Needles in a haystack: using forensic network science to uncover insider trading

Battle Count: 71

While not directly about trading strategy development, the paper is highly relevant to quantitative trading in several ways: (1) detecting coordinated insider activity can inform market surveillance and alpha generation; (2) understanding information asymmetry patterns among corporate insiders is relevant for event-driven strategies; (3) the network approach could be adapted to detect coordinated trading patterns in market microstructure; (4) regulatory enforcement actions flagged by such methods can create predictable price movements. However, the paper focuses on detection/screening rather than direct trading signal generation.

The core algorithm is relatively straightforward: computing pairwise temporal similarity using a weekly kernel, applying threshold criteria, and constructing a network. The null models require careful implementation of calibrated generative processes and constrained shuffling. Network analysis (centrality measures, OddBall) uses standard graph algorithms. The main complexity lies in data preprocessing (parsing SEC Form 4 filings, handling institutional investors, aggregating line items) and the O(n²) pairwise comparisons within each firm. The algorithm is described as scalable and explainable.

Data is publicly available from SEC EDGAR system. Code will be made available upon official publication. The methodology is fully described with explicit formulas, algorithm pseudocode, and parameter tables. Null model parameters are detailed in supplementary material. However, code is not yet available at time of preprint.