FX algo strategies: Meeting demand for smarter and faster trade execution By Chris Hall

Buy-side use of algorithms in the FX market might be in its infancy, but it’s growing up fast. Corporate and institutional clients that, five years ago, were happy to source quotes from their brokers – tapping voice-broked risk transfer for larger trades or using electronic platforms to automate smaller ‘nuisance’ transactions – are now looking to take greater control of their FX executions. Increasingly, this means broadening their use and expectations of algorithms. They are poring over execution data, identifying appropriate execution benchmarks and selecting the algorithms, parameters and venues that best suit their particular needs and objectives. Many are demanding new levels of responsiveness to changing market conditions and sensitivity to the liquidity dynamics of diverse venues. Almost all are increasing their appetite for data and willingly accepting an increase in market risk as the trade-off for reduced market impact.

FX algo strategies: Meeting demand for smarter and faster trade execution   By Chris Hall
Chris Hall

Allan Guild, global head of FX and Commodities execution services at HSBC Global Banking and Markets, sees the three key drivers of demand for smarter, faster and more transparent algorithms as the onward march of technology, regulation and time. 

First, algorithms are increasing accuracy and delivering better outcomes thanks to ongoing investment by sell-side institutions in measures to improve their performance, for example enhancing the use of feedback from past executions to inform future tactics on a more immediate basis. 

Second, regulatory concerns over systemic risk and investor protection across the financial markets have led to greater scrutiny of execution quality, with institutional investors now expected to demonstrate they are taking “all sufficient steps” to achieve best execution, under MiFID II. In this context, algorithms provide a more reliable, systematic and auditable mechanism for fulfilling execution policies than human traders. 

Allan Guild

Allan Guild

“Clients with structural, uni-directional flow might prefer to use a liquidity-seeking algorithm that can access liquidity quickly but passively,”

Third, many buy-side firms have quickly become familiar and comfortable with use of algos in the FX markets. This is partly a function of time, but institutions with extensive experience of algorithmic trading in listed instruments often leverage this expertise to demand more from their FX algos at an earlier stage in their usage. This is particularly the case in firms that have merged their trading desks and expect traders to master a range of execution channels across all asset classes.  
While institutional investors might be leading the way in this greater willingness to manage market risk via algos, not least due to their MiFID II obligations, providers note a broader swell in demand. “In the last nine months, we’ve seen a notable increase in algo usage by corporate clients. Corporate treasurers increasingly have a mandate to manage market risk via algorithms. Corporates are willing to accept that risk for a few hours rather than passing on to the sell-side immediately if they have the tools and the expertise to exert control and deliver performance,” says Alexander Barzykin, director of eFX risk at HSBC. 

Minimising market impact

At NatWest Markets, head of FX algo execution John Quayle adds a fourth factor into the mix of algo usage drivers. He asserts that the recent rapid rise in buy-side interest in algorithms is heavily influenced by the increase in trading costs experienced in recent years via established FX execution channels. “Both anecdotal and statistical evidence shows that market impact has increased across many FX pairs since 2015. When market makers are highly sensitive to even small changes in market conditions, it can become hard to get larger trades done without moving the market. Clients’ focus on limiting market impact has spurred growth in use of passive algorithms, such as NWM’s Peg Clipper” says Quayle. 

Top criteria in algo provider selection
Top criteria in algo provider selection

Having expanded steadily just after the financial crisis, FX trading volumes have largely plateaued in the past five years. FX volumes were driven 35% higher to US$5.3 trillion between 2010 and 2013 (spot volumes rose 38%), partly by the increased hedging needs of pension funds and multinational corporates, according to the Bank for International Settlements triennial survey series. In the next three-year period, volumes dipped as the FX market adjusted to the new post-crisis ‘new normal’. Between 2013 and 2016 the FX market fell 4% to US$5.1 trillion, with spot falling 19%, the first decline since 2001. This period has been accompanied by a shift in liquidity provision from the traditional sell-side toward electronic liquidity providers that are highly attuned to shifts in market conditions. While the Swiss National Bank de-pegging of January 2015 still looms large for market makers that got burned, this sector is experiencing consolidation as overall volatility levels decline, meaning providers are highly sensitive, indeed “jittery” at times. In parallel, the number of FX liquidity venues continued to increase, with falling technology costs lowering the transaction fees needed to break even, meaning consolidation may be limited. 

Alexander Barzykin

Alexander Barzykin

“Corporate treasurers increasingly have a mandate to manage market risk via algorithms.”

Kasper Folke, head of e-FX and algo quant at Nordea Markets, agrees that the FX market has become increasingly difficult to navigate and expects the trend to continue. “Trading large amounts is  proving difficult due to fragmentation and overall sparsity of liquidity. The fragmentation makes it  increasingly costly for buy-side to maintain the connections required to reach a significant portion of the available market liquidity on their own. In this regard, FX algos offer a way to leverage the  infrastructure of banks both in terms of provision of liquidity and in terms of how to get the most out of available liquidity,” he says. 

Independent evidence 

Heightened fragmentation and reduced liquidity typically results in higher market impact and higher trading costs, which can best be managed down via the decisions made by a smart algorithm rather than a smart trader with good relationships across the street. And unlike five years ago, now we can prove it. 

John Quayle

John Quayle

“Clients’ focus on limiting market impact has spurred growth in use of passive algorithms,”

BNP Paribas global head of FX algorithmic execution Asif Razaq credits the growing presence of independent TCA (transaction cost analysis) offerings in the FX market as providing buy-siders with the confidence and the evidence to take on market risk through increasingly sophisticated use of algorithms. He suggests pressures have been mounting on the buy-side for a decade to improve cost-effectiveness of FX trading operations as costs rose and budgets tightened. But few had the governance framework, execution policies and expertise to manage market risk through algorithms and other ‘self-service’ execution tools. Further, third-party TCA services were not available to monitor and validate changes in execution performance, independently from the banks that supplied the execution tools.   

These limitations explain why buy-side use of algorithms in the FX market grew only gradually between 2010 and 2017, with firms largely adopting algorithms cautiously as a response to cost pressures, i.e. trading larger orders passively rather than pay increasingly steep risk transfer prices. More recently, however, MiFID II has lent regulatory impetus, both by demanding use of tools that support best execution and encouraging supply of the means with which to measure it. With a maturing and expanding range of independent TCA offerings now providing buy-side compliance teams with auditable proof of the contribution of algorithms to better trading outcomes at lower costs, the result has been an “explosion” in buy-side FX algo usage since January, according to BNP Paribas’ Razaq. 

Greater expectations 

In response to MiFID II’s catalyst, buy-side firms have not only used algorithms to automate FX trading workflows, thus reducing operating costs and reducing market impact, but are also stretching the limits of what they expect from algorithm. Razaq reports that controls and functionality offered by BNP Paribas’ algorithms are being explored with a new vigour. “Firms are taking their algorithms off autopilot and are demanding more feedback and functionality. They want to know if they are improving performance by intervening in how the algorithms operate. They want more control and more options, including the ability to switch between modes and strategies, potentially pausing if necessary, and selecting venues based on a clear understanding of the liquidity they wish to expose themselves to,” he says.

Asif Razaq

Asif Razaq

“Firms are taking their algorithms off autopilot and are demanding more feedback and functionality.”

Razaq notes a number of trends within this highly exploratory phase. Some institutional clients are sharing their smaller trades across the algorithmic execution services of multiple rival banks as part of their assessment of providers for larger transactions, potentially as a stepping stone to migrating whole portfolios to algorithmic execution. And whilst only some FX market participants are impacted by MiFID II’s best execution rules, the increased interest in FX algos is more widespread, with many hedge funds looking at workflow efficiency and automation support. 

Already conceived as a highly adaptive algo, the latest version of BNP Paribas’ Chameleon algorithm represents a transformative step, suggests Razaq, designed to minimise market impact. Chameleon 2.0 now offers the user greater control and intelligently routes orders to venues factoring in both price and perceived market impact. 

At HSBC, Guild sees a variety of execution preferences, even within client segments. There is still a lot of demand for TWAP, he notes, but increasingly corporate clients are willing to use liquidity-seeking and implementation shortfall (IS) algorithms. “A firm might use IS to accumulate a large position if they are comfortable taking the execution risk and monitoring the market impact themselves,  as an alternative to a risk transfer price,” he explains. “Clients with structural, uni-directional flow might prefer to use a liquidity-seeking algorithm that can access liquidity quickly but passively, for example by placing orders at different levels and venues, based on estimates of fill probability and market impact. This relies on the quality of the algorithm to process the constant feedback and immediate response to interaction with other orders.”

Drivers of decision to execute trade via algo
Source: Greenwich Associates 2017 Future of FX Algos Study

These preferences are witnessed by NatWest Markets’ Quayle too, who observes that algo users are keen to get smarter in their understanding of venue liquidity dynamics, before they get faster. “Clients are analysing data to gain a deeper understanding of how to interact with liquidity effectively in different market conditions and across different venues and thus develop a more targeted approach to execution. At the more sophisticated end of the spectrum, algorithms are not only responding to the specific liquidity characteristics of individual venues; they are adjusting their behaviour in real time,” he observes. 

Kasper Folke

Kasper Folke

“..FX algos offer a way to leverage the  infrastructure of banks both in terms of provision of liquidity and in terms of how to get the most out of available liquidity,”

As demand grows for immediate responsiveness to market conditions and understanding of liquidity dynamics across different venues, banks are further exploring the potential of artificial intelligence and machine-learning capabilities. “Adaptive algos utilise AI already. But now algorithms are able to learn as they trade, responding to current liquidity behaviour in individual venues. This might mean that the algo refuses to trade in a particular venue until its performance improves. Smart order routers that learn in real time can help the client navigate through multiple venues; equally, the client can self-select venues if they have a strong view on the execution quality,” says BNP Paribas’ Razaq. 

More degrees of separation

But the goal of minimising market impact has implications beyond the design of the latest generation of algorithms. Execution service providers are keen to differentiate to customers both in terms of how they evolve their traditional role of liquidity provision and how they provide guaranteed separation of order handling from other parts of the bank to minimise information leakage. 

“We have established the execution desk as a barrier between traditional sales and trading. The execution desk effectively forms a Chinese wall and ensures information does not move between sales and trading,” says Folke at Nordea. Indeed, clear and robust separation of functions now seems established best practice across leading algo providers. However, there are limits and nuances. HSBC for example, enforces strict separation of desks to protect client interests, but its algo development function pools expertise and resource from across its FX franchise. At BNP Paribas, Razaq points out the need to balance customer protection with service quality. 

“In our experience, segregation of execution services not only increases volumes and improves relationships but helps align with regulatory requirements. Nevertheless, it is important to structure your operations to maintain client access to hybrid liquidity,” he says. 

Olivier Werenne

Olivier Werenne

“Integrating AI into the behaviour of algorithms is key to future improvements in customer service.”

With electronic liquidity providers accounting for the majority of liquidity on many third-party multilateral trading platforms, banks play a key role in helping their clients access and understand the diverse range of liquidity options across the market via their algorithms and pre- and post-trade data services. But they are also providing access to internal liquidity, either that generated by trading desks across the bank or arising from the needs of other clients. Previously, dipping into the bank-owned algorithm provider’s proprietary liquidity pool prior to exploring third-party venues was sometimes seen as more advantageous to the provider than the customer. But the advent of robust third-party TCA has given banks the independent validation to support the routing of client orders to ‘hybrid’ liquidity. 

“We will route algorithmic orders through our liquidity pool only if this will minimise market impact, as proven by independent TCA. Hybrid liquidity is a core requirement in today’s market,” says Razaq. As well as routing to BNP Paribas’ proprietary liquidity pool alongside 15 other independent venues, Chameleon 2.0 also offers a ‘get me out’ function, which offers a principal price on an anonymous basis if the client would prefer to hand over market risk for the remainder of the deal, with the whole transaction settled on a single ticket. 

Other providers also see a key role for proprietary capabilities in a continually fragmented world. “I don’t expect to see a great deal of venue consolidation, but I think proprietary flow and mid-point matching platforms will grow in importance. There is scope for further innovation,” says NatWest Markets’ Quayle.  The segregation levels at NatWest Markets are taken even further in that client orders are handled by a separate trading desk in a secure room away from the bank’s risk takers. This level of segregation has enabled the banks team to embed a large degree of Algo execution for their clients even on the simplest of order types, whilst still harnessing the full range of external and internal liquidity sources.

Firing on all cylinders

For Olivier Werenne, head of eRisk liquidity management in EMEA at HSBC, the prevailing market environment requires banks to work on many fronts to deliver to increasingly demanding clients. “Conversations with more sophisticated clients are getting increasingly technical and advanced. They want more information on how to avoid information leakage, how parameters can be adapted to changes in liquidity conditions and how to select venues,” he observes. “Integrating AI into the behaviour of algorithms is key to future improvements in customer service. But banks must combine this technology innovation with a full range of liquidity provision options, including use of balance sheet.” 

And whilst third-party TCA might be integral to growing buy-side use of FX algos, major banks and brokers see their own pre- and post-trade data services as playing a major role in improving client outcomes. According to Barzykin at HSBC, “Clients use third-party TCA to get an independent view of best price, but they also require TCA from banks, for example to get a unique view on risk transfer pricing. Increasingly, they also expect pre-trade analytics to run scenarios prior to execution. For instance, our analytics integrate with a simulation framework running on production codebase, to give clients a realistic expectation of algo performance” 

Nordea’s Folke prefers not to offer “seemingly precise” predictions of projected algo performance due to the high levels of uncertainties in such predictions. “Instead, we aim to offer advice based on current and projected general liquidity conditions in the market. Coupled with the client’s particular requirements, our execution desk has in-house developed analysis tools to suggest the algo best suited for the client,” he says. 

Conclusion 

In its recent study of FX algo use on the buy-side, Greenwich Associates cited cost reduction and execution quality as prime drivers, pointing out that some large institutions were already deploying algos for 25-30% of volume. Furthermore, execution quality, consultancy, and access to liquidity were cited as the top criteria for algo selection. In the battle for market share the smart money is on the smart algo providers.