Economist Perspective: Using AI to Analyze Financial Markets

– Blu Putnam:
Artificial Intelligence, AI, and machine learning have already made
major contributions to financial market analysis. As we have
explored the opportunities, and they are incredible, we also need to
recognize the challenges. Financial analysis is not
at all like facial recognition, where there have been no
real changes in 10,000 years. Financial analysis is not
like playing games, either. Take chess. The rules and objectives are totally clear
and they do not change. Now let's think
about financial markets. The rules change and they
may be interpreted differently. Market participants
have different objectives, different risk tolerances. Rule interpretation and
enforcement is not consistent. And some participants,
just a few, may bend the rules. In short,
with financial market analysis, there are a lot
of complications and complex feedback effects.

It gets harder. Even random variables can
develop short-term patterns, and a machine developed explicitly for
pattern recognition must learn to distinguish which patterns may have
some useful persistence, and which ones to throw away. Markets are dynamic. Market participants
act and react to every new
piece of information, adding complexity
to the feedback effects. Data from each time unit
are not of equal value, and older data may be of much
less value than newer data. Markets may be episodic, with one set of characteristics
dominating for months or even years,
and then giving a way to a new set of
environmental factors.

When applying machine learning
to financial market analysis, we've learned that a tremendous
amount of time and effort must be invested
in extensive data cleaning, checking and verification. Deep domain
expertise is required for both data cleansing, as well as choosing
the right AI or machine learning tools
just to attack a given problem and provide useful solutions that can stand
the test of time. Speaking of time, the analysis can never
be static; always dynamic. Meaning ideas such
as Bayesian Inference becoming extremely
important to consider. And then there is
the application of theory, not always from finance. Appreciating market structure can take one into
theoretical physics, behavioral finance and
disequilibrium economics, just to name a few.

All of this complexity underscores the
difficulty and challenges, yet we do not wish
to dampen the enthusiasm and the promise. We have already
made amazing strides with our new AI and
machine learning tools, coupled with
big data applications. What we've learned is
that we need diverse teams with domain expertise, statistical skills
and theoretical thinking. Collaborative teams are
likely to show us the way to effectively employ machine
learning in financial analysis. I'm Blu Putnam,
Chief Economist, CME Group..

test attribution text

Add Comment