Towards Detecting Insider Trading in Prediction Markets: Event-Aligned Multi-Scale Anomaly Detection

Jolin Zhu, Youwen Wu, INSIXIENGMAY Claire

Keywords: Time Series Anomaly Detection, Sentiment Classification, NLP

Abstract:

In recent years, the rise of prediction markets like the Polymarket and Kalshi in the United States—built upon the seminal work on such markets—seek to aggregate public opinion by allowing users to “trade their beliefs” on the market. By their very nature, such markets are
highly susceptible to insider information and insider trading. As such, detecting anomalous trading activity is paramount to preventing unscrupulous traders with non-public information from gaining an unfair advantage. We propose a hybrid anomaly detection architecture specifically built for prediction markets, which combines prior art in anomaly detection with an event-aligned system that feeds in real-world events, public sentiment, major news, and official statements from governmental bodies to obtain an anomaly detection system more robust to explainable shocks in the market.

Submission Number: 12