We believe artificial intelligence is one of the most exciting and transformative opportunities on the horizon. 

The foundation of computing was built upon the idea of people instructing a computer regarding the processes it should perform. Computers behaved like calculators, crunching data without having the ability to make decisions. In recent years, however, technology has made a giant leap, and computers are increasingly able to learn how to process data and make decisions based on patterns of human behavior. This breakthrough is broadly defined as the science of “artificial intelligence” (AI).¹ 

The notion of adaptation of computing based on what is learned from incoming data is not new. For instance, “artificial neural networks,” a programming technique that loosely mimics the way a brain solves problems with large clusters of biological neurons, has existed for decades. What has changed in recent years is the availability of data, raw computing power and superior modeling techniques³.  As of 2015, approximately 42% of the world’s population had internet access and more than 2 billion smartphones are being used in creating data, which is the raw material for AI⁴. In addition, as seen in the charts below, sophisticated technology and hardware are available at a fraction of the cost compared to just a few years ago, which has allowed for an exponential increase in raw computational power. 

Exhibit 1: Raw compute performance of global supercomputers has increased exponentially since 1993 (measured in GFLOPS, 1 GFLOP = 1 billion floating point operations/sec)⁵ 

Exhibit 2: Price per unit of compute has decreased drastically over time

AI and machine learning can be defined simply as algorithms that identify and act on repeatable persistent patterns in observed data. The observed data can be produced from a routine or recurring transaction like paying credit card bills or other forms of data produced as a result of human behavior such as consumer shopping trends when a product’s pricing changes. Typical steps for running an AI algorithm include understanding the source of data, finding a pattern within the data and forecasting the next move. An AI algorithm moves up the learning curve in its ability to make predictions as more data is fed through the algorithms by the user and additional patterns are understood. When well executed, AI accuracy in prediction increases as the quality and quantity of data increase. The Google search bar autocomplete algorithm, for instance, is based on this concept⁶.

Exhibit 3: Google search bar: Example of an autocompleting algorithm

AI has the potential to be one of the most disruptive forces of the 21st century⁷. Bank of America estimates the current $2 billion AI market will grow to $36 billion by 2020 and $127 billion by 2025. Examples of current applied AI are speech recognition/natural language processing (such as the Apple Siri, Amazon Alexa and Google photo recognition capability), gaming (such as Chess and Go), and self-driving cars; however, AI applications are expanding quickly into many sectors of the economy, including financial markets⁸. In certain tasks, like gaming and photo recognition, AI is already more accurate than humans.⁹ In other areas, like financial markets, AI is quickly gaining steam but still has some work to do. 

Exhibit 4: Estimate of the AI opportunity

Applications for the Financial Markets

Financial firms are positioning themselves to benefit from the use of AI. Some investment firms (called “robo-advisors”) use “rules-based models” that ask for few attributes like age, risk tolerance, goals, etc., to spit out an asset allocation model¹⁰. The case for using AI to make investment decisions centers around the improved speed and efficiency arising from AI’s ability to pore over massive “big data” sets. In addition, there is a potential for cost savings for execution speed and efficiency where AI is able to forecast appropriate conditions (best time, lot size, which liquidity pool, etc.) to execute, for instance. Goldman Sachs predicts the financial industry cost savings from AI implementation can be in the $34 billion to $43 billion per year range by 2025—coming from better investment decisions and quicker reaction to market events¹¹.   

Some of the largest and most sophisticated financial firms are exploring how AI can potentially improve investment decision-making and returns. Many hedge funds are leading the way by hiring some of the brightest minds in the AI field¹². In fact, there are a select number of hedge funds already using AI and machine learning technology as part of their investment strategy. 

A few examples of potential uses of AI in the investment decision-making process are:

• Estimate the impact of news announcements on a company’s stock price
• Analyze and determine patterns in a company’s financial statement releases over time
• Evaluate real-time investor sentiment from sources like Twitter and Facebook to improve timing of trades

While AI algorithms have allowed computers to beat humans repeatedly in chess, it has been harder to make the same case for computers beating the financial markets consistently. CIBC Atlantic Trust’s internal research suggests that AI investing has worked in very short investing time frames (in seconds, minutes, hours) where speed and efficiency of crunching massive amounts of data add value. In medium- to longer-term investing (months and years), AI models have not shown consistent outcomes, possibly because it is harder to predict investor behavior over such time periods. The AI model, by definition, “learns” market behavior from history and predicts an outcome that has some resemblance to the past. This model breaks down when there is a significant departure in market psyche and investor behavior. For instance, market behavior became very defensive after the great financial crisis in 2008. Most AI models would most likely have failed as they would not have “learned” about this type of environment beforehand.¹⁰ In addition, since AI has been a new area of interest for many hedge funds, there is just not enough actual performance data available to prove that AI can succeed in a consistent manner. 

In addition to following investment fundamentals ingrained over the decades, CIBC Atlantic Trust believes it is critical to understand the trends that could fundamentally alter the landscape in the future. That is why the CIBC Atlantic Trust Multi-Manager Investment Team is closely monitoring developments regarding practical uses of artificial intelligence in the financial markets. In a future edition, we will discuss more specifically some of the underlying strategies that managers are employing and our views on which are most likely to succeed.  

Jigar Patel is an investment portfolio manager in CIBC Atlantic Trust Private Wealth Management’s New York office. He leads the hedge fund research efforts within the firm’s Multi-Manager Investment Program (MMIP). With more than 15 years of industry experience, he is responsible for managing the MMIP hedge fund platform, which includes two internally managed fund of funds portfolios and a customized portfolio solutions business.

---------------------------