Automatic Detection of Market Manipulation in Cryptocurrencies

In recent times, Blockchain and Cryptocurrencies have gained significant attraction by investors. The top 10 cryptocurrencies account for about 82% of total market capitalization. Hence, the majority of cryptocurrencies (to date, there are more than 9,000 cryptocurrencies) can be considered to have a relatively low market capitalization. This makes them prone to pump and dump, a type of market manipulation where organizers mislead the public to artificially increase the price of a coin and profit from it. This type of manipulation has been there well before cryptocurrencies; it arose in par with the stock market. However, it became more pronounced over the past years with the introduction of cryptocurrencies, which are still largely unregulated. The objective of this thesis is to develop an anomaly detection system to flag cryptocurrency pump and dumps. Specifically, we use Bitcoin data on the day before the pump to train an LSTM Autoencoder. With the training results, a threshold is set and the model is tested on unseen test data. To test the performance of the model, we crawled known pump groups on Telegram and created a dataset with details for each of the pump. We achieved an overall precision of 78%, recall of 83% and F1 score of 80%. The results of this research—which improves over the state of the art—, other than being interesting on their own, can help to detect and prevent pump and dump fraud in the cryptocurrency realm, hence ultimately increasing their adoption.