Machine Learning in

Investment Management
October 2021

Having introduced concepts in artificial intelligence and different types of

machine learning (ML) models in our previous post, we will now discuss specific
applications of machine learning in investment management, the benefits of Authors
using these models, and the challenges inherent in their use. (Chart 1) Serena Li
Quantitative Associate,
Applying Machine Learning to the Quant Investment Process Systematic Global Equities
Chart
1
Traditional and ML quant investment teams
Christina Watson
may add or change data based on results Equity Intern,
Girls Who Invest Scholar
ML model adapts to new ML model adapts based on results of predictions
data and reassess existing
data in response to Sumali Sanyal, CFA
market conditions Managing Director, Senior Portfolio
Manager, Systematic Global Equities
PREDICTIVE TRADING
DATA RESULTS
MODEL DECISION Cameron McLennan, CFA
Director, Portfolio Manager,
ML used for ML used for portfolio Systematic Global Equities
data preparation construction and execution
Traditional quantinvestment process Elements of the process for which machine learning may be used

Source: J.P. Morgan Asset Management

A. Applications of Machine Learning

in Investment Management
1. Identify and analyze new forms of data
It has become increasingly difficult to generate alpha using traditional
investment strategies over the past decades. One of the reasons given for
the lack of alpha generated by these strategies, is that they rely on the same
financial data and investment factors in their construction. Realizing this, some
Chart
investment Applying
managers Machine Learning
have begun tomachine
using the Quant Investment
learning Process
algorithms to
1
identify, analyze and incorporate alternative data sources into their investment
Traditional and ML quant investment teams may add or change data based on results
strategies. Natural Language Processing (NLP) can be used to extract
ML model adapts to new ML model adapts based on results of predictions
meaningful insightsdata from news articles, social media, earnings calls etc. For
and reassess existing
example, NLP algorithms have
data been used
in response to to enhance the quantity and quality
market conditions
of data available to Environmental, Social and Governance (ESG) investors. In this
case, NLP algorithms can extract unique information from public documents
or earnings calls that companies have.1 This typePREDICTIVE TRADING
of linguistic information can
DATA RESULTS
MODEL DECISION
be particularly important for ESG investing considering the limited amount
of ESG related information most companies currently disclose in their public
ML used for ML used for portfolio
filings.data
ESG preparation
rating organizations and other data accumulators such as TruValue
construction and execution
Labs are using NLP, data scraping and other techniques to incorporate more
Traditional quantinvestment process Elements of the process for which machine learning may be used
timely company specific ESG related news and announcements, particularly
Source: J.P. Morgan Asset Management
relating to issues and controversies, to construct ESG metrics.2 Investors can use
NLP to dive deeper into third-party social media platforms, such as Yelp and
Indeed websites, to obtain more unbiased, real-time, and actionable insights on
companies. ML algorithms can also be applied to conduct sentiment analysis
and gain insights into the current environment facing a company or the real
take that management teams have on their companies’ future prospects.3

Philadelphia, PA | Durham, NC [email protected] | xponance.com 1

Machine Learning in Investment Management

2. Identify unique patterns and insights in models

Traditional fundamental factors from categories such as Value, Quality, and Momentum, and the linear
relationships between them have been the basis for quantitative investment models for several decades.
However, there is a high possibility for alpha decay as a result of crowded trades derived from using these
factors and the relationships between them. Therefore, if investment managers wish to add alpha, it is
important that they test alternative, more advanced factor research and model construction techniques.
With that in mind, investment managers are now using ML algorithms such as Generalized Additive
Models (GAM), Decision Tree models and Deep Learning to capture non-linear relationships between
factors to improve the predictive ability of their models.

Machine Learning methods are now broadly used in factor research and to develop better models for
the purposes of stock selection, asset allocation, and risk calculation. ML models can dynamically learn
changing relationships between factors and predictors and handle multicollinearity among them.
These algorithms can also take macroeconomic variables into account to detect market regime shifts.4
For example, multifactor models like Hidden Markov Models (HMM) classify market regimes using
sophisticated time-series ML algorithms.5

3. Improve the accuracy and efficiency of trading

ML algorithms are particularly effective in optimizing trading execution. Traditionally, traders manually
collected statistics to predict liquidity patterns and trading costs that were then used for trading
decisions. ML algorithms can capture the microstructure of the market in a much quicker and more
accurate way as well as efficiently process massive amounts of data for trade pattern analysis. More
importantly, using scenario analysis, these techniques allow traders to evaluate the risks related to various
trading activities. ML techniques can be used to build algorithmic trading models that reduce errors
caused by human judgment and improve trading efficiency.

Chart Machine Learning Application Process

2
Data Feature Model Optimization
Collection Engineering Design

1 3 5 7

2 4 6 8

Data Data Training Deployment

Processing Labelling & Monitoring

Source: Machine Learning Big Eight A.W. CHUNG

4. Other applications
Aside from their direct applications in investment activities, ML techniques can be used to analyze client
behavior to improve support services and resolve queries. By analyzing current market trends and web
activity responses to past advertising, sales and marketing departments can produce more effective strat-
egies to attract new clients and increase the retention of current clients. ML can also increase operational
efficiency by streamlining process automation for repetitive tasks, thus reducing costs for a business.

Philadelphia, PA | Durham, NC [email protected] | xponance.com 2

Machine Learning in Investment Management

B. Challenges of Machine Learning Techniques

Incorporating ML techniques into their factor research and model building processes is viewed by
investment managers as an opportunity for them to outperform competitors and improve returns.
However, the use of ML techniques in an investment strategy does not guarantee strong future
performance. It is important to highlight some of the challenges and limitations involved in integrating
these algorithms into the investment management process.

1. Data quality
While ML algorithms are capable of analyzing much larger amounts of data than traditional models,
there are data challenges when applying advanced techniques to investment strategies. The nature
of financial data is mostly non-stationary and messy.6 Unlike collecting user data for customized ads
on Instagram, stock market data is dynamic and volatile from one period to the next. The market has
changed significantly in the past 20 years, due in part to the increase in popularity of passive and active
quantitative strategies versus traditional fundamental investment strategies. Thus, simply feeding
time-series data into a model may not yield ideal results.7 ML models require clean and well-processed
data to generate reliable predictions. The labor intensive and time-consuming data processing step is a
particularly important aspect of building ML strategies. In addition, the experience and expertise of the
quantitative investment team is crucial in this stage for making judgments on how to process the data.

2. Overfitting and data mining bias

One of the issues highlighted with using ML algorithms in investment management is that while they
may show strong historical results using training data, they may not have strong predictive ability using
new data. This problem is known as overfitting. Machine Learning models have bias-variance trade-off.
Bias refers to the difference between prediction of the model and the actual value the model tries to
predict. A model with high bias always leads to high error on both training and test data.8 Variance refers
to the variability or sensitivity of the prediction i.e., how the model can generalize the predictors using new
data. Thus, more complex models usually come with higher variance while simple models result in higher
bias given the difficulty in accurately estimating results through the assumption of a simple relationship.

Overfitting occurs when a model fits too well during back tests and captures the noise. Specifically,
overfitting occurs when adopting over-complicated models, where the models often have low bias but
high variance. As the models sometimes extract noise instead of true signals, it is frequently referred to as
data-mining bias.9 Thus, it could be challenging to build a robust model for various investment strategies.

Chart Underfitting, Optimum, and Overfitting

3

Underfit Optimum Overfit

(high bias) (high variance)

High training error Low training error Low training error

High test error Low test error High test error

Source: IBM Cloud Learn Hub

Philadelphia, PA | Durham, NC [email protected] | xponance.com 3

Machine Learning in Investment Management

Overfitting can be addressed in several different ways. First, in the data collection step, it is crucial to
obtain good quality financial data. Second, it is important to build a rigorous framework in the back
testing process, such as cross-validation or applying regularization to penalize the complexity of the
models. Lastly, adopting feature engineering and ensemble methods can produce stabler and cleaner
outputs. To avoid data mining and obtain true insights from ML models, it is important to keep in mind
that the value of ML applications lies in the investment philosophies and theories quantitative investors
want to express instead of the algorithms themselves.10 In other words, these advanced techniques should
explain true signals and not the noise.

3. Computing power/infrastructure costs

Another challenge for ML application is that it may take long to both back test the models and deploy
them into production.11 Despite computing power increasing roughly according to Moore’s Law since
the 1970s, more advanced ML algorithms still require a lot of computation and storage capability. The
infrastructure requirements like GPU and high-intensity cores are expensive but essential for both back
testing and production purpose. The data and infrastructure required can make it challenging for smaller
investment firms to develop ML investment strategies. However, these challenges can be overcome by
leveraging open-source resources, third-party vendors, and external technology services such as Amazon
Web Services (AWS).

Machine Learning has become increasingly important in the asset management industry over the
past decade. Managers who embrace these advanced technologies will have compelling opportunities
to outperform in the long-term. BlackRock, the world’s largest asset manager, has established a new
Artificial Intelligence (AI) lab in Palo Alto, California, to focus on the potential uses of AI in the asset
management industry. Numerai, a new type of AI-powered hedge fund, has gathered tens of thousands
of data scientists to create ML algorithms to drive open-source trades.

The key components of successful ML applications are good quality of data, robust models, infrastructure,
and experienced teams. Asset management firms that use more innovative data, find unique alpha, and
incorporate machine learning techniques to enhance their investment processes will be most likely to
excel and adapt in an ever-evolving industry.

Chart AI/ML Investment of Asset Management Firms, Post COVID-19

4

Don't Know Decrease Stay the Same Increase

1%

8%

51%

40%

0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50% 55%
Source: Refinitiv 2020 AI/ML Survey • Base: all respondents (423)

Philadelphia, PA | Durham, NC [email protected] | xponance.com 4

Machine Learning in Investment Management

References:
1 “How Can Ai Help Esg Investing?” Accelerating Progress, www.spglobal.com/en/research-insights/articles/how-can-ai-help-esg-investing.
2 Aroomoogan, Kumesh. “Council Post: How Financial Organizations Can Use Ai to APPLY Esg Standards.” Forbes, Forbes Magazine, 30 Mar. 2021,
www.forbes.com/sites/forbesfinancecouncil/2021/03/30/how-financial-organizations-can-use-ai-to-apply-esg-standards/.
3 “How Financial Institutions Can Deal with Unstructured Data Overload.” InsideBIGDATA, 16 Mar. 2021, insidebigdata.com/2021/03/15/how-financial-
institutions-can-deal-with-unstructured-data-overload/.
4 “Machine Learning in Quant Investing: Revolution or Evolution?” Acadian Asset Management, www.acadian-asset.com/viewpoints/machine-learning-in-
quant-investing-revolution-or-evolution.
5 Jha, Osho. “When to ‘Buy the Dip’.” Medium, Medium, 24 Apr. 2019, medium.com/@oshojha/when-to-buy-the-dip-e2e128d737a7.
6 “Challenges in Applying Machine Learning to Finance.” CFA UK, 10 Feb. 2020, www.cfauk.org/pi-listing/challenges-in-applying-machine-learning-to-
finance#gsc.tab=0.
7 “The Right Way to Use Machine Learning.” Essentia Analytics, 16 Aug. 2019, www.essentia-analytics.com/machine-learning-for-asset-managers/.
8 Singh, Seema. “Understanding the Bias-Variance Tradeoff.” Medium, Towards Data Science, 9 Oct. 2018, towardsdatascience.com/understanding-the-bias-
variance-tradeoff-165e6942b229.
9 Harris, Michael. “Impact of Artificial Intelligence and Machine Learning on Trading and Investing.” Medium, Towards Data Science, 2 June 2021,
towardsdatascience.com/impact-of-artificial-intelligence-and-machine-learning-on-trading-and-investing-7175ef2ad64e.
10 Prado, Marcos Lopez de. Advances in Financial Machine Learning. Wiley, 2018.
11 Shah, Adarsh. “Challenges Deploying Machine Learning Models to Production.” Medium, Towards Data Science, 14 Oct. 2020, towardsdatascience.com/
challenges-deploying-machine-learning-models-to-production-ded3f9009cb3.
This report is neither an offer to sell nor a solicitation to invest in any product offered by Xponance® and should not be considered as investment advice.
This report was prepared for clients and prospective clients of Xponance® and is intended to be used solely by such clients and prospects for educational
and illustrative purposes. The information contained herein is proprietary to Xponance® and may not be duplicated or used for any purpose other than the
educational purpose for which it has been provided. Any unauthorized use, duplication or disclosure of this report is strictly prohibited.
This report is based on information believed to be correct, but is subject to revision. Although the information provided herein has been obtained from sources
which Xponance® believes to be reliable, Xponance® does not guarantee its accuracy, and such information may be incomplete or condensed. Additional
information is available from Xponance® upon request. All performance and other projections are historical and do not guarantee future performance. No
assurance can be given that any particular investment objective or strategy will be achieved at a given time and actual investment results may vary over any
given time.

Philadelphia, PA | Durham, NC [email protected] | xponance.com 5