Stochastic Discount Factors with Cross-Asset Spillovers

Battle Count: 94

Highly relevant to quantitative trading. The paper directly constructs maximum-Sharpe-ratio trading strategies using 138 firm-level signals across 138 spread portfolios and 544 bivariate sorted portfolios. The out-of-sample Sharpe ratios of 2.21 (spread) and 3.32 (bi-sort) are exceptional. The framework provides actionable portfolio weights, interpretable signal rankings, and a connection matrix revealing cross-asset information flows. The zero-cost, leverage-two implementation is directly comparable to standard factor/anomaly strategies. The factor-spanning tests with alphas of ~0.25%/month (t>11) demonstrate genuine pricing power beyond existing models.

Moderate to high complexity. The core algorithm (Algorithm 1) involves iterative ridge regressions alternating between estimating Λ and Φ, requiring construction of managed-portfolio matrices (χ_Φ and χ_Λ) of dimensions T×M and T×N² respectively. For N=544 assets and M=138 signals, the Φ vector has 544²=295,936 elements, making the ridge regression computationally demanding. The zero-cost constraint requires additional matrix operations (Θ projection). Five-fold cross-validation over a grid of 11 λ values adds computational overhead. However, the linear structure and closed-form ridge solutions make it tractable compared to deep learning approaches. The paper provides full mathematical derivations and an explicit algorithm.

The paper uses publicly available datasets (CRSP, Compustat, IBES, Jensen et al. 2023 Global Stock Returns and Characteristics dataset on WRDS). The methodology is fully specified with closed-form solutions and an iterative algorithm (Algorithm 1). Five-fold cross-validation for λ selection is described. Rolling 10-year estimation windows are clearly defined. However, no code repository is explicitly provided, and some implementation details (e.g., exact initialization for iterative algorithm) could benefit from additional specification.