Dynamic Weight Optimization for Double Linear Policy: A Stochastic Model Predictive Control Approach

Battle Count: 83

Directly addresses a core problem in quantitative trading: dynamic weight/position sizing under risk constraints. The DLP-SMPC framework provides a principled, theoretically grounded approach to adaptive trading with guaranteed survivability and positive expectation properties. Empirical results on crypto and equities demonstrate practical improvements in risk-adjusted performance. The analytical gradient derivation and efficient L-BFGS-B solver make it suitable for real-time implementation.

Requires understanding of stochastic MPC, state-space formulations, mean-variance optimization, and constrained nonlinear optimization. The analytical gradient (Theorem 4.1) involves matrix products and product-rule differentiation over the prediction horizon. The Augmented Lagrangian outer loop adds another layer. However, the algorithms are clearly specified, L-BFGS-B is a standard library routine (e.g., scipy.optimize.minimize with method='L-BFGS-B'), and the state-space dynamics are simple 2x2 diagonal systems. Main complexity lies in correctly implementing the gradient and constraint handling.

The paper provides complete algorithmic pseudocode (Algorithms 1 and 2), explicit analytical gradient formulas, hyperparameter search grids, and uses publicly available data (Yahoo Finance BTC-USD, TSLA, ETH-USD, AAPL). Training/test splits and cross-validation protocol are clearly specified. However, no code repository is provided, and some implementation details (e.g., exact initialization of w(0), specific L-BFGS-B library version) are not fully specified.