In the 1960s, Stanford computer scientist Roy Amara told his colleagues that the effect of advances in technology is never fully understood. Or in other words, ‘we overestimate the impact of technology in the short term and underestimate the effect in the long run’.
In this context, a May 2024 paper from Krissy Jones (co-founder of The Forex Library) entitled: The Future of FX Trading: Exploring the Intersection of AI, Open Innovation, and Industry Evolution, is a useful reference point for where the use of trading technology is now and where it might go.
She suggests that open innovation in FX – underpinning AI – represents a chance to revolutionise FX trading through the development of intelligent systems that may adapt to transient market conditions.
One of the important potential applications of AI in FX trading is the development of predictive models that can point at a move in the market and a trading opportunity, while another is the development of intelligent trading systems that can perform automatic trading based on predefined rules and algorithms.
According to John Stead, director of sales enablement and marketing at smartTrade Technologies, there is a problem of too much information because everything has been electronified.
“Using machine learning could help segment that data by different criteria, such as how clients are trading,” he says. “You could look for patterns and groups of similar behaviour from clients and then see where those patterns change, which means where a client’s remit or behaviour has altered, for example.”
Machine learning could also be used to look for upselling opportunities by examining all the trading data of similar clients to see what products and solutions they are using and which are suitable candidates for an upsell.
“The other buzzword at the moment is large language model co-pilots, which allow you to use natural language to query data,” says Stead. “Ahead of a meeting with their bank, this could give a firm an insight into their performance and whether anything significant had changed.”
“The more complex the technology gets, the more you need to make sure that you have safeguards that are as advanced as the technology – and if you can’t put these safeguards in place you can’t use the technology”
John Stead
Deployment gap
However, in terms of deployment he acknowledges that there is a gap between where the technology is and how the banks are using it.
“We have live platforms with a co-pilot embedded but not all clients are ready to use it because of questions around confidentiality of data and cyber compliance,” Stead says. “It is the job of the vendor to develop the technology, to see where it can go and get feedback from clients because if that is where the market is going, the sooner we start exploring the technology the better.”
Given the rate of adoption of machine learning technology, Stead reckons it will be a few years before we see widespread use of large language model co-pilots as banks will want to explore the technology and look for potential issues before they deploy it.
The global FX market is fragmented and complex, creating massive data lakes. Enabling AI could provide a mechanism to connect these vast data sets and allow for analysis of patterns or correlations which are currently lost to normal algos.
That is the view of Bart Joris, head of FX sell-side trading at LSEG, who adds that the addition of generative characteristics could also help firms continuously fine tune their models with comparisons of new data or enhanced screening.
“It will be an evolution in how market infrastructure currently operates, but FX is a market that will lend itself more easily to these new technologies given that large amounts of data are already stored electronically in a structured way – a prerequisite for AI,” he says.
Joris suggests that blockchain FX trading will create an environment for peer-to-peer transactions, reduce settlement times and act as a disintermediation tool to reduce costs and provide transparency.
Many in the FX industry are excited by the potential of big data analytics
Blockchain blow
There has been a lot of discussion about the role of blockchain technologies and smart contracts at various industry events. However, Stead says the topic rarely comes up in client meetings aside from conversations around the impact of T+1 settlement.
“The questions we hear are ‘how can I understand my clients better?’, ‘How can I deal with the multitude of information?’, and ‘How can I achieve lower latency and more competitive pricing?’,” he says. “For me, blockchain is a big open source database, although there is obviously a lot more to it than that. I don’t think there’s a big problem to solve in FX that needs the blockchain.”
There are other products where smart contracts could be very useful, for example on the equity side in relation to tokenisation of assets, adds Stead. “We provide crypto trading solutions for clients and obviously that uses various blockchains, but it is purely for crypto assets rather than FX.”
When asked how blockchain technologies and smart contracts could contribute to creating a more secure, transparent and efficient FX trading ecosystem, Joris observes that shifting outcomes is a key change to previous modelling as input and in-depth scanning will change the end result, which will create resistance due to uncertainty.
“However, I believe this will be temporary,” he says. “AI modelling comes with structures around the creation, control, accuracy, change and validation of the outcomes including the data sets. The principle of ‘garbage in leads to garbage out’ will apply here. The key will be to control the processes and deployment of AI to enable the desired logical outcomes in the end results of the model.”
“The principle of ‘garbage in leads to garbage out’ will apply here. The key will be to control the processes and deployment of AI to enable the desired logical outcomes in the end results of the model”
Bart Joris
With regards to blockchain, this technology can improve transparency due to its (open or private) ledger infrastructure, but in order to become really successful there needs to be a valid use case.
“The best use cases are probably in the back office, particularly settlement,” acknowledges Joris. “However, whilst critical, the back office is a cost centre for many banks and so investment can be minimal. So unless there is a real driver such as revenue, cost or operational risk efficiency, its ability to transform the market is limited – although T+0 or actions by central banks could initiate a shift.”
Analytics advance
Many in the FX industry are excited by the potential of big data analytics and specifically how it can be leveraged to deliver insights that will help improve trading and business performance.
The more data, the greater the insight firms have. AI will be able to do this more quickly and on a larger scale than ever before and find correlations that drive or enable FX movements which would not necessarily be visible to the human eye or algo.
“Continuous back testing of models is also possible with AI, but the migration to the cloud is also a key element,” explains Joris. “It will lead to better models and trading signals being integrated in automated trading processes or being visible to manual traders. They can interact with structured or unstructured large language models and data will further revolutionise the way markets currently operate.”
For example, via pre-trade transaction cost analysis, comparisons could driver greater competition in favour of liquidity consumers.
According to Jones, there is growing stakeholder demand for transparency and accountability in business operations. This has prompted firms to use energy efficient technologies such as low-powered computer systems and more efficient cooling systems to reduce the amount of power consumed.
“Green FX is currently only focused on whether data centres and offices are powered by renewable energy,” says Joris. “But I think this will change as the industry adopts more modern technology and diversifies its modelling. For example, firms with less focus on ESG could end up paying a wider spread than others or we may see a formation of dual markets like we have seen in debt lending.”
Blockchain technology can improve transparency due to its ledger infrastructure
Green indifference
Sceptics might suggest that much of the noise around green technology and green innovation is as hypothetical as that relating to blockchain and Stead accepts that this topic is not at the forefront of traders’ minds.
“I was at a conference a while ago where there was a discussion around data centres and their carbon footprint and the consensus was that if you had a data centre in the Nordics, for example, where they use a lot of hydro and wind power, the carbon footprint of your data centre would be quite low,” he says.
AI could provide a mechanism to connect the vast data sets in FX
Corporate social responsibility may be one of the criteria considered at the RFP stage, but it is clear that factors such as flexibility, latency and stability are the overriding considerations.
“So technically we could imagine a client saying that they would like my data centre to be located where the carbon footprint is lower,” adds Stead. “But then you would have traders expressing concerns around latency.”
As for whether there are any specific strategic issues that FX trading firms need to bear in mind when embarking on technology upgrades or technology integrations, Stead emphasises the importance of setting clear objectives and goals.
“The most successful clients have thought very clearly about what are they trying to achieve within the bank and have secured a high degree of consensus as to what the objectives are,” he says. “This means you have as few conflicting viewpoints within departments about the goals and objectives of a project as possible.”
Because replacing a platform is a complex and costly process, clients are encouraged to think about what they want to be doing in 2-5 years’ time and to work with a vendor that is committed to AI and machine learning.
On the question of whether there are any obvious risk factors in using AI and machine learning from an FX perspective, Stead observes that any change carries an element of risk.
“Staying the same for ever is not an option, but when using AI and machine learning it is sensible to understand how it is making its decisions and how it works because if you don’t you are leaving yourself open to regulatory issues when something goes wrong,” he says. “It is no different to employing someone – they have got to have proper oversight and you have got to understand what they can and can’t do and make sure that they understand.”
Safety first
Stead refers to the case of a bank that had to pull some of its models because it couldn’t prove how they were deciding on executions to the satisfaction of the regulator. While vendors can come up with very interesting technology, it has got to be verifiably safe and not using some kind of bias in a particular way that doesn’t make sense.
Then there is the security consideration – does the deployment of the technology introduce a new cyber security threat? For example, could the platform decide to trade on its own account? This level of sophistication may not be available right now, but firms need to be aware of potential operational and regulatory compliance risks.
“The more complex the technology gets, you need to make sure that you have safeguards that are as advanced as the technology – and if you can’t put these safeguards in place you can’t use the technology,” says Stead. “That is what is holding banks back in using a lot of new technology. If they can’t be confident that it is safe to use, they won’t use it.”
In general, the more data becomes available, the easier it is to identify misuse. Digital transformation also enables access to markets and information to a wider community, which increases transparency and fairness in the FX market.
But Joris cautions that there is a catch with these newer models because as they look for optimal execution, use cases such as pre-hedging could be suggested as a better result (if used correctly). So it will be important to ensure the necessary control and oversight on the models.
“AI is driving the push to better models and data creation, but this means that controls and behaviour analysis of the models will be key,” he adds. “A control framework should also be implemented with clear responsibilities, procedures and continuous monitoring of the model in production.”
