AI/ML in investment management: Current applications
Andrew Wu shows how AI and machine learning are actually applied in investment technologies and how the industry is transforming as a result of advances in data science.
Excerpt From
Transcript
Now that we have clarified the meaning
of the key concepts such as AI, deep learning, unstructured data. Let's have an overview of
how they're actually applied in the business of investment management. The investment industry is
undergoing a major transformation with advances in information and
data analytics. And this transformation is probably even
more pronounced than in the other sectors that we discussed in this fintech
series like payment and lending. This is because instead of outsiders like
tech startups trying to disrupt the game, investors themselves particularly
institutions such as mutual, pension, and
hedge funds are actively chasing and adopting these new technologies in the
competition for market bidding returns. Consequently, we're at a stage where
even tasks that are traditionally thought of as requiring a high cognitive
level such as asset allocation, security selection, trading and
economic forecasting are increasingly being outsourced fully to advanced
machine learning algorithms. Think about a traditional stock trader,
what image pops into your head? Well, maybe someone who's busy
analyzing stock charts for things like trends, tops,
bottoms, etc, right? Well, some of these
tasks like chart reading can be done much more accurately and
efficiently by computers. New machine learning tools like
signal processing and neural networks can spot much more nuanced patterns in
the data that humans cannot detect. And consequently, these high-level tasks
are increasingly being done by algorithms. This is already happening
in a lot of institutions like BlackRock which is going to replace
more than half of its human traders with machine learning algorithms. More adventurous entities like Sentient
would have hedge funds entirely run by machine learning algorithms
end-to-end from stock selection to portfolio construction and
to trade execution. For individuals,
now they're even online, quote unquote, crowdfunding platforms like Quantopian and
Numerai that aim to crowdsource new investment algorithms directly
from the investing public. Like other funding platforms, they
directly connect the individual traders on one side with capital providing
investors on the other side. These platforms will often get
their financial databases for free to the individual investors
to test their algorithms on. Then, they will pull investor
money to fund the algorithms that perform well in the back test. So overall in the past decade,
we saw AI and machine learning being firmly
established as a powerful and necessary decision support system that help us
make more informed investment decisions. And the corollary is that
data both structured and unstructured are becoming more and
more important and valuable. More complex models need more and better
data to properly train and estimate. And consequently, any new data that
can potentially provide extra insights are increasingly being sought after. And there are even now specialists
like RavenPack that gathers data in real-time such as news and
social media as soon as they came out, process these data in real-time and sell
the outputs to investment institutions. Now to the downside, there are some
common issues with using so much AI and machine learning in our
investment decisions. Perhaps the most important issue to be
aware of is the lack of transparency. A lot of the machine learning
models are well quite complex and are very difficult to understand. Even if you spend the time
to understand them, the results from a lot of these algorithms
are often very hard to replicate. And most investors are not
willing to share their secrets. This again leads to an information
asymmetry problem where you can have an institution claiming to
use advanced algorithms, but in reality are doing none of that or
worse. These types of AI scums
are on the rise and might undermine investor
confidence in AI based investing. A related point is that because
these algorithms are so complex and many are claiming to be highly automated,
many investors might actually be reluctant to turn over their trading accounts
completely over to these algorithms. Research has indicated that
many people have an aversion to too much automation
when it comes to money. I will say, well, are you really want
my money to be managed by a robot? Robots like the ones that we saw in
the movies are either dumb or evil. And I like to have some good
old-fashioned human touch to it. Rational or not, these aversions might pose some roadblocks in the future
growth of AI based strategies. Finally, even if you're
not afraid of AI and you can understand these algorithms, they
still have a really short track record. Most of the strategies launched really in
the latter half of the last decade and we're not sure if their
performance can be persistent yet. To illustrate this, let's compare the
performance of AI based hedge funds with human-based funds and the broader market. This graph uses data from
Eureka Hedges hedge fund performance indices from December
2010 to December 2019. The blue line is the cumulative
performance index of AI based hedge funds. As you can see, indeed, the AI funds
have significantly outperformed the overall hedge fund index,
which is the orange line below. However, the picture is a lot less rosy if
we superimpose the S&P 500 Index to it. As you can see, in fact, both types of funds have
underperformed the broader market. Granted, the AI fund index
is a lot less volatile but depending on which period
you choose is really hard to say which strategies have
a consistently higher sharpe ratio. And this brings us to some predictions
on what might happen in the future for the investment industry in the age
of AI and machine learning. Based on the adoption trends and
limitations, the near future will probably be a period of continued transformation
instead of truly revolutionary changes. Barring any quantum leaps in computing,
pun intended. We'll just continue our current trend. There will be more data available
every day as it has been and we'll have faster computers to
analyze them more efficiently. And we'll continue to bring more
cutting-edge algorithms that are more adapt at analyzing
large-scale financial data. By the same token,
analytics and automation will continue to play a more and more
important role in the investment industry. And the trend of automating away more
cognitive tasks will continue and probably will expand to other
areas in the financial sector. Traditional jobs that are considered,
quote unquote, human centric and safe from automation like sales and
marketing in banks, for example, are no longer that safe. New advancements in natural
language processing for instance, are rapidly enhancing computer's
ability to generate human speech. This has already been applied in many
chatterbox customer service applications and some of you might have
seen those in action. But they're now spreading to even more
advanced areas like writing contracts and generating analysts' reports, threatening
the jobs of some securities lawyers and research analysts alike. And this trend will keep expanding and
we'll see a much bigger role for data analytics, AI, and
machine learning in other investment related peripheral fields such
as regulatory compliance, fraud detection, market research,
and risk management.