ISI Cryptocurrency Research on Pump and Dump 2020

Cryptocurrency has sprung to the forefront of public consciousness in recent years, with Bitcoin(r) remains the most popular member. As the internet-based medium of exchange grows in popularity, many lesser-known, alternative cryptocurrencies, “altcoins,” have emerged, consequently leading to more fraudulent activities. Pump-and-dump aptly denotes frauds where a group of instigators uses various machinations to artificially inflate the coin’s price to attract buyers (pump), then swiftly sell their holdings to profit, leaving unsuspecting buyers stuck with a nearly worthless coin [#]. By leveraging the power LSTM-based neural network to encode time-series market data, we hope to predict the identity of the target coin for upcoming pumps. In addition, we investigate social media factors and their potential contribution to improve the task of detection.

There are two mains foldes: General_Analysis and Before_Pump_Analysis. The former focuses on the activities of the market as the coin annoucements happened while the latter focuses on the period briedly leading to the pump. There are notebooks that our analysese using Python via Jupyer Notebook.

The Before_Pump folder further explores Neural_Network based models in an attemp to predict the right coin to be pumped. We explored traditional model, LSTM based models and other anomaly detection method that it would perform well in each work.

Thanks to the help of my ISI advisors Goran Muric, Fred Morstatter and Emillio Ferrara for their contribution to this project.