海角社区

Who would you choose to manage your investments: an AI or a human?

Recent research suggests the smart answer to that question might be both. An award-winning paper from Junbo Wang, associate professor in the 海角社区 Department of Finance, coauthored with Sean Cao (University of Maryland), Wei Jiang (Emory University), and Baozhong Yang (Georgia State), demonstrated that combining human expertise and machine power may be the future of skilled labor, using the financial industry as a test case. Their work was recognized by the Journal of Financial Economics, receiving the Fama-DFA Best Paper Award in Capital Markets and Asset Pricing.

鈥淲e initiated this research driven by longstanding human curiosity about whether artificial intelligence will eventually surpass human capabilities, particularly in highly skilled professional tasks,鈥� Wang said. 鈥淥ur collaboration was sparked by our shared interest in assessing how humans can effectively work with, rather than against, AI to achieve optimal results.鈥�

 

How They Did It: DIY AI

The researchers chose the world of stock analysis as a proving ground for their experiment. Stock analysis requires precise analytical skills and intuition to develop decades of detailed forecasts and market outcomes. This type of financial output provides a clear metric to compare human and AI performance.

Once they had their arena, the researchers built their competitor: a custom AI stock analyst. The AI analyst was trained from the ground up and given access to the same array of public information a human analyst would use up until the moment they made their forecast.  This included firm-level data, macroeconomic indicators, corporate disclosures, and news articles. The AI鈥檚 12-month stock return predictions were then compared against those made by human analysts for the same stock at the same time.

鈥淲e opted to create our own AI analyst because it granted us maximum control and transparency over the process,鈥� Wang noted. 鈥淓xisting commercial AI solutions often have proprietary restrictions that limit understanding and customization 鈥� by constructing our AI analyst from scratch, we could precisely control the data, clearly analyze each input's impact, and [understand] both the strengths and weaknesses of our model.鈥�

 

Together Is Better

Over an 18-year historical sample period, the AI analyst made more accurate predictions than human analysts in approximately 54.5% of cases. However, the study did not just stop at a simple competition; it also sought to understand where each side holds a competitive advantage. The researchers found that human analysts have the edge where context is king. They performed better when judging situations that require reading between the lines, such as smaller companies, firms whose value lies in concepts and brands rather than physical assets, or businesses facing financial distress. Conversely, the AI excelled at what machines do best: processing massive volumes of data from large, established corporations with long-standing public track records.

With these strengths and weaknesses understood, the two sides were combined into a final "Man + Machine" model, where the human analyst鈥檚 forecasts were made available to the AI analyst. This "centaur analyst" further outperformed the AI-only model in nearly 55% of forecasts. Perhaps most importantly for real-world application, the hybrid model dramatically reduced the frequency of extreme prediction errors, avoiding approximately 90% of those made by historical human analysts and approximately 43% of those made by the AI alone.

鈥淚nitially, we anticipated incremental improvements when combining human judgment and AI-driven insights. However, the significant reduction in extreme errors was genuinely remarkable,鈥� Wang said. 鈥淭his outcome underscored the complementary nature of human and AI capabilities.鈥�

 

What Does This Mean for the Future of Work?

鈥� Our findings encourage us to delve deeper into understanding how [this] collaboration can be optimized 鈥� Exploring how education and professional training programs can better prepare workers to collaborate effectively with AI is essential for realizing the full potential of such hybrid approaches. 鈥�

In the finance industry, we can expect human advisors to continue playing a crucial role, while AI provides useful and timely assistance with tasks such as data processing. This hybrid approach will provide investors with timely insights and allow professionals to concentrate more on strategy and personalized guidance.

Applied more broadly, the study provides a blueprint for how skilled professionals can adapt and thrive in the new age of AI. The future, Wang and his colleagues suggest, belongs not to the lone human expert or the isolated AI, but to the professional who can effectively combine and wield each component鈥檚 strength. Instead of making human judgment obsolete, technology can amplify its importance, creating a partnership where the whole is far greater than the sum of its parts, particularly in complex tasks or risky environments. This also presents a new, critical mandate for universities and business schools: to train a generation of leaders not just in their chosen field, but in the art of understanding and collaborating with artificial intelligence.

Want to dive into the full research and data?

 

About the Researcher

Junbo Wang is an associate professor in the 海角社区 Department of Finance, where he has been a faculty member since 2015. He also serves as an Associate Editor of The Financial Review. His research focuses on financial econometrics and machine learning methods applied to asset pricing tests, the impact of artificial intelligence on firm valuation and labor markets, investor trading activities and their market impact, mutual fund performance evaluation, and option and volatility risk premia. In addition to his research, Wang is passionate about teaching and mentoring. He has taught a wide range of finance courses at the PhD, MBA, and undergraduate levels, earning multiple teaching awards. He also actively contributes to the profession through program committees, conference organizing, and refereeing for leading journals.