When panic struck the Chinese stock market last August, Man
AHL wasnt caught flat-footed. The London-based systematic
trading firm avoided losses during the massive sell-off thanks
to a machine learning algorithm that analyzed its positions and
reacted faster than traders could to the spike in volatility.
Machine learning helps us to spot patterns that humans
cant easily spot or couldnt spot at all in the
sheer amount of data being created today, says Sandy
Rattray, CEO of $19 billion Man AHL.
The firm began its life in 1987 as commodity trading adviser
AHL, the quantitative investing pioneer founded by Michael
Adam, David Harding and Martin Lueck that spawned Aspect
Capital and Winton Capital Group. Almost 30 years later, the
three rival hedge fund players are investing heavily in machine
learning and data science as they seek to capitalize on renewed
investor interest in quant strategies.
Earlier this month Man AHL announced that it would be
expanding the scope of its Oxford-Man Institute (OMI), a
quantitative finance joint venture founded in 2007 with the
University of Oxford. OMI, which aims to create a machine
learning and data analytics hub at Oxford, will also join the
universitys Department of Engineering Science in
The OMI expansion is part of a broader push by Man AHL, a
unit of U.K. alternative-investment giant Man Group, to develop
machine learning that enables better financial algorithms.
The firm trades four main strategies: classical long-term trend
following, proprietary trend following, multistrategy and
long-only. Under Rattrays leadership, Man AHL has been
developing machine learningdriven algorithms for half a
decade. The results of that work have been filtering into
select client products over the past three years, in some cases
driving significant gains.
We have spent a lot of time on machine learning; it is
the single biggest area of research spending for the
firm, says Rattray, who joined Man AHL in 2007 from
Goldman Sachs Group, where he was a managing director
specializing in quantitative trading. We are looking at
people, data, hardware its not just one aspect
that is going to bring this all together. Of Man
Groups 1,000-plus staff, 130 focus on trading strategy;
many of them are doing machine learning research that
supplements OMIs work.
Machine learningbased trading algorithms operate much
differently than their rules-based counterparts. Historically,
a portfolio manager would have created an algorithm based on a
financial model or data set and fashioned rules for how it
would behave when trading.
With machine learning, a computer sifts through billions of
data points, picking up patterns. Armed with this knowledge, it
learns trading behaviors such as buying dips or selling high
over time, based on what it has gleaned about the market from
past and present data. Man AHLs algorithms work off vast
data sets that include 1.5 trillion price ticks as well as
options information and index feeds. Billions of new price
ticks can be added to the system in a single day.
The firm has invested in high-powered hardware to process
this information. But traders and scientists must know how the
algorithms work and how to avoid overfitting data to a trading
hypothesis, warns Anthony Ledford, chief scientist at OMI.
The important thing to realize with machine learning is
that you have to understand both the problem and the data, and
fit them together carefully, Ledford explains. You
always run the risk of overfitting.
Man AHLs research effort comes after a few challenging
years for the firm, whose assets under management still
havent recovered to their peak of $24 billion before the
200809 financial crisis. Year-to-date through
April, the $4.6 billion Man AHL Diversified Programme was down
0.34 percent, according to Atlanta-based data provider
eVestment. But systematic trend follower Aspects
flagship, $4.7 billion Diversified Fund fell 4.8 percent during
the same period. At CTA Winton, the $18.5 billion Winton
Diversified Fund (Luxembourg) and the $13.3 billion Winton
Futures Fund lost 2.85 percent and 2.2 percent,
Aspect co-founder and CEO Anthony Todd started his
London-based firm in 1997 with research director and fellow
University of Oxford physics graduate Lueck after they left
AHL, which Man had acquired in stages from 1989 to 1994.
Investors are coming back into strategies like ours
because theres no yield in fixed income and they want a
diversifier alongside equities, Todd says. We were
well positioned for the sell-off that started at the end of
last year. Our programs reacted dynamically.
Aspects trading program is actually seven in one,
operating at a range of frequencies by capturing trends of
between two weeks and six months. Those different frequencies
matter in choppy conditions like the first quarter of 2016.
After markets stabilized in late February, the Aspect
strategies with the shortest-lived positions fended off
potential losses by allowing the firm to move quickly, Todd
says: We actually built up a small short position in the
dollar and went long Brent crude, he recalls. Still, the
Aspect Diversified Fund was down 4.69 percent in March,
When it comes to quantitative finance, Aspect occupies a
middle ground, blending human intuition and rigorous research
to create trading programs that arent just autonomous
algorithms. The medium-term trend follower keeps refining its
models through research and opportunistic acquisitions, Todd
In March Aspect acquired $1.4 billion, Jersey, U.K.
based rival Auriel Capital Management. Besides boosting total
assets to $6.3 billion, this takeover gave Aspect access to a
unique currency overlay for its trading programs. The firm has
also cherry-picked individual talent, bringing over Franck
Lauri, formerly of French asset manager OTEA Capital, who
specializes in statistical arbitrage, and Antonio Botelho and
Constantin Filitti from London-based hedge fund firm Capula
Investment Management. Lauri, who joined Aspect late last year,
brings 15 years experience in systematic strategies; the
Capula duo arrived in mid-2014 to run Aspect Tactical
Opportunities, a $10 million systematic multistrategy futures
Todd, who is on the lookout for other small firms to
acquire, seeks intellectual property that Aspect can use to
refine its funds with road-tested ideas. The firm is also
delving into areas like deep learning training neural
networkbased algorithms for trading purposes
examining new ways of sifting through big data and beefing up
its cloud infrastructure to add storage capacity and processing
But Aspect is hardly going all in on
artificial intelligence. Our research approach has
always been hypothesis-driven, Todd says. Of course
were developing these areas, but we are more incremental.
One of the draws of medium-term trend following is that
its intuitive, so we want to maintain that balance
between intuition and algorithms.
David Harding founded London-headquartered Winton a year
after departing AHL in 1996. Like Lueck he was always more
interested in research and trading than being part of a large
institution. Today Winton is also expanding its research
capability; to that end, the $34.5 billion firm has opened a
San Francisco data science center to tap Silicon Valley
Founder and CEO Harding, who holds a physics degree from the
University of Cambridge, sees a future in building proprietary
data sets. Winton wants top-shelf scientists to help, he told
Institutional Investor at Mays Milken Institute
Global Conference in Beverly Hills, California.
Harding is taking a little from column A and a little from
column B, matching up computing and intellect to find a new way
forward. With plans to grow Wintons San Francisco team
from six scientists to as many as 40, hes open to ideas
when it comes to building investment hypotheses and gathering
The Bay Area outpost will also house the North American arm
of Winton Ventures, a new venture capital unit that is on the
hunt for data-driven start-ups. Were interested in
companies where really understanding what can be inferred from
the data drawing valid conclusions from the data
is essential to the success of the business, Harding
says. We are also open to working with companies that
have the potential to enhance Wintons existing
For John Moody, a computational finance expert who runs his
own CTA firm in Portland, Oregon, the resurgent popularity of
quantitative trading strategies is part of a broader trend.
People have difficulty thinking statistically; they let
cognitive bias get in the way, says the founder of $152
million JE Moody & Co. The research suggests that
because of this, it can be less risky to let formulas do the
work when it comes to making investment decisions.
An invention like Googles autonomous car (see
Autonomous Vehicles index), which learns as it drives, has
made people more comfortable with that idea, Moody notes. As
society becomes more reliant on machine learning algorithms for
a wide range of decisions, Moody expects their use in finance
to keep growing a potential boon for quant shops.
Like Aspects Todd and Man AHLs Rattray, Harding
is philosophical about how far computers can take his firm.
The development of machine learning can be traced from
the 1950s all the way through to today, he says.
People talk about it like algorithms are going to run
everything, but these are really incremental advances that
weve achieved. We will have to keep up the work in order
to stay ahead, and talented people will always be at the helm
of that work.
The Rise of the Tech Model May Soon Make You Obsolete