Hermes_AI
Open Access DepositedA Transformer-based Anomaly Detection Framework for Insider Trading with Trader-based Sequencing and SEC Complaint-driven Features
Downloadable Content
In this study, we develop a novel approach to insider trading detectionusing an unsupervised Transformer→Variational Autoencoder (T→VAE) fed with trader-based sequential tensors containing features discovered using charging documents filed by the U.S. government. The new architecture and feature set are compared with the baseline, current state-of-the-art for insider trading detection using One-Class Support Vector Machines (OCSVM). Our model achieved superior performance, with a high partial Receiver Operating Characteristic — Area Under the Curve where the False Positive Rate (FPR) is constrained to 0.01 (pAUC) of the accounts trading before material announcements having insider trading charges against individuals from 2021 onward. The mean pAUC[0,0.01] across the eight test events is 0.885 versus 0.433 for the baseline OCSVM. Further, the proposed model achieved a False Negative Rate (FNR) of 0.000 versus 0.769 for the baseline OCSVM, even when the number of predicted positives is constrained to the lesser of 1% of the accounts and 30. This supports the conclusion that the proposed model is suitable for real-world applications with limited investigative resources.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.