Portfolio Analysis Based on Markowitz Stochastic Dominance Criteria: A Behavioral Perspective

Battle Count: 75

The paper is primarily focused on portfolio construction and behavioral analysis rather than algorithmic trading or market microstructure. However, the MSD/MWSD-constrained optimization framework is relevant for quantitative portfolio managers who want to construct portfolios consistent with behavioral investor preferences. The empirical findings on industry preferences (Gold, Soda, Smoke as most preferred) and diversification patterns provide actionable insights. The out-of-sample performance analysis and risk metrics (CVaR, Sortino) are directly applicable to portfolio management. The framework could inform behavioral factor investing strategies.

The MILP formulation involves multiple sets of binary variables (z, xi, zeta) and continuous variables (phi, psi, theta, delta) with big-M constraints, making the formulation non-trivial. For n=36 states and m=49 assets, the problem has O(n^2) binary variables and O(n^2) continuous variables. The big-M parameter selection requires careful tuning. Solving requires commercial MILP solvers (Gurobi). The rolling window approach with 59 periods adds computational burden. However, the linear structure of constraints makes it more tractable than direct non-linear optimization with inverse S-shaped utilities.

The paper uses publicly available Fama-French 49 industry portfolios and CRSP all-share index data. Implementation is in Python with Gurobi 11.0.3 solver. Rolling window methodology (36-month estimation, 12-month shift) is clearly specified. However, no code repository is provided, and the MILP formulations require careful big-M parameter tuning. The equal probability assumption for states is standard but simplifying.