Adoption of execution algos is accelerating rapidly for FX market players. And though it may be true that a regulatory tailwind helps, there’s also widespread industry consensus that saving money is just like making money when it comes to the bottom line.
There are basically two options when trading foreign exchange: call up a bank and leave an order, or tap into electronic channels. The latter has come into its own thanks to advantages of an automatic audit-trail at a time when proving best execution is a top priority for regulators, and being increasingly demanded by investors.
Asif Razaq, Global Head of FX Automated Client Execution in BNP Paribas’ Global Markets Unit, said that being able to reference the post-trade reporting that comes along with electronic execution is one of the major drivers of the recent surge in demand for algos.
But it’s been a slow burner to some extent. When execution algorithms first came to the fold, there was slow and cautious adoption. Hedge funds were the first to dip their toes in the water with use on somewhat of an ad hoc basis. As algos became more popular, the real money asset managers, such as pension funds, for example, started to become interested in this alternative means of execution because of proven cost savings.
One recent analysis, for example, shows that for order sizes in excess of €50 million, algorithms outperform a risk transfer price some 90% of the time.
Use of Algorithmic Model Trading – Volume-weighted proportion executed using algo tools
Who’s Using Algos?
Richard Johnson “A lot of times buy side say they are doing TCA, but when you drill down into it, I think a lot of them are relying on TCA reports that their broker sends them or internal tools built on spreadsheets…” |
Research from Greenwich Associates shows that the sell side is embracing FX algos, in part because of lessons from the equities markets where electronic trading became very competitive and low margin.
“I think the sell side is looking at the FX space and increasing adoption as a good revenue opportunity for them to expand into,” said Richard Johnson, an analyst in the Market Structure and Technology Practice for Greenwich Associates, pointing to the influx of people with equities backgrounds moving to FX desks.
Another reason for sell side adoption, he added, is that electronic tools can be offered to clients, increasing efficiency and reducing costs of services.
But it’s in the hedge fund space where use has skyrocketed in the last few years.
That’s because this group is more likely to run dedicated FX strategies, which means the ability to trade at any time of day depending on the prices in the markets, while also not being tied to the execution of any other asset class.
By contrast, an institutional money manager is likely trading a global equity portfolio, wanting to match the execution rate to FX rates, which limits timing.
BNP’s Razaq has had a front row seat to the wave of FX algo adoption across client groups. Tier 2 and Tier 3 banks are the most infrequent users, and tend to deploy on big trade order sizes, he said.
However, adoption has been boosted by a major change in liquidity conditions on the back of significant events such as Switzerland’s central bank unpegging from the euro and Brexit. As liquidity conditions became more complex, pricing began to favour electronic execution and as a result hedge funds became more active users.
More recently, the corporate space has emerged as a big adopter of FX algos, though this group tends not to directly input orders. “As more clients come to the fold of using execution algorithms, we are seeing more and more use of what we call ‘on behalf of trading’ execution algos being deployed in the market,” said Razaq.
The trend is a symptom of the complexity faced by new clients, and one of the reasons there has been slow uptake. Choosing the right algo among dozens offered by a handful of banks becomes quickly overwhelming.
BNP Paribas has gone the way of a ‘keep things simple’ philosophy, said Razaq. The bank has deliberately kept the algo family small to simplify the selection process for clients. The bank’s main strategies – Chameleon and Viper – both have the capacity to move from passive to aggressive depending on market conditions, which are calculated using proprietary technology that taps into machine learning.
Chameleon is designed to select the best execution style for a client given the amount being traded and current market conditions. Viper is designed to trade aggressively into the market place while minimising exposure and impact to market risk.
“When the client wants to execute via an algorithm, the only decision they need to make to choose between one path or the other is how much market risk they want to take,” Razaq said.
Confusion marketplace
Asif Razaq “Imagine how cool it would be that while running an execution algorithm, it’s providing real-time information to the user on current market conditions.” |
Even if the choice and use of algorithms is becoming simpler, measuring their effectiveness is the next confusion marketplace being faced by adopters. The reasons are fairly well understood: over-reliance on data from brokers providing the algo itself, a lack of consistent data across the industry, liquidity based on bilateral credit-driven agreements, and an OTC reality that does not translate well to transparency.
Because of this, transaction cost analysis (TCA) has become one of the hottest topics in FX markets, and the same people who used to watch the inner workings of investment banks are setting up independent providers to measure performance.
It’s been a relatively different development than what was observed in the equities markets, where TCA came in first and algos followed. Whereas in the FX markets, TCA and algo adoption are running parallel, said Greenwich Associates’ Johnson.
“There is a very strong link between the two, we are seeing an increased use of TCA,” he said. “The data show 27% of buy side in the US and 34% in the UK.”
He noted that more needs to be done by TCA providers in terms of coming up with better analytics and benchmarks. “You can’t really tell what percent of the market you are, which is a key element when you are looking at what your market impact was. Without that data, it is harder to come up with good benchmarks.”
The buy side has some improvements to make as well, Johnson noted. “A lot of times buy side say they are doing TCA, but when you drill down into it, I think a lot of them are relying on TCA reports that their broker sends them or internal tools built on spreadsheets or some other product like that.”
Equities markets TCA adoption does provide a roadmap for what might be expected in the FX space: broker-supplied TCA reports were found to be too limited, buy side firms shied away from building in-house TCA products and the vendor community stepped in with advanced technology.
“I think we are going to see those three things play out in the FX TCA space,” he said.
The lack of comparative performance data is presently the biggest challenge in algo execution
Advanced TCA
Guy Hopkins, Head of MFX Vector Sales at MahiFX, started working on algos and TCA with clients at Morgan Stanley as part of his role as Head of FXEM eDistribution EMEA. He is now helping to build out a buy side product for the independent tech firm.
“Between us we have decades of collective experience managing e-FX businesses. As a technology firm we can provide trading tools and analytics to clients with all the knowledge, experience and expertise of having worked on the sell side,” Hopkins said. “We can give the clients on the buy side clarity around understanding their execution costs, which then helps inform their trading decisions.”
Cost reduction for asset managers has traditionally focused on ensuring that they are quoted the tightest possible spreads. In recent years however they have also become progressively more comfortable warehousing execution risk, rather than transferring it to their liquidity providers.
Banks have risen to this challenge by effectively repackaging their eFX risk management systems and licensing them to their clients in the form of algos.
“Asset owners now routinely expect asset managers to be able to demonstrate that they have a framework for keeping their costs in check. Initially the focus was understandably on confirming things like not being charged outside the highs and lows of the day. Now the conversation is more nuanced, focusing on how different trading strategies can be deployed to get costs down still further,” says Hopkins.
Banks, he added, will always have a fundamental intermediary role to play, particularly in less liquid currency pairs and products, for example, emerging markets, FX options and NDFs (non-deliverable forwards). But for the more liquid products, like G10 spot FX that is traded on electronic venues, there’s an emerging awareness that, at least some of the time, implementation can add yield to portfolios rather than incur a cost.
“Many clients are now thinking: ‘Perhaps I can get rewarded on the spread for the trade that I am doing rather than crossing spreads all day’,” states Hopkins.
Justifying algos
Guy Hopkins “The lack of transparency around all-in performance of algos, and the consequent focus on negotiating usage fees alone is something that definitely needs to be addressed” |
For any provider, an algo offering is a major technology investment and has to generate sufficient revenues to be self-sustaining. The lack of comparative performance data, however, is presently the biggest challenge in algo execution to determine what the total costs, and savings, really are.
“Clients ask: ‘who has the best algo?’ and answering that question is, at the moment, extremely difficult to do. It inevitable focuses conversations between client and provider on the one tangible cost that can be measured, which is the usage fee,” Hopkins explained.
Opting to use one bank’s algorithm in preference to another because there’s a smaller charge for use is an approach that ultimately leads to margin compression among algo providers, but is unlikely to lead to better performance outcomes for the client.
“The lack of transparency around all-in performance of algos, and the consequent focus on negotiating usage fees alone is something that definitely needs to be addressed,” says Hopkins. “We help clients by using our analytics in identifying the underlying drivers for algo performance and then help them to use that to inform their decision-making process on the trading desk.”
In other words, MahiFX is able to tease out what’s happening with the algo: how much of the performance is the venue that the algo is operating in or specific to the strategy, or urgency level? Is the algo creating market impact?
If this kind of data is available then determining optimal choice of strategy for a given set of liquidity conditions is possible: “At this time of day, in this particular currency pair, with this level of volatility – what previous algo has given me the best result? Armed with this information, the client can then decide on the appropriate strategy.”
This could, of course, mean not using an algo at all, he added. “It may be better to transfer the risk immediately, or pick up the phone and work something with a trusted counterparty,” said Hopkins.
With a greater degree of data collection, however, there is an increasing sensitivity as to how that data is being stored and used, resulting in a demand for segregated algo execution services.
“It’s an understandable response to client demand for anonymity and confidentiality,” said Hopkins.
“Across many of the big institutions we are seeing the establishment of completely separate desks that are physically segregated from the rest of the FX business. At the very least it is expected that there are technological controls in place that govern the distribution of data to approved groups.”
“It is understood by the big users of algos that their data wouldn’t be used inappropriately. Every bank recognises that if this does occur and the client is disadvantaged by that then the relationship is over; no institution is willing to take that risk,” he said.
Cheap comes out expensive
Pete Eggleston “If you’ve got an algo that’s very cheap, asset managers may use it because they believe it will save money. But |
Another Morgan Stanley alum, Pete Eggleston, decided to set up shop measuring execution performance, including algos, and co-founded BestX, which provides independent transaction cost analytics and technology.
“In an OTC market like FX, where you don’t have a tape or an official market-mid at any one point, it means that everyone is interpreting what they feel is the best measure,” said Eggleston. “It’s one of the reasons we set the company up, because one of the things you want to be able to do is measure that performance on a completely level playing field.”
That means measuring performance using the exact same data set to compute benchmarks and using the same analytics to estimate slippage or signalling risk, or whichever metrics of best execution relate to the algo.
“At the moment, clients may be making decisions that are based purely on cost, and that can be a misleading decision to make,” he said. “If you’ve got an algo that’s very cheap, asset managers may use it because they believe it will save money. But that’s not necessarily true on a net performance basis.”
So, if clients – asset managers, hedge funds and corporates – scrutinise the algo’s performance on a level playing field, it might turn out that the cheap algo is cheap for a reason.
“There is an element of: you get what you pay for, and it might also be the case here,” Eggleston said. “It might be that the algo has a headline cost of $50 (per) million, and it may be worth paying every one of those dollars because it might give you a net performance that is considerable better than the algo that is being charged at $20 (per) million.”
What asset owners, asset managers and regulators are looking for is a process for best execution based on quantitative evidence. “This is where these analytics become so important because then you are making a completely objective decision,” he added.
The drivers associated with cost savings is linked with the need for transparency in “every shape and form” Eggleston said, pointing to scandals associated with benchmarks such as the WM/Reuters fixing.
“The market is moving to a desire for more transparency, and much clearer view on exactly what people are paying for and why they’re paying it. In such an environment, it is then critical to have a complete set of analytics to make the best decision around their choice of execution to minimise costs and deliver against any regulatory requirements,” he said.
The cold hard numbers of cost savings are difficult to assess but one estimate from a former Schroders trader, Glen Sargeant, pegs it at about one basis point savings. “When I was trading, 1.5 basis points was the expected slippage from mid-market for our book of business. The TCA we used at Schroders would return a slippage from mid-market of no more than 0.5 basis points. So one basis point saving was easily achievable by working on an agile trading desk,” he said.
Choosing the right algo among dozens offered by a handful of banks becomes quickly overwhelming
Agile algos
Sargeant is now the product manager for FlexTrade’s Buy Side FX unit, and he sees that asset managers looking to employ a strategy in today’s market conditions are seeking an algo with the agility to move from passive to aggressive simply, or as he puts it: by “just using a dial”.
Brexit was one market event in which the shift from passive to aggressive in volatile conditions could be observed, particularly in Asian markets, he added. When the initial results seemed to point to momentum for the remain camp, the GBP/USD went immediately higher. But when things turned around to the leave side, small amounts started trading on wide spreads.
Those big swings were probably opportunistic traders from non-bank liquidity providers using algo strategies in non-liquid periods, Sargeant said.
From a macro perspective, Sargeant noted that asset managers are looking to place and build positions via algorithms with concerns to the market liquidity conditions: “Asset managers now appreciate that traders, armed with a toolkit which includes algos, have an insight as to what market conditions are and whether liquidity is sufficient to complete an order.”
Gone are the days when a fund manager would hand a ticket to a trader and wait for a fill, he added.
“Traders now have to be involved at a very early stage of the investment decision. It’s all about what can a trading desk achieve with an expected slippage, and what market conditions are conducive to getting this trade on or off. Even more pertinent is: what strategy can be employed to get out of a position in adverse market conditions?”
“Lack of liquidity is definitely a consideration for a macro trading fund management team,” he says.
There’s also the trend towards multi-asset class trading across the buy side, and algos are helping to facilitate more precise FX order execution on this front, for example by reducing the delay of executing FX derived from another asset class, known as implementation shortfall.
“In a multi-asset class EMS environment that can trade across assets using dynamic rules, you can run an FX algo at the same time and place as the underlying security trade incurring FX exposure,” he said. “Then you are pretty much eradicating implementation shortfall.”
So, if an equity trade is the leading asset class and starts slowing down in adverse conditions, the FX algorithm slows down as well – one would not outpace the other if they are linked, Sargeant explained.
Man versus machine
Glen Sargeant “In a multi-asset class EMS environment that can trade across assets using dynamic rules, you can run an FX algo at the same time and place as the underlying security trade incurring FX exposure” |
There is much handwringing over the future of trading as humans increasingly rely on machines for decision-making. As algos become more widely distributed there is some concern over being priced out of a job.
MahiFX’s Guy Hopkins said that may have been an understandable concern in the past, but ultimately what the clients want is human control over a set of algorithmic execution tools that access liquidity on their behalf.
“It comes down to the human to decide what strategy to use for a given set of liquidity conditions, and humans are far more qualified to make those sorts of heuristic assessments than machines,” he said.
“Trying to develop the ultimate algo which will respond to absolutely all liquidity conditions, I think it will be a long time waiting for that. Clients prefer to be in control of a limited set of algos that do different things with a degree of predictability – passive, aggressive, time-based, etc.”
He compares the situation to driverless cars: “People just don’t trust the machine to get
it right 100% of the time. Trying to add too much complexity into algos just ends up confusing people. If they can’t understand what it is doing or indeed explain its behaviour – intended or otherwise – to their managers or clients, they end up not trusting it or using it.”
BNP Paribas’ Razaq believes that in FX, the role of a traditional market making desk in a bank is not going to diminish. For example, when order requirements call for a transfer of risk to the bank in certainty of execution, the client doesn’t have the luxury of taking market risk and improving on execution costs via an algorithm.
“The two business models provide a balanced approach to execution. At BNP Paribas we are promoting both sides of the business – principal as well as agency execution.”
Meanwhile, the rapidly increasing availability of direct market access for the buy side comes with some caveats, said BestX’s Pete Eggleston. “Increasingly, execution desks want more control around the execution process and the liquidity sources they interact with. Their own due diligence, and best execution policy, won’t allow them to blindly trust a black box.
“Their best execution committee needs to know that this algo is understood and can be managed.”
The corporate space has recently emerged as a big adopter of FX algo’s
Future forward
According to a number of sources, there is a big gap in availability of sophisticated pre-trade analytics. In other words, estimating the cost of execution before launching the trade.
The ‘beauty’ of deploying an EMS, said Sargeant, is in the access traders have to real-time data, which allows them to monitor execution as it’s happening. This in-flight monitoring addresses best execution procedure exceptions at an early stage and can then be used for post-trade regulatory and compliance requirements of best execution policies.
But the historical TCA data has value too.
“It’s advantageous to have real-time data and track the performance in front of the trader, and use that information to make a decision on a similar order on another day in similar conditions,” Sargeant said.
For example, if an order took longer than expected to fill, or didn’t fill completely at a venue, then it’s possible that the venue is subjecting an order to last look, and it may not be within a firm’s execution policy to wait.
More generally, historical TCA data can be used to analyse a liquidity provider’s performance and value and help to determine whether to change the timing, the size of the order, or the strategy, in order to improve certainty of completion.
Up to now, TCA has been hugely important in FX, but has yet to make a meaningful impact in the front office in informing the decision-making process on a trading desk, said MahiFX’s Guy Hopkins.
“The big next thing is having information presented to traders about their past execution performance in real-time, packaging and presenting empirical trading data from your past trading history to help you immediately refer back to your previous experience and not have to rely solely on your own memory or intuition. There are so many products available to the buy side now, it is essential that they are able to rigorously assess the merits of each while minimising the costs of doing so,” he said.
In the firm’s product suite, MFX Echo provides a compelling path to achieve that, said Hopkins, by taking the tools and techniques that banks have been using for many years to optimise their eFX businesses and apply them in a completely new context.
“It’s a hugely exciting time,” he added.
Sophisticated real-time analytics is the next arena that BNP Paribas is moving into, said Asif Razaq. “Imagine how cool it would be that while running an execution algorithm, it’s providing real-time information to the user on current market conditions.”
This could include live cost estimations and estimated time of execution, among other metrics, that will allow clients to alter their execution path.
“Certainly we believe that this is the direction of travel. That clients are going to get more and more hands on when it comes to execution, and take active risk executing orders themselves rather than transferring risk to the bank,” he added.
Eggleston said that in the 25 years he has been in the industry, he’s not seen so much transformation.
“There really is incredibly rapid change in foreign exchange and in OTC markets in general that’s being driven by a myriad of factors. We are seeing the move in FX to be a more order-driven market from being a quote-driven market, and all the implications that come from that. It’s a period where people are looking for help in navigating change and that is really what we are trying to achieve at BestX. In that sense, it’s very exciting,” he stated.
