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

Detecting stock price manipulation is a cat-and-mouse game. Manipulators have constantly devised new techniques to avoid detection. The majority of the related work employed supervised learning techniques, which necessitated known patterns as examples for their models recognize. To catch unknown and never-before-seen manipulation, we used unsupervised train deep neural networks detecting in order detect previously unseen manipulation. were trained recognize normal trading behaviors that expressed limit book. Anomaly actions did not follow learned identified manipulated. strength our method it does require prior knowledge about characteristics As result, best suited or types Two model architectures evaluated: autoencoder (AE) generative adversarial (GANs). They put test on six prosecuted real cases from Stock Exchange Thailand (SET). With low false-positive rate, both could identify five out cases. For practical application models, strategy called “MinManiMax” was also proposed optimize decision boundary.