Using Transformer Models for Stock Market Anomaly Detection

Abstract

Anomaly detection is an important task in financial markets. Detecting anomalies is difficult due to their rarity, multitude of parameters, and lack of labeled data for supervised learning models. Additionally, time series data used in financial models present unique challenges such as irregularity, seasonality, changing trends, and periodicity in data. While prior anomaly detection approaches have used ARIMA and LSTM models, in this paper, we employ a new transformer�based model called TranAD to compare stock market data with its predicted version, measuring deviations from normal price data for anomaly detection. We find that TranAD is an effective approach for financial anomaly detection with a high level of accuracy. We expect that this research will contribute to better detection of financial anomalies and improve market surveillance