Signal Processing - November 2017 - 176

PERSPECTIVES
Xiao-Ping (Steven) Zhang

To the Victor Go the Spoils: AI in Financial Markets

rtificial intelligence (AI) for finance
is a hot topic these days. Top hedge
funds are waging a talent war over
AI experts the way sports teams compete for pro athletes. Recently, a friend
of mine, an AI expert working at a tech
firm, got an offer from a top hedge fund.
From a financial perspective, it is "an
offer he can't refuse," as Don Corleone
says. In the sci-fi film Transcendence,
Johnny Depp applied his newly digitized
brain to the markets, pocketing hundreds
of millions overnight. We've witnessed
the invincibility of machines-AlphaGo,
Watson, DeepBlue-and can't help but
wonder what's next. In January of this
year, a new AI named Libratus joined the
club by defeating four of the world's best
professional poker players. How close
are we to AI's conquest of the game of finance? Will AI experts become the new
titans of Wall Street?
Note that AI and signal processing (SP) have a natural connection. Indeed, AI for SP recently has achieved
pioneering success in speech and image
processing through deep neural network
(DNN) technologies. There exists the
unbridled expectation people and companies might have on this connection
(SP + AI) to make money.
To properly align our expectation,
we need to understand the nature of financial markets and, within the laws of
that universe, what AI can do for us as a

Digital Object Identifier 10.1109/MSP.2017.2738038
Date of publication: 13 November 2017

176

community and for you as an individual
investor. From this perspective, I will
show that AI cannot change the fundamental structure and dynamics of the
financial market. Like all other technologies and innovations introduced to finance, AI may advance our markets and
economy, automate away jobs, and even
bring a select few vast wealth. But the
vast majority of people-no matter their
level of AI expertise-will not achieve
large excess returns. So don't plan for an
early retirement just yet.
A fundamental function of the financial market is resource allocation,
i.e., for businesses or people to secure
resources for promising investment
pro--jects. Everyone knows, or quickly
learns, that investment returns are not
guaranteed. Investors assume risk when
allocating resources because they have
imperfect knowledge of the future. A
good investment has expected rewards
that outweigh these potential losses.
Investors of high-risk projects such as
start-ups demand high expected returns.
To minimize the risk or achieve the optimum risk-return investment, people
seek information that helps them see the
future more clearly. The financial market is driven by information.
Financial market innovations often
help resources flow to the projects that
we think are the most promising and
needed. An example related to our daily
lives is the invention of the credit card,
which allows individuals easy access to
capital (with high interest/risk) when
IEEE SIGNAL PROCESSING MAGAZINE

November 2017

people are in need. Many economists
[1] believe that consumers' continuous spending using the debt accessible
through credit cards in economic down
times is an important cause for the Great
Moderation in the United States in the
1990s. It is easy to take for granted how
much technology and the capability to
evaluate personal credit through data
are a driving force for the popularity of
credit cards.
The stock market processes an incalculable amount of data, converting it to
information about the future, and boiling everything down into a single number: price. Due to the fluctuations of
security prices, many people liken the
stock and bond market to one giant casino and attempt to profit by speculating
on what the asset prices should be. The
best of quantitative hedge funds, such
as Renaissance Technologies, D.E.
Shaw, Two Sigma, and Citadel, have
minted some of the world's wealthiest
people, encouraging those of us who are
working in SP and AI to fantasize about
our shot at a big win. If we can use our
skills to predict tomorrow's stock price,
or even the direction of the market,
we could make big money. It can feel
as if we are just one DNN away from
riches. Indeed, many open-source DNN
tools are available for everyone to use in
analysis.
DNN and future AI tools undoubtedly may help us to gain new information,
(continued on page 171)
1053-5888/17©2017IEEE

Table of Contents for the Digital Edition of Signal Processing - November 2017