Financial Information Theory

Battle Count: 79

Highly relevant to quantitative trading. The paper provides actionable tools: NMI-based trading signals (Algorithm 3), entropy-adjusted VaR for dynamic risk limits, information-theoretic diversification criteria, and regime detection via KL divergence. The empirical validation on 25 years of S&P 500 data demonstrates practical utility. The NMI threshold of 0.05-0.10 for trading signals and the 4.5x VaR multiplier during COVID-19 are directly applicable. However, the paper lacks full backtesting with transaction costs and does not compare against modern ML-based approaches, which limits its immediate deployment readiness.

Moderate complexity. The core algorithms (k-NN entropy estimation, NMI calculation, KL divergence) are well-defined with clear pseudocode. The k-NN estimator requires careful implementation of distance computations, digamma function corrections, and volume of unit ball calculations. Rolling window implementation is straightforward. The main challenges are: (1) tuning k and window size, (2) handling numerical stability (Gaussian noise addition), (3) discretization for KL divergence, and (4) ensuring statistical significance through resampling. No deep learning or complex optimization is required. Standard Python libraries (scipy, sklearn, numpy) can implement most components.

The paper provides complete mathematical proofs, detailed algorithms (Algorithm 1 for Transfer Entropy, Algorithm 2 for NMI, Algorithm 3 for Trading Signals), and specific implementation parameters (w=252, k=3, sigma=10^-10, 50 bins for KL). However, no code repository is provided. The S&P 500 (SPY) data is publicly available. The k-NN entropy estimator references Kozachenko and Leonenko (1987). Reproducibility is moderate due to sensitivity of k-NN to parameter choices and lack of code.