forward.

The World Economic Forum’s

Introduction To Future of Jobs 2020 report sees the top
three job fields increasing in demand as first,
Machine Learning data analysts and scientists, second, AI and
ML specialists and third, big data specialists.

For Asset The same report sees AI and ML specialists

as the top emerging job role. i

Management What is artificial intelligence

and machine learning?
Though many people (including us at
Abstract Boosted.ai) use the terms interchangeably,
Artificial intelligence (AI) and machine they are indeed different. Artificial
learning (ML) are related, but separate terms. intelligence (AI) can be viewed as an
They are becoming increasingly present in umbrella term under which more specific
day-to-day life, including finance, but are less definitions of AI like machine learning (ML)
common in investment management. To use fall. Broadly, artificial intelligence is any
AI and ML to manage funds, institutional system that seems intelligent. Alan Turing
investors need a solution that is custom built was one of the earliest pioneers of AI all the
for finance specifically. AI and ML do not way back to the 1940s, well before
have to mean unfamiliar positions, increased computers could even function all that well –
turnover or unexplainable decisions. they could cost up to $200,000 a month to
lease and could not store commands. ii
Artificial intelligence and Today, almost all systems can be deemed
machine learning are intelligent, like rain sensing windshield
wipers, automatic light switches, and WiFi
everywhere connected appliances. Machine learning is a
People interact with artificial intelligence (AI) subset of artificial intelligence with a much
every day, but the term can still feel like higher bar to pass. To qualify as machine
science fiction. AI is very much a technology learning, a system must be able to not only
in use today, with very real-life applications. appear intelligent, but it must also be able to
Consider voice activated personal assistants learn and further still, adapt what it learns to
like Siri, Alexa and Cortana, self-driving cars, change how it operates. Our proprietary
and recommendation engines like on algorithms at Boosted.ai are machine
Amazon and Netflix. They all make use of AI learning, combining large swaths of data to
and machine learning (more on the learn and find non-intuitive links between
difference between the two coming up!) and equities to sort winning stocks from losing
are in use by humans daily. ones.
Indeed, the use of AI and machine learning
(ML) is no longer becoming a novel idea for
high-tech companies, it is a must-have for
any business looking to innovate going

Toronto | New York www.boosted.ai

How is AI and ML being used (combining fundamental experience with
quantitative techniques) being the future of
in investment management asset management.
today? Pioneers in using AI for investment
The finance industry is already making use management are seeing results. The
of AI and ML. Credit agencies are using Eurekahedge AI Index, an equally weighted
machine learning models to create risk index of funds that measures the
assessments, banks are using ML to provide performance of managers using AI and ML
transparency in their underwriting systems in their process saw return of 11.24% in 2020
and security arms are using ML to detect vi
against the Eurekahedge 50, an index of
fraud in real time. select large hedge funds, which saw 2020
Although AI and ML have already return of 4.94%. vii
transformed the world and will continue to, If it is so straightforward, why aren’t more
investment managers have been slower to managers using AI and ML today, then?
adopt it. A survey of portfolio managers by Implementing AI and ML strategies for
the CFA Institute indicated that only 10% of investment management can be more
PMs had used AI/ML techniques to improve difficult than it initially appears. Our CEO
their investment process in the last 12 Joshua Pantony puts it like this: “The current
months. iii state of AI is excellent at solving a lot of
Some worry that AI or ML will disrupt jobs. problems across different industry verticals.
Though that is likely true for some segments In a computer vision problem, for example,
of industry (for example, robotics systems identifying a cat with 95% accuracy out of an
that can draw blood or help nurses lift array of images would be considered a
patients), within investment management, success. In investment management,
many thought leaders see the best results however, with very noisy data and time
when combining human intelligence with sensitive decisions, where getting it right
artificial intelligence. In their report Smart even 55% of the time makes you a superstar,
money: AI transition from fad to future of the challenge can be more difficult.”
institutional investing, PwC says: “AI will not
supplant the need for investment firms and Roadblocks to implementing
the people who are a part of them. If AI and ML for investment
anything, it makes people even more
important, guiding AI-based processes and managers
increasing profitability with techniques that At its core, machine learning is a strategy to
will scale.” iv The Harvard Business Review incorporate as much data as possible into
says, “The bottom line is that while ML can decision making. For investment managers,
greatly improve the quality of data analysis, it starting to use ML solutions to complement
cannot replace human judgment. To utilize their portfolio managing process will be a
these new tools effectively, asset non-negotiable as pressures on the industry
management firms will need computers and like fee compression, shifting investor
humans to play complementary roles “. v We demand and other technological advances
feel strongly about quantamental investing march forward. However, data in finance can
be difficult to incorporate correctly, which is
2

Toronto | New York www.boosted.ai

why it is important that any AI or ML solution models – three standard machine learning
institutional investors opt to use must be styles (linear regression, random forest and
finance specific. generalized boosting) and three types built
specifically to handle the rigours of capital
Machine learning is only as good as the data
markets data. Those are:
it is provided. In capital markets data, there
are a number of possible issues that could - Combination Feature Tester: This is
arise to make ML models learn bad lessons a great way for both fundamental
about which stocks are winners and which asset managers and quants to test
are losers. Just some of these are: theories. It trains quickly and is
robust to noisy data. It can filter out
• survivorship bias, where it may learn bad features and finding some non-
that only stocks that “survive” are linear connections between those
winners, forgetting about delisted inputs. The results are not as
stocks accurate as our other, more vigorous
• forward looking data, where financial methods, but it is a quick and cheap
data may be amended to prior dates method to test if a new set of
(say a company revises their features has predictive power.
earnings estimates), so the model
learns based on information it
wouldn’t have had access to at the
time
• unnormalized data, where it is
challenging to compare equities
when, say, a market cap of AAPL is
not comparable to the market cap of
a small-cap stock, but there could be
information to be gleaned about the
two companies if their values were Our Combination Feature Tester model
normalized type within Boosted Insights
In order to successfully implement an AI and - Battle Royale: This model type is
ML strategy, investment management firms robust to noisy data as well. It can
must adopt technology that is specific to filter out bad feature and finding non-
their industry. linear connections between inputs
(even more so than Combination
The Boosted.ai solution Feature Tester above). In Battle
Our finance-specific algorithms account for Royale, every company is compared
the type of errors one can get when trying to to every other company in the user’s
apply artificial intelligence and machine universe. Companies are then ranked
learning to investment management. by the number of individual battles
they win. The results tend to be more
Our proprietary model types go far beyond
accurate than Combination Feature
linear regression style stock picking. Within
Tester at the cost of additional
Boosted Insights, we offer six types of
training time (in a universe like the
3

Toronto | New York www.boosted.ai

 S&P 500, it is 250,000 comparisons
per feature). In addition to these model types, built
specifically to account for the noisy data in
finance, Boosted Insights has continued to
specialize in machine learning for capital
markets. We have built out portfolio
optimizations, data cleansing, custom data
uploads, and other client requests to stay
ahead of the curve in the AI revolution in
investment management.

How to get started using AI

Our Battle Royale model type within
Boosted Insights and ML for asset management
There are a number of ways for institutional
investors to begin testing machine learning.
- Battle Royale Noise Reduction: This Three of the most frequent concerns we
model type is our most advanced hear are:
proprietary algorithm. It creates a
number of different classifiers by - that machine learning will add
training each with a subset of data unfamiliar positions to an investment
manager’s portfolio
and features. Each individual
classifier analyzes its data and - that machine learning will increase
turnover too dramatically
ranking holistically. It tends to
- that machine learning is too opaque
produce better results than our Battle
and that investment managers need
Royale model type in situations
to be able to explain all investment
where there is a large, noisy feature
decisions to their stakeholders
space or if overfitting is a concern. It
is best used when you are inputting a We had these concerns top of mind in
large number of uncertain features, creating our finance-specific solution for
where you are not sure if all of them asset managers.
have predictive power.
- In our white paper on machine
learning overlays, we added 232 bps
of alpha across different funds
without adding or subtracting a
single position, only through position
sizing.
- We have resources for how a fund
manager can keep turnover low while
improving turnover here. We show
that using our optimizers can keep
Our Battle Royale Noise Reduction model return similar while dropping turnover
type within Boosted Insights to under 100% per year.

Toronto | New York www.boosted.ai

 - Our machine learning is fully to use finance-specific AI to combat the data
explainable, with a list of features problems in capital markets data. Seeing
that influenced the machine’s results is not as simple as turning on an AI
predictions. Transparent AI is switch and leaving it be – the most
important to us and we continue to successful asset managers to use AI and ML
improve our product offering – our will see it as an amalgam of the best of the
feature importance update is another portfolio manager’s capital markets
step towards deepening a fund expertise and the machine’s advanced
manager’s understanding of why our processing capability and ability to spot non-
machine learning makes its intuitive links. Because concerns about
predictions. unfamiliar positions, increased turnover and
unexplainable positions are common among
Takeaways investor stakeholders, it will also be
Artificial intelligence and machine learning important for an asset manager to ensure
have already transformed the world. their AI and ML solution meets those needs.
Investment management is next, but its
Davison Westmoreland, Director of
adoption is still in its early stages. It will be
Customer Success
necessary for firms to adapt to this new
world, but for early adopting tech innovators, dw@boosted.ai
it is a chance to get in early. Firms will have

References

i http://www3.weforum.org/docs/WEF_Future_of_Jobs_2020.pdf
ii https://sitn.hms.harvard.edu/flash/2017/history-artificial-intelligence/
iii https://www.cfainstitute.org/-/media/documents/survey/AI-Pioneers-in-Investment-Management.ashx
iv https://www.pwc.com/us/en/industries/financial-services/library/artificial-intelligence-investing.html
v https://hbr.org/2019/12/what-machine-learning-will-mean-for-asset-managers
vi https://www.eurekahedge.com/Indices/IndexView/Eurekahedge/683/Eurekahedge-AI-Hedge-fund-Index
vii https://www.eurekahedge.com/Indices/IndexView/Eurekahedge/639/Eurekahedge-50

Toronto | New York www.boosted.ai