Nicola Tavendale

Strategies & Tactics: Discussing ways to improve FX algo trading outcomes

August 2024 in Industry Views

Asset managers recognise the value of execution quality in helping the performance of their portfolios and have therefore focused on algo and TCA adoption over the past seven to ten years, according to the findings of a recent report from Coalition Greenwich. However, outside of the largest companies, algos and TCA are not as prevalent on a corporate FX desk, but there is a strong case to support the continued adoption of algo execution, which has the potential to “deliver better outcomes for the client”, the study says. What then can FX algo providers do to further encourage buyside engagement with algos, not just in improving execution performance but helping clients to understand the results of their tactics and trading approach? Nicola Tavendale writes.

The Coalition Greenwich report, Corporate FX Trading—The Value of Relationships and Execution Quality compares practices between corporates and asset managers, discussing how the shift to electronic trading and digital innovation is transforming corporate FX trading. “In an age of digital innovation companies who are not giving FX the attention it deserves could be leaving real benefits and money on the table,” says the report’s co-author Stephen Bruel, Senior Analyst in the Market Structure & Technology group at Coalition Greenwich. Today, however, corporate treasury departments have access to digital tools such as TCA to measure the results of individual trades and the performance of the trading desk overall, as well as increasingly sophisticated algorithmic trading strategies that are adopted by asset managers and other FX market participants, the study finds.

One key consideration for both clients and algo providers is the pressing need to focus on bringing down the overall cost of execution, says Vittorio Nuti, Global Head of LD & FX Algo Trading at Deutsche Bank. He argues that clients need to look at more than the execution performance of algos and should also be considering how to streamline the costs of connecting to a variety of banks and platforms in order to access execution algos. Much of this depends on the fee that clients need to pay for using an algo, Nuti adds. For example, he explains that while the bid/offer spread on a eurodollar execution may be tiny, executing in emerging market pairs tends to incur much higher costs. “Clients also incur a variety of brokerage costs for executing FX which have remained static over the years, whereas the cost of executing algos and other more complex products have come down,” says Nuti. “If clients want to lower the overall cost of execution, understanding that process is going to be fundamental. One way to do this is for clients to start connecting their OMS directly to the banks, rather than going through some of the third-party vendors. This reduces the layers and layers of costs and the various fees and commissions which have to be charged to manage and maintain these systems. It is something Deutsche Bank is starting to offer and we believe this will be a growing development in this industry over the next three to five years.”

Insights into algo performance

Buyside clients with a reasonable amount of volume are already beginning to automate this process to bring the cost of algo execution down, according to Nuti. He adds: “Clients can see the numbers adding up to become a significant cost, so if they can automate the process and connect directly to the bank it makes more sense.” Gaining better insights into their algo execution performance is still also an important consideration, Nuti explains. “We frequently run AB testing for our clients, running pre-set parameters and testing over periods of time, then reviewing the results and helping the clients make more informed decisions on the back of those results,” he says. “Communicating with their algo provider is always going to be the best way for clients to improve algo execution performance, but a lot around the algo customisation piece can be hard to decipher, there is a lot of noise in the results. That is why we offer AB testing for our clients. It provides a stable performance result based on a set of parameters, eliminating the need for clients to perform a large number of trades to settle on the correct settings, which also means better automation of the whole process.”

Meanwhile Oleg Shevelenko, FXGO Product Manager at Bloomberg, adds that the overall success of an algorithmic execution can depend on various factors, including market conditions at a time of the trade, order size, algo strategy itself and the liquidity the algo has access to. Therefore, prior to placing a trade, the buyside needs to review relevant market data information to assess the pre-trade landscape and understand the expected trade costs, he explains. “As pre-trade market assessment is largely algo strategy agnostic, using an independent provider which is able to access the large sets of aggregated execution data may be preferred,” says Shevelenko. “For example, as a part of the Bloomberg Terminal offering, we provide access to relevant news, historic pricing and spreads, volatility and currency correlation analysis, which are all key factors which can be used to better inform trading decisions. For clients, this means that their subsequent choice of dealer and algo strategy can be based on a qualitative understanding of respective algo and quantitative predictive insights. Algo providers are certainly best placed to provide those details while platform providers need to ensure that they are readily available.” As a result, Shevelenko notes that Bloomberg has created an FXGO algo taxonomy which is intended to help clients classify the algos from different providers and encourage the algo providers to contribute their strategy descriptions into FXGO’s consolidated algo information portal. In addition, the platform also offers an integrated framework to allow liquidity providers to showcase their analytics as a part of the client order workflow on FXGO, he says.

In addition, there are several reasons for clients to utilise algo strategies over other execution methods, he explains. According to Shevelenko, this importantly includes the ability to minimise market impact to avoid adverse price movements while executing a large order. “Moreover, dealer algos allow clients to tap into additional liquidity pools which they otherwise would not have access to,” he adds. “Also, if the execution time horizon permits, the client can utilise passive strategies, which offer an ability to capture bid-ask spread and reduce execution costs.” As the FX market is bilateral and relationship based, a mutual understanding of trading requirements remains key for both clients and their dealers, Shevelenko says. He adds that a quantitative analytical framework, based on the mutually agreed data sets and parameters, makes bilateral discussions between the buyside and the sell side productive and actionable, allowing clients to adjust their trading styles and helping the dealers to offer the desired tools to meet them. “As a part of FXGO we offer such insights based on full life cycle of the trades highlighting cost of execution, cost of rejects, market impact, hit ratios among other key trading characteristics. Often such discoveries lead to adoption of algos as the most suitable trading style for the particular investment objective,” adds Shevelenko.

Exploring algo impact

Market conditions this year have also served to shine a light on the different interactions clients can have with their algo providers, says Preston Mesick, Global Head of FX Algos at Barclays. Part of this is very tactical, he observes, noting that clients openly communicating with the algo provider about what is happening in the market and the impact on the liquidity the algos can access is key. This can include what the provider is seeing in the FX market, what that provider did in the past and whether their algo is behaving as should be expected in these conditions. “This is a tactical, timely conversation that happens with the type of customer who is focused on getting the best performance from the algos,” he says. “Market conditions are changing very rapidly and having that up-to-date information can help clients to better understand their execution outcomes as a result. Another important factor is post-trade analysis, so not looking at any one execution in isolation but comparing performance across the portfolio. Liquidity is going to go up and down and the market will experience dislocations, but the reason clients need to have an execution strategy mechanism in place is to minimize the portfolio variance and to achieve more consistency in their algo performance.”

Buyside clients benefit from regular dialogue with their algo provider about the market conditions and the impact that can have on algo execution, coupled with discussions around post-trade analytics, portfolio effects and how to further adapt their execution style to get the most out of algos, adds Mesick. His colleague Ajay Kataria, Head of Electronic FX Distribution, Americas at Barclays, agrees, noting that clients also need to be wary of utilising certain peer-universe datasets offered by some third-party TCA providers. He explains that these datasets are a generic mass of data so cannot take into account the individual client, their trading strategy or what they are trying to achieve. “At Barclays, we aim to offer a much more customised, consultative approach, bespoke to each client. We can explain the data, show the results of clients who have a similar trading style and demonstrate how they achieved those results. Clients want to learn about algos and how to make them perform better and we want to help them achieve those goals,” says Kataria. 


Post-trade algo execution analysis is a must

The world of algo providers is also significantly bigger than it was 15 years ago and it can be hard for clients to make sense of all the different algo strategies that are offered by the various providers, he adds. “Clients now have so many options available to them, but it is our job on the sales and consulting side to help them figure out what they want to use to achieve their goals. We just have so much more algo usage data available than they have themselves and so we are in better position to help them better achieve their desired outcomes.” In turn, many of the default algo parameters are still the best for the majority of clients to use, notes Mesick. Clients do not have the ability to learn the nuances of 150 different algo strategies, he adds, so instead they typically aim to have a good baseline execution profile for certain algos. Also, the more clients start to make more customisations, the more difficult it can be effectively compare their overall algo performance, Mesick says. “The question then is whether the algo performance is due to the type of algo or to the customisations that were made, or if the client wants to compare providers, are the parameters the same for both. You start having a fungibility problem of comparison. So there are downsides to algo customisations and material trade-offs to consider.”

The need for fine-tuning

However, Nuti argues that the desired outcomes from algo execution can be unique to each client, so he sees a huge benefit in being able to tailor their algo accordingly, but this depends on their overall algo usage and the volume of algo executions they perform. “Ultimately, if a client is only trading a couple of 100 million a month, then they will not want to spend time tweaking and or creating special settings for an algo,” he says. “They need to be trading reasonable sizes to even go down the route of customization. It is not just about improving outcomes, which we obviously strive for all the time and our day-to-day job is improving the performance of our algo suite as a whole. Yet for clients, it only makes sense to spend significant time on customisations if they are trading large enough sizes for there to be a cost benefit to reward that time spent. Having a view of how much of the time is spent on these customisations versus how much a client is going to save in dollar value is a good starting point to evaluate if it is worth doing.” 

As an algo provider, Nuti explains that Deutsche Bank has the algo analytics data available which is already being used to effectively improve the algos all the time. Using an algo off-the shelf is a great way of just utilising this data and expertise, he adds. “It is only if a client has a specific goal then we would start looking at how to customise the algo, or if they have a non-standard use case, such as not wanting to cross the spread. Yet overall, if a client wants passive execution then they can reliably turn to our Stark passive algo, knowing it is already optimised to perform in a controlled manner with excellent results,” Nuti says. 

Expertise and market insight

Yet if one can argue that pre-trade and inflight analytics are not fully scientific or reliable and therefore maybe optional, then post-trade algo execution analysis is certainly a must as it is based on the actual traded data and the investment strategy which is likely to repeat in the future, says Shevelenko. Dealers can, for example, use post-trade data as an opportunity to discuss the benefits and drawback of in-flight order amendments and can highlight any positive or negative impacts of those to be accounted for in the future, he explains. “For quite some time participants in an algo trade were solely relying on a dealer’s reporting capabilities due to the richer set of trading and market data points available at that time. Recently, independent TCA providers have started playing an important role as they allow clients to have a systematic way to analyse algos from multiple providers using the same set of benchmarks. Bloomberg’s cross asset BTCA product offers those capabilities,” he adds. 

In addition, according to Shevelenko, many FX algo strategies or parameters which are considered standard today were born as bespoke, driven by clients’ desire to be able to address a particular situation or explore a market opportunity. “That type of continuous feedback loop between clients and algo providers is a driver for innovation and rapid progress of this segment,” he says. “But as always, there is a trade-off and clients need to acknowledge a degree of risk which custom strategies may introduce due to limited performance and test data.”

Another important aspect which Mesick highlights is that the performance of the various algo strategies will vary depending on the prevailing liquidity, time of day, currency pair etc. However, Barclays is committed to continuously improving the algos’ performance, he adds. “Over the last year and a half, we have focused on further enhancing our passive strategies to improve internalisation rates and to help clients understand that their market impact is dampened by utilising the franchise that we able to offer,” says Mesick. “All of our float, adapt and TWAP algos are taking advantage of that much more directly. We are having that conversation with customers and can demonstrate with our data the performance change over time. As the microstructure of the market changes, that dialogue is going to allow customers to appreciate when our recommendations might change, or if we introduce a new feature that we’ve implemented to improve that execution, and this will resonate more with our algo clients who understand that our changes are the result of our long-standing leadership in the FX algo space.”