Researchers from the Imperial College of London developed a machine-learning algorithm capable of automatically predicting cryptocurrency "pump-and-dump" scams before they happen.
The cryptosphere is infamous for being rife with fraud, and not for naught. The community scam jargon is rich. Exchange hacks, Pyramids, Ponzies, ICO scams, Exit scams, and Pump-and-Dumps (P&D). Many in the crypto community have fallen victim to more than one of these. Anything that can help battle crypto scams, be it man or machine, is most welcome. A good solution can save innocent victims' investments and help improve the sector's poor reputation.
The researchers studied over 200 P&D schemes over a period of five months, between July and November 2018, and built a model that can predict when a certain coin is about to get pumped, before it even starts.
The newly published paper, named "The Anatomy of a Cryptocurrency Pump-and-Dump Scheme" by Jiahua Xu and Benjamin Livshits from Imperial College London, is claimed to be the first ever detailed study of P&Ds in cryptocurrency markets. They discovered patterns in crypto markets associated with P&Ds, and claim that their model exhibits high precision and robustness in alerting against them.
Xu and Livshits recorded and investigated P&D activities on Telegram channels, which are famous for being one of the major platforms for carrying them out. One of the largest P&D channels, called "Official McAfee Pump Signals," has over 12,000 members. The researchers correlated the recorded communications with price changes and trading volumes of the tokens involved. They also searched for unexpected trades in obscure coins.
The study reveals surprising details about the behavior of this infamous type of fraud. Some might even be amazed by how fast the entire scheme operates and is over almost instantly before anyone has a chance to react.
The researchers encountered an average of two P&D scams per day, which created a trading volume of about $7M worth of transactions a month.
"After a mere 18 seconds of a manic buying wave, the coin price already skyrocketed to its peak [...] Three and half minutes after the start of the pump-and-dump, the coin price had dropped below its open price [...] After that, the trading volume dropped significantly."
The meaning of this is that anybody joining the party more than 18 seconds after the pump started, is most likely guaranteed to lose their money.
Xu and Livshits also found that the amount of the inspected tokens bought during the activity recorded was about twice the amount sold. Meaning that many buyers were left holding their bags, in hopes that they would someday become profitable again.
"Those coin holders can only expect to reverse the position in the next pump, which might never come."
Xu and Livshits later fed the amounts of recorded P&Ds' historical data to a machine-learning algorithm, in order to train it to recognize their recurring patterns. After the training phase, they then let it loose on live data.
During a period of only seven days, their algorithm alerted six times against activities it suspected as P&Ds. Five out of the six proved out to be real P&D scams.
BLOCKTV spoke with P&D analysis expert Steve McKenzie of SmartOptions trading platform, to understand if a machine could really be better than a human expert in detecting schemes, and if this system could actually save members of the crypto community from falling victim to Pump & Dumps. Although being confident in machines’ abilities to detect P&Ds, McKenzie seems skeptical that this is what would save people from becoming victims.
Xu and Livshits' work offers the potential to warn naive victims ahead of time from specific tokens that exhibit suspicious activity. For now, it is only an academic study, and not yet a viable system commercially available to the public.
Until that happens, the findings of this research can help educate the public about the scale of the problem and possibly how to avoid falling into bull traps.