Exploiting the power of a new generation of more dynamic, customisable FX algos

One of the biggest issues with FX algorithms since their inception is how to view accurate traded volume data, particularly considering the lack of order book depth data that has been available to date, says Asif Razaq, Global Head of FX Automated Client Execution at BNP Paribas. He adds that in a bid to solve this problem, the FX algo community has been working hard to produce models that can simulate or reconstruct these views of the market and the missing pieces of data, which are then given to the algorithms to use. “This concept of producing hidden data, or data that does not exist in the FX market, and reconstructing that data is becoming more and more sophisticated over time. BNP Paribas, for example, utilise machine learning and AI to help us constantly evolve the generation of data such as volume data, volatility data, spread data etc which is then fed into the algorithm. The more accurate that data is, the more efficiently the algo can execute the order. It is a constant exercise in looking at how we can fine tune that data or make it more accurate. Using AI has significantly improved the quality of data that we are able to feed the algo, which in turn makes the algo increasingly efficient in its operation,” Razaq says.

After embarking on this data modelling exercise and having shared that data with the algo, the next step for BNP Paribas was to look at how that data could also be shared with the client to further inform their algo use. The output of these AI models is now able to be offered to clients in the form of the bank’s real time TCA analytics and Insight Live tool, explains Razaq. “This is very powerful because now the client can look at that data and take an informed view,” he adds. “Based on their individual execution goals and the interpretation of the data available, they can now alter the flight path for the algorithm accordingly. They can use these tools to not only see the data, but also actively modify a strategy to slow it down, speed it up or change strategy mid-execution in response to the data provided. These data simulations are becoming a core part of client analysis by helping them to know how they may want to change the flight path of the algo.”

For example, one of the models that BNP Paribas offers to clients is based on predicting the algos observed, such as giving clients a buy/sell pressure indicator which is updated in real time. “If the client is buying and they are seeing a lot more buying pressure in the market, that is going to impact their transaction because they are on the same side as the market is trading. It also helps to explain that if the market is going against them, then it is not necessarily the algo that is driving that but the general phenomenon that we are seeing taking place in the market. So it gives clients the ability to make a much more informed decision on what is going on in the market, rather than taking a guess without having the data to back it up,” Razaq says.

Mix of parameters and variables

Meanwhile, Harshad Hariharan, FX algo product manager at JP Morgan, adds that algo suites normally consist of seven or eight distinct strategies that have a pre-defined list of default parameters which may, or may not, work for all clients. However, algo customisations form an integral part of the workflow in certain other asset classes, notably equities, futures and options. “The success we have had with algo customisation in these other asset classes pushed us to explore the use of customisable parameters in FX algos as well,” Hariharan says. Furthermore, he explains that the algo performance evaluation process has also become increasingly standardised in recent years, with algo providers reporting their data to third party TCA providers. Hariharan adds: “The increased sophistication in the use of TCA data has enabled clients to explore ways of optimising execution parameters. Clients can now experiment with thousands of execution parameter combinations before arriving at the ones that closely match their execution objectives.”

Almost every algo execution parameter can now be customised to provide a tailored experience to clients depending on their execution objectives, according to Hariharan. This customisation framework also allows algo providers to collaborate faster with clients on ideas that meet their execution needs, he explains. “In fact, over the last 18 months we have seen client use of customised algos grow steadily, with July 2023 seeing about 40% of the total JP Morgan’s client adaptive flow being customised in some manner or other,” says Hariharan. “As we move forward, we believe customisations will continue to gain popularity, with more dynamic customisations coming into effect. We may potentially end up with consolidated algo suites that offer more execution options via customisations.” However, Hariharan warns that it is important to note that fewer algo strategies does not mean fewer options. In fact, he argues that algo customisations can actually unlock thousands of configuration possibilities to help clients achieve their desired results.

However, while FX algos are undoubtedly becoming more sophisticated at recognising market conditions, algo users still need to be aware of the market trends or news events that algos cannot currently capture, says Vittorio Nuti, Head of Segregated Execution and Advanced Solutions at Deutsche Bank. “It is more about having a rich understanding of what is impacting liquidity at a given moment in time,” he adds. “We are starting to see dynamic algos that can interpret certain market conditions and more efficiently identify pockets of liquidity, allowing better deployment within the current FX landscape.” In addition, Nuti observes that although algo providers are trying to make their algo offering as dynamic and customisable as possible, it is not simply a case of one size fits all. “Speed and style of execution may vary across client needs. Listening to and partnering closely with clients to achieve the best user experience is key,” he adds.

The new generation of analytics data

Razaq adds that the use of machine learning to make sure the data which feeds the algos is as accurate as it can be is an important piece of the puzzle, as it means the algo will be more efficient in its execution. In turn, that data should also be made available to the algo user to conduct their own analysis and make an informed decision if they want to change the flight path of the strategy. Razaq adds: “The other area is more of a dark art, which is knowing how you can model market impact with this data.

Having a good market impact model helps you understand the perceived impact that your execution could have in the market, before you even launch a trade. This feeds into the concept of pre-trade tools. Yet some pre-trade tools out there only look at historical performance of strategies and extrapolate that out to provide a possible indication of how an algo is going to perform in the future, which is a very basic way to approach pre-trade TCA. We wanted to go one step further and start using our market impact model to simulate what is actually going to actually happen if we were to run this order in the market at this time of day.”

Razaq explains that by running a simulation based on the AI, market forecasts and the market impact model, BNP Paribas can now provide algo users with a more accurate picture of the expected algo performance in the current market conditions, rather than basing it on what happened in the market yesterday. “Being able to provide an enhanced level of data, coupled with a very accurate market impact model, provides clients with a much more informed understanding of how to make the algo perform better,” he adds. “Our pre-trade tool also allows clients to run simulations so they can now also test the performance of the algo in different liquidity conditions and at different times of the day. At certain times of the day there could be heightened volatility, there could be economic announcements or market data releases, and we are now able to simulate the performance of the algorithm over those periods. That makes for a very powerful offering when it comes to looking at this data.”

FX algo customisation also helps algo providers identify and invest in changes that can help clients, explains Hariharan. “Historically, you would need to build and release the same algo ticket across multiple venues,” he says. “Innovation was slow, with certain venues limited by the number of releases every year. Customisation allows providers to offer hybrid algo executions with the algo provider managing all the technology changes. If a customisation solution becomes popular enough, it naturally gets added to the main ticket.” According to Hariharan, having a scalable, rapidly iterative, robust, and transparent customisation framework should be key in meeting the ever-changing needs of algorithmic trading as the market progresses.

Achieving execution goals

Real time and post-trade TCA are very powerful tools for algo providers to develop for their clients and which also help to meet the demand for greater transparency around how the algos interact with different liquidity sources, says Nuti. He adds: “At Deutsche Bank, we are working to provide our algo clients with full transparency around how their algo is performing in real time and enabling them with tools so that they can amend their order mid-flight, adjusting to the current market conditions as required.” Clients can also work with other TCA providers, such as Tradefeedr. According to Nuti, the pre-trade tool from Tradefeedr is a great example of how innovative analytics tools can further help clients to achieve their execution goals. As a leading algo provider, Nuti believes that addressing any remaining pain points for algo clients means needing to focus on algo customisation. “Our clients are our platform, so it is about how we listen to them, take feedback and incorporate their asks as we develop our offering,” he explains.

Razaq notes that one of the major pain points that persists in algo execution is when it comes to trading non spot products, such as forwards and NDFs, because there is still a swap component attached to that. “At the moment, the majority of our clients’ exposure tends to be either an outright or an NDF,” he adds. “The spot element of the outright transaction is fairly transparent because you get all the details in the trade report, the algos are trying to seek the best price in the market and the client can see which markets we’re trading in. Yet when it comes to swaps, there is currently no electronic marketplace and so clients are then bound to trade against the bank’s house points. So clients can perform analysis on the execution of the spot points, but they have far less visibility and less transparency on the swap points.”

Continuous evolution and market development

As a result, Razaq says that clients are keen to find the same user experience of trading the swap as they are able to have with the spot element. “But in order for that to happen, there needs to be a swap market where clients can trade,” he explains. A further issue for clients is the way the swap points are traded. Typically, the algo executes the spot transaction and when it finishes executing the big block, it then goes and seeks the forward points at the end of the execution and applies that to the trade, says Razaq. “So clients feel like they’re paying a premium because if there’s any impact in the market, the swap points are still only calculated at the end of the transaction,” he adds.

To solve this issue for clients, BNP Paribas introduced a service called continuous swap hedging. This works by executing the swap points throughout the execution, says Razaq, rather than just waiting until the end. He adds: “Every time the spot execution gets to a certain size, we will then do a ‘mini-swap’ transaction with our swap desk and get the swap points locked in at that point in time. So throughout the execution, clients can now see a number of data points where we traded the swap. We then calculate the average of all of those swap transactions that we have booked against the client, so the client can see the smoothing effect this has on what the swap charge will be for them at the end of the execution.”

“In the FX market, we will also soon be seeing the arrival of the first swap ECN or swap venue where we can start trading electronically with an interbank market,” Razaq says. “When that does go live, then we can substitute that continuous swap hedging for clients so instead of exclusively trading with BNP Paribas, we can now open up and trade on these other ECNs in parallel to BNP. This will go a long way to helping solve what is quite a big problem for algo clients.” He explains that there currently is no electronic solution for trading swaps with different banks, it is only an option available when using voice trading. “BNP Paribas is also participating as a market maker in this programme. We have already committed to providing prices as an interbank market maker to this upcoming new venue and we expect other banks to join as well. This will create the first electronic liquidity pool for swaps, which in turn will help further improve transparency for algo execution in forwards and NDFs as well,” concludes Razaq.