A look ahead at what’s in store for FX algo trading in 2025

In the report, Transforming the FX Trading Desk: Strategies for Market Resilience, the decline in available liquidity was particularly evident during periods of market volatility, making it increasingly difficult to execute large trades efficiently. The rise of algorithmic trading and high-frequency trading was also claimed to have intensified the competition for information, leading to concerns about information leakage and market manipulation. A lack of execution strategy control was also noted by 54% of market leaders, emphasising the need for flexible and adaptable trading solutions to navigate rapidly changing market conditions. The critical importance of liquidity was further  underscored by the challenges faced in NDF trading, with 73% of respondents having said that access to liquidity was their primary hurdle, highlighting the need for ECNs to continuously expand liquidity pools and improve order matching algorithms.

However, the algo market has reached something of an impasse in terms of FX algo development, with the focus for bank providers now shifting from rolling out new products and features to calibrating the existing algo suite and taking a more intelligent approach in order to expand their algo footprint, says Bernard Kim, Head of Electronic Macro Business for the Americas at Credit Agricole CIB. Kim believes there are two key verticals coming into play in the algo market, with algos increasingly being used for dealer requests, with asset managers increasingly using TCA and analytics to create bespoke algo wheels to help form their execution decisions. “Clients look to independent TCA providers to help them evaluate algo products,” he says. “But we know that clients also have a ‘trilemma’ to consider when choosing which algo to execute. There are roughly six types of algo strategies and each client has a bespoke way of treating those strategies, but they all have three elements to consider to make that decision. Do I want to minimise market impact, do I want market risk and do I want to maximize execution certainty? It’s not linear. It’s not a simple choice of using a passive or aggressive algo anymore. Within that framework we need to discuss the situation with the client and evaluate accordingly. In the coming year there will be more moves in the industry towards standardisation between the data and analytics on offer to clients which will help them more fairly evaluate the performance of different algo products from the various bank platforms.” 

The second vertical identified by Kim is for the algo providers to look closely at the algos they offer and how they execute in order to differentiate in an increasingly competitive market. However, he adds that it is again very hard for clients to look at the available data, such as internalisation rates or TCA reports, to compare algo providers in an effective way, leading to the need for greater standardisation in how algo execution data and metrics are reported across the industry.  “At Credit Agricole CIB, this is where we focus on making intelligent improvements within the six different types of strategies,” Kim adds. “For example, internalisation is often seen as one way to improve fills, but we choose to partner with a peer-to-peer service such as Siege FX’s MidPool to replicate a match. Algo providers have to be smarter in order to replicate the little details that other banks have that make them perform better and we are constantly calibrating our own algo offering to continue being the algo provider of choice for our clients.”

Standing out from the competition

Kim adds that Credit Agricole CIB has also taken a smarter approach when it comes to algo development. He notes that rather than having gone down the slower and more expensive route taken by many other banks of spending years in research and development to create an algo suite in house, Credit Agricole CIB instead opted to form strategic partnerships with a selection of leading technology providers instead. This has meant the team did not need to divert resources into the actual algo build, freeing them up to focus instead of calibrating the algos to perform as required. “We compare this model of partnering with technology solution providers to the F1 Racing Team, Red Bull. Red Bull use Honda’s engines but it is the expertise of the Red Bull Engineers/Pit Crew that ultimately makes it the winning team. In the same way, we also have our own market leading team calibrating parameters and even collaborating with our clients (the drivers) to create bespoke solutions when required,” Kim adds.

In turn, Loïc Bourgeois, Head of EMEA eFIC Sales at Societe Generale agrees with the report findings adding that liquidity remains the key challenge at the moment for FX and adds that this has been the case for past year to 18 months. The biggest challenge this creates for banks is the need to source good quality liquidity for the algos to access, he explains. “We often refer to liquidity as a scarce resource, even though intraday liquidity is increasing on a regular basis,” Bourgeois says. “Liquidity is being more and more fragmentated in between numerous new players that can recycle liquidity and makes it even more sensitive to any information leakage. This can be an algo posting an order or even a simple RFQ. One needs to be careful with this liquidity illusion that creates signalling, market impact and increases client cost of execution”

This creates a challenge for algo providers such as Societe Generale, who need to curate the liquidity sources the FX algos can access to ensure they offer quality liquidity, holding the risk and not recycling the liquidity back into the market. According to Bourgeois, this will be the main challenge for algo providers in the coming months. As well as curating and fine tuning the liquidity algos can access, Bourgeois notes there has also been changes to the sources used. “We try to be as passive as we can. In the past, we were mainly using disclosed pools, but now we prefer to go darker and less disclosed and using mid books more,” he says. “We are also much more agile with the liquidity we use. For example, our most popular algo is our passive Nightjar strategy, which we used to set on an hourly basis. But when you look at the intraday liquidity available in almost most of the G10 pairs, you will notice that between 11:55am, and 12:05pm the liquidity changes rapidly, so an hourly set is not really relevant anymore. This is why a year ago we moved to minute per minute, which makes us much more agile and also cautious with the liquidity we use, to not impact the market too much or too quickly.”

Focusing on innovation

The key is managing the noise made by the algo execution, adds Patrick Guevel, Head of FX Algo Execution at Societe Generale. “This involves curating the people who we are speaking to. It is about being visible or not and one way of being visible is to trade too much. This is why we pace our algos and strictly control how much liquidity is being  added to the market so that it is not visible,” he says. “There is currently a race to passivity and into mid books in the algo market. Clients can now see the metrics between internal and external liquidity and compare all the internalisation metrics from the various banks. But we want to have undisclosed and invisible liquidity.” Guevel notes that in addition to recent enhancements to increase the size of the e-book to internalise against, Societe Generale also created direct relationships with other banks to have access to the ebook and to increase the internalisation ratio by staying almost invisible. 

Deutsche Bank have also been busy with innovation in the FX algo trading space. “We have released a new pre-trade visualisation tool called Quick Pre-Trade to our Market Colour app,” says  Vittorio Nuti, Global Head of LD & FX Algo Trading at the bank. “This is a very powerful new tool that allows clients to see the risk that they’re taking on screen and provides additional guidance on how long it will take to execute their chosen algo. For example, if I was looking to trade 100 million in NZDUSD, I can check using the tool to see all the market movements that have happened for a given execution window by choosing the last number of days to look at on screen, whether that is 30, 60, or 90. The tool then provides an indication of how long the algo would take to complete this trade size and what sort of participation rates I might see if I ran the trade at different urgencies.” 

His colleague Aled Basey, FX Workflow Solutions Director at Deutsche Bank says that “Clients are interested in making more informed decisions around their algo use, so guiding them to the optimal algo and parameters for their desired objective is fundamental in selecting the right tools for the job. We’re providing the framework for them to do that as intuitively as possible, basing decisions on real historic data augmented with projected outcomes. As the name suggests, making this quick and easy to use was a priority, such that it can fit seamlessly into their Algo workflow.”

“In recent months,” says Nuti, “we have also been partnering with some of the larger hedge funds to do ‘A/B’ testing. This will be something we will be expanding in 2025. Effectively, we are offering these clients the ability to customise their algo executions in a scheduled manner, and then the ability to run analysis against that schedule. This could start from as easy as the client wanting to explore the outcomes for different urgency parameters to compare the outcomes, or the impact of removing a particular liquidity source. The full parameter spectrum of our algos is available for this A/B testing. It’s very exciting and has taken a lot of work to get to where we are. It will massively benefit clients as they strive to be more precise with their algo executions in order to reduce costs. Historically comparing different settings had limitations on the basis that orders may have been executed in different market conditions, so moving towards a systematic approach allows conclusions to be made on a like-for-like basis. It’s an exciting area of algo development, with clients looking for additional data and fine-tuning their approach in partnership with Deutsche Bank.”

Basey says “Ultimately different clients use algos for different reasons, so what they really need is a framework that allows for continual iterative improvement. We can do this in a systematic way, so there is no burden on them. Over time, it becomes more and more tailored for what they are trying to achieve. It is industrialising the ability to customise an algo suite specifically to what the clients are looking to achieve, in the same way tech firms tailor user experiences based on usage.” 

Nuti goes on to explain that, “A/B testing is agreed before the execution and runs for a period to accrue statistically significant data sets, which then allows us to work with clients and make optimisations to align our outcomes with their objectives. Clients can always do A/B testing from a simulated perspective, but by running it in real time gives them hard data and hard evidence and neutralising any bias from differing market conditions.  Clients are also not always solving for just one thing, it’s usually multi-factorial, for example finding that sweet-spot between undue market risk and spread capture.  It is a way of changing the settings to then analyse for the outcomes. So the client might say, ‘I would like to try this’, but instead of just trying it and then having sort of two different sets of environments, because you have a pre change and then post change, they can now use A/B testing. With A/B testing, they may want to change a parameter only half the time, or for half the orders and then we can analyse the results and say, actually, this parameter change did not yield anything better in the same environment and in the same period as you were doing those changes versus what you had before. Or we might find that actually, the changes did make sense for this client who should then continue with this new parameter.This allows us to effectively provide a bespoke algo service at scale for this group of clients. This will continue to grow, as more hedge funds are getting involved in the algo space and algo usage continues to evolve. These clients want to run experiments and to have more flexibility around the product that they are using.”

“What the A/B testing is also doing,” says Basey “is accruing a statistically significant data set to allow our clients to make more informed decisions. By taking a systematic approach and trying different strategies within the same market conditions, it really moves the needle in being able to compare results on a like for like basis. Considering systematic trading houses in particular, it is very much a win-win for them.  It presents a much more economical way of rolling out A/B testing than building the programme in house, whilst also benefitting from the lessons we have already learned in this space. It is customised for them, but it also provides that scale within the parameters that we have already developed.”

Looking ahead to next year, Basey says, “we will also be focusing on the user experience side. We are continually looking at ways to enhance that in-flight experience and align our ourselves to our clients’ objectives. From an execution advisory perspective, it is very much around working with the client towards a common goal to deliver them the best outcomes on a consistent basis. What that means in practice is considering full user journey, from choosing the right tools, to in-flight experience and market colour and then post trade reviewing and engaging with the data post-trade to assess what’s working well and where there could be improvements. We’re very fortunate to have such an engaged client base, willing to partner and share their ideas around the future state of the product suite which allows our product development pipeline to continually deliver what they need.”

Measuring the cost of an algo

The algo market is becoming very crowded as more and more banks are now offering algos, says Asif Razaq, Global Head of FX Algo Execution at BNP Paribas. To entice clients to use their algo platform, these new entrants will often lower the fees for new clients, he says. “This is creating a threat to the overall functioning of the algo market because when this happens, it creates a snowball effect,” warns Razaq. “The client who was offered lower fees might then demand lower fees from their tier one algo providers, who have made significantly more of an investment or allocated significant resources to the development of these algos and the development of new features and strategies.”

If the banks are then pushed to lower their own fees, Razaq explains, then essentially the return on investment from the bank, the revenue model for the banks, reduces quite significantly. He adds: “We have seen an average 75% reduction in algo fees over the last two to three years, purely off the back of competitor price matching.” According to Razaq, the risk is this then reduces the headline revenue for the banks, which reduces return on investment, which ultimately results in the bank’s investment into this technology, into this discipline, reducing over a period of time, thus reducing the quality of the product available to the client.

“It is very much a balancing act,” he says. “We need to be demonstrating to our clients that a reduction in fees does not necessarily result in better execution. In fact, it creates an ecosystem where sub-par execution could be the norm.” The algo team at BNP Paribas counter this by having regular discussions with clients to highlight that the cost of an algo needs to be measured with three key aspects in mind. The first aspect is obviously the fee itself, says Razaq. The second is the overall performance of the algorithm. Then thirdly, which is often missed by clients, is the importance of a attributing a qualitative value score to the algo strategy, he explains. “That qualitative score could be asking do you get pre-trade analytics with these new entrants? Are you getting real-time analytics? Are you getting a dashboard where you can view the algo in flight? Are you able to modify the algorithm? Are these algo banks giving you the ability to be create bespoke strategies? If the answer to those questions is yes, then that should be taken into account when comparing the bank algo strategies against those offered by new entrants and not just comparing based on pricing levels,” Razaq adds.

Yet ultimately some clients will only look at the headline fee number and want to normalise the fees charged to equally distribute their algo order flow across those various providers, he says. 

According to Razaq, this has led to the increased use of algo wheels in the market, which try to normalise the algo fee for clients to differentiate algo routing by peer performance of the algo. “But the fee is always going to be a one way parameter, it never goes up – it always goes down,” he says. “Even if you are performing well, the algo wheel will just route more flow to you that is going to take time to bed.”

”The use of algo wheels will increase next year but at BNP Paribas we tend not to like them as they serve a purpose for those clients who do not value the qualitative nature of the service. Often clients demand high performance tools such as pre-trade analytics and in-flight controls. This is a vital quality measure which algo wheels do not cover well.”