AI Trading Agent Representation Homogeneity Risks Market Stability

A new structural model reveals that similarity in how AI trading agents represent market states can cause systemic instability. The paper, arXiv:2604.22818, constructs a multi-agent market model calibrated with high-frequency microstructural moments. AI agents use a two-layer decision architecture: a nonlinear representation layer mapping raw states into high-dimensional feature vectors, and an adaptive linear readout layer generating return forecasts for risk-controlled trading. The model separates representation homogeneity (encoding similarity) from forecast overlap (prediction similarity), which are often conflated. Results show that high representation homogeneity, even without forecast overlap, can destabilize markets by causing correlated actions. The study provides a microfoundation for understanding AI-driven market fragility.