Execution algorithms. Where innovation is driving demand in FX

June 2023 in Uncategorized

Everything is back to normal in the FX market – but it’s a new
normal. All around us, corporates are back to funding their
cross-border operations; trading desks are back to taking orders
and transacting; retail investors are back to piling in behind
every new trend. We’ve had ten-plus years of upheaval – not just
the crash but also the impact of technology on what we do; plus
all that wider IT-accelerated social, political and economic
change; lately the worst weather for centuries; all capped off
with a European political crisis that mimics 1914 rather too
accurately – and yet, behind it all, the FX market is back to
moving as it has always moved.

Which is significant. Seven years ago, Credit Suisse launched AES
FX. Take-up was sustained, as was competition to dominate the new
space. Evangelos Maniatopoulos, global head of AES FX product and
trading, Credit Suisse, says: “When Credit Suisse launched AES FX
seven years ago, it represented a new approach to executing FX
transactions. At the time, our challenge was to convince our
customers that it was a viable alternative to the more
traditional means of trading FX – click and deal, voice dealing,
et cetera. We strongly believed that the benefits of this type of
product would be appreciated by customers.” Then, AES FX was the
new game in town. Now, many (most, pretty much all) Tier 1 banks
have, or are working to have, an algorithmic execution platform
to accompany their more traditional execution capabilities.

Execution algorithms. Where innovation is driving demand in FX

The new approach to FX arrived just in time for fragmentation –
liquidity became (even more?) problematic – and a fresh
regulatory emphasis on transparency, thus TCA; Maniatopoulos
comments that work by Credit Suisse to develop pre-trade,
real-time and post-trade TCA in the FX space “significantly
contributed” to the ongoing adoption of AES FX.

Technology in general, and algorithmic execution in particular,
gave us many things over the decade just passed: (ultra) low
latency, HFT, the phrase “flash crash”, lots of speeches on
Capitol Hill, politicians adding their names to copious new
legislation. We have access, if we can use it, to whatever Big
Data is trying to tell us, and we have smart-order routing and
various forms of machine intelligence. Algorithms can trade and
transact FX faster than we can think about FX, and they can do
that effectively as well as (if you’re up for re-election)
scarily.

But the really attractive characteristic of algos, which makes
them so durable, is that they flourish just as easily in a
tightly regulated trading environment, as in the relative
free-for-all of global FX. Algorithms are good at stealth, but
they also provide an unambiguous audit trail. “Algorithmic
execution in FX is here to stay,” says Maniatopoulos,
understandably.

So what’s new about the “new normal” in algorithmic execution of
FX. The more instructive question is – what’s normal about it?
The suggestion above, that we’re “back” to normal in the FX
market, might seem to suggest that all the change of the past
decade has been temporary. In fact, the opposite is true. Best
evidence of that? Look at who is using FX algos.

Buy-side users

So which FX buy side sectors are now becoming the biggest users
of execution algorithms? Well first of all we have corporates.
They may not be the biggest yet, but they do seem to be getting
there. Asif Razaq, global head of FX algo execution at BNP
Paribas, says: “We’ve seen increasing demand in the corporate
sector. Corporates are now beginning to see the true benefits of
using execution algorithms.” This is not a finding unique to BNP
Paribas. Giving a strikingly similar answer, Maniatopoulos says:
“We have seen a great deal of interest from corporates over the
last eighteen months. Certainly, corporates are becoming more
accustomed to the concept of execution algorithms.”

To judge from these and other replies to the same question, there
is a widespread corporate awakening going on, to the benefits of
FX-algo execution. Indeed, Paul Downie, head of foreign exchange,
Royal Dutch Shell, has described algorithmic execution as the
“lynchpin for continuous improvement” in his firm’s FX activity
(citing transparency and flexibility as key drivers). So what’s
normal? Corporates have been buying and selling currencies since
the very beginning. They’ve been doing it for operational
reasons, and their presence in the market is a large part of what
makes it interesting for the rest of us. If corporates are using
algos, it follows that algos are normal as well as new.

“There is a second group,” Maniatopoulos also says: “where we
have seen significant demand for execution algorithms. It has
increased over time, and has spanned across a number of customer
segments. Most active in the space are hedge funds, investment
advisers, money managers, pension funds – the real-money
community generally.” FX-execution algorithms are, as it were,
hitting the big time across the market.

Jonathan Wykes, Head of EMEA eFX sales, Bank of America Merrill
Lynch, says: “After corporates, we see a lot of activity coming
from the real-money community. They’re typically either using
execution algorithms because they want to execute based on their
view and participate in price improvement, or they’re looking to
exit a trade at a particular level and want to do that quietly
and discreetly.”

There is a distinction to be made here. Wykes says: “Corporates
on the other hand trade in large size and with the aim of exiting
risk over a long period of time.”

The significance of corporate take-up is what it tells us about
the acceptability of algos. They’re no longer seen as “dangerous”
in the way that, let’s say, certain derivative structures came to
be seen as dangerous in the immediate aftermath of what happened
a few years back. They may be conducive to HFT, et cetera, and
appealing to adventurers, but algorithms are mainstream now. They
are no longer exotic.

Why not? Partly because trading desks have been working hard to
“normalise” FX algos, not least via innovation at the point of
access, and partly because the buy-side has been engaging much
more actively with the execution process. There has been
regulatory pressure towards greater mutual understanding between
the two sides, which has led to a mutual enthusiasm for
transparency, oversight and effective TCA (we’re coming to that),
and among other factors driving clients to take greater care have
been all manner of wars, turbulences and natural disasters. Gary
Stone, chief strategy officer, Bloomberg Tradebook, says: “We’re
seeing a lot of engagement from both sides, from liquidity
providers and from hedge funds and the buy-side. In the past,
hedge funds and the buy-side were takers; now we’re seeing them
become makers. They’re playing both sides.”

Engaging with the OBO trader

We’re all in this together, in effect. Corporates and the
real-money community alike have been incentivised to engage by
regulatory pressure and market instability (and the banks’ own
efforts, no doubt). Peter Bondesen Sales Manager EMEA FlexTrade
UK, says: “In general, the market is moving towards smarter
execution, either by utilising in-house strategies or broker
algos. The biggest push we are seeing is from the asset
managers.”

Being smart is no longer just an option; it’s a fiduciary
obligation. To jump ahead of ourselves, Bondesen’s idea of
“smart” goes beyond order routing. Bondesen says: “I can imagine
that in the future, artificial intelligence will be used to read
news quickly and make decisions based on the perception of the
story.” We would have to redefine the word “irony” if regulatory
pressure from Brussels, say, rather than just brain-power from
Silicon Valley, pushed AI to sci-fi levels of usefulness.

But we really are jumping ahead. Engagement is a two-way process,
and while the buy-side has been getting to grips with algo
execution, trading desks have been coming to meet them half-way,
as it were, by re-engaging with more traditional methodologies.
Razaq says: “We’ve seen a rise in OBO [On Behalf Of] trading. For
clients who are risk-averse and are apprehensive to invoke an
algorithm, they are now simply calling the bank, just as they
would call for a voice trade, but instead are requesting, can you
buy me €100m on Chameleon?” Note that word: massive. “The
salesperson pulls up the algo order ticket, puts in the order on
their behalf, and updates them as the order is executed,” says
Razaq.

OBO trading means clients picking up the phone. Those of us old
enough to remember rotary dials and talking to the operator can
still place our orders in the old-fashioned way, and get access
to algorithmic execution. Discussing corporate interest in
Chameleon (BNP Paribas’ passive algo, by contrast with the more
aggressive Viper), Razaq says: “Such clients still respect the
voice relationship they have with the bank and it is something
they would like to maintain. As they get more experienced, they
will eventually start submitting the orders themselves.” And
until they get to that point, they will have the salesperson’s
running commentary as the order goes through.

If today’s range of external factors (regulation, turbulence, et
cetera) represent a push towards engagement, there might be a
case for wondering whether there are any pull factors impacting
on the buy-side. And of course – there are. The sell-side is
giving the buy-side what it wants – not least, a clear audit
trail, plus a voice relationship and a GUI (to misuse the term)
that looks remarkably like an old-fashioned telephone handset.

All of this at the same time as execution itself is becoming a
much richer experience.

All TCA together now!

Okay, we can stop talking as though algo execution is something
distinct and separate from every other kind of execution, and we
can start to give up on the idea that execution itself is just a
matter of, well, executing. The buy-side wants more. Steve
Aldridge, head of macro esales, EMEA, Credit Suisse, says:
“Pre-trade and post-trade; execution-advisory has always been
part of our offering. It’s coming more into focus now, and it’s
something that clients increasingly ask for.”

We can assume that those clients recognise their own need to be
properly informed. Aldridge says: “It’s often the case that a
client will come to us with a general question, to which we can
provide specific solutions. For example, if I’m an execution
trader, how do I demonstrate that I’m doing a good job? How do I
demonstrate to the portfolio manager that my execution is better
via AES than by just hitting a risk price or using another algo
product?”

Meanwhile, Bloomberg Tradebook have recently launched their
Execution Consulting service, which is designed “to impact and
add value to each phase of a well-crafted trade cycle”, as the
website rather nicely puts it. Stone continues: “Clients are
looking at how their orders interact with the market. Are they
getting the results they are looking for? They’re asking: what
can we do with different tactical algorithms to achieve the
results we are looking for?”

These, again, are clearly well-informed clients, even if what
they’re informed about is their need for a better understanding
of what they’re doing. Next question, following on from this: is
the focus on transaction cost analysis in FX further stimulating
the use of algorithms as a key component of best-execution
toolsets? Yes, because TCA by its very nature emphasises the
transparency that is so readily achievable via algo usage. Razaq
says: “We give clients full transparency on execution, with the
Cortex iX post-trade report we break down every single
transaction the algo has traded. Furthermore, to support TCA, as
a bank we are making ourselves fully auditable to our clients. To
satisfy the best-execution requirement, we also give the client a
snapshot view of the market, including all prices and venues at
the time the algo traded.”

Clients can audit trades and trade history, and work out whether
they’ve received best execution. Oh, and algos are also handy for
what-if scenarios. On this, Peter Bondesen says: “By comparing
the market impact of similar execution algos in the same pairs
during similar hours, trading desks will over time be able to
optimise execution strategies through TCA analysis of the
strategies.” There’s a virtuous circle spinning round here;
imagine the content-rich meetings that algos enable. Bondesen,
going on to discuss the range of comparisons/analyses available,
says: “Whichever algo is being used, the most important variables
are the credit facilities of the executing broker, the liquidity
sources reached and the code itself, often using a combination of
passive and aggressive orders to maximise the likelihood of
completing the order within the timeframe without crossing the
spread on all orders.”

Scope for some really interesting conversations there. But
there’s more. FX-algo TCA is changing client behaviour. Gary
Stone describes a recent past in which hedge funds would use
tactical algos, while long-only funds would typically be much
more tentative, and contrasts this past with a very different
present. Stone says: “We’re starting to see both the buy-side and
the long-onlys work their order in the market, in an aggregator
or manual-type fashion. The algos that people are using are
reserve, I’ll iceberg the order, or I’ll peg the order, or I’ll
do different things in that realm.” We’ve spoken earlier about
the increased engagement that algos encourage; TCA is the
enabling medium for that engagement.

Note, though, that engagement is as much
strategic as tactical. Stone continues: “The innovation which I
think is coming is that, in their DMA tactical algo suite,
clients are starting more aggressively to think overall about how
they want to engage the market. New algos are going to try to
mimic some of the behaviour of traders.” For Bloomberg Tradebook,
as Stone explains, the new emphasis on the quality and nature of
the client’s market engagement is prompting a new focus on the
algo’s smart-order router and how it interacts with different
liquidity points.

Next question. What work has been done to develop more adaptive
algorithms specifically for the FX market which leave minimal
market footprints and which are a significant improvement and
refinement on the previous generation of algos?

This is an interesting one, not least because equity algos
existed before FX algos, and FX algos tend to be part of wider
product suites. “AES is available across a large number of
electronic asset classes, including equities, futures, options
and of course foreign exchange,” says Maniatopoulos. To offer a
product suite across all asset classes is not, of course, to
overlook asset-class-specific, FX-specific, development. Aldridge
says: “What we aim to demonstrate to the client, over the
hundreds of thousands of trades that AES FX has executed during
the last seven years, is that on average, they can improve their
execution quality, execute inside a spread, or beat their
benchmark.”

Adaptive algorithms leave no footprints

“We only deliver next generation adaptive algorithms,” says Asif
Razaq. “The algo market is fast-moving and we’re constantly
developing new tools.” One of these is the “feedback loop”. Razaq
says: “The algorithms talk to the client and give them a view of
what they’re seeing in the market.” Giving the example of a
client using Chameleon (passive, stealth) to execute a large
order, Razaq says: “While the algorithm is executing, we want to
show the clients the very same market signals that the algorithm
is utilising.” The algo is using nanosecond technology, says
Razaq, to work out the shape of the market, where the market’s
going, what its execution path will be.

There’s a dial. And a red button. If Chameleon
is executing rapidly in a favourable market, the client can use
the dial to slow it down (in the expectation of further
improvement in the market, say). The red button is coupled to a
heat indicator showing liquidity. “The algorithms are designed to
run on autopilot, but we want to give clients the option of
turning off autopilot and flying the execution themselves,” says
Razaq. The red button – also known as the “start rapid fill”
button – glows red when there’s liquidity about. The client hits
the button to convert a passive Chameleon into an aggressive
Viper, and then hits it again to slow things down. A client with
a hyperactive inner child can keep on hitting the button, passive
to aggressive and back again, until the button stops glowing red
and it’s time to come in from the playground.

Having started the conversation talking about “third-generation
adaptive algos”, by the mid-point Razaq has shifted to describing
“fourth-generation interactive algos”. The future is here
already, and it’s answering back.

Jonathan Wykes notes: “One of the key things
we’ve done here is make our algos more attuned to the unique
characteristics of a particular currency pair and time of day.”
This is almost an obvious point to make, although not all
algo-designers seem to be aware of it: different currencies trade
very differently. Wykes says: “So in the background a user can
set currency specific parameters for each algo, or let the algo
dynamically adapt to the current market conditions in order to
give you the best possible execution. The algo should be able to
recognise that, say, EUR:USD might trade every 25 milliseconds
but USD:ZAR only trades every 250 to 500 milliseconds, so there’s
no point in trying to replenish your liquidity in the market as
frequently.”

Time zones also impact on trading behaviour, says Wykes. “The
other point is, the algo’s engine has to recognise which
liquidity pools it should be executing, not just on historical
data, but based on what’s happening in real time.” Part of the
issue here, of course, is that the choice of liquidity pool can
determine market impact. Peter Bondesen develops this point:
“Broker algos are using a combination of external and internal
liquidity to improve overall execution, and the combination of
these multiple pools of liquidity has been optimised in some of
the latest algos.” What’s adaptive here is the size of orders
placed in each venue and how that volume is split into multiple
smaller trades. Bondesen says: “These slices are disguised as
part of the overall flow from the broker thereby not revealing
the origin or size of the parent order. By maximising the amount
done through passive orders, the direction is hidden from the
market.”

Getting it, measuring it

What about peer-group TCA? Would this, first,
reveal the benefits of using execution algorithms, and secondly,
would it lead to a significantly increased adoption rate? The
issue arises because banks, understandably enough, provide their
own proprietary forms of TCA. These, although comprehensive in
themselves, do not necessarily provide a level playing field for
comparison by, say, prospective clients or indeed regulators.
Evangelos Maniatopoulos says: “It may be that regulators will
require FX to change from OTC to an exchange-traded market.
Without agreed rules of engagement and guidelines as to how
trades are meant to be executed and reported, and what kind of
information should be disseminated to the open market, it is
difficult to reach the level of transparency that clients are
accustomed to in other asset classes.”

To answer the questions at the top of this
section, and taking into account everything else said so far –
almost certainly, and probably. That said, Jonathan Wykes sounds
a note of caution. “One of the constraints of peer-group TCA is
that while clients may execute similar trades, they could have
quite different reasons for entering into them. At the moment,
trying to incorporate personal factors into peer-group TCA is
virtually impossible to do,” he says. “Peer-group TCA also
doesn’t account for the exact amount of liquidity available at
any given time, the underlying volatility or the price action, so
its benefits are limited to giving a ‘rough guide’.”

Later in the same conversation, Wykes says: “Algorithms do what
they’re instructed to do. The more you articulate your objectives
through the parameters contained within the algos, the better its
performance.” Which is a thought-provoking observation in itself.
It also raises two further questions beyond TCA. To use the right
algo at the right time is partly a matter of effective and
ongoing pre-trade, real-time and post-trade analysis, but it also
requires dialogue. Maniatopoulos refers to the “significant
dialogue, communication, education, constant support” that AES FX
users can expect to receive, and then takes us straight to the
second of our two further questions: getting the thing plugged in
properly.

What issues are involved with integrating
execution algorithms into buy-side workflows and overcoming the
complexities of integration? Maniatopoulos says: When clients
request trading access over a new platform, you need to be able
to tick all the boxes. You need to be able to manage
connectivity, handle trading into single/multiple accounts, offer
access to a large number of currency pairs, have STP in place,
and of course do all of the above very well.” Connectivity is
key.

Pushing the parameters

In what ways may the parameterisation of
execution algorithms enable changes in their behaviour by clients
on-the-fly? Jonathan Wykes says: “We’ve recently launched an algo
called Whisper, which provides efficient direct access to our
highly coveted internal-only client liquidity. Key new features
of this tool provide traders with the ability to expedite
execution, increase customisation and streamline usage. For
example, the ‘Take Profit’ feature follows the market and
captures spread while also setting boundaries around how far the
algo should go. Similarly the ‘Expedite’ feature speeds up the
fill rate of an order if the market starts going against you.
With these enhanced capabilities in mind, we have ‘whisperised’
our other algos – by taking some of these key parameters and
applying them across our strategy suite.”

If client engagement is the future, parameterisation is the
future. Wykes says: “The clients that perform best are those that
have a strong understanding of the different parameters that are
available to them and keep an eye on what they’re doing.” Wykes
also makes a very good point about evolution: go back. “So when
you create new parameters, you not only apply them to your latest
strategies but also see if you can go back and re-tune your
pre-existing strategies to work better.”