Portfolio Optimization under Dynamic Rebalancing via Topological Data Analysis and News Sentiments

Battle Count: 78

Highly relevant to quantitative trading. The paper directly addresses portfolio construction, asset selection, dynamic rebalancing, and risk management - all core components of quantitative trading strategies. The integration of sentiment analysis with topological methods for asset screening is novel and practical. The DRMV model incorporates real-world constraints (transaction costs, short-selling, position bounds) making it directly implementable. The rolling-window framework with cross-validation mirrors practical backtesting procedures. Results show consistent outperformance across multiple rebalancing frequencies and market conditions.

High implementation complexity due to: (1) TDA computations requiring Vietoris-Rips filtration construction and persistent homology calculation for each asset's sliding windows; (2) Wasserstein distance computation between persistence diagrams; (3) Persistence landscape construction and Lp norm computation; (4) Agglomerative clustering with custom distance matrices; (5) Mixed-integer quadratic programming for DRMV with binary variables for long/short inclusion; (6) Rolling-window framework with cross-validation for lambda selection; (7) FinBERT inference for sentiment extraction; (8) Data preprocessing and standardization. The combination of TDA, NLP, and constrained optimization makes this a complex multi-stage pipeline.

The paper provides detailed algorithmic descriptions (Algorithms 1-4), parameter settings (alpha=0.95/0.99, rho=0.10, transaction costs=25bps, bounds m+=5%, M+=20%, m-=1%, M-=5%), and rolling window specifications. However, no code repository is mentioned. Data sources (LSEG Refinitiv Workspace) require commercial access. The FinBERT model is publicly available. The TDA computations (Vietoris-Rips, persistence diagrams/landscapes) are standard but computationally intensive.