A recent report from Greenwich Associates, the global financial information services firm, shows that the adoption of FX execution algorithms is steadily increasing among buy-side firms. This increase is against a backdrop of plateauing electronic FX execution volumes. The report, FX Execution: Competing in a World of Algos, shows that the share of notional FX trading volume execution electronically has hovered around 80% for the past several years.
Meanwhile, algo adoption rates across the FX market have stayed surprisingly flat, given the increased use in other asset classes. The same figures also show that institutional and corporate FX customers did not shift significant volume away from traditional voice trades to e-trading. However, in 2018, the use of execution algorithms increased by 25% to one in five FX market participants.
According to the report’s author and Principal, Satnam Sohal, the last 12 months has seen a marked increase in algo adoption among institutional investors. “Many of them used to say that they planned to use algos but the next year’s figures would stay the same. That changed in 2018. The growth was coming from asset managers, along with more sophisticated corporates, and was across the globe.”
The figures for 2019 are due out imminently and Sohal expects to see an increase in adoption among the next tier of most active managers active in FX.
Supply and demand
There is a supply and demand element to the algo adoption rates, says Sohal, with different dynamics driving the respective elements. “From a product perspective, algorithms are used in the G10 spot trade. If you see what’s happening in terms of margin pressure, banks don’t make a lot on vanilla G3 spot trades. Algos which pay an agency commission rather than a spread, offer banks a chance for more attractive risk-adjusted revenue and we see more banks coming out with their FX execution algo suite for clients.”
And in terms of client demand, liquidity dispersion in the FX market has been the big driver, says Sohal. “When buy-side participants are looking to execute large sizes, in liquid currencies, to minimise their market impact, they are turning to passive algos to work their order.”
Another contributing factor has been the development of more sophisticated transaction cost analysis (TCA) tools. “It is important for FX users to evaluate the quality of execution algos on offer and to have more transparency around an algo’s performance and to be able to measure its effectiveness.”
There is a ceiling to algo use though, says Sohal. “Algos are not applicable to everyone. From client feedback, there is clearly a volume threshold. Most algo users are looking to trade in a certain size – with a minimum threshold ranging between $10 and $50 million.”
There are other factors that influence the decision to use an algo – market conditions, the liquidity of the currency pair and the immediacy of the need to trade, says Sohal.
This is not to say that algos will not be developed or adopted for FX products beyond G10 currencies spot trades and other vanilla products. Better algos are being developed for certain EM currency pairs, as well as for non-deliverable forwards (NDFs).
The sophistication of the algos is also developing beyond the simple TWAP and VWAP algos that came out in the first generation. However, in terms of buy-side adoption of algos, there is a learning curve involved so more sophisticated algos are not likely to be used by new adopters, says Sohal. “I think the use of algos will increase but clients may also rationalise the number of algo providers that they use.”
When viewed alongside the more traditional risk-transfer alternatives, algorithmic trading is commonly considered a more sophisticated trading proposition, says Evangelos Maniatopoulos, Global Head of AES FX product and trading at Credit Suisse. “The successful algorithmic trading solutions provide a range of services and flexibility that cater to the needs of the novice and more seasoned users alike. As the FX market has continued to evolve, so have market participants by enriching their knowledge on topics such data analysis, optimal execution and the liquidity landscape. This has empowered them to take a more sophisticated view of the electronic FX market, ultimately being reflected in their decisions around how and where to best execute any given transaction,” he states.
“From bespoke trading solutions to unique execution insights, and from customised liquidity access to real-time TCA and in-flight order controls, the algorithmic trading providers with proven expertise and experience in the field are seen as great value-add partners in the execution lifecycle,” says Maniatopoulos.
Although the advantages of using FX algos are clear for large orders, Maniatopoulos maintains that algorithmic trading tools can equally benefit small as well large institutions. “Algorithmic trading strategies can be used to fulfil a range of execution goals irrespective of a given transaction’s notional size.”
Maniatopoulos says that in the 12 year history of Credit Suisse’s AES FX division, it has worked with a broad set of clients with a diverse set of execution objectives and varying algo portfolios. “Concepts such as reducing market impact and improving performance versus benchmarks ought to resonate across the investor spectrum. How to best achieve these, will of course vary based on the individual circumstances at hand as it rarely is a ‘one size fits all’. FX algos are part of any FX trader’s execution toolkit and like all other tools, they are most effective when they are used at the right time for the right deal.”
Transparency has also been a cornerstone of the Swiss bank’s AES FX platform, says Maniatopoulos. “Transparency is one of the key value propositions of algorithmic trading; transparency over how the algo executed, the venues it sourced liquidity from and how it performed versus the client’s benchmarks. Transaction cost analysis continues to play a pivotal role in this process, by providing clients the required visibility into an algo order’s lifecycle. Through their immediate feedback, real-time TCA tools offer invaluable transparency to clients and an increased level of accountability for their execution providers, algorithmic or not. Continued developments in this space will certainly contribute to further adoption of FX algos,” he says.
Another key factor that has always driven investment in FX algo offerings is customer feedback, says Maniatopoulos. “An example that demonstrates this dynamic is how the need for transparency has given rise to a series of innovations in TCA, which over a number of years have transformed the field. The highly competitive nature of algorithmic trading ought to ensure a continued stream of investment across the value-chain. As long as this is deployed to meet client demand it will lead to further growth in the use of algo tools by FX clients.”
The development of TCA has also been driven by regulatory developments such as MiFID II and best execution requirements, says Maniatopoulos. “MiFID II has encouraged the buy-side to question their execution processes, which has in turn led to more interest in both transparency and data-rich execution styles such as algorithmic trading. Although spot FX is not a MiFID II in-scope product, it is widely expected that FX dealers will apply best execution when transacting on behalf of clients in the spot market. The wealth of data and the transparency offered through the algorithmic trading process makes FX algos an ideal medium for investors seeking to prove best execution.”
Harnessing the latest technology developments is one means of demonstrating innovation and Maniatopoulos says that Credit Suisse has been actively engaged in evaluating machine learning (ML) methodologies for algo selection where the intention is to supplement human expertise with data-driven recommendations. “We believe the application of artificial intelligence and machine learning promises to deliver exciting new developments in the field of algorithmic trading. Perhaps above all though, harnessing these technologies in contributing to execution performance improvements and better client outcomes will be a catalyst in their adoption and in turn expand the appeal of trading algorithms.”
The next frontier
Advancements in electronic trading can be seen across the spectrum of products within the FX asset class and Non Deliverable Forwards (NDFs) are the next frontier in this process, says Maniatopoulos. “Although the NDF market is several years behind in the electronic stakes, tools such as FX algos are starting to prove their value to market participants that actively trade these products.”
Challenges such as broken-dated NDFs, the dearth of electronic liquidity in Latin American currencies and the regulatory hurdles associated with On-SEF venues still remain. And although broad spectrum automation in the NDF market is far from commonplace, the building blocks are in place, says Maniatopoulos, adding that Credit Suisse has recently released its first AES FX algos for APAC NDFs.
While adoption among the buy-side has increased, some participants have been more prolific algo users than others. Asset managers, hedge funds and the real money community have been especially active while a number of corporates have remained wary. But, says Maniatopoulos, corporate client engagement has shown a steady increase over the last few years.
“Corporate clients will commonly require FX transactions as a result of capital expenditure projects, merger and acquisition activity or trade financing. These tend to be less frequent in nature, larger in size and less time-sensitive, ideally positioning them to leverage the benefits of algorithmic execution,” says Maniatopoulos. “Transparency and best execution have contributed towards the engagement of corporate treasurers and are seen as important future themes. Moreover though, the continued availability of data to help in the evaluation of those larger, one-off transactions will be essential in further FX algo adoption.”
More than a decade ago, a select few clients were investing quantitative and technology resources to reap the benefits of algos, at a time when algo execution was reserved for the ‘innovators’, says Maniatopoulos. “Over the subsequent years, the barriers to algo adoption have been steadily diminishing, with a broad spectrum of clients leveraging the benefits of algorithmic trading. Looking at the most recent past, this has resulted into a higher level of algo usage with established users reserving larger shares of their portfolios for FX algo execution. Newer users will come from what is termed the ‘late majority’ in a growth cycle, clients who are now recognising the added value of execution algorithms and their broader industry adoption. The tipping point is in sight,” he says.
Asif Razaq, Global Head of FX Algo Execution at BNP Paribas, has been developing execution algorithms for more than a decade, originally for the bank’s own traders. However in recent years, his focus has been on building client-facing algos.
The very first algos were simple and prescriptive based on a single execution goal – TWAP, VWAP and so on. While these worked very well in the equities market, where centralised exchanges capture all manner of market data, there has never been the same level of data available in the FX market. This made the transition from voice to algos much harder in the early days of electronic trading because there was not enough pre-trade data to decide which algo to use, nor was there the post-trade data to measure the effectiveness of the algo.
But as the FX market became more electronic, more data became available and this has helped algos gain wider adoption among sell-side FX market participants. “In those initial days, the algos were developed for the bank’s own traders, who were all plugged into the market and its various data feeds,” says Razaq. “The algos were still prescriptive and traders would use their discretion in terms of the algo they chose to use.”
Two developments have acted as drivers to create more suitable algorithms for the buy-side and to increase adoption, says Razaq. The first of these is the evolution of the algos and the emergence of what Razaq calls third generation adaptive algorithms.
“One of our principles when we started developing algos for clients was to keep it simple. We started with just two algos –chameleon and viper. You would choose one and let the algo do the work. Since then we have added another, iguana, but we have also added the ability for the algos to be more adaptive.”
“We have used basic AI techniques to replicate human trading traits and to react to real-time data and changes in the market. The algos can interpret that data and make a decision on whether to be more active or more passive,” he says. “We developed 10 sub strategies or sub algos that live within chameleon, all with their own strategy but all able to react to real time data and to change direction.”
Razaq comes from a quantitative background, having studied AI at university and he says that traders were initially dubious about using algorithms to execute within the market in those early days. “However, they soon realised that algos were essential to their toolkit as the market has become so electronic that they cannot manage without them.”
“What has started to emerge now is the next stage of development – fourth generation interactive algos. They enable the trader to be more involved with the execution– the ‘chameleon’ strategy can be changed to a ‘viper’. You can slow it down or speed it up, put conditions into the settings or overlay your own strategy. We have also added a number of controls – for example, a ‘get me out’ button to enable traders to take a risk price and get out of a trade. We have provided traders with a whole toolkit rather than just simply an algo.”
Third party TCA
The other major driver has been the improvement in third party transaction cost analysis (TCA), says Razaq. “When we first started seven years ago, we wanted to be fully transparent and to provide a report on algo performance. It was a welcome gesture but there was a lack of independence and some people saw it as akin to marking your own homework. But then we saw the emergence of third party TCA providers.”
Now though, clients also want to see what’s under the hood. This presents algo providers with a conundrum. “We don’t want to give our intellectual property (IP) away but we also want to give clients an idea of how the algo makes decisions. This is where ‘ALiX’ comes in.”
ALiX is a digital trading assistant developed by BNP Paribas, which Razaq calls the digital assistant for financial markets. “Using AI and NLP, we have given the algo a interactive personality,” he says. “For example, it may tell you about headwinds in the market, volatility or the publication of non-farm payroll. From a management perspective, it automates a simple thing while the trader or the bank can focus on the client relationship. It is trying to build that balance. It also enables more transparency. Clients can ask ALiX to provide more information on the algo and why it hasn’t traded in the last few minutes.”
Algos were always used for mid to large orders. If you think about how they were executed in the past, it was always by voice or by risk transfer pricing.
The biggest increase in buy-side adoption has come from the ‘real money’ participants like pension funds and fund managers, says Razaq. But more corporates are becoming algo users and there is wider adoption among the buy-side. “We are seeing some systematic funds and traders also thinking of outsourcing their algos to banks – they are using these external algos to benchmark their own and to assess whether it is still worth developing their own algos. The systematic firms typically trade against the banks and want to see if the algo can outperform their own.”
While there are more advanced and interactive algos being developed, the assets to which they are applied within the FX world have remained somewhat vanilla – focused on the spot world and the major currencies. “You can only plug an algo into a market that is electronically mature,” says Razaq.
“The minute you go into the non-electronic world, it is much harder to use an algo. We are constantly monitoring markets and we have just launched algorithms for five NDF currencies. These kind of markets are a lot more fragile. Our roadmap is vast and the platform is constantly evolving.”
Regulation though has had a limited effect so far on algo adoption in the FX world, says Razaq. “MiFID II exempted FX spot so it was business as usual. That said, FX algos are a perfect fit for the reporting requirements of MiFID II and other regulations.”
Although algo adoption rates are increasing among buy-side firms, there is of course a limit to how many more participants are left to adopt. Have we reached peak algo or is there room for further adoption and innovation?
“It is a constant evolution,” says Razaq. “We are very far from the end and there is a lot more to achieve in this space. Where next? Look at how AI is used in other industries. We are learning from them. What do you see in your daily life? How e-commerce sites are constantly recommending products based on your purchasing patterns? Could we develop the same kind of service for FX algos? If a client executes in a certain manner, could we automate that via an algo?”
Taking back control
Whilst clients will still have preferred relationships, the cost reductions and control allowed by algos mean that clients are all becoming algo users, says Tom Appleton, executive director, FX Algorithmic Execution at French bank Crédit Agricole CIB. “Algorithmic trading is the best way for clients to control aspects of their execution that are important for them. Often, we will do this in consultancy with the clients and suggest changes to the algos and liquidity options.”
With most banks offering algos now, it is no longer true to say that it is only the large buy-side firms that can access effective algos, says Appleton. “Smaller regional banks are offering quality algos to their clients. And whilst it’s true that larger tickets are where the real cost savings can be achieved, there is still scope for improved prices in the smaller tickets or less liquid pairs. Also, some clients just like entering algos.”
A greater effort to introduce transparency has also helped increase adoption among the buy-side says Appleton. “Full transparency has been a key part of the service for many years. Providers who do not have connectivity to TCA providers and do not have transparency on the execution are at a clear disadvantage.”
Transparency is critical for the larger buy-side clients that are subject to best execution requirements and need to demonstrate this to their investors, says Appleton. However the impact of MiFID II and other similar regulations on buy-side adoption of algos is not the whole story says Appleton. “It is still true to say that traditional advantages of algos such as cost savings, execution anonymity, and time savings for execution desks are also driving uptake.”
Banks are also investing more in liquidity analysis tools, says Appleton. “It means clients can get a view real time of what is happening in the market. This enables them to make more informed decisions about which algo, and what aggression etc. This, in turn, gives confidence to the client that if their execution is not going so well they can pivot and make an adjustment. Further investment is being made in artificial intelligence (AI), recommender systems and speech recognition.”
Banks developing algos will be hoping that technology investment and the production of so-called next generation algorithms will also encourage greater adoption. Appleton says that there is a lot of talk among providers about the need to find the next ‘killer app’ using AI or machine learning (ML) in order to drive more algo flow through the FX market.
The problem though, says Appleton, is that the availability of useful data needed to run cutting edge FX trading techniques is still insufficient for building some FX algos. “For example, take illiquid pairs. Some of the tooling around speech recognition is interesting and driving users to try the algos of particular providers: if the whole experience is good then that is what the client remembers.”
In addition, algo developers are also looking to expand the asset class coverage within the FX market in the hope that this will also encourage greater adoption. “First mover providers have an advantage, both in terms of credibility with clients, and in terms of being able to refine the algos,” he says. “This may produce some further growth in client adoption, but one of the major drivers from a bank perspective is to provide additional service to existing clients or to provide a service to new clients who are using algos elsewhere.”
Adoption may ebb and flow between different buy-side participants but overall it is likely to continue at the current rates for now, says Appleton. Usage probably continues at current adoption rates. “There is a maturity in the financial institutions space that may lead to a slow down growth, however the corporate sector will pick up some of this slack, particularly in the mergers and acquisitions space,” he says.