[ M A R K E T
R E V I E W
funds in the market. Despite the
backing, Quantopian is opening its
arms to a more universal, less elite,
investment audience. Contributors
to Quantopian come from diverse
backgrounds ranging from data
science to financial engineering,
software development to chemistry
and academia. Each author is paid
10% of the algorithm’s net profits,
while retaining ownership of the
intellectual property.
The next frontier
Last month the company started
to deploy its first tranches of seed
capital into some 20 strategies,
with each algorithm being allocat-
ed up to $3 million—the intention
is to build this up to $50 million
by the end of the year. Thomas
Wiecki, director of data science
at Quantopian, says the invest-
ment market is moving towards
machines. In particular towards
deep learning processes. The latter
is applied to algorithms which can
discover patterns and complex
relationships without being told
what to look for.
“Machine learning can process
and correlate huge amounts of data
|
M A C H I N E
L E A R N I N G ]
using machine learning in a trading
algorithm on Quantopian, deep
learning is the next frontier. Deep
learning can successfully learn
complex features directly from the
“Everyone around the world wants to have
machine learning incorporated into the
business.”
GARY KAZANTSEV, HEAD OF THE MACHINE
LEARNING GROUP, BLOOMBERG
raw data and build higher-order
connections something which is
unfeasible for hand engineering
where the number of potential
inputs is huge.”
The application of deep learning
in quant finance is still new—but
participants believe it could be
the next step in the evolution
of machine investing. Investors
have also been looking to use the
technology for a number of other
investment purposes. Earlier this
year Axa Investment Managers
launched a nine-month trial to test
“Deep learning can successfully learn
complex features directly from the raw data
and build higher-order connections.”
THOMAS WIECKI, DIRECTOR OF DATA SCIENCE AT QUANTOPIAN
to identify predictive signals—that
definitely provides the edge when
it comes to finance,” says Wiecki.
“While there are many examples of
44
TheTrade
Summer 2017
sources of news and other data
sources to understand media
perception of stocks or companies.
AllianceBernstein, meanwhile, has
been focusing on techniques to
the machine learning technology
of MKT MediaStats which runs
a sentiment analysis service for
investors, scouring some 30,000
learn why a particular sentiment
might arise, using machine learn-
ing techniques.
But not everyone is convinced
about the market’s readiness for
machine learning. The rules of the
finance game are not constant like
chess or Go. At the same time data
financial datasets are much noisier
than those typical of the appli-
cations where machine learning
has worked best so far. These
days financial data is growing in
diversity, consisting not only of
the obvious numbers and text but
also more unusual information
sources, ranging from meteorologi-
cal diagrams and weather forecasts
to social media. While it is vastly
cheaper to access and store finan-
cial data the challenge of using
it under these strictures remains
tough. And, of course, getting an
AI system to handle the quirks of
the markets which are impacted
by human psychology, is another
question altogether. But these are
all surmountable challenges. It
seems that the market is edging
ever closer to the day when the
machines will be investing for us.