MetaPS: Adaptive Programmatic Strategy Selection for Market Agents

MetaPS is a new framework for adaptive programmatic strategy selection in financial markets, utilizing a library of executable strategies that respond to changing market conditions. It employs a simulation-guided approach to identify optimal strategy-state pairs, which are then used for supervised fine-tuning, enhancing performance across models ranging from 0.8B to 9B parameters. The framework shows significant improvements over fixed-strategy baselines and direct decision-making agents, indicating that market simulations can effectively provide targeted supervision for developing interpretable and adaptable trading strategies.