The impact of Artificial Intelligence on portfolio management

Experiencing strong growth over recent years, both in academic research and in the world of business, interest in Artificial Intelligence (AI) has grown exponentially since the launch of ChatGPT-3.5 at the end of last year. With a record adoption rate of 100 million users in less than two months⁽¹⁾ and a disconcerting ease of access, OpenAI's conversational AI system is electrifying investors, and company announcements in this field have been multiplying ever since.

Flashback

 While critical steps are currently being taken in the field of AI, they are based on decades of research and development. The concept of artificial intelligence first appeared in 1950 with the publication of Alan Turing's work⁽²⁾, in which he described his famous eponymous test designed to assess a machine's ability to match the answers a human could formulate. Given the disparity of the techniques implemented since then, defining AI has given rise to very different formulations. To describe the characteristics common to all AI systems, Yann LeCun⁽³⁾, who heads up fundamental AI research at the Meta Platforms group, notes that "artificial intelligence is the ability of a machine to perform tasks generally performed by humans: perceiving, reasoning and acting. It is inseparable from the ability to learn, as observed in living creatures".

Machine learning is often mentioned in conjunction with AI, of which it is a subset. This is the process of using mathematical data models to enable a computer to learn without explicitly programming it. Machine learning uses algorithms to identify patterns in data and develop predictive models.

Acceleration

 Several major developments have fuelled the rise of AI: firstly, the constant increase in the computing and data processing power of computers, which, according to Moore's Law⁽⁴⁾, has doubled every two years since the creation of semiconductors. In addition, the volume and scope of data (Big Data) that can be used to train AI models, as well as storage capacity, have also increased dramatically: 90% of the data currently available worldwide has been generated in the last two years alone⁽⁵⁾.

Already present in many fields, AI will continue to profoundly change most industries and the way we work in the future: 300 million jobs could eventually be impacted worldwide, but productivity gains are expected to generate global economic growth of around $7,000 billion over 10 years⁽⁶⁾.

What impact will this have on the portfolio management industry?

The adoption of AI is clearly not new within the asset management industry, and though it has developed significantly in recent years, it remains, at this stage, confined to certain specific players, mainly among hedge funds, management companies offering quantitative solutions, large research offices and fintechs⁽⁷⁾.  

The fields of application are already numerous:

• Firstly, in portfolio management, AI can be used as part of fundamental analysis. Certain AI techniques are fully in line with quantitative management techniques: based on a large volume of data that may concern the fundamentals of listed companies, the macroeconomic environment or the market environment, machine learning algorithms can identify complex non-linear relationships between the various variables, and detect trends that are not easily identified by humans.
  The predictive analysis models developed in this way are decision-making tools that aim to predict expected returns on different asset classes to generate investment ideas or arbitrage opportunities.
• Textual analysis is another example of the application of AI in fundamental analysis: based on various sources of text, such as corporate earnings reports, central bank press releases or press releases and articles, AI techniques via natural language processing are capable of highlighting economically and financially significant information and providing a quantitative and systematic measure that complements human interpretations.
• Within portfolio management, portfolio construction optimisation methods represent another field of application for AI. One theoretical framework is the modern portfolio theory of Harry Markovitz⁽⁸⁾. He defined the concept of an efficient portfolio: the aim is to optimise the allocation between the various assets making up the portfolio in such a way as to minimise risk for a given expected return.
  Here again, machine learning finds fertile ground, and can help provide better estimates of the data (expected returns, risk, correlation matrix) used in traditional portfolio construction frameworks.
  It can also be used to develop new optimisation approaches that take into consideration complex constraints.
• Risk management in portfolio management activities is another area in which AI can facilitate modelling work, particularly by requiring investors to integrate scenarios that are probable but not "desired" or even considered unrealistic.
• AI can also provide effective solutions for trading activities, where it is essential to reconcile speed and dealing with complexity. The use of AI can facilitate algorithmic trading, so called when at least part of the transaction process is automated, and which today accounts for a significant portion of the volume of transactions handled on financial markets. In the first phase, AI can help identify buy or sell signals, generally on the basis of technical analysis (as opposed to fundamental analysis) from a very large quantity of market data, mainly price trends and volumes traded. AI is then used to automatically establish and activate optimal execution strategies aimed at obtaining the best prices while minimising transaction costs.
• AI has also contributed to the emergence of robo-advisor platforms, which offer digital solutions aimed at providing a tailor-made investment advisory service. These solutions are attracting growing interest from the general public, for whom access to investment advisory services is thus possible, at an affordable rate. AI lies at the heart of the algorithms that analyse market data in real time and make investment recommendations aimed at building a portfolio deemed optimal for a given client, based on their objectives in terms of return, risk profile and any personal constraints. Robo-advisors also integrate market order execution services, as seen above.

More broadly

Beyond applications dedicated to investment, the possibilities for the development of AI within the asset management industry appear vast. But while we intuitively perceive the potential for automating certain tasks at different levels of the operational chain, it remains difficult to fully anticipate the disruptive power of AI, which is likely to give rise to new fields of application as further innovations emerge. 

Despite the technological advances and productivity gains made possible by AI, we must nevertheless be aware of its limitations or the risks it presents for certain dimensions of portfolio management. First of all, AI and machine learning techniques are dependent on the data that feeds the learning algorithms. In addition to the need for a very large volume of data, which is not always available, it is essential that this data is of high quality in terms of updates, accuracy, completeness and representativeness. Otherwise, the results obtained from predictive models lack reliability and robustness.

It is also possible that the algorithms spot irrelevant trends in the data analysed, leading to incorrect deductions. A major risk is that several market operators using the same AI algorithms could make the same mistake simultaneously, or react identically to an event, which could contribute to amplifying market movements, sharply increasing volatility and the possibility of a systemic mini-crash.

Lastly, one of the challenges faced by AI-enabled managers is the potential difficulty of explaining their results, given the complexity of the models used. This is one of the arguments put forward in a report by the European financial markets regulator (ESMA) to explain the low number of European funds currently claiming to use AI in their investment processes: 54 funds (at end-September 2022) out of 22,000 funds domiciled in the European Union⁽⁹⁾.

Today, AI still seems a long way from completely supplanting human beings in the asset management business, for which transparency, a relationship of trust and interaction between clients and management professionals remain essential criteria. However, AI brings new tools that can be used in the value chain, and whose potential could profoundly change the face of the sector.

Let's hope that human beings will be able to limit the downsides and regulate its various issues before it becomes ubiquitous in this field...

Footnotes :

(1) By way of comparison, TikTok reached the 100 million user mark in nine months, and Instagram in 2.5 years / (2) A British mathematician considered one of the founding fathers of computer science. He published "Computing machinery and intelligence" in 1950, contributing to the emergence of Artificial Intelligence. / (3) Author of "Quand la machine apprend. La révolution des neurones artificiels et de l’apprentissage profond". (When the machine learns. The revolution of artificial neurons and deep learning.) Winner of the 2018 Turing Award. / (4) Gordon Moore, one of the founders of Intel, first mentioned this empirical law in 1965. (5) Source: BofA Global Research / (6) Source: Goldman Sachs / (7) Fintech: A fintech is a company that applies technological and digital advances to the financial or banking sector. / (8) An American economist, he published his seminal article "Portfolio Selection" in 1952. / (9) Source: Agefi