Nicola Tavendale

Achieving the perfect fit: Matching FX algo intelligence with trading intentions

March 2022 in Trading Operations

The use of FX algos was already on the increase, but market conditions in the past two years have noticeably accelerated the trend and the rise looks set to continue unabated. For example, when asked what percentage of their FX trading is or will be done using algorithmic orders, traders predicted an additional 19% of their FX trading will be done via FX algos over the next two years, according to the findings of JP Morgan’s 2022 FICC e-Trading Survey. But how can users ensure that using an algo execution is really the most effective way meet their execution goals and what more can algo providers do to help with that challenge? Nicola Tavendale investigates.

An important benefit of FX algos is that they can help reduce the cost of execution

When the Global Foreign Exchange Committee (GFXC) was in the process of updating the FX Global Code, it made a number of very interesting changes around the guidance for algorithmic execution, says Patrick Guevel, Head of FX Algo Execution at Societe Generale. “One of these was the observation that customers did not have time to compile analytics and were relying on the banks or third party suppliers to construct them,” he adds. “But they were also said to need more education or clarification from the algo providers and analytics vendors in order to help them better understand what the analytics actually meant.” One of the solutions suggested by the updated version of the Code was for providers to improve their TCA reports and to publish metrics in a more standardised format for customers to make the right decisions around their algo use.

Many third party vendors already produce detailed analytics reports to help banks and algo users evaluate their algo executions. However, according to Guevel the level of detail provided might be too obscure for some users to really understand unless they are dealing with those issues on a daily basis. “For instance, it is very useful for us as an algo provider to look closely at some of the data provided to help us measure how effective an algo has been at grabbing liquidity. Yet for customers, who might perform an algo execution perhaps only once a week at most, these numbers can be harder to interpret. So when we speak to customers about how they should measure the performance of an algo, we need to provide very clear answers and be very transparent.”

Understanding and insights

Clients will naturally be happy if they see that the algo has performed at the risk transfer price or the mid arrival price, but will be less so if the next time they use the same algo they see a loss compared to the risk transfer price, Guevel says. He adds: “There can be misunderstandings about algo performance and what the analytics really show. It all comes down to this central question of whether algo execution is a good choice for the customer or not. The best way to show this is with standardised numbers from post-trade analysis, such as those provided by third-party vendors, which can help to demonstrate what their expected outcome should be.” 

As with many things in the FX marketplace, another important way to help the buyside is for FX algo providers to start being more transparent about how their algos work, adds James Singleton, Chairman and CEO at Curex Group. He argues that many algo providers instead seem to prefer to use nomenclature that avoids getting into too much detail about the actual intelligence employed by their algos. “Industry terms such as passive/aggressive; fast/moderate do not really tell the customer much in the way of useful information,” Singleton says. To really help the buyside better understand algo functionality, he argues that it would instead be more useful to supply them with the answers to a few key questions.

“The buy side should ask for a clear explanation of why algos choose to place orders outside executable, top of book prices to understand if that ‘intelligence’ is actually benefiting their trade intentions.”

James Singleton

“The first important thing to learn is the algo’s selection criteria for choosing different liquidity pools, since we know it is not simply best price available at the time of each algo child order. Our client order surveillance has confirmed that unsettling reality,” Singleton explains. He adds that buyside clients should also be asking their provider how venue brokerage impacts pool selection by the algo’s smart order router, in addition to learning how the algo interacts with mid-pools and the provider’s own internalisation interest. “Is the client’s desire to execute the trade efficiently being slowed down by the provider’s interest to execute the trade in a preferred liquidity pool? We have observed order placement issues that negatively impact trade outcomes for our buy side clients,” says Singleton. “The buy side should ask for a clear explanation of why algos choose to place orders outside executable, top of book prices to understand if that ‘intelligence’ is actually benefiting their trade intentions.”

Increasing transparency

In addition, Singleton argues that buyside customers need access to dynamic, live pre- and post-trade data that they can interact with, but warns that is not something which is widely available in the market at present. He says the type of information that would be helpful might include a breakdown of passive versus aggressive fills, order placement versus available top of book pricing and rejection rates per liquidity pool with an explanation of what each algo provider deems to be an acceptable rejection criteria. “Our Cipher platform now provides a lot of this information because its analytics are streaming, measure existing volatility and RSI, include an algo cost estimator versus risk transfer price and link post-trade performance to pre-trade conditions to deliver actual performance metrics,” Singleton adds.

Guevel agrees that “opening the blackbox” is important, with providers needing to offer more transparency about how the algorithm is working. In particular, he highlights the need for customers to have access to pre-trade analytics to know what can be expected from the algo, as well as live TCA so they can see what is happening as the algo is running. Customers should be provided with some insight during the execution of how the algo is performing against the risk transfer price, he explains, as well as when the algo execution will finish.

With some algos, like SocGen’s aggressive algo strategy Falcon, it may be difficult to show an end time to a customer, he says. But according to Guevel, if you look at the bank’s passive algorithm, which can span across 10/15/45 minutes or a few hours, then the customer can see in the middle of the execution if they are doing well against the risk transfer price so far, but they might, for example, still have 40 minutes to wait until their execution is going to complete, which is important to know. “At this stage, that customer has more elements of information to decide whether they want to accelerate, to switch the strategy type or just leave it as it is. Being able to see this information helps the customer to better understand the value of using an algo instead of a risk transfer,” he adds. 

“We are trying our best to understand how to help customers ask these important questions about the algo, as many would benefit from having a better understanding of what the algo does and what the real benefits are from using them.”

Patrick Guevel

Trading intentions and algo use

Another important benefit of FX algos is that they can help reduce the cost of execution, Guevel notes. However, he says that while algo providers have done a lot of work around building models and tools to help identify what the best times to run an algo based on the historical data, it is not a question that is typically being asked by the client base. “Understanding that the expected cost of execution can be different at different times of the day is important to our side in terms of the positioning of the algo execution as a product, but for clients this is a less of a consideration – but it should be in terms of achieving best execution,” Guevel adds. “We are trying our best to understand how to help customers ask these important questions about the algo, as many would benefit from having a better understanding of what the algo does and what the real benefits are from using them.”

He continues: “Our offering has three algos and from day one, we decided that it was complex enough and it didn’t make sense to have eight algos with funky names. We prefer to have a limited number of algos that have a bespoke purpose.” For many customers, the key question should be whether using an algo is the right choice for that particular execution, which again comes back to having the right numbers, metrics and understanding of what is going on, says Guevel. “Sophisticated customers will usually have their own desk, third-party TCA providers and will have a good understanding of what’s going on and how they should behave in the markets.” he adds. According to Guevel, there are potential risks faced by clients if their FX algo intelligence, data or metrics do not match their trading intentions, not just the monetary impact of a loss. Clients also risk wasting their time because the execution is taking longer than expected, or they risk the cost of execution being larger than anticipated, he adds.

“This is why the bank provides pre-trade analytic tools that describe the expected individual costs of the algo, the potential impact in the market and the associated costs with these on the final price,” Guevel says. “We provide live TCA that shows similar metrics to the post-trade TCA, but in an interim form, to help customers understand what is going on and decide whether they want to accelerate, decelerate, change their price etc. They can interact with the functions to pilot the order during the execution. Then, at the end of the execution, they are also provided with a detailed TCA report which describe what the algo’s performance against the mid/arrival/risk transfer price was so they can accurately assess the general cost of execution for the order.”

Pre-trade analytics can provide accurate conditions to help customers choose the right trading
option algo or risk transfer

Providing the proper tools

Different customers will ultimately have different trading intentions, Guevel explains, which is why providers will typically offer more than one algo strategy. The customer may want to go fast, or to be passive, or to pace their execution and the algos can cater to those different requirements, he adds. “The best way to maximise the power of what is offered to the customer is to offer full transparency and full control,” he says. SocGen’s live TCA offering is a good example of this, he says, as it allows customers to see the orders as they are executed in the market, at which levels, they can see them being filled live and they can even decide to switch strategy mid-flight, Guevel adds. “But this is not something that we can leave to the machine to decide, it needs to have the human being behind the order still in control,” he says. “At some stage the customer will need to provide their input about how their order will be executed by the algo. We cannot decide for the customer but we will provide them with as much information as possible to decide which algo strategy to opt for and whether they want to change that execution style during the running of the algorithm.”

In addition, it is important for buyside institutions to understand that the pursuit of best execution, which is at the heart of MiFID II, can be assisted by the use of pre- and post-trade TCA, explains Singleton. “Pilots use instrumentation to help them fly at night and in rough weather. Traders should use analytics to help them execute their FX trades effectively since they need to adapt to changing market conditions constantly,” he says. “Buyside institutions in particular need access to good data and independently derived analytics.” According to Singleton, pre-trade analytics should provide accurate conditions to help customers choose the right trading option – algo or risk transfer, for example. In turn, post-trade analysis has to reflect the outcome of the trading decision against the reality of the pre-trade market conditions, he explains. “If traders cannot access this type of information, then they are paying their TCA providers for nothing,” he warns.

Singleton adds that if the trader does not understand the FX algo intelligence being employed, then there is certainly a possibility that it does not match their trading intention. In that case, the trader may be unknowingly accepting market risk and greater execution price slippage as a result of slower or delayed execution.“At the simplest level, they risk deviating from their execution mandate and exposing their firm or their underlying investors to unintended execution risk,” he adds. “Taking on that unintended risk could lead to negative performance. The vast majority of buy side traders are mandated to save spread, not to earn spread.”