Cutting through the hype: Where AI is already making its presence felt in FX – Nokturnal

The extent to which artificial intelligence is utilised across the FX market is hard to quantify. One commonly referenced statistic is that around 90% of traders use AI and applications such as predictive analytics as part of their wider trading strategy.

What can be said with certainty is that banks are increasingly integrating the technology – and associated technologies such as machine learning and natural language processing – into their trading platforms.

For example, in May 2023 HSBC launched AI Markets, a digital services offering that uses purpose-built natural language processing to improve institutional investors’ interaction with global markets.

This functionality can be divided into three categories, the first of which is digitisation of single tasks with a rigid communication interface between machines, and between machine and humans where the challenges are to bring these functions together to build a complex system for complex tasks, and communicating with that system.

Developments such as ChatGPT and the evolution of large language models bridge this gap and make the next category of functionality – conversational interaction with data, models, and systems – achievable with the next few years according to HSBC, which expects the third category (large conversational intelligence) to become applicable to trading within the next 3-5 years.

The broker community was quick to embrace AI, with most FX brokerages having integrated the technology into their platforms over the last few years to help trading clients who don’t have access to expensive tools such as Bloomberg/Refinitiv terminals to make more informed decisions. As well as delivering real time and market analysis, AI is also being used to enable copy trading.

MT5 traders are able to use ONNX (open neural network exchange) models in their platforms, while FlexTrade Systems recently introduced AI-driven functionality in its multi-asset execution management system that traders can use to ask verbal or written questions such as ‘what is the value of my Swiss orders over 10% ADV?’

OECD paper

A paper published by the OECD in 2021 (Artificial Intelligence, Machine Learning and Big Data in Finance: Opportunities, Challenges and Implications for Policy Makers) outlined the opportunities and challenges presented by the use of these technologies in a trading environment.

In highly digitised markets such as FX markets, AI algorithms can enhance liquidity management and execution of large orders with minimal market impact by optimising duration and order size in a dynamic fashion based on market conditions. Traders can also deploy AI for risk management and order flow management purposes to streamline execution and produce efficiencies.

However, the OECD warned that the lack of explainability of AI model processes could give rise to potential pro-cyclicality and systemic risk, and could create possible incompatibilities with existing financial supervision and internal governance frameworks – possibly challenging the technology neutral approach to policymaking.

So what makes a complex global market like FX – which has many different participants reliant on various technologies – a good fit for AI? According to Tim Carmody, chief technology officer at IPC Systems, the complexity and global nature of the FX industry is the most important empowering factor.

“Even more so than other asset classes, FX is impacted by geopolitics and regulatory and business dynamics on a global scale,” he says.

Not only is it globally complex – FX does not correlate with isolated or predictable events, but is rather impacted by random combinations of a lot of different factors.

“The massive amount of fast moving data that is generated as a result makes AI analysis (and speed) highly suited to FX,” adds Carmody. “At the same time, the very liquid nature of FX markets also favours the use of AI in order execution.”

Meanwhile Alexander Culiniac, chief technology officer at SmartTrade Technologies also refers to the complexity of the FX market as one of the factors that make it an ideal environment for the application of artificial intelligence, in addition to the huge volumes of data generated and the speed with which the market moves.

“AI’s ability to process vast amounts of varied data in real-time, its predictive capabilities, and its aptitude for high frequency trading make it indispensable in this environment,” he says. “Additionally, it enhances risk management by detecting potential market shifts, automates routine tasks, and carries out market sentiment analysis through natural language processing. Its adaptive learning capabilities also allow it to respond to dynamic market changes.”

However, Culiniac cautions that while AI is a powerful tool, its effectiveness is dependent on the quality of the data it is trained on and it may not always accurately predict unforeseen events.

Exploiting the benefits

One of the obvious questions in any discussion of the use of AI in FX is whether the buy-side or sell-side is best placed to exploit the benefits of AI and what factors will influence utilisation of the technology.

Culiniac reckons this is something of a moot point in that the distinction between buy- and sell-side is increasingly a grey area with both taking and making to some extent.

“Sell-side platforms are naturally suited to exploit AI given that they can access multiple sources of information to create proprietary data sets,” he says. “Of course, there are considerations around data ownership and licences. However, the ability to access multiple sources of information and derive proprietary data sets gives these players an advantage.”

“We see our sell side clients looking at AI to provide clear actionable data. Client management teams are swamped by reports and analyses these days – the issue is not that they don’t have data, it is that they have too much information.”

AI and machine learning tools developed by SmartTrade enable banks to see how clients are segmented in ways no human can ascertain. Once segmented, clients can be targeted in terms of upselling and marketing to make sure the bank’s client interactions are relevant and offer value. Further practical examples of AI/machine learning logic include looking at how client behaviour patterns change in terms of how and when they trade, then automatically generating alerts for sales teams for clients who may need more attention. These AI generated alerts and signals can also be combined with proprietary algos hosted in the SmartTrade AlgoBox allowing a bank, if it wishes, to have automated reactions to these actionable insights.

“This automation is clearly the direction of travel of the front office market,” says Culiniac. “How long it will take to develop and to what extent full automation will prevail of course depends upon the specifics of the bank, its clients, and the types of action we are considering.”

Carmody agrees that both buy- and sell-sides can benefit from using AI, although he suggests its access to vast resources probably gives the sell-side the edge.

When asked how AI and machine learning is already being used in FX trading, Carmody says they are being deployed to synthesise data effectively from a large number of sources, with applied ‘interpretation’ such as sentiment analysis.

“This data analysis is combined with market research and historical pattern recognition,” he says. “Real time transcription of voice trading is also being adopted more widely to create another source of data to feed into trading engines, alongside algorithmic order execution.”

Culiniac says AI and machine learning have already revolutionised many aspects of FX trading.

“They are used in data analysis to process vast quantities of market data to identify patterns and predict future price movements,” he explains. “In market research, AI algorithms – particularly those involving natural language processing – analyse sentiment from diverse sources to understand market influences.”

For liquidity management, AI assists by forecasting supply and demand in the FX market, helping to pinpoint the optimal timing for trades and identify the most liquid trading pairs.

“In risk management, AI systems identify potential market shifts or volatility spikes, adjusting trading strategies accordingly to mitigate risks,” adds Culiniac, who goes on to outline the potential risks of deploying AI that echo with some of the observations made in the OECD report.

Risk considerations

“Deploying AI in loosely regulated markets such as FX carries risks associated with the concentration of power, systemic risk, and lack of transparency,” he says. “The use of sophisticated AI systems may further centralise trading power in large institutions, potentially exacerbating existing market inequalities.”

AI-driven trading can also increase systemic risk if numerous systems are trained on similar data and implement analogous strategies, which could lead to a cascade of trades amplifying market volatility during certain conditions.

“Finally, AI systems – especially those based on deep learning – are often perceived as ‘black boxes’ due to their complex and non-transparent decision making processes, which poses challenges for accountability,” continues Culiniac.

Due to the perceived relative newness of the technology, some SmartTrade Technologies clients have chosen to use AI/machine learning tools to derive actionable information from their data and to enable them to have better conversations with clients, liquidity providers and internal stakeholders rather than to fully automate mission critical processes.

“It is almost inevitable that as market acceptability increases and the technology matures we will see more and more aspects being fully automated in the front office as banks race to compete in deriving value from their data,” adds Culiniac.

As a non-regulated market there are by definition fewer systemic safeguards, such as circuit breakers that provide equities and other markets with release valves in the event of anomalistic situations.

“This means a greater risk of AI-generated events and so-called ‘flash crashes’,” warns Carmody. “The combination of a lack of regulation and vast amounts of source data also increases the risk of bad behaviours where intentionally inaccurate or misleading information is fed into the data sphere to influence the market.”

He suggests the FX industry should take a cautious approach with respect to the introduction and adoption of AI, keeping a close eye on oversight and governance.

“The rise of generative AI or Gen AI increases the risk of ‘machine mis-learning’ and resultant confusing patterns, and even the generation of new, non-deterministic patterns,” adds Carmody. “This could be amplified further as more and more counterparties rely on Gen AI – as the high frequency equities trading segment learned (the hard way) as it matured.”

Lessons from other markets

Culiniac reckons there are a number of lessons FX can learn from how AI is being applied in other markets, particularly the application of sentiment analysis.

“AI’s ability to analyse diverse textual data such as news and social media posts can help gauge market sentiment, aiding traders in predicting market trends,” he says. “This is often used as an alternate source of market probability on equity trading desks. The technology can also be applied in the FX market to understand the influence of global economic sentiment on currency values.”
In a recent article on the Refinitiv website, the company’s head of FX sell-side trading, Bart Joti, suggested teams perform better when data and AI techniques are coupled with human insight. He noted that new skill sets are becoming increasingly important so that individuals – particularly in trading and sales roles – can use AI and machine learning tools to make decisions based on market and reference data.

While the role of FX traders will continue to be augmented by more and more AI-led functions, Carmody reckons it is impossible to predict whether and how Gen AI might fully replace the physical trading function.

“More than 30 years ago when EBS and Reuters launched the first fully automated FX ‘broking’ platforms, the same concerns were expressed about the demise of the human touch in FX trading,” he says. “More recently, automated and algorithmic trading technologies have also challenged the traditional FX broker/trader model. Yet physical trading rooms and traders still very much exist as our 200,000-strong network of global trading participants will attest.”  

Conclusion

Whilst the power and potential of AI cannot be ignored – and may well end up as the most powerful technology in the trading toolbox, Carmody believes there will likely always be a need for human engagement at both a strategic and a tactical level. Culiniac agrees that despite AI’s transformative potential in FX trading, it is unlikely to completely replace human traders in the foreseeable future.

“AI, while adept at analysing vast amounts of data and identifying patterns, struggles to comprehend context as humans do and may find it challenging to interpret unprecedented events or situations outside established patterns,” he says.

Moreover, trading involves complex ethical and legal decisions that are difficult to automate and where human judgement is indispensable.

“Furthermore, even as AI becomes increasingly integral to trading, human oversight remains crucial to manage risk, ensure accountability, and make decisions when AI encounters unfamiliar situations,” concludes Culiniac. “Therefore, the future of FX trading will likely involve a synergistic partnership between AI and human traders, combining their respective strengths.”