THETRADETECH DA I LY
in-depth
THE OFFICIAL NEWSPAPER OF TRADETECH 2019
ARTIFICIAL
INTELLIGENCE
can be the buy-side’s
ACE UP THE SLEEVE
Artificial intelligence has huge potential to disrupt and improve
buy-side trading desks, but unlike the games in which such
systems can now outperform humans, there is still much work
to be done before traders could be displaced entirely.
W
hen Gideon Smith joined Rosenberg
Equites, the quant equity division of
AXA Investment Managers, in 1998, he was
given a simple instruction: “Make yourself
redundant.” Smith did just that and has never
looked back.
With his early duties at Rosenberg included
reading earnings releases on Reuters, Smith
devised an automated method for extract-
ing the key figures. The deal was that once
he had automated himself out of a job,
Rosenberg would find him something more
interesting to do.
Smith later became chief investment officer
and global head of portfolio management
at Rosenberg in London. Today, he argues,
advances in artificial intelligence (AI) mean
that it is possible to automate the process
of reading and actioning even unstructured
financial language.
Worries about job losses as a result of AI
don’t keep him awake at night. Smith sees no
signs of lower buy-side headcounts as a result
of AI and instead prefers to focus on the oppor-
tunities created by AI to generate new value.
The need for human monitoring and
evaluation will always create new jobs, he
says. Regulatory constraints set a hard limit
that is not going to disappear, so the buy-side
will continue to need to be able to explain
every trade: “We can’t just say that we did it
because the computer told us to.”
Jeff Schwartzman, head of learning and
development at Liquidnet, agrees. “The role
of traders and technology are not as inter-
changeable as some may believe,” he says.
20
THETRADETECH DAILY
“Technology won’t replace buy-side traders;
but buy-side traders who use technology will
replace those who don’t.
“It’s important that traders better equip
themselves to be better managers, leaders, col-
laborators, communicators and strategic plan-
ners,” Schwartzman says. “The technologies
available to the buy-side trader today allow this
to happen more effectively than ever before.”
Poker-faced
Machines are most efficient at “processing
data and managing structured tasks, but they
are not yet able to explore the unknown and
challenge the status quo,” Schwartzman says.
Yet there are signs that AI is able to expand
its horizons and, at the least, deal with the
‘known unknowns’.
According to a report from Lewellyn Consult-
ing, published in January, AI will increasingly
move from discrete tasks to entire functions.
Lewellyn points to the success of smart ma-
chines in progressing from beating top players
of games of perfect information (such as chess
or Go) to games such as poker, evidenced when
two machines were able to comprehensively
defeat a group of poker professionals in 2017.
In this instance, imperfect information and
blindness to the intentions and resources of
other players much more closely mimics the
situation on a real-life trading desk.
The AI programme that was used to beat
the pros, DeepStack, won by learning as it
went along. It calculated only a few steps
ahead, rather than an entire game, and re-
sponded to new information with the use of
neural networks. Smith at Rosenberg argues
that these neural networks also have the
capability to forecast market volatility spikes
– and points to kids who are now building
neural networks in their bedrooms.
Poker, of course, revolves around ‘known
unknowns’ – we know what the important
variables are, but have no clear way to work
them out. The ‘unknown unknowns’ – such
as black swan events which seem extremely
unlikely until they occur – are a different
matter entirely.
“Technology won’t replace buy-side traders;
but buy-side traders who use technology will replace
those who don’t.”
JEFF SCHWARTZMAN, LIQUIDNET