Information Networks of Stock Prices

Battle Count: 65

The paper provides structural insights into inter-stock dependencies and identifies hidden sub-structures (commodity regimes, supply chains, regional liquidity affinity) that transcend formal sector boundaries. This is relevant for pairs trading, sector rotation strategies, and risk diversification. However, the paper is primarily descriptive/diagnostic rather than predictive—it does not generate trading signals, forecast returns, or propose specific trading strategies. The MI-PMFG residual information layer could inform alpha generation by identifying non-linear co-movements, but practical implementation for trading is not demonstrated.

The framework involves multiple interacting components: correlation/MI computation, graph filtering (MST via standard algorithms, PMFG via greedy planarity-constrained construction), and four community detection algorithms. The 24-configuration grid search across 97 rolling windows requires significant computational resources. PMFG construction is non-trivial (planarity checking at each edge addition). MI estimation with adaptive binning and kNN requires careful parameter tuning. However, standard libraries exist for most components (networkx, igraph, sklearn for KNN, community detection packages).

The paper provides detailed methodological descriptions including specific parameters (k=5 for kNN, B=max(3,⌊N^(1/3)⌋) for adaptive binning, n_bootstrap=20, 300-day windows with 25-day shifts, 150 stocks). However, no code repository or data download link is explicitly provided. The Indonesian stock exchange data (IDX) is publicly available but requires specific access. The 24-configuration framework is well-described but implementation details for PMFG greedy construction and community detection are standard.