Artificial Intelligence in Asset Management: Perks and Perils

• Applicants: Professor Dr. Lars Hornuf; Dr. David Streich
• Subject Area: Accounting and Finance
• Term: since 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 552870289

Artificial intelligence in the form of large language models (LLMs) enables technically savvy laypeople to perform complex tasks in areas in which they are not experts. Due to the high demand for financial advice, retirement savings, and asset management, LLMs present a promising area of application. This potential is evident in the early adoption of LLMs by leading asset managers and advisors. The goal of this research project is to identify the potentials and risks for investors and financial advisor resulting from the use of AI-supported assistance systems, and to identify the possible need for regulation. To this end, the project will investigate the quality of the investment recommendations generated by LLMs, the determinants of user acceptance of such systems, and potential risks or impediments to the adoption in the investment context. To assess the potential of LLMs in investment advice, a distinction must be made between the performance of the recommendation itself (e.g., measured as risk-adjusted return) and the effectiveness in changing investor behavior (e.g., measured as weight of advice or the investor's adjustment towards the recommendation). Both aspects are important in evaluating the performance of LLMs. To improve individual investment behavior, an LLM must generate recommendations of sufficient quality, which in turn must be accepted and adopted to a sufficient degree by investors. The project will also investigate two central risks of LLM-based decision-support systems: algorithmic bias and data protection and data security risks. Algorithmic bias can potentially negatively affect the quality and acceptance of the investment recommendations generated by LLMs. In addition, the creation of investment recommendations requires the provision of sensitive data, which potentially negatively impacts user acceptance.

• DFG Programme: Research Grants