FinEvo introduces an ecological game formalism for evaluating multi-agent financial strategies, moving beyond isolated backtests to model trading strategies as adaptive agents interacting in a shared market. The framework integrates heterogeneous ML-based traders—rule-based, deep learning, reinforcement learning, and large language model agents—whose population evolves through three mechanisms: selection (based on forward-looking value estimates), innovation (Dirichlet-distributed mutation maintaining diversity), and environmental perturbation (stochastic shocks). The dynamics are formalized by the FinEvo stochastic differential equation, which unifies these forces and is shown to preserve simplex invariance, positivity, and existence/uniqueness under mild conditions. A macro-level variance decomposition attributes system volatility to the three mechanisms.

Experiments with artificial shocks and real-world news streams demonstrate that FinEvo reveals context-dependent outcomes: positive shocks increase diversity, negative shocks concentrate dominance, and intra-day dynamics exhibit multi-phase cycles of selection, innovation, and perturbation. Ablation studies confirm that selection drives efficiency, innovation sustains diversity, and perturbation enables adaptive reorganization. The framework also shows robustness to random seeds and parameter variations, and outperforms existing platforms like ABIDES and PyMarketSim in capturing dynamic, adaptive market evolution.

By reframing strategy evaluation as an ecological game, FinEvo provides a principled protocol for analyzing robustness, adaptation, and emergent dynamics in multi-agent financial markets. It offers a controlled environment to stress-test systemic risks and market shocks, and may serve as a tool for exploring the impact of macroeconomic policies and financial regulations on price evolution and equilibrium.