Conference Dailys TRADETech FX Daily 2018 | Page 9

A DV E RTOR IAL
out some research into the microstructure of the
market, and we found that there are three main
points that drive the performance of execution,
not necessarily the performance of an algo, but the
performance of a smart order router.
Firstly, the liquidity that we are connected to is
a simple fact of choosing venues that we want to
be connected to. That’s really about finding the
right mix of eight or nine venues, whereby con-
nection means you truly have all of the liquidity
you need without starting to access the same
liquidity via multiple access points. The second
point is around fill probabilities and dealing with
what is known in the FX market as ‘Last Look’
correctly. We are able to measure and understand
the fill probabilities on a currency pair on each of
the different venues. When we assess a price, we
ensure that we understand it correctly in terms
of the chances of executing at that price, whilst
comparing it with a slightly different price on a
different venue.
Finally, we looked at latency. Low latency market
makers typically have a competitive advantage
from building out a network that is extremely
fast and allows them to use information quicker
“Another important aspect is being able to show
the order to clients in detail and that’s where
transaction cost analysis (TCA) comes into play.”
than others. The way we are dealing with that is
not necessarily by ensuring we have the fastest
network, but by being smart about it and avoid
being picked off. We have co-located gateways
with the servers of each of our connected venues,
so for example, in London there’s EBS and in
Chicago with the CME where we trade futures.
Instead of dealing with minimising time difference
for reaching each of those servers, we send out our
order instructions to our gateways with a timer so
that they get released to the market at exactly the
same time.
How is artificial intelligence and other advance-
ments in technology influencing the FX business
and product design?
MB: With technology it’s all about learning and
assessing over time what the right decision is with
regards to execution. In terms of execution costs
for us, we take into account rejects and the impact
of resubmission of child orders. Clearly if one of
the child orders from an algo has been rejected we
need to put that back in the market with, what one
would expect, a negative outcome. We continuous-
ly measure fill rates and the cost of resubmissions
to build the data set on which our smart order
router bases execution decisions. It’s important
to understand that we may not necessarily hit the
best price on the screen, but at prices which are
probability-weighted and can get the best outcome
for the order as a whole.
CG: Probability of execution and fill rates are up-
dated continuously. They evolve and might change
depending on the time of day, currency pairs, or
how our machine learning and technology infra-
structure allows us to assess those fill ratios. The
key part here is that it is not static. The liquidity on
various venues will change over time and we are
able to pick that up reasonably quickly by measur-
ing how the fills are performing. It comes down to
a confidence of being filled compared to the best
single price on the screen, which you may not be
able to secure.
How are client expectations regarding algo liquidity
management evolving?
MB: If you go back some years, pre-algo, our cli-
ents originally wanted to know a price where they
could get a trade done. Over time, clients increas-
ingly took on some of the market time risk, as in
giving up the certainty of a fill, in the expectation
that it would improve the execution outcomes.
Equally, clients now not only want to engage with
liquidity over time, they want to access it in a fast
way. The evolution is executing on the external
venues either slowly over time to try and have
minimal market impact, or for immediate fills. So
our smart order router is directing orders to venues
based on the best possible outcome. We want to hit
prices that are real and we can access that liquidity,
as opposed to having some semblance of prices on
screen, but not being able to hit them. Our clients
truly want to understand what is adaptive liquidity
and how to access it. From our smart order router
concept, over time and at different times in the day,
we can direct child orders to different venues based
on the confidence of getting filled.
CG: Clients are increasingly trying to get a better
understanding on how liquidity is in the market on
a day or at a particular point in the day. It’s more
about the interaction. Moving away from what is
your price to deciphering the activity and depth
of the market today, and then providing valuable
feedback around that. Conversations are often
around whether clients should be executing large
orders over an extended period or as quickly as
possible, and we can have those conversations
based on the data that we are collecting on our
side and the knowledge and estimates of how our
algos execute.
Another important aspect is being able to show
the order to clients in detail and that’s where
transaction cost analysis (TCA) comes in to play.
Everything we have discussed wouldn’t stand up
against a critical client if we can’t then show them
the numbers depicting what we have done with
the order. We are able to do that using our TCA,
especially if clients are using our smart order
router. We can show our clients at which venues
they have executed, the fill ratio and at what price.
In the event that we skipped a venue that possibly
had a better price, we can show them that the data
proves it gave the better outcome.
The official newspaper of TradeTech FX Europe 2018
TheTradeNews.com
9