According to the BIS report, while the overarching goal of Execution Algorithms (EAs) is to achieve optimal execution, there are several competing dimensions to this goal which will be defined differently for different users depending on their individual objective functions. It notes that although various types of EAs can accommodate several preferences with respect to execution, no single EA can optimise all aspects simultaneously. Instead, the report argues for the need to ‘strike a balance’ between minimising market impact, minimising exposure to market risk and maximising the certainty of completing a trade: the “execution trilemma”.
The main trade-off, however, is identified as being between market impact and market risk. But in practice, what impact should this consideration have on a trading firm’s choice of execution algo?
“The choice of algo execution should be more easily answered by understanding the underlying need of the trade in question.”
“There are different strategies for different trading needs, market conditions, or even trading styles, so the choice of algo execution should be more easily answered by understanding the underlying need of the trade in question. The first and most important question for a buyside firm to answer is ‘what do I want out of this algo?’” says Kate Massey, eFX Distribution, EMEA, at Nomura. For example, Massey explains that one of Nomura’s strategies, Ninja, is a purely passive algo, meaning it never crosses a spread and is ideal for capturing the best possible levels in the market. In turn, Ninja helps to reduce market impact and is great for larger trades according to Massey as it slices up orders, distributing across a number of venues for a certain time period. This means a trader may need to be a bit more patient to complete the execution effectively and therefore accepts greater market risk, she adds. “Conversely, another strategy of ours, Shogun, is the opposite. It is fast and more aggressive with the purpose of getting it done quickly. As a consequence, Shogun would have a greater market Impact over Ninja, but potentially less market risk,” Massey says. “Depending on the circumstance and the need, the client would choose one over the other and all execution strategies should be made on a case by case basis.”
Impact on algo choice
Andrew Cole, FX Algo Product Manager at J.P. Morgan adds that the longer the algo duration, the lower the percentage of volume and therefore (on average) lower the market impact. At the same time, however, a longer duration increases the risk of a market move during the execution and therefore heightens market risk. “This trade-off is important,” Cole says. “If a buyside firm believes a market move is imminent, additional market impact is perhaps acceptable to close a position over less time and thus avoid trading through the price change. Perhaps it should impact a client’s choice of execution algo more than it does currently i.e be more important. Yet the majority of our largest algo clients traded over 75% of their volume via a single order type last year.”
“The majority of our largest algo clients traded over 75% of their volume via a single order type last year.”
Even so, the BIS report highlights the additional impact of ‘user choice’ ie user-specified settings on the level of aggression or urgency for an execution. According to Cole, this perhaps helps to explain why a single strategy can be enough for some clients; the flexibility offered by user configuration within a strategy can cover a lot of different market conditions. He continues: “J.P. Morgan doesn’t just offer an algorithm that tracks POV, for example, it allows clients to select from different levels of “urgency” which correspond to pre-configured levels for higher/lower POV. For those clients that don’t feel the POV levels programmed by J.P. Morgan adequately suit their needs, they can even customize these preset POVs to levels they prefer.” Similarly, he explains that the bank’s spread tracking algorithms simplify the selection for clients by creating multiple “speed” options which effectively target different levels within the spread, different speeds with which they react to activity in the market or multiple groupings of liquidity that they can utilize. “All our algos work to some extent in this way, while some strategies are traditionally considered more market impactful they can in each case be customized to be more or less so,” Cole adds.
Ultimately for each execution, buyside traders need to weigh up the cost of immediate execution, which typically incurs market impact, with the cost of market risk, usually from a more passive style of execution, argues Ralf Donner, Global Head of Client FX Algo Execution at Goldman Sachs. “Many clients have a conscious or subconscious utility function for market risk, which may be the result of a PM’s directive or experience trading that market,” he says. “It is an important consideration because the performance versus a chosen benchmark will depend on the choices made.” In addition, the BIS identifies a third source of risk as incomplete execution, he adds, explaining that this is mostly relevant for opportunistic trades such as take-profit limit orders or stops.
The execution algorithm trilemma
Donner notes that the modification of user-specified settings is therefore useful as it allows traders to tweak the algo’s behaviour intra-trade and adjust the balance between market impact and risk to suit market conditions. “For example, if the target had been a certain amount of market risk, but the currency pair is trending heavily against the execution, then a re-assessment of the total expected cost of execution and a switch to a more aggressive execution style may be warranted. User choice may also permit a client to adjust the types of venues being accessed to affect this optimization problem, e.g. favouring venues with softer mark outs such as internal matching engines,” he says.
“Many clients have a conscious or subconscious utility function for market risk, which may be the result of a PM’s directive or experience trading that market.”
Meanwhile, John Turney, Head of Global Foreign Exchange at Northern Trust Capital Markets, adds that overall the use of execution algorithms allows for buy side FX traders to have flexibility in execution strategy. “At its core, this could mean executing immediately, via risk transfer or an aggressive algorithm,” he says. “Central to both of these execution decisions is a firm understanding of the alpha profile of the specific order at the time of execution.” In addition, as the universe of execution algos proliferates, providers are increasingly giving traders more flexibility in execution, Turney explains. “If the first generation EAs were simplistic in terms of style, primarily focused on lengthening the execution window to reduce market impact or allowing for buy side traders to attempt to capture spread, the evolving EA offering continues to give more control to buy side traders in how they execute,” he says. “It can be a frustrating experience to see prices in the market you would be happy executing at, only to see the algorithm is not trading fast enough. At the end of the day, the algo is a tool for the trader, so providing traders with greater flexibility to an algo’s optimal pace and strategy should lead to greater algo adoption over time.”
Another important trade-off is between trading costs and opportunity costs, according to the report. Yet the reason for trading, for choosing to use an algorithm, for deciding to prioritize market risk over market impact – these are all decisions traders make on every order, argues Turney. “You can extend this to the trade-off between trading costs and opportunity costs. Specific to Northern Trust, many of our FX algo customers are executing foreign exchange off the back of security-related activity,” he says. “In this sense, they have a need to complete each order and prioritize the execution certainty that strategies such as TWAP provide.” According to Turney, the style of algorithms which focus on certainty of execution integrate seamlessly with CompleteFX, Northern Trust’s rules based, systematic approach to outsourced FX execution. “CompleteFX is built to provide customers with transparent execution, focused on minimizing market impact while reducing operational risks in an automated and highly customizable fashion,” he adds.
Taking a more nuanced approach
In addition, Massey explains that there are a multitude of algo providers and strategies on the street, so it can be a bit daunting for a user to know which one to choose and how exactly it works. “At Nomura, we have always aimed to keep the strategy choice and user settings simple without comprising on performance,” she says. “Broadly speaking, we find most users will tend to stick to the standard settings to begin with, however often some will look to amend the settings mid-flight.”
“Navigating FX liquidity also requires a dynamic approach to execution.”
Ian Daniels, Head of e-FX Distribution EMEA at Nomura, agrees, adding that because performance is largely driven by market impact and liquidity access, hybrid algos can also be useful because by their nature they will look for changes in market events and adapt themselves and their parameters accordingly. For example, hybrid algos can see what the current trading period looks like when compared to an historic time period and work out how to interact with liquidity and with what liquidity to interact. “This is where the quality and diversity of your provider’s liquidity access comes into play. Does the algo need to slow down or speed up, should we be utilising the primary market where liquidity could be deeper, or go deeper into the liquidity pools to lessen market impact at that given moment?” he says.
FX algo providers can also support trading firms to understand and address the various trade-offs highlighted by the BIS report in several other ways, Daniels adds. He believes that detailed user guides and simple to follow cheat sheets can always prove useful, as will walking clients through their initial usage while still ensuring there is a human face there should they have any concerns. “This can be enhanced through active monitoring of algos, something that we feel is key and has become increasingly important during the pandemic,” Daniels says. “This goes back to simple principles about knowing your client, having conversations about why they would use an execution algo, what are they trying to achieve, what are their benchmarks, etc. Once you understand this, you can then suggest a suitable tool from the suite of algos in your kit.” According to Daniels, the BIS report also highlighted a wide range of clients that currently utilise algos and their drivers are different. For example, he notes that while Central Banks’ main motivations are to reduce trading costs, improve productivity and access multiple liquidity pools in order to reduce market impact, small to mid-size regional banks tend to mainly target swift execution instead.
Furthermore, Coles notes that the BIS report has created the notion of a trilemma to outline the starkest decisions a firm must take when selecting their algos. However, the paper also acknowledges that user preferences don’t typically sit in the corners; clients want to reduce market risk, market impact and opportunity cost, he explains. In addition, algos that reduce market risk don’t necessarily have to create market impact; providers with high internalization ratios, such as J.P. Morgan, might allow firms to exit large quantities of risk in shorter time periods without generating significant market impact, Cole says. “Trading decisions are not without trade off, but decision making doesn’t have to be in the corner of the triangle,” he adds. “Firms should focus their algo usage on providers like us who can offer nuanced ways of operating within the trilemma, as opposed to at its outer corners.”
Help from algo providers
J.P. Morgan has also been offering a version of hybrid algorithm for many years, Cole says. For example, on the bank’s single dealer platform clients can select ‘One Cancels Other’ (OCO) strategies that combine two order types running in parallel. He adds that J.P. Morgan’s new Adaptive order is also a perfect example of a hybrid and dynamic execution within a single order. “Algo providers have to accept that client choice extends beyond what is neat or easy to provide. While additional strategies and greater flexibility can intimidate clients that are new to algorithmic trading, at the more sophisticated end of the spectrum these features are necessary to capture the exact behavior a client with extensive experience may want,” Cole explains. However, he warns that giving clients appropriate transparency around how the algo they’ve selected will act within the market context is just as important. “Tools like JPMorgan’s algo central gives the client as much visibility as possible over expected risk, cost and duration with which to weigh up the various tradeoffs cited within the BIS report,” Cole says.
“Active monitoring of algos, is key and has become increasingly important during the pandemic.”
A key source of opportunity cost for algos also arises from the use of a no worse than limit price to control an algo’s execution, argues Donner. He believes that some firms deliberately avoid making any limit-price interventions in algos to avoid potential opportunity cost, while others prefer to intervene frequently. “The jury is out on which approach is better in the long-run, although of course there are individual examples of algos that have been greatly helped or hindered by pausing execution,” he says. However, the battleground for the most popular algos has always been the hybrid algo, according to Donner. “Hybrid algos have the advantage that they never get into a situation of being completely passive in a market without any demand for liquidity or sweeping an orderbook that is not being replenished,” he explains. For example, at least one component of the hybrid algo will likely continue to function under changing market conditions and the algo can attempt to respond to changing market conditions in a way that the pegged or sweep algo may not be able to do, he says. “Overall, we believe that analytics continue to hold the key to getting the right choice of algo and algo parameters and to allow an intra-trade re-assessment of the various risk factors. We continue to invest heavily in our analytics offering, and we are currently working on a cross-asset class approach to TCA,” Donner adds.
“Navigating FX liquidity also requires a dynamic approach to execution,” warns Turney. “While the pillars of decision-making remain minimizing market impact, maximizing execution certainty as well as minimizing market risks, the optimal result for many market practitioners may lie somewhere at the intersection of these three competing values.” He believes that EA providers must continually invest and innovate to gain market share, and mind share, as traders evolve their algo expectations.
“At Northern Trust, we have addressed this with our latest algo offering, a strategy called CompTrap, which, at its heart, gives users greater flexibility in how the algo will opportunistically deviate from a TWAP baseline, defining the time, percentage of completion, or price levels which will flip the algo from a passive to a more aggressive state in an effort to outperform benchmarks such as TWAP or arrival price,” Turney says.
At the same time, because there are so many diverse participants in the FX market, it can be hard to isolate what is affecting price action. “One of the tenets of algo execution is that over time, you should be able to realize fewer trading costs by optimizing execution and reducing market impact. You would not want to draw too many conclusions from a single order or transaction,” Turney explains. “Here is where leading algo providers can still aid the buyside trader, by providing more data from both a pre-trade and post-trade perspective.” For example, pre-trade, large algo providers can use their execution history and view of current market conditions to better estimate execution costs for similar orders and how various execution strategies may perform, says Turney. “However, this may be harder for buyside firms given they do not have the same universe of algorithmic execution history to fall back on,” he concludes. “In terms of post-trade reporting, large algo providers can enhance the TCA and performance data they share with buy side traders, including things such as peer analysis which can help customers assess how their execution strategy is performing vs a group of similar peers.”