A study by the Massachusetts Institute of Technology has revealed that machine learning can identify crypto pump and dump schemes before they happen. It arrived at a figure generated by fraudulent trading which is a tiny fraction of the total trade volume.
$7 Million Monthly Volume From Pump and Dumps
The study went on to reveal that there are at least two pump and dump schemes per day which generate $7 million in trade volume per month. NewsBTC’s daily crypto market wrap often identifies a random altcoin that is pumping for no apparent reason. However, in the grand scheme of things this figure is negligible when compared to a daily trade volume of around $15 billion.
In February the US Commodity Futures Trading Commission (CFTC) issued a specific warning about these types of scam, and regulators have begun to actively pursue the perpetrators. According to the regulator these schemes are considered as securities fraud. Research carried out at Imperial College London looked into pump and dump schemes in detail in order to develop an algorithm that can detect them.
The report highlights how pump and dumps are orchestrated as the organizer quietly accumulates an obscure altcoin before announcing a pump is about to begin on social media. Once the announcement, usually on Twitter or Telegram, has been made traders load up buy orders and the price begins to spike. The organizer then sells off his stash as the pump reaches a peak resulting in a dump back to previous levels.
As an example the researchers studied a single pump and dump scheme that occurred on November 18. The Official McAfee Pump Signals channel with over 12,000 members was monitored to glean details on the scheme. The chosen altcoin was BVB, worth around 35 satoshis at the time.
“We notice that the first buy order was placed and completed within 1 second after the first coin announcement. After a mere 18 seconds of a manic buying wave, the coin price already skyrocketed to its peak,” said the researchers noting a peak of 115 sats.
The speed of the operation was key as anyone entering the trade after around 18 seconds would not have made a profit. This was just one example but the study noted 236 other pump-and-dump events that took place between July 21 and November 18. The researchers used the historical data from known pump and dumps to train a machine learning algorithm to identify signs that a one is about to occur.
Cyptocurrency scams are not likely to go away anytime soon so identifying and being aware of them is all part of the evolution of this ecosystem. A few people make a little money out of pump and dumps but the effect on the market as a whole is negligible.
Full study can be found here: https://arxiv.org/abs/1811.10109
Image from Shutterstock