CAST: A Cross-Asset State-Space Trading System for Drawdown Control in Stock Markets

Battle Count: 67

Directly addresses a core quantitative trading problem: drawdown control in stock portfolio management. Proposes a complete, deployable trading system (predictor + controller) with explicit risk penalties, online adaptation to distribution shifts, and interpretable state estimates. The modular plug-and-play design and sub-millisecond LP solving make it practically deployable. The focus on drawdown as a central design objective (rather than just returns/Sharpe) aligns with institutional risk management requirements.

Moderate complexity. No deep learning training or GPU required; the system runs on CPU. Core components are well-understood: Kalman filter recursions (matrix operations), credibility weighting (closed-form inverse-χ²), and a linear program for MPC (sub-millisecond solve). However, implementing the cross-asset joint state-space model with proper block-diagonal structure, multi-order IRW filters, and the receding-horizon MPC loop requires careful engineering. The modular design simplifies component-level implementation.

Code is publicly available on GitHub. No gradient-based training required; parameters are calibrated once in a pre-test window and remain fixed. The system runs on CPU with sub-millisecond LP solving. However, specific hyperparameter values (λ, β, L, δ) and exact calibration procedures are referenced to the code repository appendix rather than fully detailed in the paper.