The Anatomy of a CryptocurrencyPump-and-Dump Scheme

The Anatomy of a Cryptocurrency
Pump-and-Dump Scheme

Jiahua Xu University of St. Gallen
Imperial College London
   Benjamin Livshits Imperial College London
UCL Centre for Blockchain Technologies

While pump-and-dump schemes have attracted the attention of cryptocurrency observers and regulators alike, this paper is the first detailed study of pump-and-dump activities in cryptocurrency markets. We present a case study of a recent pump-and-dump event, investigate 220 pump-and-dump activities organized in Telegram channels from July 21, 2018 to November 18, 2018, and discover patterns in crypto-markets associated with pump-and-dump schemes. We then build a model that predicts the pump likelihood of a given coin prior to a pump. The model exhibits high precision as well as robustness, and can be used to create a simple, yet very effective trading strategy, which we empirically demonstrate can generate a return as high as 80% within a span of only three weeks.

cryptocurrency, trading, pump-and-dump, market manipulation, Telegram

Fig. 1: A successfully organized pump event. On the right hand side of the screenshot is the message history of a Telegram channel. The first message is the final countdown; the second message is the coin announcement; the last message presents the pump result. On the left hand side is the market movement of the corresponding coin around the pump time.

I Introduction

While pump-and-dump schemes are an old, well-trodden ruse in conventional financial markets, the old-fashioned ploy has found a new playground to thrive — cryptocurrency exchanges.

The relative anonymity of the crypto-space has lead to it becoming a fertile ground for unlawful activities, such as currency theft (e.g. the DAO hack [1]), Ponzi schemes [19], and pump-and-dump schemes that have risen in popularity in cryptocurrency markets over the last few years. Due to their end-to-end encryption, programmability, and relative anonymity, new social media tools such as Telegram111Note that not all Telegram traffic is end-to-end encrypted. and Discord have become cryptocurrency enthusiasts’ preferred communication vehicles. While pump-and-dump schemes have been discussed in the press [22], we are not aware of a comprehensive study of this phenomenon to date.

Regulation: In February 2018, the CFTC (Commodity Futures Trading Commission) issued warnings to consumers [7] about the possibility of crypto-currency pump-and-dump schemes. It also offered a substantial reward to whistle-blowers around the same time [10].

In October 2018, the SEC (Securities and Exchange Commission) filed a subpoena enforcement against an investment company trust and trustee for failure to produce documents [20].

Clearly, regulators are aiming to find perpetrators of pump-and-dump schemes and to actively prosecute them.

This paper: In this paper, we trace the message history of over 300 Telegram channels from July to November 2018, and identify 220 pump events orchestrated through those channels. We analyze features of pumped coins and market movement of coins before, during, and after pump-and-dump. We develop a predictive random forest model that gives the likelihood of each possible coin being pumped prior to the actual pump event. With an AUC (area under curve) of the ROC (receiver operating characteristic) curve of over 0.9, the model exhibits high accuracy and is indicative of the “pumpability" of coins.

Contributions: This paper has the following contributions:

  • Longitudinal study: This paper is the first study of pump-and-dump schemes in the wild, which we base on our analysis of price and volume histories across multiple crypto-exchanges, as well as Telegram groups dedicated to pump-and-dump activities.

  • Analysis: Our analysis shows that pump-and-dump activities are a lot more prevalent that previously believed. Specifically, around 100 organized Telegram pump-and-dump channels coordinate on average 2 pumps day which generates an aggregate artificial trading volume of 7 million USD a month. We discover that some exchanges are also active participants in pump-and-dump schemes.

  • Prediction: We develop a predictor that, given a pre-pump announcement can predict the likelihood of each coin being pumped with an AUC (Area Under Curve) of over 0.9 both in-sample and out-of-sample. The models confirm that market movements contain hidden information that can be utilized for monetary purposes.

  • Trading strategy: Based on this, we formulate a simple trading strategy that, based on historical data, gives us a return of 80% over a period of three weeks, even under strict assumptions.

Paper organization: The paper is structured as follows. In Section II we provide background information on pump and dump activities organized by Telegram channels. In Section III we present a pump-and-dump case study. In Section IV we investigate a range of coin features. In Section V we build a predicting model that estimates the pump likelihood of each coin for each pump, and propose a trading strategy along with the model. In Section VI we summarize the related literature. In Section VII we outline our conclusions.

Ii Background

A pump is a coordinated, intentional, short-term increase in the demand of a market instrument — in our study, cryptocurrency — which leads to a price hike. Thanks to the features of encryption and anonymity offered by chat applications such as Telegram and Discord, various forms of misconduct in cryptocurrency trading is burgeoning on those platforms.

A pump organizer, which can be an individual, or, more likely, an organized group, typically uses those social media tools to coordinate a pump event as follows.

Set-up: Firstly, the organizer creates a group or channel which is usually accessible by all users, and recruits as many group members or channel subscribers as possible by advertising and posting invitation links on major forums such as Bitcointalk, Steemit, and Reddit. The group is ready to pump once it obtains enough members (typically above 1,000). Telegram channels only allow subscribers to receive messages from the channel admin, but not post discussions in the channel. In a Telegram group, members can by default post messages, but this function is usually disabled by the group admin to avoid members’ interference.

Pre-Pump announcement: The pump organizer, who is now the group or channel admin, announces details of the next pump a few days ahead. The admins broadcast the exact time and date of the announcement of a coin which would then precipitate a pump of that coin. Other information disclosed in advance includes the exchange where the pump will take place and the pairing coin222A pairing coin is a coin that is used to trade against other coins. Bitcoin (BTC) is a typical pairing coin.. The admins advise members to transfer sufficient funds (in the form of the pairing coin) into the named exchange beforehand.

While the named pump time is approaching, the admin reminds group members by sending out countdowns, and emphasizes the pump “rules" that pump members should observe to maximize the pump profit. Typical “rules of thumb” include 1) make sure to buy fast, 2) “shill"333Crypto jargon for “advertise", “promote". the pumped coin on the exchange chat box and social media to attract outsiders, 3) “HODL"444Crypto jargon for “hold". the coin at least for several minutes to give outsiders time to join in, 4) sell in pieces and not in a single chunk, 5) only sell at a profit and never sell below the current price. The admin also gives members a pep talk, quoting historical pump profits, to boost members’ confidence and encourage their participation.

Coin announcement: At the pre-arranged pump time, the admin announces the coin, typically in the format of an OCR (optical character recognition)-proof image to hinder machine reading (Fig. 1(a)). Immediately afterwards, the admin urges members to buy and hold the coin in order to inflate the coin price. During the first minute of the pump, the coin price typically surges, increasing by several fold.

Dump: A few minutes (sometimes tens of seconds) after the pump starts, the coin price will reach its peak. While the group admin might shout “buy buy buy" and “hold hold hold" in the channel, the coin price still can’t resist dropping. As soon as the first fall in price appears, pump-and-dump participants start to panic-sell. While the price might be re-boosted by the second wave of purchasers who buy the dips (as encouraged by channel admins), chances are the price will rapidly bounce back to the start price, sometimes even lower. The coin price declining to the pre-pump proximity also signifies the end of the dump, since most investors would rather hold the coin than sell at a loss.

A few minutes later, when the coin price and trading volume recover to approximately the pre-pump level, the admin posts an analysis that showcases how much the coin price increased by the pump. It is generally known to pump participants that admins benefit the most from a pump. Most of the time, the admin is able to sell their pre-hoarded coins at an inflated price to other group members during a pump. So why are there still people enthusiastic about partaking a pump, given the risk of being ripped off by the admins? Because people believe that there are greater fools out there, who would buy the coins at an even higher price than their original purchase price. The greater fool theory also forms the foundation of many other schemes, such as pyramid scams or ponzi games [4].

(a) Tweets from @YobitExchange.

(b) Pump timer from the Yobit website.
Fig. 2: The screen-shots demonstrate that the exchange Yobit is actively involved in pump-and-dump activities.

The role of exchanges: Some exchanges are themselves directly associated with pump-and-dump. Yobit, for example, has openly organized pumps multiple times (see Fig. 2). The benefit for an exchange to be a pump organizer is threefold:

  1. With coins acquired before a pump, it can profit by dumping those coins at a higher, pumped price just as Telegram channel admins;

  2. It earns high transaction fees due to increased trading volume driven by a pump-and-dump;

  3. Exchanges are able to utilize their first access to users’ order information for front-running during a frenzy pump-and-dump.

Fig. 3: A pump attempt coordinated by multiple channels not executed due to unanticipated price movement of the to-be-pumped coin.

Failed pump-and-dump attempts: Note that not every pump attempt is successful. Fig. 3 shows that the admins decided not to carry through a pre-announced pump due to unanticipated price movements of the to-be-pumped coin. While it is unknown what caused the movements, the case evidences that the admin is aware of the coin choice before the pump (as opposed to the coin being randomly selected and immediately announced at the pump time purely by algorithm), and hence has the time advantage of purchasing the coin at a low price before the coin announcement, whereas group members only purchase the coin after the coin announcement and slow buyers risk acquiring the coin at an already (hyper)inflated price.

We can also hypothesize from this case that, someone might have worked out the pattern of the coin selection and pre-purchased a bucket of coins with high pump likelihood that happens to contain the actual to-be-pumped coin, which might explain why the admin observed peculiar movements of the coin. In the next section, we study the features of pumped coins and their price movements, to understand if it is indeed possible to predict the pumped coin.

Iii A Pump-and-Dump Case Study

We further study in depth the pump-and-dump event associated with Fig. 1(a). The pump-and-dump was organized by at least four Telegram channels, the largest one being Official McAfee Pump Signals, with a startling 12,333 members. Prior to the coin announcement, the members were notified that the pump-and-dump would take place on one of the Cryptopia’s BTC markets (i.e., BTC is the pairing coin).

Announcement: At 19:00 GMT, on Novermber 14, 2018, the channels announced the target coin in the form of a OCR-proof picture, but not quite simultaneously. Official McAfee Pump Signals was the fastest announcer, having the anouncement message sent out at 19:00:04. Bomba bitcoin “cryptopia" was the last channel that broadcast the coin, at 19:00:23.

The target coin was BVB, a dormant coin that is not listed on CoinMarketCap. The launch of the coin was announced on Bitcointalk on August 25, 2016.555 The coin was claimed to be made by and for supporters of a popular German football club, Borussia Dortmund (a.k.a. BVB). The last commit on the associated project’s source code on GitHub was on August 10, 2017.666

Although it has an official Twitter account, @bvbcoin, its last Tweet dates back to 31 August, 2016. The coin’s rating on Cryptopia is a low 1 out of possible 5. This choice highlights the preference of pump-and-dump organizers to go after coins associated with projects that are not active and cannot resist the pump-and-dump activity.

During the first 15 minutes of the pump, BVB’s trading volume exploded from virtually zero to 1.41 BTC (illustrated by the tall grey bar towards the right end of the price/volume chart), and the coin price increased from 35 Sat777One Satoshi (Sat) equals  Bitcoin (BTC). to its threefold, 115 Sat (illustrated by the thin grey vertical line inside the tall grey bar).

Fig. 4: Tick-by-tick movement of the BVB/ BTC market during the first four minutes after the coin announcement.

Price fluctuations: Further dissecting the tick by tick transactions (Fig. 4), we notice that the first buy order was placed and completed within 1 second after the first coin announcement. With this lightning speed, we conjecture that such an order might have been executed by automation. After a mere 18 seconds of a manic buying wave, the coin price already skyrocketed to its peak. Note that Bomba bitcoin “cryptopia” only announced the coin at the time when the coin price was already at its peak, making it impossible for investors who solely relied on their announcement to make any money.

Not being able to remain at this high level for more than a few seconds, the coin price began to decrease, with some resistance in between, and then plummeted. Three and half minutes after the start of the pump-and-dump, the coin price had dropped below its open price. Afterwards, transactions only occurred sporadically.

Fig. 5: Gap between buy volume and sell volume caused by the BVB pump-and-dump. The figure shows a timeline of 48 hours before up to 1 hour after the pump-and-dump. For the illustration purposes, the timeline is scaled with non-linear transformation to better display the development of volume gaps during the pump-and-dump.

Volume: Fig. 5 shows that the pump-and-dump induces fake demand and inflates buy volume. While every pump-and-dump participant would hope for a quick windfall gain during a minute-long pump, the majority would not manage to act fast enough to sell at a high price. Those investors would either end up selling coins at a loss, or, if reluctant to sell low, would hold the virtually worthless coins. This can be demonstrated by Fig. 5 which shows that the buy volume exceeds the sell volume, whether measured by the target coin BVB or by BTC. The figure also shows small volume movements shortly before the pump-and-dump, also observable in Fig. 4(a), which can be indicative of organizers’ pre-purchase conduct. Note again that the BVB blockchain is not being actively maintained and the coin itself is extremely illiquid, so any market movement can be deemed unusual.

Fig. 5 illustrates that the total buy volume (also including the pre-purchased volume, though negligible) in BTC associated with the pump-and-dump amounts to 1.06 BTC, the sell volume only 0.58 BTC; the total buy volume measured in BVB is 1,619.81 thousand BVB, the sell amount 1,223.36 thousand BVB. This volume discrepancy between the sell and the buy sides indicates a higher trading aggressiveness on the buy side.888Note that Cryptopia is a peer-to-peer trading platform which lets users trade directly with each other; the exchange takes no risk position and only profits from charging trading fees. Therefore, buying volume implies that the trade is initiated by the buyer, which typically drives the market price up; similarly, sale volume is initiated by the sell side and would drive the price down. This further suggests that many investors may be “stuck” with BVB which they are unwilling to liquidate at the low market price after the pump-and-dump. Those coin holders can only expect to reverse the position in the next pump, which might never come.

Low transaction volume: It is worth noting that the total count of trading transactions associated with this pump-and-dump is merely 322. That number appears very low compared to the 1,376 views of the coin announcement message, let alone the over 10,000 channel members. This indicates that the majority of group members are either observers, who want no skin in the game, or have become aware of the difficulty in securing profit from a pump-and-dump.

Fig. 6: Cumulative counts of pumps and pumped coins over time on four exchanges.

Iv Analyzing Pump-and-Dump Schemes

In this section we explain how we obtain data from both Telegram and the various exchanges, which allows us to analyze and model pump-and-dump schemes.

Iv-a Data Collection

Telegram channels are the primary medium for pump-and-dump activity and announcements.

Our primary source on pump-and-dump Telegram channels and events is provided by PumpOlymp,999 a pump-and-dump information website. PumpOlymp gleans pump-and-dump information from across Telegram channels and publishes pump-and-dump agendas on a continuous basis. They also host a comprehensive directory of hundreds of pump-and-dump channels.

With the channel list from PumpOlymp as the starting point, we use the official Telegram API to retrieve message history from those channels regarding their status and activity examination. We also employ hand collection (by searching channels with keyword “pump”) to to cross-check for missing and incorrect data.

Telegram channels: In the end, we arrive at a list of 358 Telegram channels that exhibit traits of being a pump-and-dump organization, e.g. the word “pump" contained in screen name or user name. Among those channels, 43 have been deleted from the Telegram sever, possibly due to inactivity for an extended period of time. Among the existing ones, over half (168/315) have not been active for a month. This might be because cautious admins delete pump-and-dump messages to not leave any trace behind. This might also imply that the Telegram channels have the “hit-and-run" characteristic. As described in the section above, one learns from participation in pump-and-dump activities that quick bucks are not easy to get. Therefore, curious newcomers might be fooled by pump-and-dump organizers’ advertising and lured into the activity. After losing money a few times, participants may lose faith and interest, and cease partaking. This forms a vicious circle, since with fewer participants, it would be more difficult to pump a coin. Therefore, channel admins might desert their channel when the performance declines, and start new ones to attract the inexperienced.

Exchange Volume (30d) No. markets Launch Country
Binance $21,687,544,416 385 Jul 2017 China
Bittrex $1,168,276,090 281 Feb 2014 U.S.A.
Cryptopia $107,891,577 852 May 2014 New Zealand
YoBit $797,593,680 485 Aug 2014 Russia
Fig. 7: Exchanges involved in pump-and-dump schemes, sorted by 30-day volume: No. markets is the number of trading pairs (eg. DASH/BTC, ETC/USDT) in the exchange. Volume and No. markets were extracted from CoinMarketCap on November 5, 2018.

Pump-and-dump history: PumpOlymp also has a continuously updated web page that lists Telegram organized pump-and-dump’s not older than 3 months. By scraping the coin pump history listed on PumpOlymp multiple times, we acquire an initial list of historical pump-and-dump activities that include the pumped coin, the target exchange, the organizing Telegram channel, and the start time. We run plausibility checks to ensure each record’s qualification as a pump-and-dump. For example, if an alleged pump-and-dump is recorded to have started at a time that is far from a full hour (6:00, 7:00, etc.) or a half hour, then we would be suspicious, because an organizer would normally not choose a random time for a pump-and-dump. If there is no significant increase in volume or high price around the pump time, we would also be skeptical. In such a circumstance, we manually check the message history to make a final judgment. In most cases, the message either discusses the potential of a coin (as opposed to a coin announcement in a pump-and-dump) or the record is simply a mistake.

In total, we trace 236 pump-and-dump coin announcements from July 21, 2018 to November 18, 2018, each of which is characterized by a series of messages similar to those presented in Fig. 1(a). One pump-and-dump can be co-organized by multiple channels; if two coin announcements were broadcast within 3 minutes apart from each other and they target the same coin at the same exchange, then we consider them to be one pump-and-dump event. In total, we have collected 220 unique pump-and-dump events.

Excluded data points: All the pumped coins in our sample are paired with BTC. We also observed and manually collected a few ETH-paired pumps, most of which took place in other exchanges.101010For example, PLX on October 10 in CoinExchange, ETC on April 22 in Bibox. Inclusion of those cases would demand data collection with other methods and resources. Due to their rarity, we do not consider ETH-paired pump-and-dump’s in our study.

Pump-and-dump distribution by exchange: Among the 220 pump-and-dump activities, 24 (11%) took place in Binance, 8 (4%) in Bittrex, 148 (67%) in Cryptopia and 40 (18%) in Yobit. In aggregate, 25.9% (56/220) of the time, the selected coin has been pumped in the same exchange before (see Fig. 6).

Fig. 8: Aggregate trading volume of pumped coins before and during a pump.

Iv-B Obtaining Coin Data

Apart from consulting the online pump-and-dump information center PumpOlymp, we retrieve additional information on features and price movements of coins from other sources, in order to establish a connection between the information and the pump-and-dump pattern.

Specifically, we use the public API from CryptoCompare111111 for coins’ hourly OHLC (open, high, low, close) and volume data on 189 exchanges, including Binance, Bittrex, Cryptopia and Yobit. The website also provides historical minute-level data but they are restricted to a 7-day time window and thus not utilized. We further use the public API from CoinMarketCap to collect coins’ market data. While it might be useful to also collect coins’ historical market cap before each pump-and-dump, we have not found a public source that provides this type of data. Therefore, we purposefully chose to retrieve the data at 08:42 GMT, November 5, when we believe the market cap would not be influenced by any Telegram organized pump-and-dump, since they typically start on the hour or the half hour and last only a few minutes.

In addition to market trading data, we also retrieve coins’ non-financial features. Specifically, we use exchanges’ public API121212 for Binance, for Bittrex, for Cryptopia, and for Yobit. to collect information on coins’ listing status, algorithm, and total supply. We also collect coins’ launch date using CryptoCompare’s public API. For information that is not contained in the API but viewable online (such as coins’ rating data on Cryptocurrency), we use either page source scraping or screen scraping, depending on the design of the desired webpage. All our data are from publicly accessible sources.

(a) Pump and dump activities from July to November 2018

(b) Enlarged section of the highlighted area in (a) that shows one of the most recent pump-and-dump
Fig. 9: Pump and dump timeline. A green bar represents price increase through pump, calculated as ; a red bar represents price drop after pump, calculated as . All prices are denominated in BTC, and from a 3-hour window around pump activities. Visit for the full, up-to-date, interactive chart.

Iv-C Role of Exchanges

Pump-and-Dump schemes take place within the walled gardens of crypto-exchanges. Binance, Bittrex, Cryptopia, and Yobit are among the most popular exchanges used by pumpers (see Fig. 7). While those exchanges differ vastly in terms of their volume, markets, and user base, each of them has its own appeal to pumpers. Large exchanges such as Binance and Bittrex have a large user base, and abnormal price hype caused by pump activities can quickly attract a large number of other users to the exchange. Smaller exchanges such as Cryptopia and Yobit tend to host esoteric coins with low liquidity, whose price can be more easily manipulated compared to mainstream coins such as Ether (ETH) or Litecoin (LTC).

Fig. 10: Views of coin announcement message versus coin price increase during the pump. The figure illustrates the relationships between coin price increase through pump, views of coin announcement message, pump volume, and pump exchange.

Comparing exchanges: Fig. 8 compares the aggregate three-hour trading volume in BTC of pumped coins before and during a pump-and-dump and the artificial trading volume generated by those pump-and-dump activities is astonishing: 4,238 BTC (95% of which come from Binance), roughly equivalent to 25 million USD, of trading volume during the pump hours, 14 times as much as the pre-pump volume (294 BTC), and that only over a period of three and half months.

Fig. 9 illustrates the occurrence and the effectiveness of individual pump-and-dump activities. In terms of frequency, Bittex is most rarely chosen; Binance started to gain traction only since September, but still witnesses much less pump-and-dump occurrence than Yobit and Cryptopia. Comparing Yobit with Cryptopia, we find that the former is becoming more inactive with the passage of time, while the latter is increasingly gaining more traffic. In terms of percentage of coin price increase, pumps in both Yobit and Cryptopia appear to be more powerful than those in Bittrex and Binance. What goes hand in hand with price surge is price dip: coin prices also drop more dramatically during the dump in Yobit and Cryptopia compared to their peer exchanges.

Coin announcement views: While investigating the degree of exposure in coin announcement messages distributed by Telegram channels, we find a negative correlation (-0.162) between number of views of coin announcement and pump gain, which is quite counter-intuitive, because one would think that more views would indicate more participation, which would result in higher pump gain. Two extreme examples: the coin announcement of the pump on MST has 325 views and the pump gain was 12.6%; another coin announcement of the pump on PARTY had only 18 views, and the pump gain was a whopping 533.3%. This suggests the use of bots to read the coin announcement message (which does not require membership of the group) is involved in trading.

Price increase: We further notice that although significantly more people participated in pump-and-dump in Binance (participation proxied by views of coin announcement message) — because of its large user base — and generated more trading volume during the pump hour,131313A pump hour refers to the clock hour during which a pump occurs. Due to constraints with data availability, we are only able to obtain hourly market data. coin price increase through pumps is generally at a much smaller scale than its equivalent in Cryptopia and Yobit (Fig. 10). This is possibly caused by high bid and sell walls on the order book that are typical to large crypto exchanges like Binance, which prevent the price from fluctuating significantly even at coordinated pump-and-dump events.

Fig. 11: Arbitrage opportunities: coin price (highest during the pump hour) in pumped exchange versus price in other exchanges

Arbitrage: Pump-and-dump activities not only engender abnormal returns within the pumped exchange, but also arbitrage opportunities across different exchanges. Fig. 11 shows the presence of a price discrepancy of the same coin during the pump hour across different exchanges. Interestingly, coin price can sometimes be higher in exchanges other than the pumped one. It is also worth noting that most coins pumped in Cryptopia are also listed in Yobit but not in Bittrex or Binance, and vice versa. This is because the former two have more conservative coin listing strategies, which results in a different, more mainstream portfolio of listed coins compared to the latter two. While there may be trading strategies resulting from these arbitrage opportunities, they are outside the score of this work.

Fig. 12: Distribution of coin market caps. Market cap information was extracted from CoinMarketCap on November 5, 2018.

Iv-D Capturing the Features

Fig. 12 presents the market cap distribution of coins pumped in different exchanges. Pumped coins’ market cap ranges from 1 BTC (Royal Kingdom Coin (RKC), pumped in Cryptopia) to 27,600 BTC (TrueUSD (TUSD), pumped in Yobit). Half of those coins have a market cap below 100 BTC, most of which were pumped in Cryptopia.

Pump-and-Dump organizers who favor Cryptopia are attracted by the wide range of coins with low market cap listed on the exchange. Unsurprisingly, small coins are more likely to be associated with scams, leading to potential delisting. As of November 15, 2018, 20 (21%) coins among the 99 pumped in Cryptopia between July 21, 2018 and November 18, 2018, have already been delisted. Coins pumped in other exchanges remain listed.141414ICONOMI (ICN), a coin that has been pumped on Binance, was delisted from the exchange for a brief period before being re-listed. The delisting occurred after the announcement that ICN would be converted into a security token, which is not allowed to be traded on Binance. As far as we know, no coin-specific reasons are quoted for the delisting decisions on Cryptopia.

Fig. 13 depicts time series of coin returns up to 48 hours before and 3 hours after a pump. We detect unusual return signals even an hour before the announcement of the pumped coin. The return signal before the pump is the strongest with Cryptopia, where in numerous pumps, coin prices were elevated to such an extent that the hourly return before the pump even exceeds the hourly return during the pump. This can be explained by the assumption that pump organizers might utilize their insider information to purchase the to-be-pumped coin before the coin announcement, causing the coin price elevation and usual return volatility before the pump. The analysis above provides grounds for predicting the pumped coin before coin announcement using coin features and market movement.

Fig. 13: Time series of coin returns before and after pump. In each subplot, the hourly log return of each pumped coin before and shortly after the pump is superimposed on each other. The vertical red line represents the pump hour during which the coin was announced.
Feature Description Notation
Market cap Market cap information extracted from CoinMarketCap at 08:42 GMT, November 5, 2018 when no pump-and-dump activity in Telegram channels was observed
Pumped times before Number of times the coin been pumped in Cryptopia before
Last price before pump Open price of the coin one hour before the coin announcement
Returns before pump -hour log return of the coin within the time window from  hours to 1 hour before the pump
Volumes in coin before pump Total amount of the coin traded within the time window from  hours to 1 hour before the pump
Volumes in BTC before pump Total trading volume of the coin measured in BTC within the time window from  hours to 1 hour before the pump
Return volatilities before pump Volatility in the hourly log return of the coin within the time window from  hours to 1 hour before the pump
Volume volatilities in coin before pump The volatility in the hourly trading volume in coin within the time window from  hours to 1 hour before the pump
Volume volatilities in BTC before pump The volatility in the hourly trading volume in BTC within the time window from hours to 1 hour before the pump
Time since existence The time difference between the time when the first block of the is mined and the pump time
Coin rating Coin rating displayed on Cryptopia, 0 being the worst, 5 being the best. The rating considers the following criteria wallet on {Windows, Linux, Mac, mobile, web, paper}, premine ratio, website and block explorer
Withdraw fee Amount of coin deducted when withdrawing the coin from Cryptopia
Minimum withdraw Minimum amount of coin that can be withdrawn from Cryptopia
Maximum withdraw Daily limit on the amount of coin that can be withdrawn from Cryptopia
Minimum base trade Minimum base trade size of the coin
Fig. 14: Features included in the prediction model. . .
Pumped? Training Validation Test Total
TRUE 78 28 27 133 (0.3%)
FALSE 27,681 10,078 9,728 47,487 (99.7%)
Total 27,759 10,106 9,755 47,620 (100.0%)
Fig. 15: Sample split.

V Predicting Pump-and-Dump Schemes

V-a Feature Selection

Based on the preliminary analysis in the last section, we believe it would be possible to reverse engineer pump-and-dump organizers’ coin selection criteria using coin features and market movements. For the ease of standardization of data and due to its high pump-and-dump frequency, we focus on predicting coins pumped in Cryptopia.

For each coin at each pump point, we predict whether it will be pumped (TRUE) or not (FALSE). The formula for the prediction model is:

where the dependent variable is a binary variable that equals 1 (TRUE) when the coin is selected for the pump, and 0 (FALSE) otherwise. Fig. 14 lists the features considered in the prediction model.

Previous analyses indicate unusual market movements prior to the pump-and-dump might signal organizers’ pre-pump behavior, which could consequently give away the coin selection information. Therefore, we place great emphasis on features associated with market movements, such as price, returns and volatilities covering different lengths of time span. Those features, 46 in total, account to 85% of all the features considered.

V-B Model Application

Sample specification: We consider all the coins listed on Cryptopia at each pump-and-dump event. On average, we have 358 coin candidates at each pump, out of which one is the actual pumped coin. The number of coins considered varies for each event due to constant listing/delisting activities on the part of exchanges. The full sample contains 47,487 pump-coin observations, among which 133 are pumped cases,151515Due to missing data on several delisted coins, this number deviates from the total number of 148 pump events in Cryptopia, as presented in Fig. 6. accounting for 0.3% of the entire sample population. The sample is apparently heavily skewed towards the unpumped class and needs to be handled with care at modelling.

For robustness tests, we split the whole sample into three chronologically consecutive datasets: training sample, validation sample and and test sample (Fig. 15). The training sample covers the period of July 21 to October 10 and consists of 27,759 data points (58.3% of full sample), among which 78 are pumped cases; the validation sample covers October 10 to October 29 and consists of 10,106 data points (21.2% of full sample), among which 28 are pumped cases; the test sample covers October 29 to November 18 and consists of 9,755 data points (20.5% of full sample), among which 27 are pumped cases.

Model selection: We test both classification and logit regression models for the prediction exercise. Specifically, for the classification model, we choose random forest (RF) with stratified sampling; for the logit regression model, we apply generalized linear model (GLM). Both RF and GLM are widely adopted in machine learning and each has its own quirks.

RF is advantageous in handling large quantities of variables with no overffiting issues. In addition, RF is resilient to correlations, interactions or non-linearity of the features, and one can be agnostic about the features. On the flip side, RF relies upon a voting mechanism based on a large number of bootstrapped decision trees, which can be time-consuming, and thus challenging to execute. In addition, RF provides information on feature importance, which is less intuitive to interpret than coefficients in GLM.

GLM is a highly interpretable model [21] that can uncover the correlation between features and the independent variable. It is also highly efficient in terms of processing time, which is a prominent advantage when coping with large datasets. However, the model is prone to overfitting when fed with too many features, which potentially results in poor out-of-sample performance.

Hyperparameter specification: Due to the heavily imbalanced nature of our sample, we stratify the dataset when using RF, such that the model always includes TRUE cases when bootstrapping the sample to build a decision tree. Specifically, we try the following three RF variations:

Sample size per tree Number
Model TRUE FALSE Total of trees
RF1 60 20,000 20,060 5,000
RF2 60 5,000 5,060 20,000
RF3 60 1,000 1,060 40,000

We fix the number TRUE’s at 60 for each RF variation, so that the model may use the majority of TRUE’s to learn their pattern when building each tree. Model RF1 stays loyal to our sample’s original TRUE/FALSE ratio, with 0.3% of TRUE’s contained in each tree-sample. RF2 and RF3 raise the TRUE/FALSE ratio to 1.2% and 6%, respectively. Note that while the sample size per tree decreases from RF1 to RF2 to RF3, we are mindful to increase the number of trees accordingly to ensure that whichever model we use, every input case is predicted a sufficient number of times. We use the R package randomForest to model our data with RF1, RF2 and RF3.

With conventional binomial GLM, problems can arise not only when the dependent variable has a skewed distribution, but also when features are skewed. With heavy-tailed coin price distribution and market cap distribution, conventional binomial GLM can be insufficient to handle our sample. Therefore, we apply LASSO (least absolute shrinkage and selection operator) regularization to the GLM models. After preliminary testing, we choose to focus on three representative LASSO-GLM models with various shrinkage parameter values (). Higher values of causes elimination of more variables. We use the R package glmnet to model our data with GLM1, GLM2, and GLM3.

Model Shrinkage parameter ()
1.31 1.65 2.52
2.14 2.02 1.66
8.52 8.53 7.67
4.60 6.65 7.30
2.88 3.83 4.62
2.89 3.22 3.88
2.53 2.59 2.63
2.45 2.84 3.68
3.84 3.95 4.17
2.65 2.71 3.10
2.55 2.95 3.60
2.22 2.45 2.84
1.53 1.34 1.21
1.58 1.41 1.14
1.70 1.60 1.38
1.85 1.68 1.37
1.84 1.69 1.41
1.93 1.81 1.53
1.95 1.87 1.66
2.64 3.27 3.19
1.85 1.87 1.60
1.86 1.70 1.45
2.23 2.07 1.79
2.42 2.18 1.87
2.31 2.19 1.86
2.40 2.30 2.01
2.81 2.51 2.16
2.02 2.30 3.07
2.08 1.94 1.87
2.17 1.96 1.73
2.22 1.99 1.78
2.44 2.10 1.69
2.39 2.17 1.80
2.30 2.09 1.67
1.39 1.34 1.31
1.57 1.42 1.16
1.65 1.51 1.25
1.75 1.55 1.21
1.81 1.56 1.22
1.79 1.56 1.25
1.81 1.66 1.33
1.86 2.06 1.96
1.77 1.74 1.52
2.10 1.94 1.70
2.16 1.94 1.65
2.12 1.96 1.64
2.15 1.99 1.67
2.26 2.04 1.70
1.77 1.64 1.37
0.73 0.71 0.63
1.02 1.03 0.98
0.43 0.36 0.28
0.00 0.00 0.00
2.20 1.88 1.69
Fig. 16: Features’ importance indicated by mean decrease in Gini coefficient. Higher importance is marked by darker cell color.
0.69 0.66 -
-91.74 - -
0.00 - -
2.76 4.75 5.02
-0.04 - -
1.08 - -
-4.81 - -
1.41 0.11 -
3.64 2.33 -
0.07 - -
1.21 - -
0.00 - -
-0.00 - -
- - -
- - -
- - -
0.00 - -
- - -
- - -
1.61 - -
5.99 - -
- - -
- - -
- - -
- - -
-2.88 - -
-0.49 - -
3.94 - -
4.41 - -
-9.39 - -
10.40 - -
9.10 - -
-12.57 - -
-3.93 - -
-0.00 - -
0.00 - -
- - -
- - -
- - -
- - -
-0.00 - -
-7.46 - -
1.32 - -
-9.96 - -
-2.13 - -
18.83 - -
- - -
8.65 - -
-0.16 - -
-0.00 - -
-0.00 - -
0.00 - -
- - -
0.00 - -
(Intercept) -5.43 -6.15 -5.95
Fig. 17: Variable coefficients using GLM. Coefficients of variables not selected by the model are shown as “-".
(a) Training sample.
(b) Validation sample.
Fig. 18: Model performance measured by Precision and at different threshold levels.

Variable assessment: By applying the specified models on the training sample, we are able to assess the features’ relevance to coin prediction. Fig. 16 presents features’ importance based on mean decrease in Gini coefficient with RF models. We find that:

  • Coin market cap and last hour return before the pump appear to the two most important features in predicting pumped coin using RF models;

  • Features describing market movements shortly before the pump, e.g. , and , appear to be more important than features describing longer-term movements.

  • Among all the features related to market movements, return features are generally more important than volume or volatility features.

  • Exchange-specific features including , , , and are least important.

Fig. 17 presents the estimated coefficients of variables with GLM models, from which we can form several findings in line with what is indicated by RF models above. Specifically, we notice that:

  • When only one variable is included, appears to have the highest explanatory power on coins’ pump likelihood;

  • The positive coefficients of return features imply that the higher the return a coin shows before the pump, the more likely the coin is to be pumped;

  • The positive coefficient of implies that pumped coins are more likely to get pumped again.

V-C Assessing Prediction Accuracy

Both the random forest model and GML are able to predict whether a given coin will be pumped as a likelihood ranging between 0 and 1. We apply thresholding to get a binary TRUE/FALSE answer.

Fig. 18 depicts the change in precision and the measure, as the threshold value changes. Fig. 18(a) describes models’ in-sample fitting with the training sample and Fig. 18(a) their out-of-sample accuracy with the validation sample.

Precision represents the number of true positive divided by number of predicted positive, and the precision line ends when the denominator equals zero, i.e. when no TRUE prediction is produced. Fig. 18 shows that, among the three RF models, the threshold value at which the line ends is the lowest with RF1, and highest with RF3. This indicates that absent balanced bootstrapping, an RF model tends to systematically underestimate pump likelihood, leading to zero predicted TRUE cases even when the threshold value is small.

Similarly among the GLM models, GLM3 (with highest ) seems to underestimate pump likelihood more severely than GLM2 and GLM1. Between the latter two, GLM2 exhibits higher Precision than GLM3.

In terms of measure, RF models in general appear superior to GLM models both with the training sample and the validation sample. Among the three RF models, the RF1 performs best at a low threshold range (), while RF3 performs best at a high threshold range (). RF2 resides in between.

(a) Training sample.
(b) Validation sample.
Fig. 19: Model performance measured by ROC AUC at different threshold levels.

The RF models’ superiority is further demonstrated by the ROC (Receiver operating characteristic) curve in Fig. 19. While GLM1’ decent in-sample performance () is comparable to RF models, its diminished out-of-sample performance () indicates overfitting. This mirrors Fig. 18, where GLM1 is observed to fit the training sample relatively well, but poorly with the validation sample. Among the three RF models, no discernible difference can be found in terms of ROC AUC.

V-D Testing an Investment Strategy

To explore the model’s practical utility, we devise a simple investment strategy. At each pump, we check which coin’s normalized vote surpasses a predetermined threshold, and we will purchase all those coins before the actual coin announcement (if no coin’s vote exceeds the threshold, we will not pre-purchase any coin). Note that if we had the ability to short or use margin trading on the exchanges we use, potentially more options would open up for us.

Strategy: Specifically, for each coin that we pre-purchase, we buy the coin at the open price one hour before the coin announcement with the amount of BTC equivalent to times the vote where is a constant. That is to say, with all the coins we purchase, the investment, measured in BTC, on each coin is proportionate to its vote supplied by the random forest model. This is logical because a higher vote implies a higher likelihood of being pumped, and thus worth a higher investment.

We further assume that among all the coins we purchased, those coins that do not get pumped (false positive, “false alarm") will generate a return of zero, i.e. their price will remain at the same level as the purchase price; those coins that get pumped (true positive, “hits") will be sold at an elevated price during the pump. To be conservative, we assume that we only obtain half of the pump gain

with each purchased coin that gets pumped.

(a) Training sample.
(b) Validation sample.
Fig. 20: Model performance measured by return on trading investment at different threshold levels.

Returns: Fig. 20 presents the relationship between the aggregate return and the threshold choice. The figure shows that, in general, the higher the threshold, which means we buy coins with higher pump likelihoods and disregard others, the higher the return.

Note that the line shape for each model has a striking resemblance with the model’s Precision line from Fig. 18. This is due to the fact that the strategy is prediction-based (so is the denominator of Precision) and rewards true positive scenarios (so is the numerator of Precision). This gives us guidance on the selection of performance measurements that should be aligned with the end goal. In our specific case, maximizing Precision is more important than maximizing the or AUC.

As already indicated by Fig. 18, every model has its own optimal threshold value and one should be mindful that if the threshold is set too high (e.g., greater than 0.9 with RF1, or greater than 0.8 with GLM3), then the investor might end up not buying any coins, and consequently gaining no profit. In terms of the magnitude of the profit, with the right combination of threshold and model, investors would theoretically enjoy a return of 140% with the training sample cases (covering 11 weeks), and a return of 100% with the validation sample cases (covering  3 weeks).

V-E Final Test

Based on the training and validation results of specified models, we need to select one model and an accompanying threshold value to apply to the test sample. Our ultimate goal to maximize the trading profit using the selected model in combination with the proposed trading strategy (as opposed to e.g. maximizing ). Therefore, we base our decision primarily on Fig. 20.

As already established, all the specified models but GLM1 seem able to deliver high return with an appropriately chosen threshold value. For demonstration purposes, we arbitrarily choose RF2 as our final model. While Fig. 20(a) shows that with RF2, the return reaches its highest at the threshold of 0.8, Fig. 20(b) shows that with out-of-sample validation, the model further underestimates pump likelihood, and the optimal threshold would be around 0.7. While we understand the higher threshold is associated with higher return, to be conservative and to make sure that TRUE predictions will be produced, we choose a slightly lower threshold of 0.6. Fig. 21 displays the confusion matrix of the model prediction.

Actual TRUE 5 22 27
FALSE 1 9,727 9,728
Total 6 9,749 9,755
Fig. 21: Confusion matrix of RF2 with threshold value 0.6 applied to test sample.

Fig. 21 shows that the model suggests us to purchase 6 coins in total, among which 5 are actually pumped, and 1 not. Fig. 22 lists those 6 coins, their respective investment weight and assumed profit. The return on the investment amounts to 79.5% (3.21/4.04) over the test sample period of 3 week. The result is very similar to that with the validation sample as shown in Fig. 20(b), confirming the model’s robustness.

Invest. Pump Assumed Assumed
Coin Date Pumped? weight gain gain profit
8BIT Oct 31 TRUE 0.61 117% 58% 0.36
DRPU Nov 1 TRUE 0.72 396% 198% 1.43
ERY Nov 1 TRUE 0.72 57% 28% 0.20
EZT Nov 6 TRUE 0.75 224% 112% 0.84
TAJ Oct 30 TRUE 0.63 120% 60% 0.38
XWC Nov 4 FALSE 0.61 - - -
Total 4.04 3.21
Fig. 22: Purchased coins based on pump likelihood predicted by RF2. Only coins with predicted pump likelihood of greater than 0.6 are purchased. Investment weight equals pump likelihood.

Vi Related Work

To the best of our knowledge, the only existing study with the same research subject — Telegram organized pump-and-dump activities — is a recent working paper by Li et al. [17], which, different from our study, focuses on the impact of pump-and-dump on the liquidity and price of cryptocurrencies.

Our paper is closely linked to the limits of literature on crypto trading. Gandal et al. [14] demonstrates that the unprecedented spike in the USD-BTC exchange rate in late 2013 is possibly caused by price manipulation. Makarov et al. [18] probe arbitrage opportunities in crypto markets. Aune et al. [2] highlights potential manipulation in the blockchain market resulting from the exposure of the footprint of a transaction after its broadcast and before its validation in a blockchain, and proposes a cryptographic approach for solving the information leakage problems in distributed ledgers.

Our paper is also akin to existing literature on cryptocurrencies’ market movements. The majority of related literature still presses the focus on Bitcoin. Many scholars use GARCH models to fit the time series of Bitcoin price. Among them, Dyhrberg et al. [11] explore the financial asset capabilities of Bitcoin and suggests categorizing Bitcoin as something between gold and US Dollar on a spectrum from pure medium of exchange to pure store of value; Bouoiyour et al. [6] argues that Bitcoin is still immature and remains reactive to negative rather than positive news by the time of their writing; 2 years later, Conrad et al. [8] presents the opposite finding that negative press does not explain the volatility of Bitcoin; Dyhrberg [12] demonstrate that bitcoin can be used to hedge against stocks; Katsiampa [16] emphasize on modelling accuracy and recommend the AR-CGARCH model for price retro-fitting. Bariviera et al. [3] compute the Hurst exponent by means of the Detrended Fluctuation Analysis method and conclude that the market liquidity does not affect the level of long-range dependence. Corbet et al. [9] demonstrates that Bitcoin shows characteristics of an speculative asset rather than a currency also with the presence of futures trading in Bitcoin.

Among the few research studies that also look into the financial characteristics of other cryptocurrencies, Fry et al. [13] examine bubbles in the Ripple and Bicoin markets; Baur et al. [5] investigates asymmetric volatility effects of large cryptocurrencies and discover that in the crypto market positive shocks increase the volatility more than negative ones. Jahani et al. [15] assess whether and when the discussions of cryptocurrencies are truth-seeking or hype-based, and discover a negative correlation between the quality of discussion and price volatility of the coin.

Vii Conclusions

This paper presents the first detailed study of pump-and-dump schemes in cryptocurrency markets. We start by presenting the anatomy of a typical attack and then investigate a variety of aspects of real attacks on crypto-coins over a period of July 21, 2018 — November 18, 2018 on four crypo-exchanges. The study demonstrates the persisting nature of pump-and-dump activities in the crypto-market that are the driving force of tens of millions of dollars of phony trading volumes each month. The study reveals that pump-and-dump organizers can easily use their insider information to take extra gain at a pump-and-dump event at the sacrifice of fellow pumpers.

Through market investigation, we further discover market movements prior to a pump-and-dump events that frequently contains information on witch coin will be pumped. Using LASSO regularized GML and random forests, we build various models that are predicated on the time and venue (exchange) of a pump-and-dump broadcast in a Telegram group. Multiple models display high performance (AUC>0.9 for both the training and the testing samples), implying that pumped coins can be predicted based on market information.

We further propose a simple but powerful trading strategy that can be used in combination with the predicting models. Out-of-sample tests show that a return of as high as 80% over three weeks can be consistently exploited even under conservative assumptions. The study thus sheds light on the application of machine learning for crypto-trading.


  • [1] Nicola Atzei, Massimo Bartoletti, and Tiziana Cimoli. A Survey of Attacks on Ethereum Smart Contracts. Proceedings of the 6th International Conference on Principles of Security and Trust, 10204:164–186, 2017.
  • [2] Rune Tevasvold Aune, Adam Krellenstein, Maureen O’hara, and Ouziel Slama. Leakage in Distributed Ledgers Trading and Information Footprints on a Blockchain. Journal of Trading, 12(3):5–13, 2017.
  • [3] Aurelio F. Bariviera, María José Basgall, Waldo Hasperué, and Marcelo Naiouf. Some stylized facts of the Bitcoin market. Physica A: Statistical Mechanics and its Applications, 484:82–90, 2017.
  • [4] Massimo Bartoletti, Salvatore Carta, Tiziana Cimoli, and Roberto Saia. Dissecting Ponzi schemes on Ethereum: identification, analysis, and impact. 2017.
  • [5] Dirk G. Baur and Thomas Dimpfl. Asymmetric volatility in cryptocurrencies. Economics Letters, 173:148–151, 2018.
  • [6] Jamal Bouoiyour and Refk Selmi. Bitcoin: A beginning of a new phase? Economics Bulletin, 36(3):1430–1440, 2016.
  • [7] CFTC. Customer Advisory: Beware Virtual Currency Pump-and-Dump Schemes. 2018.
  • [8] Christian Conrad, Anessa Custovic, Eric Ghysels, Christian Conrad, Anessa Custovic, and Eric Ghysels. Long- and Short-Term Cryptocurrency Volatility Components: A GARCH-MIDAS Analysis. Journal of Risk and Financial Management, 11(2):23, 2018.
  • [9] Shaen Corbet, Brian Lucey, Maurice Peat, and Samuel Vigne. Bitcoin Futures – What use are they? Economics Letters, 172:23–27, 2018.
  • [10] Crypto Insider. CFTC offers $100,000+ bounty for crypto pump and dump whistleblowers.
  • [11] Anne Haubo Dyhrberg. Bitcoin, gold and the dollar - A GARCH volatility analysis. Finance Research Letters, 16:85–92, 2015.
  • [12] Anne Haubo Dyhrberg. Hedging capabilities of bitcoin. Is it the virtual gold? Finance Research Letters, 16:139–144, 2016.
  • [13] John Fry and Eng-Tuck Cheah. Negative bubbles and shocks in cryptocurrency markets. International Review of Financial Analysis, 47:343–352, 2016.
  • [14] Neil Gandal, J. T. Hamrick, Tyler Moore, and Tali Oberman. Price manipulation in the Bitcoin ecosystem. Journal of Monetary Economics, 95(4):86–96, 2018.
  • [15] Eaman Jahani, Peter M. Krafft, Yoshihiko Suhara, Esteban Moro, and Alex Pentland. ScamCoins, S*** Posters, and the Search for the Next Bitcoin™: Collective Sensemaking in Cryptocurrency Discussions. Proceedings of the ACM on Human-Computer Interaction, 2:79, 2018.
  • [16] Paraskevi Katsiampa. Volatility estimation for Bitcoin: A comparison of GARCH models. Economics Letters, 158:3–6, 2017.
  • [17] Tao Li, Donghwa Shin, and Baolian Wang. Cryptocurrency Pump-and-Dump Schemes. 2018.
  • [18] Igor Makarov and Antoinette Schoar. Trading and Arbitrage in Cryptocurrency Markets. 2018.
  • [19] SEC. Ponzi schemes Using virtual Currencies Ponzi Schemes Generally. Investor Alert, (153), 2017.
  • [20] SEC. SEC Files Subpoena Enforcement Against Investment Company Trust and Trustee for Failure to Produce Documents. 2018.
  • [21] Lin Song, Peter Langfelder, and Steve Horvath. Random generalized linear model: a highly accurate and interpretable ensemble predictor. BMC bioinformatics, 14:5, 2013.
  • [22] Oscar Williams-Grut. ’Market manipulation 101’: ’Wolf of Wall Street’-style ’pump and dump’ scams plague cryptocurrency markets. Business Insider, 2017.


Group name Link to the group Last message sent
1 SAFE BUNKER SIGNAL 2018-11-21 11:06:41
2 Palm Venice Beach 2018-11-21 10:56:40
3 Trading Crypto Guide 2018-11-21 10:36:24
4 SUB 4 SUB . 2018-11-21 10:21:42
5 TRADING CRYPTO KING 2018-11-21 10:18:58
6 Signal Express 2018-11-21 10:05:40
7 Signal Expert 2018-11-21 09:25:59
8 Trading Crypto Coach 2018-11-21 08:58:42
9 Moonbow Signal 2018-11-21 07:24:43
10 Moon Team 2018-11-21 07:24:42
11 Moonlight Signal 2018-11-21 07:24:42
12 OFFICIAL SIGNAL EXPERT 2018-11-21 05:50:24
13 Premium Pump 2018-11-21 05:41:46
14 Pump Up 2018-11-21 05:41:44
15 World Pumps 2018-11-21 05:41:39
16 Shark Pump Signals 2018-11-21 05:41:28
17 Wall Street PUMPS 2018-11-21 05:41:27
18 Optimus Pump 2018-11-21 05:41:26
19 Yo Crypto Signals 2018-11-21 04:10:17
20 Meg Blue Whale 2018-11-21 01:27:18
21 Trading Crypto Assistant 2018-11-21 00:17:49
22 Whales Crypto Guide 2018-11-20 23:48:14
23 Vip Trading Crypto 2018-11-20 22:00:05
24 Official McAfee Pump Signals 2018-11-20 20:57:49
25 Trades For Crypto Community 2018-11-20 20:42:34
26 Crypto Coins 2018-11-20 20:33:14
27 Crypto Experts Signal 2018-11-20 19:52:52
28 Coopet 2018-11-20 19:48:54
29 Bomba bitcoin "cryptopia" 2018-11-20 19:48:04
30 Captain’s signal - Bittrex 2018-11-20 19:04:39
31 Crypto Trade 2018-11-20 18:44:23
32 VICTOR’S - PUMP SIGNALS FOR THE CLASSES 2018-11-20 18:19:01
33 Team Pumps 2018-11-20 18:06:00
34 Bittrex Super Signals 2018-11-20 18:03:53
35 Cryptonians (Free Signals and News Update) 2018-11-20 17:48:46
36 Binance Profit Signal 2018-11-20 17:07:13
37 2018-11-20 17:07:13
38 Knh cp nht chin lc giao dch alt coin 2018-11-20 17:04:28
39 WhalesCRYPTOOFFICIAL 2018-11-20 17:04:25
40 TOP TRADING 2018-11-20 16:42:11
41 Mega Pumps 2018-11-20 16:34:19
42 Private Group For Pump 2018-11-20 16:34:03
43 Crypto God 2018-11-20 16:34:03
44 Crypto Profits 2018-11-20 16:34:03
45 Crypto Pump Island 2018-11-20 16:23:50
46 Super Pump Bros 2018-11-20 16:20:00
47 Bitcoin Pump group 2018-11-20 16:18:45
48 LEAK Signal 2018-11-20 15:49:22
49 PumpMyWallet 2018-11-20 15:34:20
50 Crypto pumpers 2018-11-20 15:34:09
51 CRYPTO DIAMOND SIGNALS 2018-11-20 15:25:45
52 VSSN TOKEN 2018-11-20 15:04:02
53 Is Pump 2018-11-20 13:01:00
54 A+ Signals 2018-11-20 11:34:48
55 WhalePump Reborn : Best signals ! (FREE) 2018-11-20 10:47:40
56 Crypto Green 2018-11-20 09:27:07
57 Crypto Lion’s 2018-11-20 09:04:55
58 PumpinTime 2018-11-20 08:03:40
59 Bitcoin Pump VIP 2018-11-20 08:03:37
60 BIG PUMP SIGNAL 2018-11-20 08:03:36
61 Big signal 2018-11-20 08:03:34
62 Crypto Coinsultants 2018-11-20 08:03:23
63 MoneyPumps 2018-11-20 08:03:21
64 Hot Signals Binance Bittrex 2018-11-20 08:03:06
65 Big Pump channel 2018-11-20 08:03:05
66 SIG_NALS Fenix 2018-11-20 07:54:12
67 MEGA PUMP 2018-11-20 07:54:11
68 Pump Club - Yobit 2018-11-20 07:54:09
69 CryptoPump Empire 2018-11-20 07:54:08
70 Alarm! YOBA 2018-11-20 07:54:07
71 Cryptopia Signal 2018-11-20 07:30:50
72 CryptoCoinRankings Private Signals 2018-11-20 07:00:44
73 STRONG SIGNALS 2018-11-20 05:09:10
74 Bigpump24 2018-11-20 04:10:43
75 Crypto Insiders 2018-11-20 02:42:50
76 Crypto Elite Signals 2018-11-20 02:42:50
77 Crypto Signals | Smart Investments 2018-11-20 02:42:50
78 PUMPandCASH .com | Crypto coin pumps ALERTS 2018-11-20 01:10:51
79 Top Pump 2018-11-19 21:49:10
80 Crypto Single 2018-11-19 19:58:11
81 Fast Trading Signal 2018-11-19 17:21:12
82 SPARTA PUMP TEAM 2018-11-19 17:14:19
83 Creative Signals 2018-11-19 14:45:23
84 DAFT PUMP 2018-11-19 14:42:25
85 MasonPump 2018-11-19 13:09:55
86 Exposure Pumps 2018-11-19 11:59:50
87 Moon Crypto Team 2018-11-19 11:16:27
88 Crypto Rocket 2018-11-19 07:45:45
89 Crypto Signals & Pumps 2018-11-19 07:45:41
90 Z (Zsignal) 2018-11-19 05:03:07
91 CAMP(Crypto Analysis & Market Perdictions) 2018-11-19 02:18:49
92 Crypto Yoda Signals 2018-11-18 23:54:35
93 Hit Pump Signals & Strategies 2018-11-18 23:53:26
94 Donald Pump 2018-11-18 23:19:59
95 Signal Plug 2018-11-18 23:02:18
96 Fast Crypto Signals. 2018-11-18 19:37:26
97 True Crypto Guru. 2018-11-18 17:56:53
98 Crypto Experts Signal 2018-11-18 15:43:25
99 Super Pumps & Signal Kings 2018-11-18 14:41:50
100 The pumping army 2018-11-17 23:49:32
101 Trading signals for crypto 2018-11-17 21:54:15
102 European Pumps 2018-11-17 20:43:43
103 *Partners Of Cryptos* 2018-11-17 18:16:56
104 CoinExchange Whale 2018-11-17 07:39:47
105 Binance Pump Whales 2018-11-17 04:36:52
106 Pump’N’Pump - 2018 2018-11-16 19:55:29
107 Caesar’s Scalpers 2018-11-16 18:47:05
108 Daily Crypto Profits 2018-11-16 18:34:36
109 YoBit Pump Team 2018-11-16 16:53:11
110 Extreme_Pumps_CE 2018-11-15 20:34:47
111 Pump Signal 2018-11-15 19:55:27
112 Super Signals 2018-11-15 19:48:01
113 Inner Signals Circle - BITTREX / UPBIT 2018-11-15 18:54:31
114 Crypto Pump 2018-11-15 18:01:29
115 Biggest n Crypto ICO pools 2018-11-15 06:25:18
116 Anonymous Signals ( Bittrex / Upbit ) 2018-11-15 02:59:33
117 Vote Pump 2018-11-14 21:02:42
118 Rocket BINANCE Pump 2018-11-14 21:02:32
119 Whale Crypto Signals 2018-11-14 19:43:14
120 OSNOVA PUMP 2018-11-14 18:08:58
121 Seales crypto 2018-11-14 17:56:01
122 Tornado Signals 2018-11-13 19:10:53
123 Crypto Advisor 2018-11-13 13:07:34
124 Ultra Profit Signal 2018-11-13 12:48:50
125 Crypto experts signal testimonials 2018-11-13 09:24:42
126 Smart Crypto Trading 2018-11-13 02:42:21
127 Crypto Warrior - Free Binance & Bittrex Signals 2018-11-13 02:42:21
128 PumpKings 2018-11-12 04:03:14
129 Coinexchange_Pump 2018-11-11 16:30:29
130 CryptoMoon Pumps 2018-11-11 01:47:54
131 Crypto Family Pumps 2018-11-10 10:32:03
132 MONEYMAKER 2018-11-10 06:02:26
133 COINEXCHANGE Pumping Group 2018-11-09 22:31:51
134 CryptoPump 2018-11-09 19:55:10
135 PumpIt To The Moon 2018-11-09 17:06:21
136 Pump Masters 2018-11-09 17:06:20
137 MassivePump 2018-11-09 17:06:17
138 Crypto Pumpers 2018-11-09 17:06:15
139 BTTM 2018-11-09 17:06:12
140 CRYPTO BILLIONAIRE 2018-11-09 09:33:55
141 The Crypto Analyst 2018-11-08 21:22:14
142 Official Moonwalker Signals 2018-11-06 11:47:47
143 Supreme Pumps 2018-11-05 19:32:46
144 Cryptopia Pump Squad 2018-11-05 00:41:59
145 Crypto Bulls Pump 2018-11-03 19:06:22
146 Crypto Toros 2018-11-03 19:06:22
147 Crypto Pump 2018-11-03 19:06:21
148 Crypto Bulls Pump 2018-11-03 19:05:56
149 BULL PUMP 2018-11-03 01:57:26
150 HITBTC PUMP 2018-11-01 22:37:24
151 Big Pump Signals 2018-11-01 17:41:04
152 TO THE MOON 2018-10-31 05:48:24
153 Crypto Pump 2018-10-29 19:37:35
154 Crypto pump up channel 2018-10-29 07:28:01
155 Big Pump Signal 2018-10-27 00:40:22
156 Moon Crypto Pump Signals & Strategies 2018-10-26 20:12:51
157 Robin HOOD 2018-10-26 17:51:03
158 Mega Pump Group 2018-10-26 15:33:59
159 Crypto Trading (Free Signals) 2018-10-26 14:10:25
160 Crypto trading. 2018-10-26 12:19:15
161 Ministry of Coins 2018-10-26 12:19:11
162 My Bitmex Paradise 2018-10-26 10:18:23
163 Dynamic Signals 2018-10-26 04:46:11
164 John McAfee Yobit Pump 2018-10-24 23:55:10
165 Korean Super Pumps 2018-10-24 20:11:15
166 Crypto Trading 2018-10-24 19:00:21
167 NY Crypto Adviser 2018-10-24 11:36:15
168 Smart Investor 2018-10-24 11:36:15
169 PUMP OFFICIAL CRYPTO 2018-10-24 00:39:12
170 Pump family 2018-10-23 20:14:08
171 Omega Calls 2018-10-23 16:00:32
172 Tele Pumps 2018-10-20 18:08:10
173 PowerPumP 2018-10-20 13:59:44
174 YoBit Pumps 2018-10-19 08:50:28
175 Super pumps 2018-10-18 19:32:39
176 Yobit Pumping Crazy Community 2018-10-18 19:13:45
177 Smarty Signals 2018-10-18 10:58:50
178 Money Machanics 2018-10-17 14:26:50
179 Whales Forex signals 2018-10-15 22:24:43
180 WORLD CRYPTO COMMUNITY 2018-10-14 13:48:58
181 VIPSIGNAL Strategy 2018-10-14 13:48:54
182 HawkEye Bittrex Signals 2018-10-14 13:48:51
183 Best pump group 2018-10-11 16:54:40
184 Yobit-2ch-Pumper 2018-10-11 16:48:49
185 Genuine_Callz 2018-10-09 22:33:14
186 DUTCH CRYPTO PUMPS! 2018-10-09 21:54:08
187 Pump Coin Today 2018-10-09 21:54:08
188 Ultra Pump Channel 2018-10-09 13:44:32
189 Binance Daily Signals! 2018-10-09 11:25:12
190 Goat Pumps 2018-10-08 23:53:33
191 BULLS PUMP 2018-10-08 20:50:06
192 Great Big Pumps 2018-10-07 15:07:48
193 YoBit-Pump-Community 2018-10-07 00:52:20
194 EAGLE PUMPS 2018-10-06 03:43:55
195 Yobit and Cyptopia pump group 2018-10-06 03:43:54
196 USS CALLISTER - PUMPS 2018-10-06 03:43:53
197 YoPumps 2018-10-06 03:43:53
198 The Pump Room 2018-10-06 03:43:52
199 crypto pump 2018-10-06 03:43:10
200 Crypto God’s 2018-10-06 03:43:09
201 F14sH_Pumps 2018-10-06 03:42:58
202 Binance Pump Signal 2018-10-05 22:53:34
203 ToTheMoonPumps 2018-10-05 18:07:41
204 BWP(baby whale pump) 2018-10-02 19:00:40
205 Elite Crypto Group 2018-10-02 00:13:59
206 Franklin Pump. 2018-09-30 18:22:46
207 PUMPS MASTER ESP 2018-09-30 02:41:15
208 Legendary Pumps 2018-09-29 18:27:03
209 Pump Signals 2018-09-28 19:09:08
210 Crypto signal channel 2018-09-28 18:51:49
211 YOBIT Bitcoin Pumps 2018-09-25 15:05:00
212 Profitable Crypto Signals 2018-09-25 13:03:52
213 COINLANCER PUMPER 2018-09-25 11:42:05
214 All Link new bot 2018-09-24 16:05:01
215 WePUMP 2018-09-23 23:28:18
216 Crypto Pump 2018-09-22 01:26:14
217 ELITE PUMP GLOBAL CHANNEL 2018-09-21 18:42:49
218 Crypto blasters 2018-09-21 10:30:27
219 World Pump Association 2018-09-18 20:30:02
220 Kings Of Pump VIP 2018-09-17 22:32:43
221 TOP PUMP VIP 2018-09-17 19:18:23
222 GAINS Private Group (G.P.P.G) 2018-09-15 15:40:35
223 SkyMoon Crypto Signals 2018-09-15 12:08:59
224 PumpNationz 2018-09-13 20:17:00
225 The Alt Pump 2018-09-10 17:03:57
226 Trust PUMP 2018-09-10 11:41:01
227 Bitmex Pro Signals 2018-09-08 21:21:39
228 Altcoins Booster 2018-09-06 11:59:03
229 Altcoins Booster Community 2018-09-06 11:58:31
230 Big Crypto Pump 2018-09-02 04:35:43
231 Superb Pumps 2018-09-01 18:48:04
232 Trading Crypto Coach Backup 2018-09-01 10:32:19
233 20X Pump Actions 2018-08-30 07:28:53
234 McAfee Alt Signals 2018-08-26 05:24:19
235 MoonShot Pump 2018-08-24 11:18:12
236 Central Pumps 2018-08-23 13:00:20
237 Fairwin Crypto News/Pump Signals 2018-08-20 18:23:43
238 Pump Latam 2018-08-18 04:33:10
239 Crypto of the Day 2018-08-16 21:36:35
240 Swiss Signals 2018-08-11 22:08:18
241 Cryptology 2018-08-10 16:57:47
242 Cryptonary VIP 2018-08-09 06:09:30
243 Call Of Pumps 2018-08-08 22:48:13
244 Arabic Big Pump 2018-08-08 14:07:02
245 Pumper_chat 2018-07-29 08:18:28
246 PUMP MASTERS 2018-07-24 17:37:57
247 V-LA SGNALS 2018-07-17 23:14:24
248 Crypto_mania 2018-07-17 13:23:52
249 Crypto Watch 2018-07-17 04:08:21
250 Icenter LTC bot 2018-07-14 23:28:08
251 Pump BTC 2018-07-14 14:03:25
252 Bulls Eye Signals 2018-07-12 14:17:26
253 CryptoWorld 2018-07-09 12:44:42
254 Majestic Pumps 2018-07-05 19:00:20
255 Mega Pump Group 2018-06-28 19:17:29
256 Phoenix Cryptopia Team 2018-06-27 20:35:34
257 Neon Pumps 2018-06-27 12:08:44
258 Big Pump Signals 2018-06-25 11:42:30
259 Yobit international 2018-06-23 18:22:32
260 Mighty Whales 2018-06-21 16:00:11
261 Spartan pump group 2018-06-21 15:07:21
262 ALL PUMPS 2018-06-21 06:51:34
263 PUMP 2 Group 2018-06-18 13:00:23
264 Crypto Pump 2018-06-17 21:24:37
265 Monster Signalz 2018-06-15 19:59:59
266 Binance and Cryptopia Pumps 2018-06-14 17:59:59
267 MidEarthCrypto Community 2018-06-12 14:08:57
268 Phoenix Pump Team 2018-06-08 20:44:56
269 Whales & BTC 2018-06-05 20:17:48
270 Pump signal 2018-06-05 19:27:08
271 PIRATES PUMPS 2018-06-03 18:04:46
272 Coin To The Moon 2018-05-31 09:58:35
273 PumpKings 2018-05-28 18:02:58
274 Crypto Future Signs 2018-05-28 10:25:49
275 ITpump 2018-05-26 15:41:32
276 PumpingHard 2018-05-25 22:26:34
277 Dragon Signals 2018-05-23 17:57:56
278 Dragon Pumps & Signals 2018-05-23 17:57:50
279 Explosive Pumps 2018-05-22 18:37:20
280 PumpYobit24 2018-05-21 14:04:09
281 Unicorn Magic Signals 2018-05-20 23:04:19
282 Cryptopia Family Pumps 2018-05-20 18:00:14
283 Invest in Brokers Info 2018-05-17 09:54:54
284 Wealthy Whale Pumps 2018-05-15 22:12:42
285 Moon Pump 2018-05-14 08:45:57
286 100x Cryptocurrency Signals 2018-05-14 08:44:13
287 Easy Money 2018-05-13 10:48:44
288 YObit Pump Network 2018-05-11 10:16:45
289 Y0 PumBit 2018-05-11 09:26:03
290 Crypto Signals Official 2018-05-11 09:25:32
291 Smashing PumpKings 2018-05-11 00:08:23
292 USA PUMP 2018-05-07 14:27:09
293 PumpWhales 2018-05-06 18:58:53
294 Binance And Cryptopia Pumps espaol English 2018-05-06 18:58:52
295 2 PUMPS EVERY DAY 2018-05-06 18:58:51
296 WEB Pump YoBit 2018-05-06 12:47:21
297 Crypto Pump Signals 2018-05-05 18:39:23
298 MPERIAL PUMP 2018-05-04 18:30:18
299 Golden Ticket Pumps 2018-05-03 17:05:42
300 Insane pumps 2018-05-01 12:39:38
301 Cryptoverse 2018-05-01 00:17:54
302 Pump Coin Signal 2018-04-30 20:53:45
303 0x 2018-04-27 11:17:34
304 CryptoHunter 2018-04-26 07:00:34
305 YoPump 2018-04-26 00:11:51
306 Crypto Pumps 2018-04-26 00:11:50
307 Altcoin Pumps 2018-04-26 00:11:49
308 YoBit/Bittrex Pumps 2018-04-26 00:11:48
309 Good Pump Channel. 2018-04-23 17:54:42
310 Private Pump Signals 2018-04-21 05:06:46
311 Pablo Pumps 2018-04-21 01:22:05
312 Eternal Crypto Pumps 2018-04-19 14:27:32
313 GosPump 2018-04-18 21:20:11
314 Pump Channel 2018-04-18 04:22:03
315 PUMPED! 2018-04-18 00:13:30
316 McAfee group 2018-04-17 13:06:49
317 X5-PUMPING 2018-04-17 08:52:14
318 Great pump 2018 2018-04-13 17:04:06
319 YObit pump 2018-04-13 10:56:56
320 GALAXY PUMP 2018-04-13 10:50:05
321 Robin Hood’s Pump 2018-04-10 21:07:08
322 Pump King 2018-04-10 13:00:25
323 - Announcements 2018-04-09 19:54:31
324 Cryptopia Pump 2018-04-08 20:00:03
325 Largest Pump Community 2018-04-08 11:12:42
326 The Pumpers 2018-04-07 15:08:57
327 Sunrise Trend 2018-04-07 14:59:49
328 VIVUCoin Channel: Crypto Insight 2018-02-15 17:57:13
329 Crypto Mega Pumps Yobit 2017-10-17 13:31:32
330 BEST PUMP Account not found
331 UMP. Account not found
332 Altcoin Mega Pumps Account not found
333 Big Pump Signal Account not found
334 Big Pump Signal Account not found
335 Binance and Cryptopa Pumps Account not found
336 Bitcoin Pump Signal Account not found
337 Bullish signals Account not found
338 CRYPTO BLACK Account not found
339 Crypto La MAFIA Account not found
340 Crypto Pros Account not found
341 Crypto Pump Cage Account not found
342 Crypto Universe Trading Account not found
343 Crypto VIP Signal Account not found
344 Crypto Whales Account not found
345 Crypto experts signal Account not found
346 CryptoTradingWorld. Account not found
347 Exchange Signals Account not found
348 Mcafee pump Account not found
349 Pumpimm Account not found
350 Rocket signals Account not found
351 Signals Lion Account not found
352 Signalsgram Account not found
353 Tech News Account not found
354 Techlab Trading Account not found
355 Whale Club Account not found
356 Whale Pumps Club Account not found
357 s Account not found
358 Cryptopia Pump Community Account not found
TABLE I: Pump-and-Dump channels on Telegram. Messages retrieved through Telegram API on November 21, 2018.
Comments 0
Request Comment
You are adding the first comment!
How to quickly get a good reply:
  • Give credit where it’s due by listing out the positive aspects of a paper before getting into which changes should be made.
  • Be specific in your critique, and provide supporting evidence with appropriate references to substantiate general statements.
  • Your comment should inspire ideas to flow and help the author improves the paper.

The better we are at sharing our knowledge with each other, the faster we move forward.
The feedback must be of minimum 40 characters and the title a minimum of 5 characters
Add comment
Loading ...
This is a comment super asjknd jkasnjk adsnkj
The feedback must be of minumum 40 characters
The feedback must be of minumum 40 characters

You are asking your first question!
How to quickly get a good answer:
  • Keep your question short and to the point
  • Check for grammar or spelling errors.
  • Phrase it like a question
Test description