See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/350020789

An Empirical Analysis of Pool Hopping Behavior in the Bitcoin Blockchain

Conference Paper · May 2021

DOI: 10.1109/ICBC51069.2021.9461118

CITATIONS READS
21 1,176

4 authors, including:

Natkamon Tovanich Nicolas Soulié

École Polytechnique Institut Mines-Télécom
17 PUBLICATIONS 166 CITATIONS 23 PUBLICATIONS 233 CITATIONS

SEE PROFILE SEE PROFILE

Nicolas Heulot
IRT System X
12 PUBLICATIONS 347 CITATIONS

SEE PROFILE

All content following this page was uploaded by Natkamon Tovanich on 12 March 2021.

The user has requested enhancement of the downloaded file.

 An Empirical Analysis of Pool Hopping Behavior
in the Bitcoin Blockchain
Natkamon Tovanich Nicolas Soulié Nicolas Heulot Petra Isenberg
IRT SystemX and Université Paris-Saclay, IRT SystemX Université Paris-Saclay,
Université Paris-Saclay, LISN Univ Evry, IMT-BS, LITEM CNRS, Inria, LISN
Palaiseau, France Evry-Courcouronnes, France Palaiseau, France Gif-sur-Yvette, France
natkamon.tovanich@irt-systemx.fr nicolas.soulie@imt-bs.eu nicolas.heulot@irt-systemx.fr petra.isenberg@inria.fr

Abstract—We provide an empirical analysis of pool hopping join Mining Pool

behavior among 15 mining pools throughout Bitcoin’s history.
reward
Mining pools have emerged as major players to ensure that the share
Bitcoin system stays secure, valid, and stable. Individual miners join
join mining pools to benefit from a more predictable income. integration
reward REWARDS
Many questions remain open regarding how mining pools have share
evolved throughout Bitcoin’s history and when and why miners blocks mined by
join or leave mining pools. We propose a heuristic algorithm to join
the pool Blockchain
extract the payout flow from mining pools and detect the pools’ reward
share
migration of miners. Our results showed that payout schemes Miners
and pool fees influence miners’ decisions to join, change, or
exit from a mining pool, thus affecting the dynamics of mining Fig. 1. The overview of Bitcoin mining activity. The diagram shows the
pool market shares. Our analysis provides evidence that mining interactions between individual miners, mining pools, and the Bitcoin protocol.
activity becomes an industry as miners’ decisions follow classical
economic rationale.
Index Terms—Bitcoin, Bitcoin mining, mining pools, pool
hopping, visual analytics Nowadays, mining pools dominate the Bitcoin mining
activity as known mining pools contribute ≈99% of the total
I. I NTRODUCTION hash rate [6]. Bitcoin mining is an economic activity with
Bitcoin mining as a term refers to Bitcoin’s proof-of- three main agents: mining pools, individual miners, and the
work protocol. Nakamoto proposed the protocol to solve Bitcoin network as shown in Fig. 1. First, mining pools are in
the double-spending problem in digital currencies and to competition with one another to attract more miners in order to
prevent individuals from tampering with the blockchain [1]. gain a higher market share and have a better chance to obtain a
Bitcoin miners compete to solve a computation-intensive task mining reward. Mining pools distribute the rewards they collect
to propose a new block in the network. They receive financial to their individual miners and keep some amounts of rewards
rewards for each block they successfully mine. These rewards for their profits (pool fee). Second, individual miners invest
consist of the block reward fixed by the protocol and transaction in computational resources to receive a mining reward. They
fees from transactions in the mined block. Practically, individual decide to enter a pool (new miners), move to another pool
miners receive a reward only occasionally, relative to their (pool hoppers), exit from mining activities (dropout miners),
computational power. As more miners join the network [2] or even participate in multiple pools (cross-pooling miners).
and with faster mining hardware available [3], the total The migration of miners directly affects mining pools’ market
computational power of miners (called the hash rate) has been shares. Third, the Bitcoin network regulates the mining activity
growing rapidly [4]. The mining difficulty is set by the protocol by automatically adjusting the mining difficulty every 2,016
relative to the hash rate [5] so the probability of mining a blocks (≈2 weeks) according to the total hash rate in order to
new block with the same hardware becomes lower as hash maintain the block discovery rate at 10 minutes. The network
rate increases. Therefore, the expected reward of an individual also pays the reward to the successful miners and sets the
miner diminishes with more competition. In order to overcome block reward. The block reward started with 50 BTC per block
this problem, mining pools have emerged in which miners from the genesis block (block 0) and is reduced by half every
combine computational resources to gain a more stable and 210,000 blocks (≈4 years).
predictable income. Documenting the evolution of mining pools is of major
interest for evaluating the future of Bitcoin, and more generally,
This research work has been carried out under the leadership of the Institute of blockchain-based technology. A better understanding of the
for Technological Research SystemX, and therefore granted with public funds impacts on pools’ growth or decline of both external (e. g.,
within the scope of the French Program Investissements d’Avenir.
This is the author’s version of the article that will appear in 2021 IEEE market price, regulation) and internal (e. g., payout scheme,
International Conference on Blockchain and Cryptocurrency (ICBC). pool fee, share of transaction fee) factors is critical to assess
the viability of this industry, and thus the sustainability of III. P OOL H OPPING D ETECTION
proof-of-work cryptocurrencies such as Bitcoin. Our research approach involved extracting members of
Mining pools are major players in Bitcoin mining that mining pools and measuring pool hopping behavior. Our
influence how stable, secure, and trustworthy the currency is. process includes three steps: (A) we obtain the mining reward
Joining or leaving a pool is a decision miners make not only to for each block and attribute it to a known mining pool; (B–C)
increase their short-term mining rewards but also to counteract for each coinbase transaction, we extract the reward payout
the possible domination of pools in the network. In return, flow and detected pool members (miners) of the mining pool;
migration flows affect how mining pools set their policies (e. g., and (D) we identify miners who participated in 15 mining pools
payout schemes and pool fees) to compete in the market. The over time and migrated between pools. We provide the data
combination of these activities and behaviors related to mining about mining pool attribution and miner’s migration at [11].
are not yet well understood and few methods exist that allow to
study mining pools. We contribute to a new approach to detect A. Mining pool attribution, market shares, and characteristics
miners’ migration (called pool hopping) among mining pools. Mining pool attribution. First, we identified the mining
First, we developed an algorithm to extract reward payout pool that mined each block in the Bitcoin blockchain. When a
flows based on the concept of transaction purity. Then, we mining pool mines a block, it receives the mining reward from
obtained a list of miners for each payout flow and detected the coinbase transaction of the block. A coinbase transaction
the miners that migrated between pools. Next, we derived combines the block reward from the Bitcoin network and
quantitative measurements to evaluate miners’ migration flows transaction fees from every transaction in the mined block. It
(e. g., pool hopping and cross-pooling). Finally, we visualize also includes a coinbase string inserted by the miner. For
miners’ migration among 15 mining pools throughout Bitcoin’s each coinbase transaction, we attributed the mining pool
history. We highlight the existence of regular patterns of miners’based on the address matching or coinbase string pattern. We
entry, hopping and cross-pooling behaviors in comparison to initially used the dataset from Romiti et al. [9] that compiled
different payout schemes and pool fees. known mining pool tagging until block 556,400 (2018-12-31).
After this block, we continued their procedure and tagged
II. R ELATED W ORK pools until block 650,731 (2020-09-30) with the datasets from
Blockchain.info [12] and BTC.com [13]. We labeled the blocks
Some previous theoretical work related to pool hopping that did not match any known mining pool as “unknown.” We
analysis has applied game theory to explain the motivation stored this data as an “attribution table.” Table I shows the
behind pool selection and miners’ migration [7]. In contrast, total number of blocks we found for each mining pool.
our work relates more closely to other empirical work that Mining pool market shares: We define the market share
has attempted to detect miners in mining pools and to analyze of a mining pool as the percentage of the blocks it mined
miners’ migration patterns among pools in the Bitcoin network. compared to the total blocks mined in a month. We used our
attribution table to calculate each pool’s monthly market share.
Belotti et al. [8] investigated pool hopping between KanoPool
Mining pool characteristics: We obtained information
and SlushPool between April 6–20, 2016. The authors found
about pool characteristics, in particular payout schemes and
that a few miners tried to exploit the time difference of reward
pool fees, from the Bitcoin Wiki page [14] on the topic. We
payout between two pools with diverse strategies to gain a
downloaded the page’s edit history and manually cleaned the
small profit gain. Romoti et al. [9] presented reward payout
data for each month by comparing it with the information from
flow patterns of three pools: BTC.com, AntPool, and ViaBTC,
the Bitcoin Forum [15]. As a result, we constructed panel data
between block 510,000 and 514,032 (≈4 weeks) and detected
that includes all changes in pool characteristics over time.
overlapped miners in those pairs. They detected high cross-
Finally, we merged the mining pool market share data with
pooling between BTC.com and AntPool probably because both
the Wiki data. For many mining pools in the attribution table,
pools are owned by the same company, Bitmain. Xia et al. [10]
the Bitcoin Wiki did not contain additional data. Therefore,
developed a visualization tool showing the internal address
we selected the top 15 mining pools that we found in both
networks of mining pools and the estimated number of pool
the Wiki data and the attribution table for our study on pool
hoppers. This past work is closely related to ours but we deviate
hopping behavior (highlight in green color in Table I).
in several areas. We propose a miners’ migration flow model
and measurements to detect different types of pool hoppers B. Transaction flow and transaction purity
over long time intervals. Our extraction method is also less
computationally expensive approach although it shares a similar We introduce transaction flow graphs and the transaction
underlying concept to detect miners. purity definitions before applying them to our payout flow
model and heuristic algorithm.
Overall, compared to the majority of past work, we look
empirically at a higher number of mining pools across Bitcoin’s Definition 1. A transaction flow is a directed graph of Bitcoin
mining history. Moreover, we compare our result with external transactions from a seeding transaction. Each node represents
information (e. g., market shares, payout schemes, and pool a transaction tx in the transaction flow. A transaction has a
fees) to help explain pool hopping behavior. timestamp attribute time. Each directed edge corresponds to a
 TABLE I Algorithm 1: Reward payout flow extraction
L IST OF MINING POOLS THAT RECEIVED MORE THAN 1,000 BLOCK
REWARDS . ROWS HIGHLIGHTED IN GREEN COLOR ARE THE 15 MINING
Input : txpayout is a payout transaction as a seeding
POOLS WE STUDIED IN THIS WORK . node of the payout flow.
Output : edges is the edge list of the the payout flow.
Mining Pool Blocks Total Rewards First Block Last Block
F2Pool 58,174 1,101,804 2013-05-05 still active
queue ← P riorityQueue([txpayout ]);
AntPool 51,211 883,692 2013-12-07 still active edges ← List();
SlushPool 33,640 755,910 2012-01-26 still active while queue is not empty do
BTC Guild 32,936 1,010,779 2011-11-10 2015-06-30
BTC.com 31,814 403,201 2016-09-05 still active tx ← queue.pop();
DeepBit 31,107 1,508,254 2011-02-25 2013-11-28 if purity(tx) = 1 and valid(tx) = True then
GHash.IO 23,083 579,128 2013-08-04 2016-10-24
BitFury 20,901 420,027 2013-11-14 2020-03-25 for edge in tx.out do
ViaBTC 18,640 239,882 2016-06-05 still active edges.append(edge);
BTCC Pool 18,036 363,493 2014-10-21 2018-09-25
BTC.TOP 15,896 211,808 2016-12-11 still active
queue.append(edge.spend);
Poolin 15,142 175,971 2018-07-02 still active end
BW.COM 12,733 250,044 2015-01-29 2018-11-09 end
Eligius 11,430 338,236 2011-06-14 2017-11-22
50BTC 7,859 198,651 2012-12-18 2014-06-02 end
KnCMiner 7,477 185,427 2014-02-25 2016-09-12
BitMinter 6,464 205,382 2011-11-07 2019-08-12
EclipseMC 6,024 212,395 2012-02-10 2016-03-02
Huobi 5,904 65,243 2015-06-01 still active
Bixin 5,753 80,640 2016-06-12 2019-09-01 C. Mining pool payout flows
BitClub Network 5,672 88,892 2015-04-02 2019-11-06
OzCoin 4,845 187,123 2011-12-30 2014-09-24 After a mining pool receives the mining reward from a
ASICMiner 3,146 79,279 2013-05-20 2014-04-09 coinbase transaction, the pool has to distribute the reward to
1THash&58COIN 3,067 30,635 2019-08-25 still active
okpool.top 2,750 27,820 2018-12-27 still active pool members. Even though mining pools distribute the reward
Bitcoin.com 2,465 32,944 2016-09-21 still active to individual miners in different patterns [9], [18], we introduce
KanoPool 2,432 46,359 2014-10-14 2020-07-26
GBMiners 2,093 28,980 2016-08-30 2018-04-15
the payout flow model as a transaction graph consisting of four
DPOOL 1,918 24,398 2018-03-31 2019-05-11 transaction types: coinbase ( txcoinbase ), payout ( txpayout ),
1Hash 1,895 29,268 2016-03-04 2017-12-07 intermediate ( txinter ), and miner ( txminer ). Examples of
Telco 214 1,830 40,109 2014-12-19 2017-08-23
CloudHashing 1,824 45,745 2013-10-04 2015-02-14 reward payout flows are shown in Fig. 2.
21 Inc. 1,508 37,996 2015-04-09 2016-03-01
WAYI.CN 1,306 16,364 2018-02-14 still active 1) A mining pool receives mining rewards from coinbase
Polmine 1,290 32,365 2013-04-12 2015-06-07 transactions txcoinbase and collects them in a payout
transaction txpayout before distributing it to miners.
2) A mining pool distributes the reward from txpayout
value transfer from a transaction to another. Therefore, whether to intermediate transactions txinter before splitting
it is the input or the output of a transaction depends on the rewards to pool member (miner) addresses.
direction of the edge. An edge contains the information about 3) Pool members receive the reward from txinter and
the amount of transferred value, and the public-key address spend it in a transaction we call miner transaction
of the owner. Each edge contains references to the receiving txminer . We assumed that pool members receive the
transaction node receive and spending transaction node spend. reward from this flow and then combine it with other
Bitcoin values outside the flow to spend in txminer .
We adopted a transaction purity measure to determine how Therefore, the purity of txminer is < 1.
much Bitcoin value in the transaction is received from the
Based on this model, the reward payout flow is the Bitcoin
seeding transaction. This measure is commonly used for taint
transaction flow from a payout transaction txpayout to
analysis in Bitcoin (e. g., [16], [17]).
pool members txminer . We considered txpayout as the
Definition 2. Let tx.in and tx.out be sets of receiving (inputs) seeding transaction because it collects every mining reward
and spending (outputs) edges of a transaction tx respectively. and distributes it to pool members.
The transaction purity is recursively defined as being the Extracting reward payout flows. We devised Algorithm 1
average purity of the input transactions weighted by their to automatically extract payout flows from the coinbase
respective values. The purity of a transaction tx can be transactions in the Bitcoin blockchain. We used the BlockSci
expressed as follows: API [19] to access the transaction data. We initiated the list of
txpayout from all outputs of txcoinbase as inputs to the
P algorithm. For each txpayout , we traversed the transaction
purity(e.receive) · e.value txpayout which has purity = 1 until the
purity(tx) = e∈tx.in P (1) graph from
e∈tx.in e.value transaction has puritytx < 1 (i. e. txminer ). The algorithm
returns a directed edge list that represents the payout flow.
The purity of a transaction without inputs is 1 because it is We added two additional termination criteria valid(tx) that
the root transaction in the transaction flow. stop following the current transaction tx flow: (1) when the
(a) DeepBit (2011-08-03) Transaction Type Coinbase Payout Intermediate Miner (c) F2Pool (2014-10-31)
1 txs 1 txs 2435 txs
(b) SlushPool (2013-01-01)
₿49 ... ₿1,302 ₿855
1 txs 42 txs 24 txs
31 txs
₿49 ₿1,053 1 txs
₿84
1 txs ₿47 1 txs 37 txs
1 txs 2 txs 1 txs
₿49 ₿2,314 ₿1,011 ₿100
1 txs 1 txs 1 txs ₿51 ₿51 1 txs 1 txs
1 txs
1 txs ₿1,260 ...
₿50 ₿50 ₿50
₿4 1 txs 18 txs
1 txs
₿59 ₿11

(d) BTCC Pool (2016-02-18) (e) AntPool (2016-04-23) (f) BTC.com (2018-04-06)
11 txs 1 txs
₿735 ... ₿1,099 ... 25 txs
16 txs 4 txs ₿139 ...
36 txs 1 txs 2 txs
₿745 ₿1,099
7 txs 4 txs ₿456 ₿851 ₿278
15 txs 6 txs 1 txs
₿872 ₿7 ₿1,100 26 txs 962 txs
1 txs 43 txs 1 txs ...
65 txs 94 txs ₿1,329 ₿285 ₿121
1 txs 1 txs
₿1,000 ₿96 ₿1,0881 txs ₿1,100 96 txs
40 txs 1 txs
80 txs 92 txs ₿873 ₿478 ₿10
₿1,008 ₿1,008 ₿165 ₿1,100 ₿9 967 txs
1 txs 5 txs 1 txs ₿214
₿8 ₿12

Fig. 2. Representative examples of reward payout flow patterns from the mining pools in our study. The flows were extracted using our algorithm and are
represented as node-link diagrams. Here, we sampled the payout flow in the month where the mining pool had the highest market share. Each node represents
a transaction type with branches of similar patterns grouped together. The color of the node indicates the transaction type. The total value of transactions in
each node is encoded by circle size in proportion to the txpayout value. The number of transactions and their combined value are the top and the right labels
for each node respectively. We omitted labels for combined values below 1 BTC.

time difference between txpayout and tx is more than one multiple miners at each step, e. g. AntPool (e), BTC.com (f),
day and (2) when the tx.value is < 0.001 BTC—as most BTCC Pool, and Poolin; and (2) the indirect payout to miners
mining pools have a minimum payout value [8], [10]. on an txinter , e. g. F2Pool (c). We also noticed that F2Pool
(c) and BTC.com (f) usually send half of their payout back to
Identifying individual miners. For each edge list obtained
their addresses as a reserve to pay miners in the next payout.
from Algorithm 1, we constructed a payout flow graph using
the NetworkX library [20]. Representative payout flow patterns D. Miners’ migration between mining pools
that we obtained from the algorithm are shown in Fig. 2. Next, To analyze miner migration between pools, we compared
we extracted the txminer and derived the list of miners the list of miners who received rewards from each mining pool
from each payout flow graph. in a set time interval and calculated the intersection of miners
between pools. We set the time interval to months to be able
Definition 3. Miner transaction txminer is a transaction
to analyze detailed patterns for the entire mining pool history.
in the payout flow graph that does not have any output in
the payout flow graph |txminer .out| = 0. We tagged all input Definition 4. Let t be a time interval where t ∈ T = {t0 , ..., t−
t
edge(s) of txminer as owner edges. The list of miners who 1, t, t + 1, ..., tn }. The set of miners in the mining pool Mpool
received the reward from txpayout is defined as Mtxpayout . is the summation of the miner list Mtxpayout for all payout
transactions of a mining pool pool at time t.
Some txminer transactions may be connected to the pool
wallet to keep the represented value as profits for the pool or The miner’s migration flow is modelled as a diagram in
as deposits for the next payout, as illustrated in Fig. 2 (c) and Fig. 3. For each time interval t, the list of miners that migrate
t
(f). We detected txminer input edges that have the same from/to a mining pool pool, Mpool , is divided into 7 miner
owner addresses as the mining pool and assigned them as groups as follows:
txpayout to extract further reward payout flows. • New (Dropout) miners are miners that enter (exit) the min-

Payout flow patterns: We visualized payout flow patterns ing activity at time t, annotated as Mnew|pool (Mdrop|pool ).
t
for all mining pools in our study and show representative • Same before (Same after) miners are in Mpool but are
t−1 t+1
flows patterns in Fig. 2. Early mining pools, operated from also in Mpool (Mpool ).
t
2011 until mid-2015, distributed the reward directly after every • Hopping in (Hopping out) miners are in Mpool but move
t−1 t+1
block it mined. We found two payout patterns in this period: from (to) other pools Mothers|pool (Mothers|pool ).
(1) the long chain of payout flows distributing the reward to a t
• Cross-pooling miners are in Mpool but also receive a
single miner at each step, e. g., DeepBit (a), BTC Guild, and t
reward from other pools at the same t (Mothers|pool ).
GHash.IO; and (2) the direct payout to miners after receiving We estimated the quantity of miners’ migration as the
the mining reward, e. g. SlushPool (b) and Eligius. percentage of the total value for each miner group. We report
After mid-2015, most mining pools tended to collect the the percentage of value rather than the number of addresses
mining rewards in their wallet and to distribute the reward because it gives more weight to miners that have a high
to miners regularly (i. e. daily). We also observed two payout contribution to the pool and therefore the measure is more
patterns (1) the chain of payout flow distributing the reward to robust regarding small or occasional miners.
 includes input transactions from outside the flow. The algorithm
will make a false classification when a miner simply forwards
the reward using a transaction without further inputs. In this
New Dropout case, the algorithm will calculate that the transaction purity is
Same before Same after 1, assign it as txinter , and follow all outputs from txinter .out.
To detect migration patterns, we also assume that miner
addresses are reused. We are aware, however, that miners can
Hopping in Hopping out
always generate new addresses. As a result, our percentage of
new and dropout miners is an upper bound. In our data, each
Cross pooling
miner address received a reward from a pool on average 18.1
[15.4, 21.4] times. We summarize the miners’ migration flows
into net gain or loss metrics to reduce the impact of miners
who change their addresses within the same month.
Although the basic address clustering method [21] is an
effective method to group the addresses that are likely to belong
to the same entity [22], we found that it led to false-positive
clusters. For example, the method may group different miners
Fig. 3. The miners’ migration flow model of pool at time interval t. t − 1
(t + 1) is the time interval before (resp. after) t. Mnew (Mdrop ) is the list in the same cluster because they used the same exchanges or
of miners not in any pool at t − 1 (resp. is not found in any mining pool at mixing services. We expect that miners would participate in
t + 1). The union of the list of miners from other pools is Mothers . 1–2 pools at a time. We report the average number of mining
pools that miners participated as the average weighted by their
total reward. During the first halving (second halving) period,
Definition 5. The percentage of the total value of miners (X)
t we found that miners receive the reward from 3.92 [3.12,
is the total value of Mpool associated with Mx , where x is a
4.74] (3.06 [2.47, 3.68]) different pools per month compared
set of miners from miner groups. We defined this measure as:
to 1.46 [1.39, 1.56] (1.30 [1.24, 1.37]) pools per address.
The percentage of cross-pooling per month using address
P
m∈M t ∩Mx m.value
t
X(Mpool , Mx ) = P pool (2) clustering is on the average of 25.9% [24.2%, 27.6%] (31.6%
m∈Mpoolt m.value
[29.7%, 33.7%]) higher than using solely miner addresses, with
For example, the percentage of hopping in (hopping pairwise comparison for the same pool and month. Therefore,
out) miners is annotated as X(Mpool t t−1
, Mothers|pool ) (resp. we decided to use miner addresses to avoid adding errors from
X(M t
,M t+1
)). the address clustering to the results.
pool others|pool
For each mining pool, we obtained the monthly percentage B. Miners addresses association with known entities
of miners’ migration for each miner group. To understand
Since there is no ground truth to evaluate the identity
the flow of miners in a mining pool, we summarized miners’
of individual miners, we indirectly validated whether our
migration flows into a net gain or loss metric for the pool from
approach can identify individual miners correctly. We posited
different flow types with (1) New and dropout flow: the percent
the assumption that miners should receive a mining reward
difference between new and dropout miners; (2) Hopping in
(input address) from the mining pool and keep it in their wallet
and out flow: the percent difference between hopping in and
(miner address) before spending it (output address) on services
hopping out miners; and (3) Cross-pooling: the percentage of
(e. g. exchange, mixer, or marketplace). We used a known entity
cross-pooling miners.
dataset from WalletExplorer.com with entity type classification
Additionally, we calculated the percentage of cross miners’
from Zola et al. [23]. We report the percentage of addresses
rewards from the pool as the total reward that cross miners
and Bitcoin values for each entity type in Table II. We studied
received from the pool divided by the total reward that cross
the payout flows that spent between 2013-01-01 and 2016-12-
miners received from all mining pools. A higher percentage
32 because the website stopped updating more known entities
implies that miners dedicated more computational resources to
from 2016 [24]. In summary, we found:
this particular pool. It also indicates the attractiveness of the
pool compared to other pools at the same time interval. 1) Miners detected from our algorithm mostly cannot be
associated with any known address (“unknown” type in
IV. D ISCUSSION ON THE P OOL H OPPING D ETECTION Table II) (84.8%) as well as input and output address
In this section, we discuss the validity and quality of our (96.1% and 91.5% resp.). However, when we measured
approach and compare it to related work. All confidence the total value for each entity type, we found that
intervals are 95% bootstrap CIs. 84.8% of miner rewards are from unknown addresses,
compared to 44.0% for input and 68.6% for output
A. Assumptions and limitations of the approach addresses. Therefore, we show that our algorithm can
Our approach rests on the assumption that individual miners detect individual miners because they are largely not
who receive a reward share will spend it in a transaction that associated with any known Bitcoin entities.
 TABLE II A. Pools’ competition, fees and pool hopping
T HE PERCENTAGE OF ADDRESSES AND TOTAL B ITCOIN VALUES
ASSOCIATED WITH KNOWN ENTITIES FROM 2013-01-01 TO 2016-12-31 In the competition to attract miners, payout schemes and pool
fees are major pool characteristics that directly impact miners’
Addresses Total Value
Type
Input Miner Output Input Miner Output
income. We illustrate that fee and payout schemes exhibit the
Unknown 96.1 84.8 91.5 44.0 84.8 68.6 usual economic evolution observed in the competition context in
Mining pool 1.38 0.555 7.43e-2 45.9 6.02 0.371 Fig. 4 (a). Our previous work [26] showed that the Proportional
Exchange 1.42 8.36 4.80 0.35 6.52 18.0
Wallet 0.428 4.38 2.67 7.62 2.48 12.6 payout scheme was used at the beginning of Bitcoin. Over time,
Marketplace 0.665 1.19 0.567 2.14 0.162 0.398 mining pools switched progressively to PPS and PPLNS payout
Gambling 3.46e-2 0.609 0.347 1.86e-4 3.25e-2 5.07e-2
Mixer 1.15e-3 5.51e-2 3.69e-2 1.16e-5 6.57e-3 3.57e-2 schemes. As PPS and PPLNS are more robust to pool hopping
Lending 5.03e-3 4.54e-2 2.86e-2 3.68e-5 1.50e-3 1.51e-3 than the proportional reward [27], these payout schemes are
more attractive for pool managers. Our empirical result is in
line with prior work as Proportional payout scheme disappeared
2) Miners tend to receive a reward from known mining pool in 2013. After that, PPS and PPLNS became the dominant
addresses (45.9% of the total value) followed by unknown payout schemes among the large pools.
addresses (44%). This result aligns with our assumption A second explanation for the growing use of PPS and PPLNS
that miners should receive the money from txinter of relies on their different but complementary risk/return ratios.
the mining pool. For unknown addresses, mining pools PPS pools pay miners in proportion to their contribution to the
may use external addresses that are undetected in the pool and thus provides risk-free, low income. All the risk is
known entity dataset to pay miners. supported by the pool, which needs then to create a reserve
3) Miners spent 68.6% of their total value using unknown of money to be able to pay the miners during ‘bad luck’
output addresses. We also detected that some miners spent periods. In comparison, PPLNS pools pay only those miners
their reward on exchanges (18% of the total value) and who contributed to the last N shares in a given time window.
wallet services (12.6%). This result provides evidence that Miners who contribute but leave the pool before a block has
regular miners convert mining rewards to fiat currencies been mined might not get any reward. Therefore, PPLNS lefts
or deposit them to their Bitcoin wallets. all the risk to the miners, and the expected reward variance
is higher compared to PPS [27]1 . These two payout schemes
Our approach differs from Xia et al.’s work [10] as we do can be viewed then as two different financial assets. For this
not filter out known entities after we extract the payout flow. We purpose, it is noticeable that the fees applied to these two
have three reasons for this choice: (1) Xia et al. focus on only financial assets follow the classical two-parameter financial
a 1-month time frame. The WalletExplorer dataset, however, asset pricing model [28]. In financial markets, risky assets
includes 30,167,518 labeled addresses. It is computationally must have a higher expected return to be attractive. In the case
expensive to linearly scan for addresses in every transaction; of Bitcoin mining, Fig. 4 (a) is consistent with this scheme
(2) WalletExplorer does not update new entity labels after as the more risky asset (PPLNS) is likely to have a lower fee
2016 [24]. Hence, it cannot be applied to recent reward payout (≈0%) compared to the risk-free one (PPS, ≈2-3%).
flows; and (3) Our measurements based on the percentage of Pool fees are used as a competitive advantage for mining
value are tolerant to possible misclassification of miners. pools. Within each payout scheme type, new pools tend to apply
a lower fee than the incumbents. For instance, DeepBit applied
a relatively high fee for PPS (10%) as the first dominant mining
V. E CONOMIC A NALYSIS OF P OOL H OPPING B EHAVIOR
pool between 2011–2012. In 2012, mining pools, such as BTC
Guild or OzCoin, applied lower PPS fees (5%) to attract new
We calculated miners’ migration statistics for 15 pools that miners (Fig. 4 (b)) and hopping-in miners (Fig. 4 (c)), probably
adopted three main payout schemes: Proportional, Pay-Per- from DeepBit which had more hopping-out miners in the same
Share (PPS), and Pay-Per-Last-N-Shares (PPLNS). We explain period. We see the same pattern in 2013 when F2Pool (4%,
miners’ behavior in the Bitcoin network based on rational named Discus Fish at the time) or 50BTC appeared (3%), then
behavior in economic theory and provide visual evidence in 2014 with AntPool (2.5%) or BTCC (2%), and in 2016
that some characteristics of mining pools (e. g. market shares, with BTC.com (1.5%). This competition led to a decrease in
payout schemes, pool fees) affect miners’ mobility. the average PPS fees implemented by pools which stabilized
Bitcoin mining has become an industry where miners gather around 2% from 2016. The same dynamics occurred for PPLNS
into pools to maximize their investment in mining devices [25].
1 Following Rosenfeld’s article [27], the expected rewards of PPS and PPLNS
Choosing a pool becomes a strategic economic decision for
are both equal to (1 − f )pB, where B is the block reward, f is pool fee, and
miners as a pool’s characteristics greatly affect a miner’s p is the probability of a share to be a valid one. However, as we showed in this
income. First, we focus on the competition between pools paper, the PPLNS fee is lower than the PPS fee. PPLNS tend then to generate,
based on payout schemes and transaction fees. Then, we in the long run, higher income than PPS. Moreover, PPLNS reward variance
pB 2
investigate market entry and the expected revenue of new can be approximated following [27] by N
, using previous notations and N
is the total number of share in a round, while each share sends to the pool in
miners. Finally, we analyze miners’ cross-pooling behavior the PPS scheme is rewarded by a fixed amount, leading to no or insignificant
that helps to diversify income and risks. reward variance.
(a) Market share, payout scheme and pool fee (b) New and dropout flow (c) Hopping in and out flow
Proportional & Score (only SlushPool) Proportional & Score (only SlushPool) Proportional & Score (only SlushPool)
DeepBit ▲ DeepBit ▲ DeepBit ▲

OzCoin ▲ OzCoin ▲ OzCoin ▲

SlushPool ▲ SlushPool ▲ SlushPool ▲

SlushPool ◼ SlushPool ◼ SlushPool ◼

2012 2013 2014 2015 2016 2017 2018 2019 2020 2012 2013 2014 2015 2016 2017 2018 2019 2020 2012 2013 2014 2015 2016 2017 2018 2019 2020

PPS PPS PPS

DeepBit ▲ DeepBit ▲ DeepBit ▲
Eligius ▲ Eligius ▲ Eligius ▲

Eligius ◼ Eligius ◼ Eligius ◼

BTC Guild ▲ BTC Guild ▲ BTC Guild ▲

BTC Guild ◼ BTC Guild ◼ BTC Guild ◼

OzCoin ▲ OzCoin ▲ OzCoin ▲

OzCoin ◼ OzCoin ◼ OzCoin ◼

EclipseMC ▲ EclipseMC ▲ EclipseMC ▲

50BTC ▲ 50BTC ▲ 50BTC ▲

Polmine ◼ Polmine ◼ Polmine ◼

F2Pool ▲ F2Pool ▲ F2Pool ▲

F2Pool ◼ F2Pool ◼ F2Pool ◼

AntPool ▲ AntPool ▲ AntPool ▲

AntPool ◼ AntPool ◼ AntPool ◼

BTCC Pool ▲ BTCC Pool ▲ BTCC Pool ▲

BTC.com ▲ BTC.com ▲ BTC.com ▲

BTC.com ◼ BTC.com ◼ BTC.com ◼

Poolin ◼ Poolin ◼ Poolin ◼

2012 2013 2014 2015 2016 2017 2018 2019 2020 2012 2013 2014 2015 2016 2017 2018 2019 2020 2012 2013 2014 2015 2016 2017 2018 2019 2020

PPLNS PPLNS PPLNS

BTC Guild ▲ BTC Guild ▲ BTC Guild ▲
BTC Guild ◼ BTC Guild ◼ BTC Guild ◼

BitMinter ◼ BitMinter ◼ BitMinter ◼

50BTC ▲ 50BTC ▲ 50BTC ▲

GHash.IO ◼ GHash.IO ◼ GHash.IO ◼
AntPool ▲ AntPool ▲ AntPool ▲
AntPool ◼ AntPool ◼ AntPool ◼

2012 2013 2014 2015 2016 2017 2018 2019 2020 2012 2013 2014 2015 2016 2017 2018 2019 2020 2012 2013 2014 2015 2016 2017 2018 2019 2020

Market share (%) Pool fee (%) Difference (%) Miner gain vs. loss Difference (%) Miner hop in in vs. hop out
10 20 30 40 0 1 2 3 4 5 6 10 20 40 60 80 100 NEW > DROPOUT DROPOUT > IN 10 20 30 40 50 60 IN > OUT OUT > IN

Fig. 4. Mining pool characteristics and miners’ migration statistics over time. We divided into three main payout schemes: Proportional (and Score only for
Slush Pool), Pay-Per-Share (PPS), and Pay-Per-Last-N-Shares (PPLNS). Each row in a graph represents the mining pool and the shape encodes whether the
pool kept transaction fees for itself (squared ) or shared with its miners (triangle N). We separated the same mining pool in different rows and facets but
provided the shadow colors (blue in a, grey in b and c) to highlight the continuity of the pool with different payout schemes and transaction fee policies. (a)
Market shares and payout schemes. The market share of the mining pool is represented as the size of the circle for each month. Pool fees are encoded as the
color scale. (b) New and dropout flow. The size of the circle represents the absolute difference between new and dropout miners. The positive (or negative)
flow of new miners are encoded as green (or red) color. (c) Hopping in and out flow. The size of the circle represents the absolute difference between hopping
in and hopping out miners. The color indicates whether hopping in miners are more than hopping out miners (green) or vice versa (red). Three grey vertical
lines in each chart indicate halving days on 2012-11-28, 2016-07-09, and 2020-05-11.

pools. While BTC Guild has applied a 3% fee since 2011, Bitcoin experienced 8 local high valuation periods before
50BTC created in 2012 applied a lower fee (2.5%). This trend 2020: 06/2011 ($19), 04/2013 ($130), 12/2013–03/2014 ($800),
got stronger with GHash.io (0%) in 2013 or AntPool (0%) in 06/2014 ($600), 12/2015 ($420), 06/2016 ($630), 05–12/2017
2014. When these pools appeared with lower fees, new miners ($15,000) and 06–12/2019 ($10,500) [29]. These periods are
were attracted by those pools (Fig. 4 (b)) and hopped out from characterized by many new miners entering pools and even
older pools (Fig. 4 (c)). pool creations. Bitcoin mining became an economic investment
Summary: The market share of mining pools is a confounding as demonstrated in Prat and Walter [25] and acquisition of
factor with miner flows. Mining pools that gain market share mining hardware tends to increase when its expected return
tend to attract new and hopping-in miners. Miners drop out and rises. The corollary is that miners exit the market when Bitcoin
hop-out from pools that lose market share. This feedback loop value decreases. Fig. 4 (b) shows indeed large miner dropout
probably explains the domination of a few mining pools at a close to halving days, which correspond to periods of sharp
time. The main driver of pool-hoping we observe in this article decrease in mining revenue.
is the gap between pools fee for a given reward scheme. New
successful pools adopted lower fees to attract miners while the C. Bitcoin values, income optimization and cross-pooling
older ones declined or stopped operating if they did not follow Cross-pooling allows to diversify risks and optimize income.
this trend. After 2015, pool fees tended to converge for eachFig. 5 (c) provides interesting insights into cross-pooling prac-
reward scheme, and pool-hoping decreased. tice. Three periods of intense cross-pooling can be observed:
04/2013–11/2014, 06/2014–12/2014, and 11/2016–01/2017.
B. Bitcoin value and mining market entry These three periods exhibited an economic rationale where
Another evidence for the economic rationale of mining miners seem to diversify their risk between risk-free pools
activities comes from the incoming flow of miners during (PPS) and more risky ones (PPLNS).
bitcoin’s high valuation periods. Fig. 4 (b) provides infor- In particular, the first two periods corresponded to a similar
mation about the new miners joining Bitcoin mining pools. pattern, which is the rise of cross-pooling from PPS pools to-
 a b c to extract payout flows with different payout patterns. We also
DeepBit
propose the miner’s migration flow to measure the mobility of
Eligius

BTC Guild
miners who enter, exit, hop, or cross between mining pools.
BitMinter Based on these algorithms and metrics, we provide an
OzCoin

SlushPool
analysis of miners’ migration among 15 mining pools across
EclipseMC Bitcoin’s history. The visualizations allow us to highlight
50BTC
regular patterns of miners’ entry, hopping, and cross-pooling
Polmine

F2Pool
behaviors. These regularities are consistent with classical
GHash.IO economic behaviors under competition. Our work provides new
AntPool
empirical evidence that miners and mining pools behave as
BTCC Pool

BTC.com
typical economic agents, seeking to maximize their profits. We
Poolin show in particular that pools’ competition is based especially
2012 2013 2014 2015 2016 2017 2018 2019 2020
on pool fee and payout schemes. The most popular current
Avg % miner reward from the pool % cross-pooling miners
payout schemes, namely Pay-per-Share (PPS) and Pay-per-Last-
20 40 60 80
0 20 40 60 80 100 N-Shares (PPLNS), can be seen as two different financial assets
from a miner’s viewpoint: a free-risk scheme (PPS) and a more
Fig. 5. Cross-pooling miners. Each row represents the percentage of cross- risky one (PPLNS). Consistent with the economic rationale,
pooling for each mining pool over time. The percentage of reward that cross-
pooling miners obtained from the pool is encoded as the color scale. Three the more risky one is associated with more expected returns
selected periods that we focused on are highlighted in the grey background. due to the lower pool fee associated with PPLNS. Moreover,
miners tend to perform cross-pooling with pools applying
different reward schemes seemingly for risk diversification and
ward PPLNS ones. The first period (04–11/2014) corresponded income optimization. This is especially the case during high
to the switch from PPS toward PPLNS scheme for BTC Guild. Bitcoin value periods where expected income from mining
BTC Guild applied PPS until February 2013, then proposed gets higher, in particular for the PPLNS payout scheme. These
PPS and PPLNS until March 2014, and after offered uniquely periods are then associated with important levels of new miner
PPLNS. Before that time, cross-pooling was very low. It seems entries and cross-pooling from PPS pools toward PPLNS
that the availability of PPLNS reward has attracted miners ones. Additionally, pool fees are major drivers of the pool’s
operating on pools using PPS (F2Pool or Eligius) or Score competition. We show that new pools tend to apply a lower
(SlushPool) as shown in Fig. 6 (a). fee with respect to incumbents. It leads to a decrease in the
The second period (06–12/2014) followed a similar pattern, average pool’s fee along time and is correlated with important
except for AntPool (a newly created pool). Before the creation pool hopping decisions toward new pools. Our result provides
of AntPool, cross-pooling was limited. Once AntPool launched an insight into the dynamics of mining pools, which is crucial
the PPLNS payout scheme (in addition to PPS), cross-pooling for improving regulations and policies in cryptocurrencies.
rose considerably. Fig. 6 (b) shows that cross-pooling occurred The data obtained from our approach provides rich and
between the main PPS (Eligius and F2Pool) and PPLNS pools detailed information on miners’ migration among pools. The
(AntPool, BTC Guild and GHash.IO). data is publicly available at [11]. For future research, we are
The last period of intense cross-pooling (11/2016–01/2017) planning to develop a quantitative model to explain miners’
followed the same rationale but the timing was different. In decisions and dynamics of mining pool competitions. In
this period, neither the creation nor the switch toward a PPLNS particular, the econometric analysis might be useful to check
pool led to cross-pooling. However, the apparition of a large if the trends highlighted in this empirical work still hold when
PPS pool (BTCC Pool) generated a lot of cross-pooling with controlling for confounding variables. On the other hand, we
an already existing PPLNS pool (AntPool). Fig. 6 (c) shows are developing a visual analytics system to explore and monitor
that a large flow of cross-pooling existed between AntPool and Bitcoin’s mining activities in multiple coordinated views. The
BTCC Pool, and also other PPS pools (BTC.com or F2Pool). tool will allow mining pool managers, individual miners, and
In all the cases reviewed above, we demonstrated that cross- Bitcoin data analysts to analyze the miner’s migration flow
pooling is used to diversify miners’ risk and leads them to in granular details, and to relate the information to pool
combine mining in risk-free pools (PPS) and more risky ones characteristics and other relevant Bitcoin indicators.
(PPLNS). In this respect, individual miners seem to act as a
portfolio manager who optimizes their income concerning the
ACKNOWLEDGMENT
risk associated with each type of asset.
We thank Antoine Durand, Lorenzo Candeago, Suprawee
VI. C ONCLUSION AND F UTURE W ORK Mekrungruangkul, Lucas Khornelord, Kahlifa Toumi, and
We contribute a new approach to extract reward payout Kalpana Singh for their valuable feedback. We also express
flows of mining pools and to detect individual miners from our gratitude toward Malte Möser who maintains BlockSci
the payout flow. We provide rationale and evidence that our and Aleš Janda (WalletExplorer), Marc Jourdan, and Francesco
approach based on the general payout flow model can be used Zola for providing the known entity dataset.
(a) March 2013 - November 2013 (b) June 2014 - December 2014 (c) November 2016 - February 2017
BTC Guild F2Pool AntPool
50BTC GHash.IO F2Pool
GHash.IO BTC Guild BTCC Pool
SlushPool AntPool SlushPool
Eligius Eligius BTC.com
BitMinter SlushPool BitMinter
EclipseMC BTCC Pool Eligius
OzCoin Polmine 0k 20k 40k

AntPool

F2Pool
BTCC Pool

SlushPool

BTC.com
BitMinter

Eligius
F2Pool BitMinter Total Reward

DeepBit EclipseMC
Miner type
Polmine OzCoin same cross Cross pool

0k 200k 400k 50BTC

BTC Guild
50BTC

Eligius
GHash.IO
SlushPool

BitMinter
EclipseMC
OzCoin
F2Pool
DeepBit
Polmine
% Cross pooling
Total Reward 0k 100k 200k

F2Pool

BTC Guild

Eligius

50BTC
GHash.IO

AntPool

SlushPool
BTCC Pool

Polmine
BitMinter

EclipseMC
OzCoin
Total Reward 0 10 20 30
Cross pool

Cross pool

Fig. 6. Cross-pooling among mining pools during three periods: (a) March–November 2013, (b) June–December 2014, and (c) November 2016–February 2017.
The stacked bar chart shows the total reward distributed from each mining pool during the period as a proxy for the pool size. Mining pools are sorted from
the highest to the lowest value. Each bar is separated by the total amount of non-cross-pooling miners (in blue) and cross-pooling miners (in orange). The
heatmap shows the percentage of the total cross-pooling miners from mining pools on the y-axis to other pools on the x-axis.

R EFERENCES for Security. Springer, 2019, pp. 1–12. [Online]. Available:

https://doi.org/10.1007/978-3-030-15465-3 1
[1] S. Nakamoto, “Bitcoin: A peer-to-peer electronic cash system,” Tech. [18] X. F. Liu, X.-J. Jiang, S.-H. Liu, and C. K. Tse, “Knowledge discovery
Rep., 2008. [Online]. Available: http://bitcoin.org/bitcoin.pdf in cryptocurrency transactions: A survey,” IEEE Access, 2021.
[2] L. Wang and Y. Liu, “Exploring miner evolution in bitcoin network,” in [19] H. Kalodner, M. Möser, K. Lee, S. Goldfeder, M. Plattner, A. Chator,
Passive and Active Measurement. Springer, 2015, pp. 290–302. and A. Narayanan, “Blocksci: Design and applications of a blockchain
[3] M. B. Taylor, “The evolution of bitcoin hardware,” Computer, vol. 50, analysis platform,” in USENIX Security Symposium (USENIX Security),
no. 9, pp. 58–66, Sep 2017. Aug 2020, pp. 2721–2738.
[4] Blockchain.com, “Total hash rate of the bitcoin network,” 2020, [20] A. A. Hagberg, D. A. Schult, and P. J. Swart, “Exploring network
Accessed on: Dec 12, 2020. [Online]. Available: https://www.blockchain. structure, dynamics, and function using networkx,” in Proceedings of
com/charts/hash-rate the Python in Science Conference, 2008, pp. 11–15.
[5] BTC.com, “Bitcoin mining difficulty,” 2020, Accessed on: Dec 12, 2020. [21] F. Reid and M. Harrigan, “An analysis of anonymity in the bitcoin
[Online]. Available: https://btc.com/stats/diff system,” in Security and Privacy in Social Networks. Springer, 2013,
[6] ——, “Mining pool statistics,” 2020, Accessed on: Dec 12, 2020. pp. 197–223.
[Online]. Available: https://btc.com/stats/pool [22] M. Harrigan and C. Fretter, “The unreasonable effectiveness of address
[7] Z. Liu, N. C. Luong, W. Wang, D. Niyato, P. Wang, Y.-C. Liang, and clustering,” in Proceedings of the Conference on Ubiquitous Intelligence
D. I. Kim, “A survey on applications of game theory in blockchain,” Computing, Advanced and Trusted Computing, Scalable Computing and
arXiv preprint arXiv:1902.10865, 2019. Communications, Cloud and Big Data Computing, Internet of People, and
[8] M. Belotti, S. Kirati, and S. Secci, “Bitcoin pool-hopping detection,” in Smart World Congress (UIC/ATC/ScalCom/CBDCom/IoP/SmartWorld).
IEEE 4th International Forum on Research and Technology for Society IEEE, 2016, pp. 368–373.
and Industry (RTSI). IEEE, 2018, pp. 1–6. [23] F. Zola, J. L. Bruse, M. Eguimendia, M. Galar, and R. Orduna Urrutia,
[9] M. Romiti, A. Judmayer, A. Zamyatin, and B. Haslhofer, “A deep dive “Bitcoin and cybersecurity: Temporal dissection of blockchain data to
into bitcoin mining pools: An empirical analysis of mining shares,” arXiv unveil changes in entity behavioral patterns,” Applied Sciences, vol. 9,
preprint arXiv:1905.05999, 2019. no. 23, p. 5003, Nov 2019.
[10] J.-z. Xia, Y.-h. Zhang, H. Ye, Y. Wang, G. Jiang, Y. Zhao, C. Xie, X.-y. [24] A. Janda, “Information about walletexplorer.com,” 2020, Accessed on:
Kui, S.-h. Liao, and W.-p. Wang, “SuPoolVisor: a visual analytics system Dec 12, 2020. [Online]. Available: https://www.walletexplorer.com/info
for mining pool surveillance,” Frontiers of Information Technology & [25] J. Prat and B. Walter, “An equilibrium model of the market
Electronic Engineering, vol. 21, no. 4, pp. 507–523, 2020. [Online]. for bitcoin mining,” CESifo, Tech. Rep. 6865, 2018. [Online].
Available: https://doi.org/10.1631/FITEE.1900532 Available: https://www.cesifo.org/en/publikationen/2018/working-paper/
[11] N. Tovanich, N. Soulié, N. Heulot, and P. Isenberg, “Dataset: An empirical equilibrium-model-market-bitcoin-mining
analysis of pool hopping behavior in the bitcoin blockchain (version 1),” [26] N. Tovanich, N. Soulié, and P. Isenberg, “Visual analytics of bitcoin
Dec. 2020. [Online]. Available: https://doi.org/10.5281/zenodo.4342747 mining pool evolution: on the road toward stability?” in International
[12] Blockchain.info, “Blockchain known pools,” 2020, Accessed on: Dec 12, Workshop on Blockchains and Smart Contracts (BSC), Apr 2021.
2020. [Online]. Available: https://raw.githubusercontent.com/blockchain/ [Online]. Available: https://hal.archives-ouvertes.fr/hal-02902465v2
Blockchain-Known-Pools/master/pools.json [27] M. Rosenfeld, “Analysis of bitcoin pooled mining reward systems,” arXiv
[13] BTC.com, “Blockchain known pools,” 2020, Accessed on: Dec 12, preprint arXiv:1112.4980, 2011.
2020. [Online]. Available: https://raw.githubusercontent.com/btccom/ [28] W. F. Sharpe, “The sharpe ratio,” The Journal of Portfolio Management,
Blockchain-Known-Pools-BCH/master/pools.json vol. 21, no. 1, pp. 49–58, 1994.
[14] Bitcoin Wiki contributors, “Comparison of mining pools,” 2020, [29] CoinMarketCap, “Bitcoin market price,” 2020, Accessed on: Dec 12,
Accessed on: Dec 12, 2020. [Online]. Available: https://en.bitcoin.it/ 2020. [Online]. Available: https://coinmarketcap.com/currencies/bitcoin/
wiki/Comparison of mining pools
[15] Bitcointalk contributors, “Bitcoin mining pools discussion forum,”
2020, Accessed on: Dec 12, 2020. [Online]. Available: https:
//bitcointalk.org/index.php?board=41.0
[16] G. Di Battista, V. Di Donato, M. Patrignani, M. Pizzonia, V. Roselli,
and R. Tamassia, “Bitconeview: visualization of flows in the
bitcoin transaction graph,” in Symposium on Visualization for
Cyber Security (VizSec). IEEE, 2015, pp. 1–8. [Online]. Available:
https://doi.org/10.1109/VIZSEC.2015.7312773
[17] M. Ahmed, I. Shumailov, and R. Anderson, “Tendrils of crime:
Visualizing the diffusion of stolen bitcoins,” in Graphical Models

View publication stats