension and hedge funds have developed machine learning (ML) models to split and execute metaorders optimally across different trading venues and times in order to minimise their market impact and transaction costs.
Pre-trade also has potential as AI models can be deployed  to analyse signals in asset prices and identify investment opportunities.
However, studies have shown that that most algorithmic trading that banks and large non-bank market makers workflows are still built around relatively transparent rules-based models.
This is changing with many big proprietary trading firms integrating ML models in their trading algorithms, the report added.
As for post trade, ML is still more popular than AI to for example, predict the probability of a trade not being settled and how to optimally distribute the liquidity.
“While ESMA is right to highlight that AI and ML models can be used to improve efficiencies in post-trade processing, the reality is that the market is some way off from realising the benefits,” said Daniel Carpenter, CEO of Meritsoft, a Cognizant company
He added,” This is partly because the insights banks need to run AI and ML is often housed across numerous legacy systems and in different data formats.
Before any meaningful analysis can be done, these disparate data sets need to be normalised, digitised and made available centrally. Only once this has been done can banks begin to start thinking about a meaningful application of AI and ML.”
©Markets Media Europe 2023
