Overview and details of the sessions of this conference. Please select a date or room to show only sessions at that day or location.
Please select "Show Presentations" and "List View" or a single session for detailed view (with abstracts, discussants, downloads). PDF downloads are accessible to registered participants only (after Login).
Ranking concerns or reference points: The impact of communicating expected payoffs in experimental studies
Sebastian Krull, Matthias Pelster
Universität Duisburg-Essen, Germany
Discussant: Lars Hornuf (Dresden University of Technology)
Online platforms such as CloudResearch, Amazon MTurk, or Prolific require researchers
to communicate average expected payoffs to participants prior to experiments. We show
that knowledge of expected payoffs introduce confounding effects in experimental studies
on risk-taking. We use the literature that shows that individuals with lower ranks take
higher risks as a playground. Our experimental design disentangles this ranking effect
from a reference point effect introduced by the average expected payoff. Holding the rank
constant, risk-taking is 18.23% higher below the reference point on average. Holding the
reference point distance constant, ranks have no significant effect on risk-taking. These
results are robust to nonsocial settings and alternative risk-taking measures.
11:30am - 12:00pm
Will AI replace or enhance human intelligence in investment management?
Vikram Nanda1, Sanghyun Hugh Kim2
1University of Texas at Dallas, United States of America; 2Wilfrid Laurier University, Canada
Discussant: Sebastian Krull (Universität Duisburg-Essen)
Using unique data from LinkedIn profiles, we measure the adoption of AI technologies by mutual fund managers. Compared to low-AI funds, high-AI funds generate superior returns and incur lower expenses. AI outperformance is particularly strong among discretionary funds, which rely on human judgment. The greater the AI adoption, the more pronounced the time-varying skill of fund
managers across different market conditions. The stock-picking abilities of high-AI funds improve with the availability of big data, such as satellite imagery of parking lots. The local availability of AI skills is a key determinant of cross-sectional variation in mutual fund AI investment. Our findings are robust to using geographic variation in AI supply as an instrument for AI utilization by mutual funds.
12:00pm - 12:30pm
Making GenAI smarter: Evidence from a portfolio allocation experiment
Lars Hornuf1, David Streich2, Niklas Töllich2
1Dresden University of Technology; 2Catholic University Eichstaett-Ingolstadt
Discussant: Vikram Nanda (University of Texas at Dallas)
We evaluate the performance implications of providing various types of domain-specific information to large language models (LLMs) in a simple portfolio allocation task. We compare the recommendations of seven state-of-the-art LLMs in various experimental conditions against a benchmark of professional financial advisors. Our main result is that the provision of domain-specific information does not unambiguously improve recommendations. We find that LLM recommendations underperform recommendations by human financial advisors in the baseline condition. Providing firm-specific information improves historical performance in LLM portfolios and closes the gap to human advisors. Performance improvements are achieved through higher exposure to market risk and not through an increase in mean-variance efficiency within the risky portfolio share. Risk-averse investors are recommended substantially riskier portfolios when firm-specific information is provided.
