FinRL-X: An AI-Native Modular Infrastructure for Quantitative Trading

Battle Count: 139

This paper is directly and fundamentally about quantitative trading infrastructure. It addresses the critical gap between research backtesting and live deployment, provides a complete end-to-end trading system architecture, integrates DRL-based portfolio allocation, supports multiple allocation paradigms (Mean-Variance, Min-Variance, Equal-Weight, DRL), includes timing and risk overlay modules, and validates through both backtesting and broker-integrated paper trading. It is a systems-level contribution to the quantitative trading field.

The modular architecture with clear interface contracts (weight vectors) simplifies individual component implementation. However, building a full end-to-end system requires integrating data pipelines, strategy modules, backtesting engines, broker APIs, monitoring, and state management. The framework reduces complexity through composability and standardized interfaces, but production deployment still requires significant engineering for reliability, fault tolerance, and operational monitoring. The open-source codebase lowers the barrier to entry.

The paper provides an open-source implementation on GitHub (Apache 2.0 license), standardized backtesting protocols with fixed transaction costs (10 bps per side), explicit data sources (FMP), persistent storage of raw snapshots and processed features, rolling out-of-sample validation for learning-based models, and runnable examples. The weight-centric interface ensures consistent semantics across all layers. Paper trading validation on Alpaca further demonstrates reproducibility.