Covid-19 crisis: FX algos prove their worth in volatile markets

June 2023 in Industry Views

How well did FX algos perform in these choppy markets and are client perceptions of algo use now changed forever? Nicola Tavendale investigates.

When first and even second generation FX algos were introduced to the market, they were designed to follow a very rigid set of rules which meant they did not perform that well in changing market conditions, especially volatile conditions, says Asif Razaq, Global Head of FX Algo Execution at BNP Paribas. But Razaq adds from inception, BNP has worked hard to engineer its third generation adaptive algos, which are designed to read markets in real time, understand the changes in the market and then formulate a dynamic strategy accordingly. “Adaptive algos actually have performed really well in these recent difficult market,” he adds. “Our expectation was that clients were likely to avoid using  algos during this volatile period, but we instead saw our volumes more than double during in the month of March. We have observed record volumes during the month of March.”

While the surge in the use of FX algos was quite surprising, Razaq believes there are a number of factors that led to that increase. Firstly, the comfort level of BNP’s algo clients, many of whom have now been using algos for some time now and so have a very sophisticated understanding of how they work, how to look at the performance metrics and are fully aware of the saving that using algos can add to their execution chain. Secondly, because liquidity was an issue the spreads of the interbank market started to be two or three times what they normally are. According to Razaq, this then results in a bigger increase in the prices that clients are getting from their banks because the banks are always going to price in the risk of trading in volatile and illiquid markets.

Industry Views

“So the typical spread 30 million dollar ticket had increased by a factor of 2 or 3. This made the cost of execution expensive when  clients were looking to trade. So they look at alternatives,” he explains. “When clients turned to utilising  our algorithms, they found that the algos were consistently outperforming this risk transfer price because they were able to source liquidity in the market without having to pay the spread. In fact they were able to capture spread in the market that gave clients a considerable cost saving against the risk transfer price, which is the other key reason why we saw volumes just explode in March.”

Tools for changing times

Ian Daniels, Head of e-FX Distribution EMEA at Nomura, agrees, adding that at Nomura they saw that algos have performed well in general but that it is the dynamic algos which have been the clear winners. “In challenging market conditions, trying to stick to static, historical trading profiles is a bad idea,” he says. “Instead, you need an algo that can dynamically react to what happens in the market. It should be able to speed up when opportunities arise or back off when liquidity evaporates, rather than chasing the market away.”  Volatile market conditions can also create liquidity issues, which may have serious implications for the safe and effective operations of FX algos. Although liquidity conditions have been tough, Nomura believes diversified pools and algos which respect deteriorating conditions can help, explains Daniels. He adds: “We have therefore not seen any problems with execution because of liquidity and have been pleased that the continuous tweaks and enhancements we make have paid dividends – even in less liquid times for G10 such as in the Asia morning session, the algos have been optimised to still seek out liquidity and execute efficiently.”

“In challenging market conditions, trying to stick to static, historical trading profiles is a bad idea. Instead, you need an algo that can dynamically react to what happens in the market.”

Ian Daniels

Among the bank’s client base, they also appear to have clearly preferred the passive pegging and liquidity seeking algos during this volatile period, according to Daniels. He says: “It seems clear clients found that capturing spread in these conditions and using algos that can intelligently seek out pockets of liquidity was the best option for them. Clients had already become accustomed to deploying these algos and trust in their results.” Similarly, Mary Leung, Global Head of Client Algos at State Street, says that it has long been discussed whether the increased use of algos in recent years was partly driven by low volatility and whether clients would revert to use risk transfer for execution certainty in times of volatility remained to be seen. Yet the backdrop of the Covid-19 crisis has since provided some very good evidence that algo usage is solid during volatile conditions, including at State Street where the bank’s algo volumes also rose significantly in March.

Industry Views
Source: Credit Suisse AES FX. Date ranges: Pre-COVID-19 (Jan 2020 – Feb 2020), COVID-19 (Mar 2020), Post COVID-19 (Apr 2020)

“In speaking with clients, they have told us that the combination of increased market volatility, wider market spreads, reduced liquidity and the unchartered territory of trading from home have caused them to use algos more to increase efficiency of execution,” Leung says. “Some clients have increased their percentage of algo execution volume five-fold and that the experience has been so positive that going forward they will maintain this level of algo usage. In March, we also saw that our clients increased their usage of liquidity seeking algos for faster execution, as well as their use of hybrid passive algos for saving spreads as compared to schedule algos.”

Understanding the trends

After having referenced both State Street’s own internal TCA as well as bestX TCA, the bank’s FX algo executions in March have outperformed the risk transfer benchmark by a few times compared to that of pre-crisis months, accompanied by the same magnitude of increase in the standard deviation, Leung notes. “In addition, the slippage to arrival mid has also increased by a few times as expected by the volatility in the market. Yet both benchmark metrics have a better Sharpe, which shows the efficiency of algo execution during the volatile period,” she says. “In April, we achieved further improvement in Sharpe measures over March as we maintained similar outperformance against the risk transfer benchmark and improved slippage to arrival mid, as the standard deviation is much reduced.”

“The increase in forward/swap spreads during the crisis was much wider than spot and still has not returned to pre-crisis levels. As FX algo usage continues to increase, this will also be an important factor in choosing an algo provider.”

Mary Leung

During the COVID-19 peak volatility period in March, Evangelos Maniatopoulos, Global Head of Advanced Execution Services (AES®) FX Product and Trading at Credit Suisse says that algorithmic trading volumes in AES FX had also more than doubled versus the immediately preceding months. He believes the rise was fuelled both by the overall increased FX activity as well as a visibly increased appetite for algorithmic execution, adding that FX algos have been well positioned to help traders navigate these turbulent market conditions. According to Maniatopoulos fast-moving markets, coupled with the desire to work orders around price levels and the overall challenging liquidity environment, have all contributed to a notable increase in both opportunistic and hybrid strategies as clients became more liquidity seeking. “This trend was particularly evident in EM trading, where spread widening and liquidity squeeze have been most prominent. Within the G10 space (in addition to opportunistic and hybrid algorithms), clients have continued to favour the use of passive as well as benchmark-tracking algorithms, both of which have yielded consistent outperformance vs. risk price on average,” he adds.

Industry Views

Clients are looking for feedback and colour on what they should be doing in volatile markets

However, Maniatopoulos says it is also important to highlight that the overall surge in use of AES FX algorithms is not isolated to this latest market crisis. “Looking at previous crises, it becomes evident that the currencies with the strongest direct impact from the event are the ones where the liquidity seeking activity is concentrated the most,” he explains. In addition, Maniatopoulos believes that an interesting observation in algo usage has also been a notable increase in the usage of aggressive algorithms for the execution of small sized orders by FX market professionals. He continues: “Though it may sound unexpected given pre-crisis trends, we have seen clients executing orders in the sub $5 million range through algorithms. It is possible that this may stem from an increased need for control of the execution, the requirement to target specific price points in such a dynamic environment and the benefit offered through liquidity aggregation. To us it proves that there is no transaction too small for an FX algo, as the available liquidity and customisation options can contribute to the bottom line.”

Safeguarding the markets

Razaq also highlights an interesting phenomenon that developed on the back of this surge in FX algo use during March. One of the key features of BNP’s algo platform is a service called BNP Internal Exchange (BIX), which allows the bank to match off two clients if they’re trading in the opposite direction to each other at market mid, without ever having to go and touch the external market. “For me that’s the utopia trade, because clients get an instant block fill at the current market mid, which in volatile markets is a really attractive trade to win,” says Razaq. “Before we launched BIX we found that orders would take  20 to 30% longer to execute. However we found that the increased volume of flow during March, led to a 50% increase in  BIX matches, all because there were more clients trading in the marketplace in opposite directions that allowed us to internalise the flow. Effectively, through the algo market itself, we had built a new liquidity pool which never existed before.”

“It’s essential to let clients know that even in these difficult times they can still access our adaptive algo strategies on various electronic trading platforms, but that they also have access to our dedicated algo traders to help them through the execution process.”

Asif Razaq

As a result of this surge in the use of internal matching , BNP’s clients were very impressed with the quality of execution they can get from using the bank’s algo strategies, reports Razaq. Yet even with the greatly increased number of users and the stressed market conditions, BNP reported the smooth operation of their algos and their risk management measures, with no need to deploy circuit breakers or interrupt trading on the platform. Razaq explains that during volatile periods, the bank remains very proactive in assessing the performance of the FX algo strategies and even in normal business conditions, continually tweak the internal parameters of the algos to make sure they are efficiently working in current market conditions. Then in terms of circuit breakers, that are designed to cater for flash events, where there is a significant move in the market in one particular direction, rather than very choppy and volatile markets, Razaq says. “So the circuit breakers didn’t trigger because there wasn’t a really big, sudden move in the market,” he adds. “This again allowed our clients to be continually plugged into the market rather than be switched off from it. For everything else the adaptive nature of the algos are able to self-adjust to fit the liquidity conditions, even in extremely choppy markets.”

At Nomura, despite execution systems obviously also having been pushed hard there were also no circuit breakers triggered during that period of market volatility, adds Daniels. He explains: “There is, or was, a school of thought that algos would generally not perform well in periods of heightened volatility. However, with spreads widening across the board, we have seen some great spread capture from passive algos, such as Ninja, which have taken advantage of the liquidity sources available to us.” He believes it is also vital to understand that FX algos are only as good as the data you feed them with. Newer technologies such as machine learning and AI have definitely played a part in rationalising this data, according to Daniels, while low latency, stability of the connection and quality of the data in times of stress have all been of paramount importance when it comes to algos that have to analyse the market to hunt for opportunities.

Industry Views

&Algos need to continue to adapt quickly to changing market conditions

No turning back

What then are the key lessons why FX algo providers can take away from the crisis which can be applied to the development of execution algos going forward? According to Daniels, ensuring that algos can move away from static execution paths and having algos use additional signals from the market will be key for best execution in the future. “Algos need to continue to adapt quickly to changing market conditions,” he says. “If they do this, then I think we can use this ‘event’ as a lesson and an educational tool to show the value of algo execution to a wider client base. It feels like the ‘new normal’ will continue for some time, which I think will actually aid further adoption globally and especially across the Asian client ecosystem where adoption is lighter.”

Leung agrees, adding that State Street strongly believes that consistent, competitive pricing and quality service are especially key in volatile times and this is no exception in the algo execution space. “We have always opted for ‘less is more, quality over quantity’ liquidity from the beginning by connecting directly to liquidity providers and a smaller number of tried and tested high quality ECNs, where we have deep analysis of their liquidity and technical capabilities, rather than to a large number of ECNs,” she says. “While other algo providers might have to turn off venues due to unreliable pricing during this period, our liquidity providers and ECNs have remained strong and consistent.” Apart from external liquidity, she notes that State Street’s FX algos are also connected to our eFX market making desk for internal liquidity. As a result, the algos have increased internalisation fills through the bank’s e-FX franchise, as well as through its Interest Match feature, which provides natural liquidity matching opportunities from the internal franchise to algo clients – and between algo clients.

However, Leung says that the increase in forward/swap spreads during the crisis was also much wider than spot and still has not returned to pre-crisis levels. “As FX algo usage continues to increase, this will also be an important factor in choosing an algo provider,” she predicts. “At State Street, our substantial core franchise client base of large institutional Investment managers, our credit rating and commitment of balance sheet to our clients provides us with the ability to price clients consistently in the forward space, something our clients expect and measure as part of their TCA.”

Learn from the data

Razaq believes in turn that BNP is already ahead of the curve in terms of FX algo development on this and already offers a number of highly innovative toolsets unique to the bank, such as Insight Live – which enables clients to see the market through the algo’s eyes and adjust parameters mid-flight if needed – and Alix, a virtual trading assistant which provides clients with live commentary on the execution while the algo is running. On the back of the algo experience during the crisis, he argues that many more providers will now also have realised they need to start providing some form of live analytics to their clients while they’re executing as well. “One of the lessons learnt should be that clients are looking for feedback and colour on what they should be doing in volatile markets and whether FX algos should be used at all,” he says. “We’re obviously in unprecedented times. People are trading from home but they can’t get hold of their regular sales person with whom they normally trade with. When clients are in these difficult situations, where do they turn to? It’s essential to let clients know that even in these difficult times they can still access our adaptive algo strategies on  various electronic trading platforms, but that they also have access to our dedicated algo traders to help them through the execution process.”

In turn, one of the key areas for Credit Suisse’s FX algo business has always been the continued focus on optimising both its market data consumption infrastructure as well as the liquidity venue execution handlers. “Both these components form the backbone on which a successful algo trading engine is built and have positioned us well for the challenging market conditions, allowing us to manage the increasing speed of market data updates and to achieve positive execution performance across our algorithmic trading platform,” Maniatopoulos says. “Then from a liquidity point of view, complementing our network of external connections with Credit Suisse’s internal liquidity also allows clients to elect hybrid liquidity access and benefit from the best of both worlds.”

Evangelos Maniatopoulos

“Though it may sound unexpected given pre-crisis trends, we have seen clients executing orders in the sub $5 million range through algorithms.”

Evangelos Maniatopoulos

He adds that in normal conditions with significantly tighter spreads, this aggregated liquidity is what gives the bank an edge over its peers, but in the conditions seen in March and previous crises, the additional liquidity in the bank’s ecosystem proved substantially more impactful. “Looking forward, as we are integrating more machine learning areas of research in our business, we are also conscious that more complex models and assumptions are required to make sense of an ever-evolving financial markets environment,” Maniatopoulos concludes. “Each crisis is personal for us as traders, exposed to the news and fundamentals driving price moves. However, underlying patterns in market conditions and algo behaviours do repeat themselves and we want to look into what the past few months can teach us about our performance in a high volatility environment through the lens of our data.”