Anna Reitman

Harnessing the full power of algorithmic FX trading strategies

June 2023 in Previous Features

Anna Reitman looks at why trading firms and investors should look to deploy FX algorithms as a means to make their FX execution more efficient.

Over the years, algorithmic execution strategies have advanced in technology and capabilities. And in an FX market that is progressively electronic, there’s plenty to learn about making the most of another tool in the evolution of best execution.

FX has been a screen-based market for years, and the latest figures show that globally, 64% of investors use some form of electronic trading, while in the US and Europe that goes up to 74% and 79% respectively, according to Greenwich Associates.

In terms of algo trading specifically, of 200 institutional FX traders recently surveyed by J.P. Morgan, 12% reported time was spent with algos versus 83% on click-to-trade. But there are also expectations that this will change over 2017, with 38% of users planning on increasing algo usage.

The rise of algos, however, is a new development and has a lot to do with data, said Richard Johnson, an analyst in the Market Structure and Technology Practice at Greenwich Associates. “The advantages are that it enables investors to, in the same way that happens in other asset classes, algorithmically look at a lot more data in one go, more than a human trader can, and make decisions based on what they’re seeing from a number of different liquidity pools. And execute orders more quickly, and according to the specific type of strategy that the investor chooses.”

Richard Johnson

“Randomisation of order timing, randomisation of order sizing, and intelligent use of order types and liquidity venues will minimise any observable footprint the algo makes.”

Regulation too has its part to play in helping the rise of algos in addressing transparency demands in FX, notably, MiFID II and the FX Global Code of Conduct.

Given the evolution of regulation, and in particular best execution requirements, Credit Suisse believes algos have a big role in helping clients adhere to systems and controls, ultimately increasing efficiency across the lifecycle of FX execution.

It’s becoming increasingly clear, that on a trade-by-trade basis, a set of transparent and standardised metrics will be required, said Ben Robson, Vice President AES FX and EFX Sales at Credit Suisse.

“We have seen a migration in FX spot execution away from voice dealing towards electronic venues. In 2017, we believe this shift will be exacerbated further as clients ensure that trade data are more easily captured and efficiently distributed,” he states.

Although click and deal transactions allow liquidity providers to compete easily and transparently, an increasing number of order types are migrating to algorithmic execution platforms. Algos, which typically run over a longer timeframe, allow for the capture of more data points and better benchmarking of individual fills, as well as increasing transparency around how orders interact with market liquidity, Robson explained.

There’s been an observable reduction in real liquidity over the last 18 months, he added, not only around month- and quarter-end, but also on economic releases and events intraday.

“Our clients, especially the corporate and real money institutions, are increasingly adopting algos as a way to execute more efficiently by reducing market impact on their bigger tickets,” said Robson. “Reduced liquidity also means that relying on any one source can lead to inefficiencies.”

That’s where algorithmic trading solutions step in: clients can optimise their execution workflow by leveraging a suite of trading behaviours while accessing aggregated liquidity, which not only allows for better execution but also helps with the adherence of regulatory requirements.

The rise of algos has a lot to do with data


Ben Robson

“Our clients, especially the corporate and real money institutions, are increasingly adopting algos as a way to execute more efficiently by reducing market impact on their bigger tickets.”

Corvil, a FinTech firm that works with platform and market data providers among its FX clients, saw the first significant amount of business heading towards FX clients some three years ago, followed by a spike in growth.

In many of the tier 1 banks, the FX teams started to use the same infrastructure that was developed for the equities teams and leveraging some of the same algos, said Donal Byrne, CEO of Corvil, and the big change this year is the arrival of the tier 2 and tier 3 banks.

“That to me signifies the continuing migration and adoption of all things electronic for the FX market,” said Byrne.

Moreover, many of the players who became electronically proficient are looking for new liquidity opportunities and have moved into FX markets: “You see a lot of the so-called old high frequency traders that became market makers now participating in FX markets,” he added.

Another reason accounting for that spike in growth is efficiency: “Within a market place that was arguably behind its time in the adoption of technology, it’s now modernising. Our systems are used in quite a few mixed environments, where people are trading equities, derivatives, FX and now fixed income on the same infrastructure with multiple desks.”

The big challenge in the FX space, said Byrne, is concern about stale quotes: “If they’re not getting up-to-date time critical market quotes, then they won’t be efficient. So, what our customers use our monitoring systems for is to bring transparency and visibility and to assure their end-clients that they’re not working off stale quotes, that they have the latest data and are providing the best execution possible in the FX market.”


On the buy side, there are some reservations against using algos, even for large asset managers that applaud the dominance of electronic trading, and work with third-party vendors for transaction cost analysis (TCA).

In part, it’s because the buy side “seems fixated on this RFQ idea and they don’t quickly adopt, or sometimes not at all, other trading protocols”, and FX is no exception, said the head of trading from a large asset manager based in the US. More specifically, suspected information leakage is one reason that he stopped using algos a few years ago.

The reality is that proprietary trading firms are always going to be looking for some information advantage in the market, and banks and vendors need to ensure their algos are optimised to minimise information leakage, said Greenwich’s Richard Johnson.

Asif Razaq

“We are now focusing on providing clients real-time analytics, mid-execution, which includes a live update of estimated performance based on real-time data.”

“The key to this is removing any predictability in how the algo trades: randomisation of order timing, randomisation of order sizing, and intelligent use of order types and liquidity venues will minimise any observable footprint the algo makes. This applies in all asset classes including FX,” he noted.

Other major concerns identified by the head of trading are potential conflicts of interest, particularly that algo providers to most buy side firms are both principal and agent.

“You’re sending your information into the bank and not really knowing what’s really happening after the fact, and I think it’s just really good practice to take as much control of the order as you can,” the source said.

It’s a concern that BNP Paribas has found to be pervasive as they enlist clients.

“We are getting questionnaires from clients asking us how our algorithms work as part of their due diligence process. They want to understand what actually happens under the hood so they can make an informed decision on whether to use algorithms,” said Asif Razaq, Global Head of FX Automated Client Execution in BNP Paribas’ Global Markets Unit.

“We get many buy side firms giving us questionnaires we need to answer before they can even start using our products, understand what level of transparency we provide and what level of segregation we have. These regulatory demands are actually increasing the demand of execution algorithms quite significantly over the last year or so,” he said.

But transparency wasn’t just built to address regulatory demands, rather, it was also about avoiding the stigma of “black box technology”.

“It gives clients an audit trail, so they can go back and audit the bank, checking what happened at this particular time with any particular order because we see our market impact and then, because this is technology based, we can go back and look at our logs and give our clients an answer,” said Razaq.

One of the major changes, pushed by regulatory demands, is getting a view of what’s happening inside the bank itself. The question regulators are asking is: how are banks organised internally, and if a customer submits a very large order via an algorithm, who in the bank can see it?

This is where a product like execution algos fits in nicely, said Razaq, as BNP Paribas has set up a completely segregated trading desk to manage algorithmic execution orders away from the traditional risktaking desks.

“This gives clients confidence that when they execute an order, that information is not being disseminated across the bank,” he added.

Algos have a big role in helping clients adhere to systems and controls


Evangelos Maniatopoulos

“Increasingly, the demand we are seeing is for the provision of pre-trade TCA solutions,”

In the beginning, clients had FX execution algorithms and black box units: there was an execution, a fill came back, and that was the end of the story. Now post-trade analysis in FX markets is extending the narrative. Greenwich Associates data show that 63% of traders are using brokers’ proprietary tools, 38% are using a thirdparty vendor system, and 21% are using internal systems.

Compare that to equities markets where 86% of traders are using a vendor product.

“In the TCA space, you see people relying on internal solutions first before deciding to invest in a third-party solution,” said Richard Johnson from Greenwich. “There are some difficulties around doing TCA for FX because there’s no consolidated tape. It’s hard to know how much of a footprint you really have in the market when you don’t really have a full viewpoint of everybody else’s trading.”

When BNP Paribas launched its FX algorithmic execution service, the bank built an interface that gives live feedback around what algos were doing in the markets, so that clients felt more plugged in rather than just looking at a black screen.

After that, there was a focus on post-trade analysis: running analytics for a TCA report that detailed all of the benchmarks, exactly what the algorithm is doing, price action in the market, how the algo traded, and whether it traded aggressively or passively. In other words, the microlevel detail of every transaction to the millisecond.

Since then, a dedicated TCA team was assembled to develop a more holistic view to look at a portfolio level of algo trades over a period of months, a quarter, or a year, depending on client requirements.

A portfolio analysis results in feedback on what’s working, and what’s not. For example, it may become evident after looking at the execution analysis overall that algo performance is lagging in a particular currency pair at a specific time of day when there is market impact. By moving the trading time period forward or back to a better time, there could be significant cost savings in the execution, Razaq explained.

Last year, BNP Paribas launched their pre-trade TCA tool, Cortex Insight. Before launching a trade, clients can run meaningful analytics, look at options, and estimate costs of execution in a simulation environment.

This year, the bank’s going further with real-time TCA:

David Mechner

“Trading systematically permits averaging results across orders, which in turn results in a greater likelihood of better outcomes when compared to trading in a more ad hoc manner,”

“Now that we have given our clients a view of our forecasts on a pre-trade basis, and a view of the realised performance by looking at the post-trade TCA, we are now focusing on providing clients real-time analytics, mid-execution, which includes a live update of estimated performance based on real-time data,” says Razaq. When looking at the evolution of FX algorithmic trading over the last decade, it’s hardly surprising how advances on the execution side have been matched with an equal pace of development in the trade analytics space, said Evangelos Maniatopoulos, Global Head of AES FX Product and Trading at Credit Suisse.

And while the initial focus was firmly on analysing execution performance on a post-trade basis, the increased sophistication of execution platforms as well as the client base has given rise to a number of innovations that span the pre-, intra-, and post-trade analysis spectrum.

“Increasingly, the demand we are seeing is for the provision of pre-trade TCA solutions,” Maniatopoulos said.

As popular as it may be though, pre-trade TCA should not be viewed in isolation.

“The wealth of available data has meant that post-trade TCA is as important as ever, and the successful solutions will combine pre-trade estimates with post-trade results, creating a feedback loop that enables TCA providers to continuously improve their models and ultimately empower clients to extract important insights from the data,” states Maniatopoulos.

TCA, he added, is often considered “more art than science”; the complex dynamics driving execution performance are too difficult to show in a single average, cost estimate, or even a table of segmented data. “Instead, we use sophisticated tools to bring patterns to life, help understand multiple drivers of cost and focus traders on the data points and patterns that matter,” Maniatopoulos said.

An example to illustrate this point is to look at the Implementation Shortfall analysis for a trading desk’s set of orders, broken down in hourly segments. While a set of tabular data would provide a quantitative assessment of these results, it does not necessarily allow the reader to easily frame them within the broader context.

By making use of visualisation techniques, extra dimensions (for example: liquidity profile, average spread, and algo usage distribution) can be plotted alongside the original dataset so clients can easily extract insights such as “my algo selection during the less liquid New York afternoon could be impacting my performance”, or, “my volume allocation is optimally distributed to benefit from tight spreads and deep liquidity”.


The fragmented nature of the market, where there are 20+ venues for spot FX, and managing these different liquidity venues, is one of the big challenges influencing the effectiveness of algos, BNP Paribas’ Razaq added.

To address this, the bank has built in another level of artificial intelligence into the order router: “We have built a learning engine into our smart order router such that every time the algo trades in the market, it will log and capture its experience. If it has a negative experience, it will remember, and next time it will move away from that venue and trade where it sees more reliable behaviour,” Razaq said.

Fragmented electronic markets are what make algorithms an “indispensable tool” for trading, said David Mechner, co-founder and CEO of Pragma Securities, a provider of multiasset algorithmic trading tools for major banks, brokers, and hedge funds. Pragma integrates neutrally with every ECN, all the big dealers and all the thirdparty aggregation systems.

“There’s so much data and complexity coming from a fragmented electronic market – human traders need tools to master that complexity, and that’s what algorithms do; they automated best practices that traders might do manually if there wasn’t so much data and so many different things to deal with,” said Mechner. “As every electronic market evolves and becomes more mature, algorithmic trading is an inevitable endpoint.”

Mechner identified three core benefits to algo execution: breaking up a large order into multiple smaller pieces means, on average, paying less than trading in a block; building algorithms on top of an aggregated liquidity pool effectively narrows the spreads being traded on; and automation itself means fewer people can handle more orders, more reliably.

From a buy side perspective, using an algorithm alone doesn’t guarantee any better execution than an RFQ. In practice, however, algorithms can aid transparency.

“It’s more likely that you are going to have time stamps on orders and executions, and with more data from more individual executions, you can better compare against benchmark prices,” said Mechner. “Trading systematically permits averaging results across orders, which in turn results in a greater likelihood of better outcomes when compared to trading in a more ad hoc manner, and that lines up with algorithmic trading methods.”

Active trading styles have greater potential for market impact


While TWAP and VWAP are sometimes considered firstgeneration algorithms, they perform well and remain the workhorse. Even in extremely mature developed markets like equities, they represent almost half of all algorithmic usage, said Mechner.

In FX markets, he noted that an increasingly popular algorithm is the “float” or “provide” strategy, which is a passive strategy that provides liquidity on platforms. “That’s an empowering feeling for the buy side, to know that they can participate in providing liquidity and they don’t always have to be takers.”

Algo strategy deployment, said Credit Suisse’s Ben Robson, will depend on the institutions in question and their respective execution objectives and goals.

Historically, hedge funds and market participants with a short-term investment profile have tended to use strategies on the aggressive end of the spectrum, while corporates and real money institutions, which are typically less sensitive to levels and limits, have opted for strategies that run over a longer period of time in favour of passive execution and spread capture.

Across the industry, the overwhelming majority of flow is passive with a share of 81% compared to 19% for aggressive strategies, according to Greenwich Associates.

n the J.P. Morgan survey of 200 institutional FX traders, 42% identified liquidity-seeking/ limit-based orders as the most popular algo, with 80% ranking it in their top three. This was followed by market trading/ pegged algo with 31% giving it the top spot, with 75% identifying it in their top three.

Donal Byrne

“The big challenge in the FX space is concern about stale quotes”

Irrespective of algo selection, common themes have included the deployment of execution strategies that minimise market impact and the use of diverse pools of liquidity, said Robson: “Our client discussions have revolved around the use of algo strategies that intelligently access both internal and external pools to optimise execution, which has resulted in increased interest for the use of hybrid algos that employ a variety of execution components in a dynamic manner, as market conditions change.”

The execution profile and resulting cost outcome of deploying an FX algo can vary greatly based on a number of factors that range from the client’s execution goals, tolerance to risk, and algorithm selection to the order’s speed of execution, as well as total notional amount and how it compares to market volumes during the transacted period, Robson added.

The more active the trading style employed by the trader, the greater the potential for market impact from such a transaction.

“Traders that employ these execution behaviours will likely be looking to enter and exit positions quickly, have a directional view and will employ opportunistic algorithms with tight limits for further protection,” said Credit Suisse’s Maniatopoulos. “Although they will be willing to accept a degree of execution shortfall, algorithms such as TWAP or VWAP will not feature in their arsenal as the timing risk associated with them will greatly diminish any alpha in the transaction.”

On the other hand, firms that take a longer-term view to currency trading, or perhaps need to hedge their currency exposure alongside a primary transaction in the equity market can greatly benefit from employing algorithms that seek to achieve a more uniform execution over a longer trading horizon. The cost of execution in this case will most likely be measured against an interval TWAP or VWAP calculation, instead of targeting an arrival price benchmark.


Market-specific factors will also play a significant role in the effectiveness of FX algo execution.

First and foremost, explained Maniatopoulos, is the liquidity environment that the algorithm is operating in. During periods of enhanced liquidity, larger transactions can be absorbed by the market over shorter periods of time with minimal impact, opening up an array of execution options. But during liquidity gaps, market participants need to weigh-up a more cautious approach against the time risk of the market naturally trending against them.

Volatility too is an important factor that can influence FX algo execution effectiveness. FX algos that are designed to quickly react to market signals have often been proven to perform exceptionally well during periods of high volatility due to their ability to effectively capture liquidity at opportune price points.

A case in point, Maniatopoulos said, was seen during the eurozone sovereign debt crisis in May 2010. When analysing overall performance across the most popular AES FX strategies, the most aggressive algos achieved the best relative performance. In such trending and volatile markets, these strategies were also shown to often reduce lost alpha due to trend costs, as their average duration was almost half that of other strategies.

A more recent example of opportunistic algos performing well during a market dislocation event was when the Swiss National Bank depegged from the euro in January 2015, he added.

Most popular FX algos


This year, Credit Suisse is looking to “super-charge” its TCA framework by introducing a number of enhancements that “put clients in the driving seat”.

“Our focus spans the full spectrum of data analytics, including pre-trade, realtime, and post-trade TCA, all delivered through a streamlined and easy to access interactive portal,” said Maniatopoulos. “Our new tools will enable clients to transparently visualise the execution progression as it takes place, better understand the impact of algorithmic parameter changes and include new metrics for measuring execution cost,” he states.

With analytics being a big part of algo trading, Johnson from Greenwich Associates thinks that execution management systems are going to get increased visibility, with some companies creating portals for a better understanding of what algos are doing.

The next generation of algos, said BNP Paribas’ Razaq, will go past being adaptive to becoming interactive. Currently, the firm has three flavours of adaptive algorithms: Viper, which runs a fast execution with minimal market risk; Iguana, which is designed for clients with a requirement to trade over a fixed time-period; and Chameleon, which is a dynamic hybrid strategy.

Chameleon is by far the most popular because it gives “a bit of everything” and can morph into other execution strategies by itself due to its adaptive nature. Clients can adapt their strategy manually if they want to, or leave the algo to self-navigate in the market as there is a high level of AI built into it, Razaq noted.

“Clients are relying less on black box algos that do all the work for them, they want to have something from the algo, they want to be able to speed it up, slow it down, dynamically switch into a different algo strategy mid-execution,” he said. “These control factors are also going to be a key part of the evolution of  algos in the market.”