Anastassia Fedyk (UC Berkeley, Haas): ChatGPT and Perception Biases in Investments: An Experimental Study

Please join us for a seminar of the Advanced Financial Technologies Laboratory (AFTLab):

Time: Thursday, September 26, 2024, 5:00pm-6:00pm

Location: Huang 305

Speaker: Anastassia Fedyk, UC Berkeley, Haas School of Business

Title: ChatGPT and Perception Biases in Investments: An Experimental Study

Abstract: Does generative AI accurately capture demographic heterogeneity in investment preferences? We survey humans and GPT4 side-by-side and show that AI correctly predicts preferences by income, gender, and age (70% correlation). The overlap includes higher stock ratings among men and high earners. Human and AI-generated responses reflect similar reasoning: both refer to "risk" and "return" in general and, for stocks, to "knowledge" and "experiences." However, when not seeded with demographics, algorithmic bias emerges: AI responses predominantly mirror young, high-incomes males. Finally, AI generates more transitive rankings than humans. Our results highlight the promise of generative AI for applications such as robo-advising.

Bio: Anastassia Fedyk is an Assistant Professor of Finance and a Barbara and Gerson Bakar Faculty Fellow  at the Haas School of Business at UC Berkeley. Her research lies at the intersection of behavioral finance and innovation, with a specific emphasis on using big data techniques to understand firm news and valuations. She studies how information from financial news and individual employment records influences asset prices and explores the impact of firms’ investments in technology and skilled human capital. Anastassia holds a Ph.D. in Business Economics from Harvard University and a B.A. in Mathematics with honors from Princeton University. Prior to pursuing her academic career, Anastassia Fedyk was a researcher and portfolio manager at Goldman Sachs Asset Management. Anastassia serves as the Chief Economist at the AI for Good Foundation and is a co-founder of Economists for Ukraine.