Artificial intelligence (AI) and machine learning (ML) are emerging technologies that seem destined to define the future of business operations — including investment management.
The financial services and asset management industries are increasingly shaped by the use of AI and ML solutions. However, despite the widespread attention given to them, there are still many misconceptions surrounding AI and ML, which may be holding firms and fund managers back from leveraging these tools for portfolio performance, risk analysis and customer service.
As more funds and other financial institutions prioritize the use of AI and machine learning, let’s take a closer look at AI and ML, how they’re deployed and what machine learning investment strategies are being used by funds and others.
Defining artificial intelligence and machine learning
Data has become the lifeblood for investment management companies. While data has always been essential to investment decisions — what are price, earnings and P/E but data points? — funds and other market participants now benefit from much higher quality and more robust data. Asset managers can view real-time market data, run complex algorithms to find patterns and consult alternative data.
The tools for processing big data and producing insights have also come a long way: enter AI and machine learning. Investment companies have more computing power than ever to power their data analytics.
However, AI and ML are often erroneously used interchangeably, so it’s important to have a definition for each to understand their utility in investment processes and decisions:
- Artificial intelligence is a branch of computer science that aims to replicate human decision-making capacity with machines. It’s important to also define different AI approaches. Not every AI tool is built to replace humans entirely — something that in theory is called “strong” AI. In fact, most all currently used solutions are examples of “weak” AI, which is AI meant for automation of mundane manual processes. As BlackRock explains, AI works by machines processing inputs through various functions or algorithms to produce an output, or a computer-generated decision. The functions an AI tool uses may be rules-based, mathematical or a combination.
- Machine learning is a subset of AI. The key difference is that whereas AI is programmed with specific functions, ML tools are meant to learn from the data they process and, through trial-and-error, find deeper, more relevant insights without being explicitly programmed to do so. While machine learning programs need to be built from something, as Investopedia explains, “when new or additional data becomes available, the algorithm automatically adjusts the parameters to check for a pattern change.”
In effect, ML systems “learn” over time as they process ever more data. This is the crucial factor that separates ML from AI in general — and which makes the former so intriguing for real-life portfolio investment and investment decision applications.
As JPMorgan summarizes, it’s this core differentiator that makes machine learning so ripe and advantageous in portfolio management. By virtue of processing such enormously large and diverse data sets in a fraction of the time it would take humans, ML tools can help managers, analysts and traders:
- Build more accurate predictive models and uncover predictive signals with greater efficiency.
- Continuously monitor markets for change — like proactively identifying risks or volatility.
- Adapt trading strategies and techniques.
How can ML be used in investment management?
Machine learning investment strategies are gaining greater buy-in as more funds and firms adopt AI and ML for investment decision-making and asset management, among other functions.
At a high level, there are four functions of asset management in which AI and machine learning, specifically, can have value. According to a post from FirmAI, an online AI research resource and advocacy group, these four domains include:
- Asset price predictions — forecasting future prices using an ML methodology that can be used to power factor trading strategies, among other approaches.
- Event prediction — predicting “hard” and “soft” financial events such as mergers and acquisitions activity, corporate defaults and leadership changes.
- Value estimation — estimating value and factors beyond price.
- Optimization — constructing portfolios using machine learning techniques to optimize allocations, weightings and sizings.
The article also went on to mathematically describe various quantitative trading strategies that are informed by machine learning, including one for trading the VIX and another that leverages earnings call surprises.
To get an even better understanding of the practical potential in ML tools, let’s look at some typical portfolio management use cases:
Analyzing unstructured/alternative data
Market data is structured data: i.e., it can be easily ordered, organized, formatted and searched — like price. You can go back and search the trading price of a security at exact points in history with relative ease. However, fund managers increasingly in search of an edge are turning to alternative sources of data, which are largely unstructured. A 2020 Refinitiv survey found just 3% of financial firms do not use alternative data, which is significantly down from the 30% who said the same in 2018. Alternative data sources include satellite imagery, news items, central bank monetary policy announcements, social media trends and earning call transcripts. Such datasets are so expansive and unstructured as to be impossible to parse using manual efforts. Finding patterns in unstructured data is much easier with machine learning tools, allowing funds to leverage deeper insights that expose market, performance and price drivers that fly under the radar. For example, ML tools can analyze data from satellite imagery that gives fund managers insight into crop yields or consumer retail sentiment.
One of the leading machine learning investment strategies is leveraging such tools to generate alpha from diverse streams and with more consistency. In short, alpha investing is focused on generating returns that beat the market using active asset management methods. Trading strategies seeking alpha can be more efficiently and effectively implemented with the help of ML tools, JPMorgan says: “ML systems decipher change and can even adapt the time frames of their measurements and price predictions to potentially enhance alpha generation across different market environments.”
Uncovering arbitrage opportunities
Arbitrage trading strategies are built around the approach of exploiting small discrepancies in the margins to drive large-scale returns. Essentially, arbitrage involves the concurrent buying and selling of assets across markets, profiting off market inefficiencies that allow traders to sell an asset in one market at a price and buy the same asset in a different market at a different price. The value of machine learning techniques to such a trading methodology is clear: Humans only have so much time, attention and effort they can dedicate to scouring markets for arbitrage opportunities. The process of identifying such openings can be markedly reduced and streamlined with the help of ML tools.
Mitigating human bias
In the end, some trades or investment decisions may be made on gut instincts or emotion. According to Harvard Business Review, machine learning tools can be relied upon to reduce biases, to a degree, or bring them to light: “ML can be employed to interrogate the historical trading record of portfolio managers and analyst teams to search for patterns manifesting these biases. Individuals can then double check investment decisions fitting into these unhelpful patterns.” Asset managers can depend on machine learning systems to provide objective intelligence or predictive analytics to make optimal investment decisions. Such automation tools could be particularly useful amid crisis or extreme volatility when a rapid allocation reweighting may be needed.
Who uses AI/ML in investment management?
Despite the clear advantages of AI and ML, adoption of such tools is still relatively low in investment management, even if the future points to greater uptake.
A 2020 survey CFA Institute members from across the globe found that of those in investment management (more than half of the respondent pool), 49% did not use AI in any respect. The CFA Institute reasoned that the low number occurred “likely because [investment management] respondents are mainly portfolio managers who focus on asset allocation, stock selection, and portfolio construction tasks rather than risk management.”
However, as we’ve seen, ML tools can indeed be used to solve problems or otherwise optimize and streamline operations in these areas of asset management. Of investment management respondents:
- 14% were investigating the use of AI tools.
- 7% were in the process of implementing tools.
- 16% were already using AI in a limited capacity.
- 3% had achieved widespread use of AI.
- 12% were unsure about their use or plans.
In comparison, just 32% of risk management respondents said they didn’t use AI and indicated they were more likely to be investigating use cases (23%) and already making limited or widespread use of AI (28%).
Even though adoption and current use of advanced tools is low, there is substantial room for growth. In a 2020 FactSet survey of asset managers and executives from investment management firms, just one-third of executives said they were happy with their technology strategy. As a result, 75% said they need to invest more in technology. That’s a viewpoint largely present across the asset management industry, with just one-third of asset management firms using advanced tech like AI, ML and natural language processing (NLP).
The survey from Refinitive found that as appetite for AI and Ml tools grows among financial firms, the barriers to implementation are also receding, creating a hospitable business environment for investment and deployment of such tools. In 2018, 33% of respondents agreed finding the right data talent was a considerable barrier to their adoption of AI and machine learning; but in 2020, that number dropped to 23%. The same trend was seen in other barriers including:
- Funding for technology investment (cited by 38% as a top challenge in 2018 vs. 22% in 2020).
- Choices for technology tools and platforms (30% vs. 21%).
However, the survey did find an uptick in the share of respondents who saw poor data quality (43% vs. 54%) and data availability (38% vs. 45%) as barriers.
AI/ML and the future of finance
One other interesting finding from the Refinitiv survey was that COVID-19 is likely to accelerate investment in technology and adoption of AI/ML in the financial sector: 30% of those surveyed said the pandemic and its impacts made AI and Ml more important.
Yet no matter the value that ML tools can deliver funds, implementation of machine learning investment strategies requires a strategic approach to technology and resources dedicated to keeping ML models current and accurate.
Although Refinitiv survey respondents said COVID-19 made AI/ML more important to financial organizations:
- 26% said models need updating.
- 20% said models need to be made more dynamic to reflect change.
- 17% said AI and machine learning techniques/processes need to be changed.
There are many considerations, and at the top of the list, perhaps, is data quality — the No. 1 barrier to effective use of ML. As a best practice, BlackRock recommends that “all data inputs should be robustly tested to ensure models are performing analysis on accurate data sets, and
periodic review procedures should be in place to ensure that no investment process is out of date.” This is because ML models themselves may contain biases, depending on the quality of data used. As a result, data scientist will become a fundamental role in the operation of any fund, handling the inputs and analyses so that portfolio managers can make meaningful use of the insights derived.
Other issues to be aware of, as raised by JPMorgan, include:
- Overfitting: There is such a thing as fine-tuning a model too far. Overfitting is a modeling error whereby an ML function produces an outcome that responds too closely with the inputs. That is, the model finds past patterns too well to the extent that it does not take into account future uncertainty.
- Shock: Machine learning systems must learn. So when they encounter an unprecedented event or scenario, there may be a learning curve for the machine.
The solution for both problems, in many cases, is more robust training data being fed to the model. For example, running a simulation that produces a probabilistic distribution of outcomes can help keep an ML model inclusive of future uncertainty in projections for asset prices.
How Magma approaches dynamic investing
At the heart of machine learning investment strategies is the ambition to do more with less, to trade on the cutting edge. Magma Capital Funds embodies this vision of dynamic investing with how we treat risk and volatility — i.e., as useful signals rather than alarm bells.
Want to learn more about how Magma invests? Contact us today.
This article is for informational and educational purposes only and is not an offer to sell or a solicitation of an offer to buy any securities or other instruments through Magma or by any other means. The information contained herein is not intended to provide, and should not be relied upon for investment, accounting, legal or tax advice. This blog post does not purport to advise you personally concerning the nature, potential, value or suitability of any particular sector, geographic region, security, portfolio of securities, transaction, investment strategy or other matter. No consideration has been given to the specific investment needs or risk-tolerances of any recipient as part of this publication. The recipient is reminded that an investment in any security is subject to a number of risks including the risk of a total loss of capital, and that discussion herein does not contain a list or description of such relevant risk factors. Please be advised that past performance is no guarantee of future results. The recipient hereof should make an independent investigation of the information described herein, including consulting its own tax, legal, accounting and other advisors about the matters discussed herein. This report does not constitute any form of invitation or inducement by Magma to engage in investment activity or to invest in any Magma product offerings.