Do News and Social Media Tell the Same Story? Constructing and Comparing Sentiment Spillover Networks.

Battle Count: 71

The paper provides valuable insights into how sentiment information flows between technology companies through different media channels, identifying information hubs (IBM, AVGO, TSLA, CRM) and dominant spillover paths. This can inform sector rotation strategies, event-driven trading, and risk management. However, it is primarily descriptive/analytical rather than predictive, and does not propose a direct trading signal or backtest a strategy. The identification of regime changes and information hub companies could be useful for portfolio managers monitoring tech sector contagion risk.

Implementation requires: (1) transfer entropy calculation with quantile discretization and Markov bootstrap for significance testing, (2) rolling window network construction for 34 companies (1122 pairwise TE values per window), (3) network analysis including PageRank, degree centrality, and Maximum Spanning Arborescence algorithms, (4) Jaccard similarity comparison, (5) exponential decay imputation for missing data. The bootstrap procedure is computationally intensive. The paper references the R package RTransferEntropy and NlinTS for implementation.

The methodology is well-described with clear mathematical formulations, but the data source (Bloomberg Terminal) is proprietary and not publicly available. The paper provides detailed parameter settings (rolling window of 200 days, step size of 10 days, 3 quantile states, lag 1, 90% confidence interval, damping factor 0.85). No code or repository is provided. The Markov bootstrap procedure is described but implementation details for parallel computing are not specified.