Signed network models for portfolio optimization

Battle Count: 75

The paper directly addresses portfolio construction and risk management, which are core quantitative trading concerns. The signed network approach provides an intuitive framework for identifying hedge relationships between assets. However, the empirical results show mixed performance (sometimes outperforming, sometimes underperforming benchmarks), and the method is primarily a dimensionality reduction technique rather than a complete trading strategy. The connection to quantum computing for portfolio optimization adds forward-looking relevance. The approach is practical for reducing computational complexity of portfolio optimization problems.

The core algorithm is relatively straightforward: (1) compute daily signed graphs by comparing returns to means, (2) calculate hedge scores as frequency of negative edges, (3) sort and select top-K assets, (4) apply standard Markowitz or 1/N. Time complexity is O(N log N) for the selection step. The main complexity lies in the Markowitz optimization step and handling edge cases in correlation estimation. No specialized libraries beyond standard numerical computing are required.

The paper provides detailed mathematical formulations (Definitions 1-2, Theorem 1, Equations 1-5), specifies datasets (Kaggle Market Champions and S&P500 aligned with Google), and describes the algorithm clearly. However, no code repository is provided, and some implementation details (e.g., exact threshold values, rolling window specifics) could be more explicit. The backtesting methodology is described but not fully reproducible without code.