The Bank for International Settlements (BIS) is launching Project Logos, which aims to enable central banks to observe and analyse the behaviour of large language model (LLM)-based agents acting as portfolio managers in a simulated financial market environment.
The project is a collaboration between the BIS Innovation Hub London Centre and Eurosystem Centre, the Bank of England and the Deutsche Bundesbank.
As applications of artificial intelligence (AI) in financial services continue to grow, common AI infrastructure across the financial sector, such as LLMs, could lead to greater homogeneity in decision-making, notably in financial markets. Additionally, these emerging generations of AI may behave in ways that are difficult to anticipate using existing frameworks.
This project will develop an agent-based modelling (ABM) environment to allow these risks to be explored in the context of portfolio allocation. It will compare portfolio allocation decisions of heuristic (rules-based) and LLM-based asset managers under controlled conditions, with a focus on how LLM-based agents interpret information, allocate capital and respond to constraints over time, and the conditions that might amplify or dampen correlated decision-making.
The project will support central banks in developing practical understanding of how LLM-based agents may behave in traded market settings, and to provide a reusable foundation for future exploratory analysis.
