Robo-Advising in Motion: A Model Predictive Control Approach

Battle Count: 257

Highly relevant to quantitative trading and portfolio management. The paper directly addresses dynamic asset allocation, a core problem in quantitative finance. The SMPC framework with HMM-BL forecasting is directly applicable to systematic portfolio management. The mean-variance and risk-budgeting criteria are standard in institutional and retail wealth management. The practical constraints (turnover, transaction costs, target allocations) reflect real-world trading considerations. However, the focus is on monthly rebalancing robo-advising rather than high-frequency or intraday trading strategies.

Moderate to high complexity. Requires implementation of: (1) HMM with Baum-Welch estimation and Viterbi decoding, (2) Black-Litterman Bayesian updating with regime-specific priors, (3) SMPC rolling horizon optimization loop, (4) Convex optimization via cvxpy for MV and iteratively linearized MRB, (5) Multiple constraint handling (long-only, budget, turnover, target portfolio). The convex nature of all sub-problems and availability of off-the-shelf solvers (cvxpy) reduce practical difficulty. Computation time is on the order of milliseconds per strategy, making it scalable. The iterative convexification for MRB adds complexity but is well-documented in literature.

The paper uses publicly available data (Yahoo Finance, ycharts.com), open-source Python libraries (hmmlearn v0.3.2, cvxpy v1.5), and specifies all parameter ranges and calibration choices. The computational setup (Python 3.13.7, specific hardware) is described. However, no code repository is explicitly linked, and some parameter calibration is described as 'more of an art.' The HMM uses Viterbi and Baum-Welch algorithms via hmmlearn. All optimization problems are convex, enabling use of standard solvers.