Artificial Intelligence (AI) is rapidly becoming an integral part of our modern world, transforming industries and shaping our daily lives. Its growing importance lies in its ability to enhance efficiency, productivity, and decision-making across various industries, from healthcare and finance to transportation and entertainment. There is also concern that AI may displace humans in the work that they do- including in the financial sector. Many believe that AI can already make better decisions when it comes to stock predictions, but is that really the case? During the Inquire Europe autumn seminar Wei Jiang presented her research ‘From Man vs. Machine to Man + Machine: The Art and AI of Stock Analyses’ which addresses these issues.
For those who were unable to attend her presentation, a summary has been made:
“Everyone has heard about the legendary chess game when the IBM supercomputer, Deep Blue, beat world chess champion Garry Kasparov. What is less known about the story is that this game played a crucial role in inspiring figures like Kasparov to lead the way in developing the idea of Man + Machine matches. In these matches, a human chess player, helped by AI (referred to as a “centaur” player), faces off against artificial intelligence. Remarkably, to this day, the centaur player has consistently outperformed machines. What’s even more encouraging is that the availability of affordable AI-powered chess programs has led to a growing number of highly skilled human chess players.
This, in many ways, represents the notion of our research. This study explores the profession of stock analysis, who play a critical role as information intermediaries and have also been impacted by AI’s ability to make fast and powerful predictions at a low cost.
To bridge the transition from “Man vs. Machine” to “Man + Machine,” we developed our own AI model for 12-month stock predictions. The AI model focuses on target prices, as earnings forecasts can be influenced by managerial discretion. The AI analyst was trained on current machine-learning tools using publicly available data and textual information from firms’ disclosures, excluding analyst forecasts.
In the study we found that the AI analyst outperforms human analysts in 55.9% of target price predictions. This advantage could be due to the AI’s superior information processing and lack of human biases. Even when human analyst forecasts are “debiased” using machine learning, AI still outperforms them in 48.2% of cases, suggesting that bias correction explains a significant portion of the Man-Machine performance gap. AI especially has an advantage when firms are complex, and when information is highly-dimensional, transparent, and voluminous.
However, the study also identifies scenarios in which human analysts outperform AI. This occurs for smaller, more illiquid firms and those with asset-light business models. Human analysts tend to remain competitive when critical information requires institutional knowledge, and when the situation is unusual. Moreover, the edge of AI over human analysts declines over time, as analysts gain access to alternative data and to in-house AI resources.
We conclude that human and AI can complement each other, creating a “Man + Machine” model that outperforms the AI-only model. This partnership helps avoid extreme errors and can be beneficial in various skilled professions, including for stock analysts. Finally, the study shows that analysts using AI-processed alternative data can improve their performance, particularly when affiliated with companies with strong AI capabilities”.
Inquire Europe members can access the research and the presentation slides via: https://www.inquire-europe.org/event/autumn-seminar-2023/