Stock Price Manipulation Detection Using Deep Unsupervised Learning: The Case of Thailand

Abstract

Detecting stock price manipulation is a cat-and-mouse game. Manipulators have continuously invented new techniques to avoid being caught. Most of the related work used supervised learning techniques, which require known manipulation patterns as examples for their models to recognize. To catch unknown and never-before-seen manipulation, we implemented unsupervised learning to train deep neural networks for detecting stock price manipulation. The models were trained to recognize normal trading behaviors that were expressed on a limit order book. Anomaly trading actions, which did not conform to the learned patterns were identified as manipulated. The strength of our method is that it does not require prior knowledge about the characteristics of manipulation. Therefore, it is best used for catching new or unknown types of manipulation. Two model architectures were tested: autoencoder (AE) and generative adversarial networks (GANs). They were tested on six prosecuted real manipulation cases in the Stock Exchange of Thailand (SET). Both models could identify five out of six cases with a low false-positive rate. A strategy to optimize the decision boundary, called "MinManiMax", was also proposed for practical usage of the models.

A Survey on Stock Market Manipulation Detectors Using Artificial Intelligence

“…The authors then enhanced the design of their detector by introducing an improved version of the LSTM-GAN and a new LSTM-AE network [81]. The same limit order book data, containing a set of 15 features were used as the input to the network.…”