The paper introduces RD-Agent(Q), a pioneering multi-agent framework designed to automate the full research and development pipeline for quantitative finance strategies. The system addresses key limitations in current quantitative research, including limited automation, weak interpretability, and fragmented coordination between factor mining and model innovation. By decomposing the quant process into iterative Research and Development stages connected through a feedback mechanism, RD-Agent(Q) enables coordinated factor-model co-optimization that dynamically sets goal-aligned prompts, formulates hypotheses based on domain knowledge, and implements task-specific code through the Co-STEER code-generation agent.

The framework employs a multi-armed bandit scheduler for adaptive direction selection and incorporates real-market backtesting with comprehensive evaluation of experimental outcomes. This data-centric approach allows the system to continuously refine strategies based on performance feedback. Empirical results demonstrate that RD-Agent(Q) achieves significantly higher annualized returns than classical factor libraries while using substantially fewer factors, and outperforms state-of-the-art deep time-series models in real market conditions. The joint optimization of factors and models provides an effective balance between predictive accuracy and strategy robustness, representing a significant advancement in automated quantitative research methodology.