Krishna Neupane, PhD

This study identifies a significant informational friction in the SEC Form 144 regime termed predictive decoupling. It argues that the current reporting inversion—where Form 4 precedes Form 144—violates Pareto efficiency. Utilizing an event-study framework, the machine learning audit reveals a 52.4% opacity rate in aborted sale signals. The study documents a large-cap significance paradox and proposes a mandatory Form 144-A to restore informational symmetry.

Traditional models of market efficiency assume that equity prices incorporate information based on content alone, often neglecting the structural influence of reporting timing and cadence. This study introduces the Autonomous Disclosure Regulator, a multi-node AI framework designed to audit the intersection of disclosure complexity and filing unpredictability. Analyzing a population of 484,796 regulatory filings, the research identifies a structural Strategic Gap: a state where companies use confusing language and unpredictable timing to slow down how fast the market learns the truth by 60%.

This study integrates Shapley Values (SHAP) and Causal Forests (CF) to detect Unlawful Insider Trading (UIT) and uncover the latent causal structures of fraudulent behavior. Utilizing a high-dimensional feature space, the architecture identifies statistically significant relationships between UIT and director status, price-to-book ratios, and market beta. The framework demonstrates high classification accuracy and provides robust causal significance assessments, facilitating the discovery of unique relationships within trade, finance, and governance data.

Detecting Unlawful Insider Trading (UIT) remains a critical yet challenging task. The sheer volume of transactions, often numbering in the thousands per minute, necessitates the development of sophisticated algorithms to augment manual review. While previous studies have demonstrated the potential of Machine Learning (ML) methods, such as Random Forest (RF), for UIT detection, this study significantly improves upon existing approaches. We expand the feature set from 25 previously used features to 110, incorporating 85 novel features that prove highly informative for UIT identification. Features were result of merging of multiple database for the first time to best of our knowledge. Furthermore, we substantially increase the training dataset size to 3,984 transactions. Consequently, our RF model achieves a UIT detection accuracy over 95 percent, a substantial improvement over previously reported results. Our key findings also highlight the significant influence of corporate governance factors on UIT and underscore the importance of data quality and feature engineering for optimal model performance. This research demonstrates the effectiveness of ML tools in automating UIT detection, thereby reducing reliance on manual analysis and enhancing efficiency.