Beyond Co-Movement: Locality by Exposures Enables a Joint Factor–Graph Framework for Portfolio Diversification

Battle Count: 74

Highly relevant to quantitative trading and portfolio management. The paper directly addresses portfolio diversification, covariance estimation, and allocation strategies. The MINGLE framework provides a principled way to construct better-conditioned covariance estimates and graph-based allocation (Contagion Cut) that consistently outperforms MVO, HRP, and correlation-based CutV across multiple transaction cost levels and market regimes. The exposure-similarity graph concept is directly applicable to risk parity, sector rotation, and systematic allocation strategies. The backtest spans 2019-2026 including crisis periods, providing practical evidence.

High implementation complexity. The ADMM framework requires solving multiple coupled subproblems: a Sylvester equation for exposure profiles (O(p^3)), a Sherman-Morrison solve for graph topology, gradient steps for factor dynamics, proximal L1 steps for sparsity, Stiefel manifold projections, and quadratic-root solves for degree variables. The non-convex nature requires careful initialization (SVD-based) and hyperparameter tuning (8 parameters in Table 1). The Contagion Cut allocation adds graph-based weighting logic. Multiple regularization terms (smoothness, sparsity, degree barrier, decorrelation) must be balanced. No reference implementation is provided.

The paper provides detailed hyperparameters (Table 1), the full ADMM algorithm (Algorithm 1), initialization procedures, and dataset description (100 most actively traded constituents from S&P 500, Nikkei 225, STOXX 600). However, no code repository is mentioned, and the data is sourced from Bloomberg Terminal (proprietary). The backtest period (Jan 2019–March 2026) and specific asset selection criteria are described, but exact constituent lists are not provided. The ADMM subproblem solutions reference external algorithms (Sylvester solve, Sherman-Morrison, Stiefel manifold projection).