Density Estimation for Financial Fraud Detection: A Multivariate Kernel-Based Approach

Abstract

The detection of insider trading has become increasingly challenging due to the complexity of modern financial markets. This paper introduces a novel approach using Volume-Weighted Multivariate Kernel Density Estimation to identify potential insider trading activities. By integrating volume-weighting with the Kernel Density Estimation framework, this method effectively captures abnormal price-volume relationships, providing a sophisticated means to flag irregular trading behaviour. This paper proposes a robust approach using Volume-Weighted Multivariate Kernel Density Estimation to detect abnormal trading patterns linked to insider trading and this approach improves traditional Kernel Density Estimation by incorporating trading volume as a weighting factor, allowing it to capture the joint distribution of stock returns and trading volumes. The model’s application is tested on data from companies targeted by Hindenburg Research, including Nikola Corporation, Clover Health, and Adani Enterprises. The results show clear pre-event anomalies, including increased trading volumes and price movements preceding the report releases. By utilizing adaptive bandwidth selection, Volume-Weighted Multivariate Kernel Density Estimation balances bias and variance to refine detection of insider trading activities in both dense and sparse regions of the data. The emergence of reports from activist short-sellers like Hindenburg Research has triggered significant market reactions, often accompanied by allegations of insider trading. This methodology is highly sensitive to abnormal volumes that may signal preemptive insider trading ahead of major market events, such as the release of damaging reports. This research demonstrates that VW-MKDE offers enhanced sensitivity to volume-weighted deviations, enabling the detection of abnormal trading behavior associated with potential insider information. Our findings highlight the tool’s capacity to assist regulators and analysts in identifying suspicious market activities in real-time, particularly during periods of heightened market volatility and post-report market responses.