Detection of Stock Price Manipulation

This study applies a new approach in detecting trade-based pump-and-dump manipulations on the market. Through the usage of Neural Networks (NN), I attempt to generalize heuristics based on Message Flow data (a detailed but highly expensive type of market data) on Open-High-Low-Close-Value (OHLCV) data (less detailed but free). My experiments show promising results, with NN achieving 91% Precision-Recall-AUC score. While the approach still requires additional testing, this is potentially a very important result. The inaccessibility of detailed market data to researchers is one of the main obstacles in developing sufficient manipulation detection algorithms, and the approach suggested in this work can help to solve this problem. It implies that one can use smaller datasets of expensive data and instead propagate the algorithms on larger and cheaper ones, at least for some types of market manipulations, which, in turn, can make these algorithms more accessible for every participant of the market.