Single-Period Portfolio Selection via Information Projection

Battle Count: 78

The paper provides a novel information-theoretic framework for single-period portfolio selection under CRRA utility, with a practical algorithm (Info-Proj EG) that shows empirical convergence advantages in low risk-aversion regimes. However, it is primarily theoretical, limited to single-period settings, assumes known market law, and does not address estimation, transaction costs, or dynamic strategies. The connection between Rényi order and risk aversion coefficient is conceptually valuable for quantitative portfolio construction.

The Info-Proj EG algorithm involves: (1) computing the symmetric covering set, (2) constructing the risk-tilted measure, (3) alternating between a closed-form auxiliary law update and an EG/Armijo portfolio step. The closed-form r-update simplifies one block, but the overall alternating structure, Armijo line search, and handling of different ρ_u regimes (min-min vs. min-max) add moderate complexity. The payoff matrix construction and symmetrization step require careful implementation.

The paper provides detailed update rules (Eq. 26-36), algorithm descriptions, and numerical experiment parameters (k=100, m=50, Dirichlet(10) state probabilities). However, no code repository is mentioned. The theoretical derivations are complete with proofs. Reproduction would require implementing the Info-Proj EG, Naive EG, and Cover's method from the described formulas.