Cryptocurrency Trading: A Comprehensive Survey: March 2020

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/340172901
Cryptocurrency Trading: A Comprehensive Survey
Preprint · March 2020
CITATIONS READS
0 22,206
8 authors, including:
Fang Fan Carmine Ventre

King's College London King's College London
4 PUBLICATIONS 3 CITATIONS 65 PUBLICATIONS 430 CITATIONS
SEE PROFILE SEE PROFILE
Lingbo Li
University College London
19 PUBLICATIONS 96 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
DASSE project View project
All content following this page was uploaded by Carmine Ventre on 10 January 2021.
The user has requested enhancement of the downloaded file.

Fan Fang a,∗ , Carmine Ventrea , Michail Basiosb , Hoiliong Kongb , Leslie Kanthanb ,
David Martinez-Regob , Fan Wu b and Lingbo Li b,∗
a King’s College London, UK
b Turing Intelligence Technology Limited, UK
ARTICLE INFO ABSTRACT

Keywords: In recent years, the tendency of the number of financial institutions including cryptocurrencies in
trading, cryptocurrency, machine learn- their portfolios has accelerated. Cryptocurrencies are the first pure digital assets to be included by
ing, econometrics asset managers. Even though they share some commonalities with more traditional assets, they have a
separate nature of its own and their behaviour as an asset is still under the process of being understood.
arXiv:2003.11352v3 [q-fin.TR] 7 Jan 2021
It is therefore important to summarise existing research papers and results on cryptocurrency trading,
including available trading platforms, trading signals, trading strategy research and risk management.
This paper provides a comprehensive survey of cryptocurrency trading research, by covering 126
research papers on various aspects of cryptocurrency trading (e.g., cryptocurrency trading systems,
bubble and extreme condition, prediction of volatility and return, crypto-assets portfolio construction
and crypto-assets, technical trading and others). This paper also analyses datasets, research trends and
distribution among research objects (contents/properties) and technologies, concluding with some
promising opportunities that remain open in cryptocurrency trading.
1. Introduction
Cryptocurrencies have experienced broad market accep-
tance and fast development despite their recent conception.
Many hedge funds and asset managers have begun to in-
clude cryptocurrency-related assets into their portfolios and
trading strategies. The academic community has similarly
spent considerable efforts in researching cryptocurrency trad-
ing. This paper seeks to provide a comprehensive survey of
the research on cryptocurrency trading, by which we mean
any study aimed at facilitating and building strategies to
trade cryptocurrencies.
Figure 1: Cryptocurrency Trading Publications (cumulative)
As an emerging market and research direction, cryp-
during 2013-2019
tocurrencies and cryptocurrency trading have seen consid-
erable progress and a notable upturn in interest and activ-
ity [103]. From Figure 1, we observe over 85% of papers
have appeared since 2018, demonstrating the emergence of • Emergent trading technologies including economet-
cryptocurrency trading as a new research area in financial ric methods, machine learning technology and other
trading. emergent trading methods;
The literature is organised according tosix distinct as- • Portfolio and cryptocurrency assets including research
pects of cryptocurrency trading: among cryptocurrency co-movements and crypto-asset
• Cryptocurrency trading software systems (i.e., real- portfolio research;
time trading systems, turtle trading systems, arbitrage
• Market condition research including bubbles [106] or
trading systems);
crash analysis and extreme conditions;
• Systematic trading including technical analysis, pairs
• Other Miscellaneous cryptocurrency trading research.
trading and other systematic trading methods;
fan.fang@kcl.ac.uk ( Fan Fang);
In this survey we aim at compiling the most relevant re-
fan.fang@kcl.ac.uk ( Fan Fang); search in these areas and extract a set of descriptive indica-
carmine.ventre@kcl.ac.uk ( Carmine Ventre); tors that can give an idea of the level of maturity research in
mike@turintech.ai ( Michail Basios); justin@turintech.ai this area has achieved.
( Hoiliong Kong); leslie@turintech.ai ( Leslie Kanthan);
david@turintech.ai ( David Martinez-Rego); We also summarise research distribution (among research
fan@turintech.ai ( Fan Wu); lingbo@turintech.ai ( properties, categories and research technologies). The dis-
Lingbo Li) tribution among properties and categories identifies classifi-
www.cvr.cc,cvrŽsayahna.org ( Fan Fang) cations of research objectives and contents. The distribution
ORCID (s): 0000-0002-5086-5614 ( Fan Fang);
0000-0002-3073-1352 ( Lingbo Li)
among technologies identifies classifications of methodol-
First Author et al. Page 1 of 30

ogy or technical methods in researching cryptocurrency trad-

ing. Specifically, we subdivide research distribution among
categories and technologies into statistical methods and ma-
chine learning technologies. Moreover, We identify datasets
and opportunities (potential research directions) that have
appeared in the cryptocurrency trading area. To ensure that
our survey is self-contained, we aim to provide sufficient
material to adequately guide financial trading researchers
who are interested in cryptocurrency trading.
There has been related work that discussed or partially
surveyed the literature related to cryptocurrency trading. Kyr-
iazis et al. [166] surveyed efficiency and profitable trading
opportunities in cryptocurrency markets. Ahamad et al. [4] Figure 2: Tree structure of the contents in this paper
and Sharma et al. [221] gave a brief survey on cryptocur-
rencies. Ujan et al. [191] gave a brief survey of cryptocur-
rency systems. Ignasi et al. [186] performed a bibliometric
analysis of bitcoin literature. These relate work outcomes
focused on specific area in cryptocurrency, including cryp-
tocurrencies and cryptocurrency market introduction, cryp-
tocurrency systems / platforms, bitcoin literature review, etc.
To the best of our knowledge, no previous work has pro-
vided a comprehensive survey particularly focused on cryp- Figure 3: Workflow of Blockchain transaction
tocurrency trading.
In summary, the paper makes the following contribu-
tions: 2.1. Blockchain
2.1.1. Blockchain Technology Introduction
Definition. This paper defines cryptocurrency trading and
Blockchain is a digital ledger of economic transactions
categorises it into: cryptocurrency markets, cryptocur-
that can be used to record not just financial transactions, but
rency trading models, and cryptocurrency trading strate-
any object with an intrinsic value. [232]. In its simplest
gies. The core content of this survey is trading strate-
form, a Blockchain is a series of immutable data records
gies for cryptocurrencies while we cover all aspects
with timestamps,which are managed by a cluster of ma-
of it.
chines that do not belong to any single entity. Each of these
Multidisciplinary Survey. The paper provides a compre- data blocks is protected by cryptographic principle and bound
hensive survey of 126 cryptocurrency trading papers, to each other in a chain (cf. Figure 3 for the workflow).
across different academic disciplines such as finance Cryptocurrencies like Bitcoin are made on a peer-to-
and economics, artificial intelligence and computer peer network structure. Each peer has a complete history
science. Some papers may cover multiple aspects and of all transactions, thus recording the balance of each ac-
will be surveyed for each category. count. For example, a transaction is a file that says “A pays
X Bitcoins to B” that is signed by A using its private key.
Analysis. The paper analyses the research distribution, datasets This is basic public-key cryptography, but also the build-
and trends that characterise the cryptocurrency trading block on which cryptocurrencies are based. After being
ing literature. signed, the transaction is broadcast on the network. When a
peer discovers a new transaction, it checks to make sure that
Horizons. The paper identifies challenges, promising re-
the signature is valid (this amounts to use the signer’s public
search directions in cryptocurrency trading, aimed to
key, denoted as the algorithm in Figure 3). If the verification
promote and facilitate further research.
is valid then the block is added to the chain; all other blocks
Figure 2 depicts the paper structure, which is informed added after it will “confirm” that transaction. For example,
by the review schema adopted. More details about this can if a transaction is contained in block 502 and the length of
be found in Section 4. the blockchain is 507 blocks, it means that the transaction
has 5 confirmations (507-502) [218].
2. Cryptocurrency Trading 2.1.2. From Blockchain to Cryptocurrencies

This section provides an introduction to cryptocurrency Confirmation is a critical concept in cryptocurrencies;
trading. We will discuss Blockchain, as the enabling tech- only miners can confirm transactions. Miners add blocks
nology, cryptocurrency markets and cryptocurrency trad- to the Blockchain; they retrieve transactions in the previ-
ing strategies. ous block and combine it with the hash of the preceding
block to obtain its hash, and then store the derived hash into

the current block. Miners in Blockchain accept transactions,

mark them as legitimate and broadcast them across the net-
work. After the miner confirms the transaction, each node
must add it to its database. In layman terms, it has become
part of the Blockchain and miners undertake this work to
obtain cryptocurrency tokens, such as Bitcoin. In contrast
to Blockchain, cryptocurrencies are related to the use of to-
kens based on distributed ledger technology. Any transac-
tion involving purchase, sale, investment, etc. involves a
Blockchain native token or sub-token. Blockchain is a plat-
form that drives cryptocurrency and is a technology that acts Figure 4: Total Market Capitalization and Volume of cryp-
as a distributed ledger for the network. The network creates tocurrency market, USD [238]
a means of transaction and enables the transfer of value and
information. Cryptocurrencies are the tokens used in these
networks to send value and pay for these transactions. They cies are theoretically immune to government control and in-
can be thought of as tools on the Blockchain, and in some terference.
cases can also function as resources or utilities. In other As of December 20, 2019, there exist 4,950 cryptocur-
instances, they are used to digitise the value of assets. In rencies and 20,325 cryptocurrency markets; the market cap
summary, Cryptocurrencies are part of an ecosystem-based is around 190 billion dollars [78]. Figure 4 shows histor-
on Blockchain technology. ical data on global market capitalisation and 24-hour trad-
ing volume [238]. The total market cap is calculated by
2.2. Introduction of cryptocurrency market aggregating the dollar market cap of all cryptocurrencies.
2.2.1. What is cryptocurrency? From the figure, we can observe how cryptocurrencies expe-
Cryptocurrency is a decentralised medium of exchange rience exponential growth in 2017 and a large bubble burst
which uses cryptographic functions to conduct financial trans- in early 2018. But in recent years, cryptocurrencies have
actions [90]. Cryptocurrencies leverage the Blockchain tech- shown signs of stabilisation.
nology to gain decentralisation, transparency, and immutabil- There are three mainstream cryptocurrencies: Bitcoin
ity [187]. In the above, we have discussed how Blockchain (BTC), Ethereum (ETH), and Litecoin (LTC). Bitcoin was
technology is implemented for cryptocurrencies. created in 2009 and garnered massive popularity. On Oc-
In general, the security of cryptocurrencies is built on tober 31, 2008, an individual or group of individuals oper-
cryptography, neither by people nor on trust [194]. For ex- ating under the pseudonym Satoshi Nakamoto released the
ample, Bitcoin uses a method called ”Elliptic Curve Cryp- Bitcoin white paper and described it as: ”A pure peer-to-
tography” to ensure that transactions involving Bitcoin are peer version of electronic cash that can be sent online for
secure [246]. Elliptic curve cryptography is a type of public- payment from one party to another without going through
key cryptography that relies on mathematics to ensure the a counterparty, ie. a financial institution.” [193] Launched
security of transactions. When someone attempts to circum- by Vitalik Buterin in 2015, Ethereum is a special Blockchain
vent the aforesaid encryption scheme by brute force, it takes with a special token called Ether (ETH symbol in exchanges).
them one-tenth the age of the universe to find a value match A very important feature of Ethereum is the ability to create
when trying 250 billion possibilities every second [118]. new tokens on the Ethereum Blockchain. The Ethereum net-
Regarding its use as a currency, cryptocurrency has the same work went live on July 30, 2015, and pre-mined 72 million
properties as money. It has a controlled supply. Most cryp- Ethereum. Litecoin is a peer-to-peer cryptocurrency created
tocurrencies limit the supply of tokens. E.g. for Bitcoin, by Charlie Lee. It was created according to the Bitcoin pro-
the supply will decrease over time and will reach its final tocol, but it uses a different hashing algorithm. Litecoin
quantity sometime around 2,140. All cryptocurrencies con- uses a memory-intensive proof-of-work algorithm, Scrypt.
trol the supply of tokens through a timetable encoded in the Scrypt
Blockchain. Figure 5 shows percentages of total cryptocurrency mar-
One of the most important features of cryptocurrencies ket capitalisation; Bitcoin and Ethereum occupy the vast
is the exclusion of financial institution intermediaries [125]. majority of the total market capitalisation (data collected on
The absence of a “middleman” lowers transaction costs for 8 Jan 2020).
traders. For comparison, if a bank’s database is hacked or
damaged, the bank will rely entirely on its backup to recover 2.2.2. Cryptocurrency Exchanges
any information that is lost or compromised. With cryp- A cryptocurrency exchange or digital currency exchange
tocurrencies, even if part of the network is compromised, (DCE) is a business that allows customers to trade cryp-
the rest will continue to be able to verify transactions cor- tocurrencies. Cryptocurrency exchanges can be market mak-
rectly. Cryptocurrencies also have the important feature of ers, usually using the bid-ask spread as a commission for
not being controlled by any central authority [217]: the de- services, or as a matching platform, by simply charging fees.
centralised nature of the Blockchain ensures cryptocurren- Table 1 shows the top or classical cryptocurrency ex-

Table 1
Cryptocurrency exchanges Lists
Exchanges Category Supported currencies Fiat Currency Registration country Regulatory authority
CME Derivatives BTC and Ethereum [71] USD USA [73] CFTC [72]
CBOE Derivatives BTC [59] USD USA [58] CFTC [60]
BAKKT (NYSE) Derivatives BTC [15] USD USA [16] CFTC [15]
BitMex Derivatives 12 cryptocurrencies [31] USD Seychelles [32] -
Binance Spot 98 cryptocurrencies [27] EUR, NGN, RUB, TRY Malta [181] FATF [26]
Coinbase Spot 28 cryptocurrencies [76] EUR, GBP, USD USA [37] SEC [77]
Bitfinex Spot > 100 cryptocurrencies [28] EUR, GBP, JPY, USD British Virgin Islands [29] NYAG [30]
Bitstamp Spot 5 cryptocurrencies [33] EUR, USD Luxembourg [34] CSSF [35]
Poloniex Spot 23 cryptocurrencies [213] USD USA [213] -
a set of predefined rules to buy and sell on cryptocurrency

markets.
2.3.2. Advantages of Trading Cryptocurrency

The benefits of cryptocurrency trading include:
Drastic fluctuations. The volatility of cryptocurrencies are
often likely to attract speculative interest and investors.
The rapid fluctuations of intraday prices can provide
traders with great money-earning opportunities, but it
also includes more risk.
24-hour market. The cryptocurrency market is available
24 hours a day, 7 days a week because it is a de-
centralised market. Unlike buying and selling stocks
Figure 5: Percentage of Total Market Capitalisation [79]
and commodities, the cryptocurrency market is not
traded physically from a single location. Cryptocur-
rency transactions can take place between individuals,
changes according to the rank list, by volume, compiled
in different venues across the world.
on “nomics” website [199]. Chicago Mercantile Exchange
(CME), Chicago Board Options Exchange (CBOE) as well Near Anonymity. Buying goods and services using cryp-
as BAKKT (backed by New York Stock Exchange) are reg- tocurrencies is done online and does not require to
ulated cryptocurrency exchanges. Fiat currency data also make one’s own identity public. With increasing con-
comes from “nomics” website [199]. Regulatory authority cerns over identity theft and privacy, cryptocurren-
and supported currencies of listed exchanges are collected cies can thus provide users with some advantages re-
from official websites or blogs. garding privacy. Different exchanges have specific
Know-Your-Customer (KYC) measures used to iden-
2.3. Cryptocurrency Trading tify users or customers [3]. The KYC undertook in
2.3.1. Definition the exchanges allows financial institutions to reduce
Firstly we give a definition of cryptocurrency trading. the financial risk while maximising the wallet owner’s
Definition 1. Cryptocurrency trading is the act of buying anonymity.
and selling of cryptocurrencies with the intention of making
Peer-to-peer transactions. One of the biggest benefits of
a profit.
cryptocurrencies is that they do not involve financial
The definition of cryptocurrency trading can be broken down institution intermediaries. As mentioned above, this
into three aspects: object, operation mode and trading strat- can reduce transaction costs. Moreover, this feature
egy. The object of cryptocurrency trading is the asset being might appeal to users who distrust traditional systems.
traded, which is “cryptocurrency”. The operation mode of Over-the-counter (OTC) cryptocurrency markets of-
cryptocurrency trading depends on the means of transaction fer, in this context, peer-to-peer transactions on the
in the cryptocurrency market, which can be classified into Blockchain. The most famous cryptocurrency OTC
“trading of cryptocurrency Contract for Differences (CFD)” market is “LocalBitcoin [176]”.
(The contract between the two parties, often referred to as
Programmable “smart” capabilities. Some cryptocurren-
the “buyer” and “seller”, stipulates that the buyer will pay
cies can bring other benefits to holders, including lim-
the seller the difference between themselves when the po-
ited ownership and voting rights. Cryptocurrencies
sition closes [11]) and “buying and selling cryptocurrencies
may also include a partial ownership interest in phys-
via an exchange”. A trading strategy in cryptocurrency trad-
ical assets such as artwork or real estate.
ing, formulated by an investor, is an algorithm that defines

3. Cryptocurrency Trading Strategy and investors in making optimal trading and investment de-
cisions [116]. Pairs trading is a systematic trading strat-
Cryptocurrency trading strategy is the main focus of this
egy that considers two similar assets with slightly different
survey. There are many trading strategies, which can be spreads. If the spread widens, short the high stocks and buy
broadly divided into two main categories: technical and fun- the low stocks. When the spread narrows again to a certain
damental. They are similar in the sense that they both rely
equilibrium value, a profit is generated [94]. Papers shown
on quantifiable information that can be backtested against
in this section involve the analysis and comparison of tech-
historical data to verify their performance. In recent years,
nical indicators, pairs and informed trading, amongst other
a third kind of trading strategy, which we call quantitative, strategies.
has received increasing attention. Such a trading strategy is
similar to a technical trading strategy because it uses trad- 3.3. Emergent Trading Technologies
ing activity information on the exchange to make buying or Emergent trading strategies for cryptocurrency include
selling decisions. Quantitative traders build trading strate- strategies that are based on econometrics and machine learn-
gies with quantitative data, which is mainly derived from ing technologies.
price, volume, technical indicators or ratios to take advan-
tage of inefficiencies in the market and are executed auto- 3.3.1. Econometrics on Cryptocurrency
matically by trading software. Cryptocurrency market is dif- Econometric methods apply a combination of statistical
ferent from traditional markets as there are more arbitrage and economic theories to estimate economic variables and
opportunities, higher fluctuation and transparency. Due to predict their values [244]. Statistical models use mathe-
these characteristics, most traders and analysts prefer using matical equations to encode information extracted from the
quantitative trading strategies in cryptocurrency markets. data [152]. In some cases, statistical modeling techniques
can quickly provide sufficiently accurate models [24]. Other
3.1. Cryptocurrency Trading Software System methods might be used, such as sentiment-based prediction
Software trading systems allow international transactions, and long-and-short-term volatility classification based pre-
process customer accounts and information, and accept and diction [64]. The prediction of volatility can be used to
execute transaction orders [50]. A cryptocurrency trading judge the price fluctuation of cryptocurrencies, which is also
system is a set of principles and procedures that are pre- valuable for the pricing of cryptocurrency-related deriva-
programmed to allow trade between cryptocurrencies and tives [147].
between fiat currencies and cryptocurrencies. Cryptocur- When studying cryptocurrency trading using economet-
rency trading systems are built to overcome price manip- rics, researchers apply statistical models on time-series data
ulation, cybercriminal activities and transaction delays [21]. like generalised autoregressive conditional heteroskedastic-
When developing a cryptocurrency trading system, we must ity (GARCH) and BEKK (named after Baba, Engle, Kraft
consider the capital market, base asset, investment plan and and Kroner, 1995 [96]) models to evaluate the fluctuation
strategies [190]. Strategies are the most important part of an of cryptocurrencies [55]. A linear statistical model is a
effective cryptocurrency trading system and they will be in- method to evaluate the linear relationship between prices
troduced below. There exist several cryptocurrency trading and an explanatory variable [196]. When there exists more
systems that are available commercially, for example, Cap- than one explanatory variable, we can model the linear re-
folio, 3Commas, CCXT, Freqtrade and Ctubio. From these lationship between explanatory (independent) and response
cryptocurrency trading systems, investors can obtain pro- (dependent) variables with multiple linear models. The com-
fessional trading strategy support, fairness and transparency mon linear statistical model used in the time-series analysis
from the professional third-party consulting companies and is the autoregressive moving average (ARMA) model [69].
fast customer services.
3.3.2. Machine Learning Technology
3.2. Systematic Trading Machine learning is an efficient tool for developing Bit-
Systematic Trading is a way to define trading goals, coin and other cryptocurrency trading strategies [185] be-
risk controls and rules. In general, systematic trading in- cause it can infer data relationships that are often not di-
cludes high frequency trading and slower investment types rectly observable by humans. From the most basic perspec-
like systematic trend tracking. In this survey, we divide tive, Machine Learning relies on the definition of two main
systematic cryptocurrency trading into technical analysis, components: input features and objective function. The def-
pairs trading and others. Technical analysis in cryptocur- inition of Input Features (data sources) is where knowledge
rency trading is the act of using historical patterns of trans- of fundamental and technical analysis comes into play. We
action data to assist a trader in assessing current and pro- may divide the input into several groups of features, for ex-
jecting future market conditions for the purpose of making ample, those based on Economic indicators (such as, gross
profitable trades. Price and volume charts summarise all domestic product indicator, interest rates, etc.), Social indi-
trading activity made by market participants in an exchange cators (Google Trends, Twitter, etc.), Technical indicators
and affect their decisions. Some experiments showed that (price, volume, etc.) and other Seasonal indicators (time of
the use of specific technical trading rules allows generat- day, day of the week, etc.). The objective function defines
ing excess returns, which is useful to cryptocurrency traders the fitness criteria one uses to judge if the Machine Learn-

ally, crypto asset denotes a digital asset (i.e., cryptocurren-

cies and derivatives). There are some common ways to build
a diversified portfolio in crypto assets. The first method is
to diversify across markets, which is to mix a wide vari-
ety of investments within a portfolio of the cryptocurrency
market. The second method is to consider the industry sec-
Figure 6: Process of machine learning in predicting cryp- tor, which is to avoid investing too much money in any one
tocurrency category. Diversified investment of portfolio in the cryp-
tocurrency market includes portfolio across cryptocurren-
cies [175] and portfolio across the global market including
ing model has learnt the task at hand. Typical predictive stocks and futures [140].
models try to anticipate numeric (e.g., price) or categorical
(e.g., trend) unseen outcomes. The machine learning model 3.5. Market Condition Research
is trained by using historic input data (sometimes called Market condition research appears especially important
in-sample) to generalise patterns therein to unseen (out-of- for cryptocurrencies. A financial bubble is a significant in-
sample) data to (approximately) achieve the goal defined by crease in the price of an asset without changes in its intrinsic
the objective function. Clearly, in the case of trading, the value [48]. Many experts pinpoint a cryptocurrency bubble
goal is to infer trading signals from market indicators which in 2017 when the prices of cryptocurrencies grew by 900%.
help to anticipate asset future returns. In 2018, Bitcoin faced a collapse in its value. This signif-
Generalisation error is a pervasive concern in the appli- icant fluctuation inspired researchers to study bubbles and
cation of Machine Learning to real applications, and of ut- extreme conditions in cryptocurrency trading.
most importance in Financial applications. We need to use
statistical approaches, such as cross validation, to validate 4. Paper Collection and Review Schema
the model before we actually use it to make predictions.
In machine learning, this is typically called “validation”. The section introduces the scope and approach of our
The process of using machine learning technology to pre- paper collection, a basic analysis, and the structure of our
dict cryptocurrency is shown in Figure 6. survey.
Depending on the formulation of the main learning loop,
4.1. Survey Scope
we can classify Machine Learning approaches into three
We adopt a bottom-up approach to the research in cryp-
categories: Supervised learning, Unsupervised learning and
tocurrency trading, starting from the systems up to risk man-
Reinforcement learning. Supervised learning is used to de-
agement techniques. For the underlying trading system, the
rive a predictive function from labeled training data. La-
focus is on the optimisation of trading platforms structure
beled training data means that each training instance in-
and improvements of computer science technologies.
cludes inputs and expected outputs. Usually, these expected
At a higher level, researchers focus on the design of
outputs are produced by a supervisor and represent the ex-
models to predict return or volatility in cryptocurrency mar-
pected behaviour of the model. The most used labels in
kets. These techniques become useful to the generation of
trading are derived from in sample future returns of assets.
trading signals. on the next level above predictive mod-
Unsupervised learning tries to infer structure from unla-
els, researchers discuss technical trading methods to trade
beled training data and it can be used during exploratory
in real cryptocurrency markets. Bubbles and extreme condi-
data analysis to discover hidden patterns or to group data
tions are hot topics in cryptocurrency trading because, as
according to any pre-defined similarity metrics. Reinforce-
discussed above, these markets have shown to be highly
ment learning utilises software agents trained to maximise
volatile (whilst volatility went down after crashes). Portfo-
a utility function, which defines their objective; this is flex-
lio and cryptocurrency asset management are effective meth-
ible enough to allow agents to exchange short term returns
ods to control risk. We group these two areas in risk man-
for future ones. In the financial sector, some trading chal-
agement research. Other papers included in this survey in-
lenges can be expressed as a game in which an agent aims
clude topics like pricing rules, dynamic market analysis,
at maximising the return at the end of the period.
regulatory implications, and so on. Table 2 shows the gen-
The use of machine learning in cryptocurrency trading
eral scope of cryptocurrency trading included in this survey.
research encompasses the connection between data sources’
Since many trading strategies and methods in cryptocur-
understanding and machine learning model research. Fur-
rency trading are closely related to stock trading, some re-
ther concrete examples are shown in a later section.
searchers migrate or use the research results for the latter
to the former. When conducting this research, we only con-
3.4. Portfolio Research
sider those papers whose research focuses on cryptocurrency
Portfolio theory advocates diversification of investments
markets or a comparison of trading in those and other finan-
to maximize returns for a given level of risk by allocating
cial markets.
assets strategically. The celebrated mean-variance optimisa-
Specifically, we apply the following criteria when col-
tion is a prominent example of this approach [182]. Gener-
lecting papers related to cryptocurrency trading:

Table 2 Table 3
Survey scope table Paper query results. #Hits, #Title, and #Body denote the
Trading System number of papers returned by the search, left after title filter-
Prediction (return) ing, and left after body filtering, respectively.
Trading (bottom up)
Prediction (volatility) Key Words #Hits #Title #Body
Technical trading methods [Crypto] + Trading 555 32 29
Risk management Bubble and extreme condition [Crypto] + Trading System 4 3 2
Porfolio and Cryptocurrency asset [Crypto] + Prediction 26 14 13
Others [Crypto] + Trading Strategy 22 9 8
[Crypto] + Risk Management /
120 14 14
[Crypto] + Portfolio
1. The paper introduces or discusses the general idea of Query - - 66
cryptocurrency trading or one of the related aspects of Snowball - - 60
cryptocurrency trading. Overall - - 126
2. The paper proposes an approach, study or framework
that targets optimised efficiency or accuracy of cryp-
tocurrency trading. 4.3. Collection Results
3. The paper compares different approaches or perspec- Table 3 shows the details of the results from our paper
tives in trading cryptocurrency. collection. Keyword searches and snowballing resulted in
126 papers across the six research areas of interest in Sec-
By “cryptocurrency trading” here, we mean one of the terms tion 4.1.
listed in Table 2 and discussed above. Figure 7 shows the distribution of papers published at
Some researchers gave a brief survey of cryptocurrency [4, different research sites. Among all the papers, 45.24% pa-
221], cryptocurrency systems [191] and cryptocurrency trad- pers are published in Finance and Economics venues such
ing opportunities [166]. These surveys are rather limited in as Journal of Financial Economics (JFE), Cambridge Centre
scope as compared to ours, which also includes a discussion for Alternative Finance (CCAF), Finance Research Letters,
on the latest papers in the area; we want to remark that this Centre for Economic Policy Research (CEPR) and Journal
is a fast-moving research field. of Risk and Financial Management (JRFM); 4.76% papers
are published in Science venues such as Public Library Of
4.2. Paper Collection Methodology Science one (PLOS one), Royal Society open science and
To collect the papers in different areas or platforms, we SAGE; 15.87% papers are published in Intelligent Engi-
used keyword searches on Google Scholar and arXiv, two neering and Data Mining venues such as Symposium Se-
of the most popular scientific databases. We also choose ries on Computational Intelligence (SSCI), Intelligent Sys-
other public repositories like SSRN but we find that almost tems Conference (IntelliSys), Intelligent Data Engineering
all academic papers in these platforms can also be retrieved and Automated Learning (IDEAL) and International Con-
via Google Scholar; consequently, in our statistical analysis, ference on Data Mining (ICDM); 4.76% papers are pub-
we count those as Google Scholar hits. We choose arXiv as lished in Physics / Physicians venues (mostly in Physics
another source since it allows this survey to be contempo- venue) such as Physica A; 10.32% papers are published in
rary with all the most recent findings in the area. The in- AI and complex system venues such as Complexity and In-
terested reader is warned that these papers have not under- ternational Federation for Information Processing (IFIP); 17.46%
gone formal peer review. The keywords used for searching papers are published in Others venues which contains inde-
and collecting are listed below. [Crypto] means the cryp- pendently published papers and dissertations; 1.59% papers
tocurrency market, which is our research interest because are published on arXiv. The distribution of different venues
methods might be different among different markets. We shows that cryptocurrency trading is mostly published in Fi-
conducted 6 searches across the two repositories just before nance and Economics venues, but with a wide diversity oth-
October 15, 2019. erwise.
- [Crypto] + Trading
4.4. Survey Organisation
- [Crypto] + Trading system
We discuss the contributions of the collected papers and
- [Crypto] + Prediction
a statistical analysis of these papers in the remainder of the
- [Crypto] + Trading strategy paper, according to Table 4.
- [Crypto] + Risk Management The papers in our collection are organised and presented
- [Crypto] + Portfolio from six angles. We introduce the work about several dif-
To ensure high coverage, we adopted the so-called snow- ferent cryptocurrency trading software systems in Section
balling [250] method on each paper found through these 5. Section 6 introduces systematic trading applied to cryp-
keywords. We checked papers added from snowballing meth- tocurrency trading. In Section 7, we introduce some emer-
ods that satisfy the criteria introduced above until we reached gent trading technologies including econometrics on cryp-
closure. tocurrencies, machine learning technologies and other emer-

Table 4
Review Schema
Classification Sec Topic
5.1 Trading Infrastructure System
5.2 Real-time Cryptocurrency Trading System
Cryptocurrency Trading Software System
5.3 Turtle trading system in Cryptocurrency market
5.4 Arbitrage Trading Systems for Cryptocurrencies
5.5 Comparison of three cryptocurrency trading systems
6.1 Technical Analysis
Systematic Trading 6.2 Pairs Trading
6.3 Others
7.1 Econometrics on cryptocurrency
Emergent Trading Technologies 7.2 Machine learning technology
7.3 Others
8.1 Research among cryptocurrency pairs and related factors
Portfolio and Cryptocurrency Assets
8.2 Crypto-asset portfolio research
9.1 Bubbles and crash analysis
Market condition research
9.2 Extreme condition
Others 10 Others related to Cryptocurrency Trading
11.1 Timeline
11.2 Research distribution among properties
Summary Analysis of Literature Review
11.3 Research distribution among categories and technologies
11.4 Datasets used in cryptocurrency trading
5. Cryptocurrency Trading Software Systems

5.1. Trading Infrastructure Systems
Following the development of computer science and cryp-
tocurrency trading, many cryptocurrency trading systems/bots
have been developed. Table 5 compares the cryptocurrency
trading systems existing in the market. The table is sorted
based on URL types (GitHub or Official website) and GitHub
stars (if appropriate).
Capfolio is a proprietary payable cryptocurrency trading
system which is a professional analysis platform and has an
advanced backtesting engine [51]. It supports five different
Figure 7: Publication Venue Distribution cryptocurrency exchanges.
3 Commas is a proprietary payable cryptocurrency trad-
ing system platform that can take profit and stop-loss orders
gent trading technologies in the cryptocurrency market. Sec- at the same time [1]. Twelve different cryptocurrency ex-
tion 8 introduces research on cryptocurrency pairs and re- changes are compatible with this system.
lated factors and crypto-asset portfolios research. In Sec- CCXT is a cryptocurrency trading system with a unified
tion 9 we discuss cryptocurrency market condition research, API out of the box and optional normalized data and sup-
including bubbles, crash analysis, and extreme conditions. ports many Bitcoin / Ether / Altcoin exchange markets and
Section 10 introduces other research included in cryptocur- merchant APIs. Any trader or developer can create a trading
rency trading not covered above. strategy based on this data and access public transactions
We would like to emphasize that the six headings above through the APIs [61]. The CCXT library is used to con-
focus on a particular aspect of cryptocurrency trading; we nect and trade with cryptocurrency exchanges and payment
give a complete organisation of the papers collected under processing services worldwide. It provides quick access to
each heading. This implies that those papers covering more market data for storage, analysis, visualisation, indicator de-
than one aspect will be discussed in different sections, once velopment, algorithmic trading, strategy backtesting, auto-
from each angle. mated code generation and related software engineering. It
We analyse and compare the number of research papers is designed for coders, skilled traders, data scientists and fi-
on different cryptocurrency trading properties and technolo- nancial analysts to build trading algorithms. Current CCXT
gies in Section 11, where we also summarise the datasets features include:
and the timeline of research in cryptocurrency trading.
I. Support for many cryptocurrency exchanges;
We build upon this review to conclude in Section 12 with
II. Fully implemented public and private APIs;
some opportunities for future research.
III. Optional normalized data for cross-exchange analysis
and arbitrage;

IV. Out-of-the-box unified API, very easy to integrate. CryptoSignal is a professional technical analysis cryp-
tocurrency trading system [86]. Investors can track over 500
Blackbird Bitcoin Arbitrage is a C++ trading system
coins of Bittrex, Bitfinex, GDAX, Gemini and more. Auto-
that automatically executes long / short arbitrage between
mated technical analysis includes momentum, RSI, Ichimoku
Bitcoin exchanges. It can generate market-neutral strate-
Cloud, MACD, etc. The system gives alerts including Email,
gies that do not transfer funds between exchanges [36]. The
Slack, Telegram, etc. CryptoSignal has two primary fea-
motivation behind Blackbird is to naturally profit from these
tures. First of all, it offers modular code for easy implemen-
temporary price differences between different exchanges while
tation of trading strategies; Secondly, it is easy to install with
being market neutral. Unlike other Bitcoin arbitrage sys-
Docker.
tems, Blackbird does not sell but actually short sells Bitcoin
Ctubio is a C++ based low latency (high frequency)
on the short exchange. This feature offers two important
cryptocurrency trading system [87]. This trading system can
advantages. Firstly, the strategy is always market agnostic:
place or cancel orders through supported cryptocurrency ex-
fluctuations (rising or falling) in the Bitcoin market will not
changes in less than a few milliseconds. Moreover, it pro-
affect the strategy returns. This eliminates the huge risks of
vides a charting system that can visualise the trading ac-
this strategy. Secondly, this strategy does not require trans-
count status including trades completed, target position for
ferring funds (USD or BTC) between Bitcoin exchanges.
fiat currency, etc.
Buy and sell transactions are conducted in parallel on two
Catalyst is an analysis and visualization of the cryp-
different exchanges. There is no need to deal with transmis-
tocurrency trading system [57]. It makes trading strategies
sion delays.
easy to express and backtest them on historical data (daily
StockSharp is an open-source trading platform for trad-
and minute resolution), providing analysis and insights into
ing at any market of the world including 48 cryptocurrency
the performance of specific strategies. Catalyst allows users
exchanges [227]. It has a free C# library and free trading
to share and organise data and build profitable, data-driven
charting application. Manual or automatic trading (algo-
investment strategies. Catalyst not only supports the trading
rithmic trading robot, regular or HFT) can be run on this
execution but also offers historical price data of all crypto
platform. StockSharp consists of five components that offer
assets (from minute to daily resolution). Catalyst also has
different features:
backtesting and real-time trading capabilities, which enables
I. S#.Designer - Free universal algorithm strategy app, users to seamlessly transit between the two different trad-
easy to create strategies; ing modes. Lastly, Catalyst integrates statistics and machine
II. S#.Data - free software that can automatically load and learning libraries (such as matplotlib, scipy, statsmodels and
store market data; sklearn) to support the development, analysis and visualiza-
III. S#.Terminal - free trading chart application (trading tion of the latest trading systems.
terminal); Golang Crypto Trading Bot is a Go based cryptocur-
IV. S#.Shell - ready-made graphics framework that can be rency trading system [117]. Users can test the strategy in
changed according to needs and has a fully open source sandbox environment simulation. If simulation mode is en-
in C#; abled, a fake balance for each coin must be specified for
V. S#.API - a free C# library for programmers using Vi- each exchange.
sual Studio. Any trading strategies can be created in 5.2. Real-time Cryptocurrency Trading Systems
S#.API. Amit et al. [21] developed a real-time Cryptocurrency
Freqtrade is a free and open-source cryptocurrency trad- Trading System. A real-time cryptocurrency trading system
ing robot system written in Python. It is designed to support is composed of clients, servers and databases. Traders use
all major exchanges and is controlled by telegram. It con- a web-application to login to the server to buy/sell crypto
tains backtesting, mapping and money management tools, assets. The server collects cryptocurrency market data by
and strategy optimization through machine learning [108]. creating a script that uses the Coinmarket API. Finally, the
Freqtrade has the following features: database collects balances, trades and order book informa-
tion from the server. The authors tested the system with an
I. Persistence: Persistence is achieved through SQLite experiment that demonstrates user-friendly and secure expe-
technology; riences for traders in the cryptocurrency exchange platform.
II. Strategy optimization through machine learning: Use
machine learning to optimize your trading strategy pa- 5.3. Turtle trading system in Cryptocurrency
rameters with real trading data; market
III. Marginal Position Size: Calculates winning rate, risk- The original Turtle Trading system is a trend following
return ratio, optimal stop loss and adjusts position size, trading system developed in the 1970s. The idea is to gen-
and then trades positions for each specific market; erate buy and sell signals on stock for short-term and long-
IV. Telegram management: use telegram to manage the term breakouts and its cut-loss condition which is measured
robot. by Average true range (ATR) [144]. The trading system will
V. Dry run: Run the robot without spending money; adjust the size of assets based on their volatility. Essen-
tially, if a turtle accumulates a position in a highly volatile

Table 5
Comparison of existing cryptocurrency trading systems. #Exchange, Language, and
#Popularity denote the number of the exchanges that are supported by this software,
programming language used, and the popularity of the software (number of the stars in
Github).
Name Features #Exchange Language Open-Source URL #Popularity
Capfolio Professional analysis platform, 5 Not mentioned No Official website [51]
Advanced backtesting engine
3 Commas Simultaneous take profit and 12 Not mentioned No Official website [1]
stop loss orders
CCXT An out of the box unified API, 10 JavaScript / Python / PHP Yes GitHub [61] 13k
optional normalized data
BlackBird Strategy is market-neutral 8 C++ Yes GitHub [36] 4.7k
strategy not transfer funds between exchanges
StockSharp Free C# library, 48 C# Yes GitHub [227] 2.6k
free trading charting application
Freqtrade Strategy Optimization by machine learning, 2 Python Yes GitHub [108] 2.4k
Calculate edge position sizing
CryptoSignal Technical analysis trading system 4 Python Yes GitHub [86] 1.9k
Ctubio Low latency 1 C++ Yes GitHub [87] 1.7k
Catalyst Analysis and visualization of system 4 Python Yes GitHub [57]
1.7k
seamless transition between live
and back-testing
GoLang Sandbox environment simulation 7 Go Yes GitHub [117] 277
market, it will be offset by a low volatility position. Ex- 495.44% when arbitraging to buy in Cryptopia exchange
tended Turtle Trading system is improved with smaller time and sell in Binance exchange. Another three well-traded ar-
interval spans and introduces a new rule by using exponen- bitrage signals (profit expectation around 20% mentioned by
tial moving average (EMA). Three EMA values are used to the author) were found on 25 May 2018. Arbitrage Trading
trigger the “buy” signal: 30EMA (Fast), 60EMA (Slow), Software System introduced in that paper presented general
100EMA (Long). The author of [144] performed backtest- principles and implementation of arbitrage trading system
ing and comparing both trading systems (Original Turtle and in the cryptocurrency market.
Extended Turtle) on 8 prominent cryptocurrencies. Through
the experiment, Original Turtle Trading System achieved an 5.5. Comparison of three cryptocurrency trading
18.59% average net profit margin (percentage of net profit systems
over total revenue) and 35.94% average profitability (per- Real-time trading systems use real-time functions to col-
centage of winning trades over total numbers of trades) in lect data and generate trading algorithms. Turtle trading sys-
87 trades through nearly one year. Extended Turtle Trad- tem and arbitrage trading system have shown a sharp con-
ing System achieved 114.41% average net profit margin and trast in their profit and risk behaviour. Using Turtle trading
52.75% average profitability in 41 trades through the same system in cryptocurrency markets got high returns with high
time interval. This research showed how Extended Turtle risk. Arbitrage trading system is inferior in terms of revenue
Trading System compared can improve over Original Turtle but also has a lower risk. One feature that turtle trading sys-
Trading System in trading cryptocurrencies. tem and arbitrage trading system have in common is they
performed well in capturing alpha.
5.4. Arbitrage Trading Systems for
Cryptocurrencies
Christian [205] introduced arbitrage trading systems for
6. Systematic Trading
cryptocurrencies. Arbitrage trading aims to spot the differ- 6.1. Technical Analysis
ences in price that can occur when there are discrepancies in Many researchers have focused on technical indicators
the levels of supply and demand across multiple exchanges. (patterns) analysis for trading on cryptocurrency markets.
As a result, a trader could realise a quick and low-risk profit Examples of studies with this approach include “Turtle Soup
by buying from one exchange and selling at a higher price pattern strategy” [233], “Nem (XEM) strategy” [236], “Amaz-
on a different exchange. Arbitrage trading signals are caught ing Gann Box strategy” [234], “Busted Double Top Pat-
by automated trading software. The technical differences tern strategy” [235], and “Bottom Rotation Trading strat-
between data sources impose a server process to be organ- egy” [237]. Table 6 shows the comparison among these
ised for each data source. Relational databases and SQL five classical technical trading strategies using technical in-
are reliable solution due to the large amounts of relational dicators. “Turtle soup pattern strategy” [233] used a 2-day
data. The author used the system to catch arbitrage oppor- breakout of price in predicting price trends of cryptocurren-
tunities on 25 May 2018 among 787 cryptocurrencies on 7 cies. This strategy is a kind of chart trading pattern. “Nem
different exchanges. The research paper [205] listed the best (XEM) strategy” combined Rate of Change (ROC) indica-
ten trading signals made by this system from 186 available tor and Relative Strength Index (RSI) in predicting price
found signals. The results showed that the system caught trends [236]. “Amazing Gann Box” predicted exact points
the trading signal of “BTG-BTC” to get a profit of up to of increase and decrease in Gann Box which are used to

Table 6
Comparison among five classical technical trading strategies
Technical trading strategy Core Methods Tecchnical tools/patterns
Turtle Soup pattern [233] 2-daybreakout of price Chart trading patterns
Rate of Change indictor (ROC)
Nem (XEM) [236] Price trends combined ROC & RSI
Relative strength index (RSI)
Predict exact points of rises and falls Candlestick, boxcharts with
Amazing Gann Box [234]
in Gann Box (catch explosive trends) Fibonacci Retracement
Bearish reversal trading pattern that
Busted Double Top Pattern [235] Price chart pattern
generates a sell signal
Pick the bottom before the reversal
Bottom Rotation Trading [237] Price chart pattern, box chart
happens
catch explosive trends of cryptocurrency price [234]. Tech- data. Overall, the model was able to achieve a 3% monthly
nical analysis tools such as candlestick and box charts with profit in Miroslav’s experiments [105]. Broek [47] ap-
Fibonacci Retracement based on golden ratio are used in this plied pairs trading based on cointegration in cryptocurrency
technical analysis. Fibonacci Retracement uses horizontal trading and 31 pairs were found to be significantly cointe-
lines to indicate where possible support and resistance lev- grated (within sector and cross-sector). By selecting four
els are in the market. “Busted Double Top Pattern” used a pairs and testing over a 60-day trading period, the pairs trad-
Bearish reversal trading pattern which generates a sell sig- ing strategy got its profitability from arbitrage opportunities,
nal to predict price trends [235]. “Bottom Rotation Trading” which rejected the Efficient-market hypothesis (EMH) for
is a technical analysis method that picks the bottom before the cryptocurrency market. Lintihac et al [174] proposed an
the reversal happens. This strategy used a price chart pattern optimal dynamic pair trading strategy model for a portfolio
and box chart as technical analysis tools. of assets. The experiment used stochastic control techniques
Sungjoo et al. [122] investigated using genetic program- to calculate optimal portfolio weights and correlated the re-
ming (GP) to find attractive technical patterns in the cryp- sults with several other strategies commonly used by practi-
tocurrency market. Over 12 technical indicators including tioners including static dual-threshold strategies. Thomas et
Moving Average (MA) and Stochastic oscillator were used al. [171] proposed a pairwise trading model incorporating
in experiments; adjusted gain, match count, relative mar- time-varying volatility with constant elasticity of variance
ket pressure and diversity measures have been used to quan- type. The experiment calculated the best pair strategy by
tify the attractiveness of technical patterns. With extended using a finite difference method and estimated parameters
experiments, the GP system is shown to find successfully by generalised moment method.
attractive technical patterns, which are useful for portfolio
optimization. Hudson et al. [130] applied almost 15, 000 6.3. Others
to technical trading rules (classified into MA rules, filter Other systematic trading methods in cryptocurrency trad-
rules, support resistance rules, oscillator rules and channel ing mainly include informed trading. Using USD / BTC ex-
breakout rules). This comprehensive study found that tech- change rate trading data, Feng et al. [104] found evidence
nical trading rules provide investors with significant pre- of informed trading in the Bitcoin market in those quantiles
dictive power and profitability. Corbet et al. [82] analysed of the order sizes of buyer-initiated (seller-initiated) orders
various technical trading rules in the form of the moving are abnormally high before large positive (negative) events,
average-oscillator and trading range break-out strategies to compared to the quantiles of seller-initiated (buyer-initiated)
generate higher returns in cryptocurrency markets. By using orders; this study adopts a new indicator inspired by the vol-
one-minute dollar-denominated Bitcoin close-price data, the ume imbalance indicator [93]. The evidence of informed
backtest showed variable-length moving average (VMA) rule trading in the Bitcoin market suggests that investors profit
performs best considering it generates the most useful sig- on their private information when they get information be-
nals in high frequency trading. fore it is widely available.
6.2. Pairs Trading

Pairs trading is a trading strategy that attempts to ex-
7. Emergent Trading Technologies
ploit the mean-reversion between the prices of certain secu- 7.1. Econometrics on cryptocurrency
rities. Miroslav [105] investigated the applicability of stan- Copula-quantile causality analysis and Granger-causality
dard pairs trading approaches on cryptocurrency data with analysis are methods to investigate causality in cryptocur-
the benchmarks of Gatev et al. [115]. The pairs trading strat- rency trading analysis. Bouri et al. [41] applied a copula-
egy is constructed in two steps. Firstly, suitable pairs with quantile causality approach on volatility in the cryptocur-
a stable long-run relationship are identified. Secondly, the rency market. The approach of the experiment extended the
long-run equilibrium is calculated and pairs trading strategy Copula-Granger-causality in distribution (CGCD) method
is defined by the spread based on the values. The research of Lee and Yang [170] in 2014. The experiment constructed
also extended intra-day pairs trading using high frequency two tests of CGCD using copula functions. The paramet-

ric test employed six parametric copula functions to dis- ence of jumps as well as structural breaks except the Dash
cover dependency density between variables. The perfor- market. Four GARCH-type models (i.e., GARCH, APARCH,
mance matrix of these functions varies with independent IGARCH and FIGARCH) and three return types with struc-
copula density. Three distribution regions are the focus of tural breaks (original returns, jump-filtered returns, and jump-
this research: left tail (1%, 5%, 10% quantile), central re- filtered returns) are considered. The research indicated the
gion (40%, 60% quantile and median) and right tail (90%, importance of jumps in cryptocurrency volatility and struc-
95%, 99% quantile). The study provided significant evi- tural breakthroughs.
dence of Granger causality from trading volume to the re- Some researchers focused on long memory methods for
turns of seven large cryptocurrencies on both left and right volatility in cryptocurrency markets. Long memory meth-
tails. Elie et al. [42] examined the causal linkages among ods focused on long-range dependence and significant long-
the volatility of leading cryptocurrencies via the frequency- term correlations among fluctuations on markets. Chaim et
domain test of Bodart and Candelon [38] and distinguished al. [63] estimated a multivariate stochastic volatility model
between temporary and permanent causation. The results with discontinuous jumps in cryptocurrency markets. The
showed that permanent shocks are more important in ex- results showed that permanent volatility appears to be driven
plaining Granger causality whereas transient shocks dom- by major market developments and popular interest levels.
inate the causality of smaller cryptocurrencies in the long Caporale et al. [52] examined persistence in the cryptocur-
term. Badenhorst [13] attempted to reveal whether spot rency market by Rescaled range (R/S) analysis and frac-
and derivative market volumes affect Bitcoin price volatility tional integration. The results of the study indicated that
with the Granger-causality method and ARCH (1,1). The the market is persistent (there is a positive correlation be-
result shows spot trading volumes have a significant posi- tween its past and future values) and that its level changes
tive effect on price volatility while the relationship between over time. Khuntin et al. [154] applied the adaptive market
cryptocurrency volatility and the derivative market is uncer- hypothesis (AMH) in the predictability of Bitcoin evolving
tain. Elie et al. [45] used a dynamic equicorrelation (DECO) returns. The consistent test of Dominguez and Lobato [89],
model and reported evidence that the average earnings equi- generalized spectral (GS) of Escanciano and Velasco [98]
librium correlation changes over time between the 12 lead- are applied in capturing time-varying linear and nonlinear
ing cryptocurrencies. The results showed increased cryp- dependence in bitcoin returns. The results verified Evolving
tocurrency market consolidation despite significant price de- Efficiency in Bitcoin price changes and evidence of dynamic
clined in 2018. Furthermore, measurement of trading vol- efficiency in line with AMH’s claims.
ume and uncertainty are key determinants of integration. Katsiampa et al. [150] applied three pair-wise bivariate
Several econometrics methods in time-series research, BEKK models to examine the conditional volatility dynam-
such as GARCH and BEKK, have been used in the liter- ics along with interlinkages and conditional correlations be-
ature on cryptocurrency trading. Conrad et al. [81] used tween three pairs of cryptocurrencies in 2018. More specifi-
the GARCH-MIDAS model to extract long and short-term cally, the BEKK-MGARCH methodology also captured cross-
volatility components of the Bitcoin market. The technical market effects of shocks and volatility, which are also known
details of this model decomposed the conditional variance as shock transmission effects and volatility spillover effects.
into the low-frequency and high-frequency components. The The experiment found evidence of bi-directional shock trans-
results identified that S&P 500 realized volatility has a nega- mission effects between Bitcoin and both Ether and Lit-
tive and highly significant effect on long-term Bitcoin volatil- coin. In particular, bi-directional shock spillover effects are
ity and S&P 500 volatility risk premium has a significantly identified between three pairs (Bitcoin, Ether and Litcoin)
positive effect on long-term Bitcoin volatility. Ardia et al. [8] and time-varying conditional correlations exist with positive
used the Markov Switching GARCH (MSGARCH) model correlations mostly prevailing. In 2019, Katsiampa [149]
to test the existence of institutional changes in the GARCH further researched an asymmetric diagonal BEKK model
volatility dynamics of Bitcoin’s logarithmic returns. More- to examine conditional variances of five cryptocurrencies
over, a Bayesian method was used for estimating model pa- that are significantly affected by both previous squared er-
rameters and calculating VaR prediction. The results showed rors and past conditional volatility. The experiment tested
that MSGARCH models clearly outperform single-regime the null hypothesis of the unit root against the stationar-
GARCH for Value-at-Risk forecasting. Troster et al. [239] ity hypothesis. Once stationarity is ensured, ARCH LM
performed general GARCH and GAS (Generalized Auto- is tested for ARCH effects to examine the requirement of
regressive Score) analysis to model and predict Bitcoin’s re- volatility modeling in return series. Moreover, volatility
turns and risks. The experiment found that the GAS model co-movements among cryptocurrency pairs are also tested
with heavy-tailed distribution can provide the best out-of- by the multivariate GARCH model. The results confirmed
sample prediction and goodness-of-fit attributes for Bitcoin’s the non-normality and heteroskedasticity of price returns in
return and risk modeling. The results also illustrated the cryptocurrency markets. The finding also identified the ef-
importance of modeling excess kurtosis for Bitcoin returns. fects of cryptocurrencies’ volatility dynamics due to major
Charles et al. [65] studied four cryptocurrency markets in- news. Hultman [131] set out to examine GARCH (1,1),
cluding Bitcoin, Dash, Litecoin and Ripple. Results showed bivariate-BEKK (1,1) and a standard stochastic model to
cryptocurrency returns are strongly characterised by the pres- forecast the volatility of Bitcoin. A rolling window approach

is used in these experiments. Mean absolute error (MAE), ing method. The algorithm operates by constructing a large
Mean squared error (MSE) and Root-mean-square deviation number of decision trees during training and outputting the
(RMSE) are three loss criteria adopted to evaluate the de- average consensus as predicted class in the case of classi-
gree of error between predicted and true values. The re- fication or mean prediction value in the case of regression
sult shows the following rank of loss functions: GARCH [173]. GB produces a prediction model in the form of an
(1,1) > bivariate-BEKK (1,1) > Standard stochastic for all ensemble of weak prediction models [111].
the three different loss criteria; in other words, GARCH(1,1) Clustering Algorithms. Clustering is a machine learn-
appeared best in predicting the volatility of Bitcoin. Wavelet ing technique that involves grouping data points in a way
time-scale persistence analysis is also applied in the predic- that each group shows some regularity [137]. K-Means is a
tion and research of volatility in cryptocurrency markets [202]. vector quantization used for clustering analysis in data min-
The results showed that information efficiency (efficiency) ing. K-means stores the 𝑘-centroids used to define the clus-
and volatility persistence in the cryptocurrency market are ters; a point is considered to be in a particular cluster if it is
highly sensitive to time scales, measures of returns and volatil- closer to the cluster’s centroid than any other centroid [245].
ity, and institutional changes. Adjepong et al. [202] con- K-Means is one of the most used clustering algorithms used
nected with similar research by Corbet et al. [85] and showed in cryptocurrency trading according to the papers we col-
that GARCH is quicker than BEKK to absorb new informa- lected.
tion regarding the data. Regression Algorithms. We have defined regression
as any statistical technique that aims at estimating a con-
7.2. Machine Learning Technology tinuous value [164]. Linear Regression (LR) and Scatter-
As we have previously stated, Machine learning technol- plot Smoothing are common techniques used in solving re-
ogy constructs computer algorithms that automatically im- gression problems in cryptocurrency trading. LR is a lin-
prove themselves by finding patterns in existing data with- ear method used to model the relationship between a scalar
out explicit instructions [128]. The rapid development of response (or dependent variable) and one or more explana-
machine learning in recent years has promoted its applica- tory variables (or independent variables) [164]. Scatterplot
tion to cryptocurrency trading, especially in the prediction Smoothing is a technology to fit functions through scatter
of cryptocurrency returns. plots to best represent relationships between variables [110].
Deep Learning Algorithms. Deep learning is a modern
7.2.1. Common Machine Learning Technology in this take on artificial neural networks (ANNs) [257], made pos-
survey sible by the advances in computational power. An ANN is
Several machine learning technologies are applied in cryp- a computational system inspired by the natural neural net-
tocurrency trading. We distinguish these by the objective set works that make up the animal’s brain. The system “learns”
to the algorithm: classification, clustering, regression, rein- to perform tasks including the prediction by considering ex-
forcement learning. We have separated a section specifically amples. Deep learning’s superior accuracy comes from high
on deep learning due to its intrinsic variation of techniques computational complexity cost. Deep learning algorithms
and wide adoption. are currently the basis for many modern artificial intelli-
Classification Algorithms. Classification in machine gence applications [231]. Convolutional neural networks
learning has the objective of categorising incoming objects (CNNs) [168], Recurrent neural networks (RNNs) [188],
into different categories as needed, where we can assign Gated recurrent units (GRUs) [70], Multilayer perceptron
labels to each category (e.g., up and down). Naive Bayes (MLP) and Long short-term memory (LSTM) [67] networks
(NB) [216], Support Vector Machine (SVM) [247], K-Nearest are the most common deep learning technologies used in
Neighbours (KNN) [247], Decision Tree (DT) [109], Ran- cryptocurrency trading. A CNN is a specific type of neu-
dom Forest (RF) [173] and Gradient Boosting (GB) [111] ral network layer commonly used for supervised learning.
algorithms habe been used in cryptocurrency trading based CNNs have found their best success in image processing
on papers we collected. NB is a probabilistic classifier based and natural language processing problems. An attempt to
on Bayes’ theorem with strong (naive) conditional indepen- use CNNs in cryptocurrency can be shown in [143]. An
dence assumptions between features [216]. SVM is a su- RNN is a type of artificial neural network in which con-
pervised learning model that aims at achieving high mar- nections between nodes form a directed graph with possi-
gin classifiers connecting to learning bounds theory [256]. ble loops. This structure of RNNs makes them suitable for
SVMs assign new examples to one category or another, mak- processing time-series data [188] due to the introduction of
ing it a non-probabilistic binary linear classifier [247], al- memory in the recurrent connections. They face neverthe-
though some corrections can make a probabilistic interpre- less for the vanishing gradients problem [203] and so dif-
tation of their output [153]. KNN is a memory-based or ferent variations have been recently proposed. LSTM [67]
lazy learning algorithm, where the function is only approx- is a particular RNN architecture widely used. LSTMs have
imated locally, and all calculations are being postponed to shown to be superior to nongated RNNs on financial time-
inference time [247]. DT is a decision support tool algo- series problems because they have the ability to selectively
rithm that uses a tree-like decision graph or model to seg- remember patterns for a long time. A GRU [70] is another
ment input patterns into regions to then assign an associ- gated version of the standard RNN which has been used in
ated label to each region [109]. RF is an ensemble learn-

crypto trading [91]. Another deep learning technology used phas" [141].) Vo et al. [243] applied RFs in High-Frequency
in cryptocurrency trading is Seq2seq, which is a specific cryptocurrency Trading (HFT) and compared it with deep
implementation of the Encoder–Decoder architecture [251]. learning models. Minute-level data is collected when util-
Seq2seq was first aimed at solving natural language process- ising a forward fill imputation method to replace the NULL
ing problems but has been also applied it in cryptocurrency value (i.e., a missing value). Different periods and RF trees
trend predictions in [226]. are tested in the experiments. The authors also compared F-
Reinforcement Learning Algorithms. Reinforcement 1 precision and recall metrics between RF and Deep Learn-
learning (RL) is an area of machine learning leveraging the ing (DL). The results showed that RF is effective despite
idea that software agents act in the environment to maximize multicollinearity occurring in ML features, the lack of model
a cumulative reward [230]. Deep Q-Learning (DQN) [120] identification also potentially leading to model identifica-
and Deep Boltzmann Machine (DBM) [219] are common tion issues; this research also attempted to create an HFT
technologies used in cryptocurrency trading using RL. Deep strategy for Bitcoin using RF. Maryna et al. [260] inves-
Q learning uses neural networks to approximate Q-value tigated the profitability of an algorithmic trading strategy
functions. A state is given as input, and Q values for all pos- based on training an SVM model to identify cryptocurren-
sible actions are generated as outputs [120]. DBM is a type cies with high or low predicted returns. The results showed
of binary paired Markov random field (undirected probabil- that the performance of the SVM strategy was the fourth be-
ity graphical model) with multiple layers of hidden random ing better only than S&P B&H strategy, which simply buys-
variables [219]. It is a network of randomly coupled random and-hold the S&P index. (There are other 4 benchmark
binary units. strategies in this research.)The authors observed that SVM
needs a large number of parameters and so is very prone
7.2.2. Research on Machine Learning Models to overfitting, which caused its bad performance. Barnwal
In the development of machine learning trading signals, et al. [18] used generative and discriminative classifiers to
technical indicators have usually been used as input fea- create a stacking model, particularly 3 generative and 6 dis-
tures. Nakano et al. [193] explored Bitcoin intraday tech- criminative classifiers combined by a one-layer Neural Net-
nical trading based on ANNs for return prediction. The ex- work, to predict the direction of cryptocurrency price. A
periment obtained medium frequency price and volume data discriminative classifier directly models the relationship be-
(time interval of data is 15min) of Bitcoin from a cryptocur- tween unknown and known data, while generative classifiers
rency exchange. An ANN predicts the price trends (up and model the prediction indirectly through the data generation
down) in the next period from the input data. Data is pre- distribution [198]. Technical indicators including trend, mo-
processed to construct a training dataset that contains a ma- mentum, volume and volatility, are collected as features of
trix of technical patterns including EMA, Emerging Markets the model. The authors discussed how different classifiers
Small Cap (EMSD), relative strength index (RSI), etc. Their and features affect the prediction. Attanasio et al. [10] com-
numerical experiments contain different research aspects in- pared a variety of classification algorithms including SVM,
cluding base ANN research, effects of different layers, ef- NB and RF in predicting next-day price trends of a given
fects of different activation functions, different outputs, dif- cryptocurrency. The results showed that due to the hetero-
ferent inputs and effects of additional technical indicators. geneity and volatility of cryptocurrencies’ financial instru-
The results have shown that the use of various technical in- ments, forecasting models based on a series of forecasts ap-
dicators possibly prevents over-fitting in the classification peared better than a single classification technology in trad-
of non-stationary financial time-series data, which enhances ing cryptocurrencies. Madan et al. [179] modeled the Bit-
trading performance compared to the primitive technical trad- coin price prediction problem as a binomial classification
ing strategy. (Buy-and-Hold is the benchmark strategy in task, experimenting with a custom algorithm that leverages
this experiment.) both random forests and generalized linear models. Daily
Some classification and regression machine learning mod- data, 10-minute data and 10-second data are used in the
els are applied in cryptocurrency trading by predicting price experiments. The experiments showed that 10-minute data
trends. Most researchers have focused on the comparison gave a better sensitivity and specificity ratio than 10-second
of different classification and regression machine learning data (10-second prediction achieved around 10% accuracy).
methods. Sun et al. [229] used random forests (RFs) with Considering predictive trading, 10-minute data helped show
factors in Alpha01 [141] (capturing features from the his- clearer trends in the experiment compared to 10-second back-
tory of the cryptocurrency market) to build a prediction model. testing. Similarly, Virk [242] compared RF, SVM, GB and
The experiment collected data from API in cryptocurrency LR to predict the price of Bitcoin. The results showed that
exchanges and selected 5-minute frequency data for back- SVM achieved the highest accuracy of 62.31% and preci-
testing. The results showed that the performances are pro- sion value 0.77 among binomial classification machine learn-
portional to the amount of data (more data, more accurate) ing algorithms.
and the factors used in the RF model appear to have different Different deep learning models have been used in find-
importance. For example, “Alpha024” and “Alpha032” fea- ing patterns of price movements in cryptocurrency markets.
tures appeared as the most important in the model adopted. Zhengy et al. [258] implemented two machine learning mod-
(The alpha features come from paper “101 Formulaic Al- els, fully-connected ANN and LSTM to predict cryptocur-

rency price dynamics. The results showed that ANN, in sult showed that the seq2seq model exhibited demonstra-
general, outperforms LSTM although theoretically, LSTM ble improvement over the ARIMA model for Bitcoin-USD
is more suitable than ANN in terms of modeling time series prediction but the seq2seq model showed very poor perfor-
dynamics; the performance measures considered are MAE mance in extreme cases. The authors proposed performing
and RMSE in joint prediction (five cryptocurrencies daily additional investigations, such as the use of LSTM instead
prices prediction). The findings show that the future state of of GRU units to improve the performance. Similar models
a time series for cryptocurrencies is highly dependent on its were also compared by Stuerner et al. [228] who explored
historic evolution. Kwon et al. [165] used an LSTM model, the superiority of automated investment approach in trend
with a three-dimensional price tensor representing the past following and technical analysis in cryptocurrency trading.
price changes of cryptocurrencies as input. This model out- Samuel et al. [206] explored the vector autoregressive model
performs the GB model in terms of F1-score. Specifically, (VAR model), a more complex RNN, and a hybrid of the two
it has a performance improvement of about 7% over the GB in residual recurrent neural networks (R2N2) in predicting
model in 10-minute price prediction. In particular, the ex- cryptocurrency returns. The RNN with ten hidden layers is
periments showed that LSTM is more suitable when classi- optimised for the setting and the neural network augmented
fying cryptocurrency data with high volatility. Alessandretti by VAR allows the network to be shallower, quicker and
et al. [5] tested Gradient boosting decision trees (includ- to have a better prediction than an RNN. RNN, VAR and
ing single regression and XGBoost-augmented regression) R2N2 models are compared. The results showed that the
and the LSTM model on forecasting daily cryptocurrency VAR model has phenomenal test period performance and
prices. They found methods based on gradient boosting de- thus props up the R2N2 model, while the RNN performs
cision trees worked best when predictions were based on poorly. This research is an attempt at optimisation of model
short-term windows of 5/10 days while LSTM worked best design and applying to the prediction on cryptocurrency re-
when predictions were based on 50 days of data. The rel- turns.
ative importance of the features in both models are com-
pared and an optimised portfolio composition (based on ge- 7.2.3. Sentiment Analysis
ometric mean return and Sharpe ratio) is discussed in this Sentiment analysis, a popular research topic in the age of
paper. Phaladisailoed et al. [207] chose regression mod- social media, has also been adopted to improve predictions
els (Theil-Sen Regression and Huber Regression) and deep for cryptocurrency trading. This data source typically has to
learning-based models (LSTM and GRU) to compare the be combined with Machine Learning for the generation of
performance of predicting the rise and fall of Bitcoin price. trading signals.
In terms of two common measure metrics, MSE and R- Lamon et al. [167] used daily news and social media
Square (R2 ), GRU shows the best accuracy. Fan et al. [100] data labeled on actual price changes, rather than on positive
applied an autoencoder-augmented LSTM structure in pre- and negative sentiment. By this approach, the prediction
dicting the mid-price of 8 cryptocurrency pairs. Level-2 on price is replaced with positive and negative sentiment.
limit order book live data is collected and the experiment The experiment acquired cryptocurrency-related news arti-
achieved 78% accuracy of price movements prediction in cle headlines from the website like “cryptocoinsnews” and
high frequency trading (tick level). This research improved twitter API. Weights are taken in positive and negative words
and verified the view of Sirignano et al. [224] that univer- in the cryptocurrency market. Authors compared Logistic
sal models have better performance than currency-pair spe- Regression (LR), Linear Support Vector Machine (LSVM)
cific models for cryptocurrency markets. Moreover, “Walk- and NB as classifiers and concluded that LR is the best
through” (i.e., retrain the original deep learning model it- classifier in daily price prediction with 43.9% of price in-
self when it appears to no longer be valid) is proposed as creases correctly predicted and 61.9% of price decreases
a method to optimise the training of a deep learning model correctly forecasted. Smuts [225] conducted a similar bi-
and shown to significantly improve the prediction accuracy. nary sentiment-based price prediction method with an LSTM
Researchers have also focused on comparing classical model using Google Trends and Telegram sentiment. In
statistical models and machine/deep learning models. Rane detail, the sentiment was extracted from Telegram by us-
et al. [214] described classical time series prediction meth- ing a novel measure called VADER [132]. The backtest-
ods and machine learning algorithms used for predicting ing reached 76% accuracy on the test set during the first
Bitcoin price. Statistical models such as Autoregressive In- half of 2018 in predicting hourly prices. Nasir et al. [195]
tegrated Moving Average models (ARIMA), Binomial Gen- researched the relationship between cryptocurrency returns
eralized Linear Model and GARCH are compared with ma- and search engines. The experiment employed a rich set
chine learning models such as SVM, LSTM and Non-linear of established empirical approaches including VAR frame-
Auto-Regressive with Exogenous Input Model (NARX). The work, copulas approach and non-parametric drawings of time
observation and results showed that the NARX model is the series. The results found that Google searches exert signif-
best model with nearly 52% predicting accuracy based on icant influence on Bitcoin returns, especially in the short-
10 seconds interval. Rebane et al. [215] compared tradi- term intervals. Kristoufek [162] discussed positive and neg-
tional models like ARIMA with a modern popular model ative feedback on Google trends or daily views on Wikipedia.
like seq2seq in predicting cryptocurrency returns. The re- The author mentioned different methods including Cointe-

gration, Vector autoregression and Vector error-correction plied supervised machine learning algorithms such as logis-
model to find causal relationships between prices and searched tic regression, Naive Bayes and support vector machines,
terms in the cryptocurrency market. The results indicated etc. on Twitter Sentiment Analysis for cryptocurrency trad-
that search trends and cryptocurrency prices are connected. ing. Garcia et al. [113] applied multidimensional analy-
There is also a clear asymmetry between the effects of in- sis and impulse analysis in social signals of sentiment ef-
creased interest in currencies above or below their trend val- fects and algorithmic trading of Bitcoin. The results veri-
ues from the experiment. Young et al. [156] analysed user fied the long-standing assumption that transaction-based so-
comments and replies in online communities and their con- cial media sentiment has the potential to generate a posi-
nection with cryptocurrency volatility. After crawling com- tive return on investment. Zamuda et al. [254] adopted new
ments and replies in online communities, authors tagged the sentiment analysis indicators and used multi-target portfo-
extent of positive and negative topics. Then the relation- lio selection to avoid risks in cryptocurrency trading. The
ship between price and the number of transactions of cryp- perspective is rationalized based on the elastic demand for
tocurrency is tested according to comments and replies to computing resources of the cloud infrastructure. A gen-
selected data. At last, a prediction model using machine eral model evaluating the influence between user’s network
learning based on selected data is created to predict fluctu- Action-Reaction-Influence-Model (ARIM) is mentioned in
ations in the cryptocurrency market. The results show the this research. Bartolucci et al. [19] researched cryptocur-
amount of accumulated data and animated community ac- rency prices with the “Butterfly effect”, which means “is-
tivities exerted a direct effect on fluctuation in the price and sues” of the open-source project provides insights to im-
volume of a cryptocurrency. prove prediction of cryptocurrency prices. Sentiment, po-
Phillips et al. [212] applied dynamic topic modeling and liteness, emotions analysis of GitHub comments are applied
Hawkes model to decipher relationships between topics and in Ethereum and Bitcoin markets. The results showed that
cryptocurrency price movements. The authors used Latent these metrics have predictive power on cryptocurrency prices.
Dirichlet allocation (LDA) model for topic modeling, which
assumes each document contains multiple topics to different 7.2.4. Reinforcement Learning
extents. The experiment showed that particular topics tend Deep reinforcement algorithms bypass prediction and
to precede certain types of price movements in the cryp- go straight to market management actions to achieve high
tocurrency market and the authors proposed the relation- cumulated profit [126]. Bu et al. [49] proposed a combina-
ships could be built into real-time cryptocurrency trading. tion of double Q-network and unsupervised pre-training us-
Li et al. [172] analysed Twitter sentiment and trading vol- ing DBM to generate and enhance the optimal Q-function in
ume and an Extreme Gradient Boosting Regression Tree cryptocurrency trading. The trading model contains agents
Model in the prediction of ZClassic (ZCL) cryptocurrency in series in the form of two neural networks, unsupervised
market. Sentiment analysis using natural language process- learning modules and environments. The input market state
ing from the Python package “Textblob” assigns impactful connects an encoding network which includes spectral fea-
words a polarity value. Values of weighted and unweighted ture extraction (convolution-pooling module) and temporal
sentiment indices are calculated on an hourly basis by sum- feature extraction (LSTM module). A double-Q network
ming weights of coinciding tweets, which makes us com- follows the encoding network and actions are generated from
pare this index to ZCL price data. The model achieved this network. Compared to existing deep learning models
a Pearson correlation of 0.806 when applied to test data, (LSTM, CNN, MLP, etc.), this model achieved the high-
yielding a statistical significance at the 𝑝 < 0.0001 level. est profit even facing an extreme market situation (recorded
Flori [107] relied on a Bayesian framework that combines 24% of the profit while cryptocurrency market price drops
market-neutral information with subjective beliefs to con- by -64%). Juchli [138] applied two implementations of rein-
struct diversified investment strategies in the Bitcoin mar- forcement learning agents, a Q-Learning agent, which serves
ket. The result shows that news and media attention seem to as the learner when no market variables are provided, and
contribute to influence the demand for Bitcoin and enlarge a DQN agent which was developed to handle the features
the perimeter of the potential investors, probably stimulating previously mentioned. The DQN agent was backtested un-
price euphoria and upwards-downwards market dynamics. der the application of two different neural network architec-
The authors’ research highlighted the importance of news tures. The results showed that the DQN-CNN agent (convo-
in guiding portfolio re-balancing. Elie et al. [39] compared lutional neural network) is superior to the DQN-MLP agent
the ability of newspaper-based metrics and internet search- (multilayer perceptron) in backtesting prediction. Lucarelli
based uncertainty metrics in predicting bitcoin returns. The et al. [177] focused on improving automated cryptocurrency
predictive power of Internet-based economic uncertainty- trading with a deep reinforcement learning approach. Dou-
related query indices is statistically stronger than that of ble and Dueling double deep Q-learning networks are com-
newspapers in predicting bitcoin returns. pared for 4 years. By setting rewards functions as Sharpe
Similarly, Colianni et al. [80], Garcia et al. [113], Za- ratio and profit, the double Q-learning method demonstrated
muda et al. [254] et al. used sentiment analysis technol- to be the most profitable approach in trading cryptocurrency.
ogy applying it in the cryptocurrency trading area and had
similar results. Colianni et al. [80] cleaned data and ap-

7.3. Others 8. Portfolio and Cryptocurrency Assets

Atsalakis et al. [9] proposes a computational intelligence
technique that uses a hybrid Neuro-Fuzzy controller, namely 8.1. Research among cryptocurrency pairs and
PATSOS, to forecast the direction in the change of the daily related factors
price of Bitcoin. The proposed methodology outperforms Ji et al. [135] examined connectedness via return and
two other computational intelligence models, the first be- volatility spillovers across six large cryptocurrencies (col-
ing developed with a simpler neuro-fuzzy approach, and the lected from coinmarketcap lists from August 7 2015 to Febru-
second being developed with artificial neural networks. Ac- ary 22 2018) and found Litecoin and Bitcoin to have the
cording to the signals of the proposed model, the investment most effect on other cryptocurrencies. The authors followed
return obtained through trading simulation is 71.21% higher methods of Diebold et al. [88] and built positive/negative re-
than the investment return obtained through a simple buy turns and volatility connectedness networks. Furthermore,
and hold strategy. This application is proposed for the first the regression model is used to identify drivers of various
time in the forecasting of Bitcoin price movements. Topo- cryptocurrency integration levels. Further analysis revealed
logical data analysis is applied to forecasting price trends of that the relationship between each cryptocurrency in terms
cryptocurrency markets in [155]. The approach is to har- of return and volatility is not necessarily due to its mar-
ness topological features of attractors of dynamical systems ket size. Adjepong et al. [201] explored market coherence
for arbitrary temporal data. The results showed that the and volatility causal linkages of seven leading cryptocurren-
method can effectively separate important topological pat- cies. Wavelet-based methods are used to examine market
terns and sampling noise (like bid–ask bounces, discrete- connectedness. Parametric and nonparametric tests are em-
ness of price changes, differences in trade sizes or infor- ployed to investigate directions of volatility spillovers of the
mational content of price changes, etc.) by providing theo- assets. Experiments revealed from diversification benefits to
retical results. Kurbucz [163] designed a complex method linkages of connectedness and volatility in cryptocurrency
consisting of single-hidden layer feedforward neural net- markets. Elie et al. [43] found the presence of jumps was
works (SLFNs) to (i) determine the predictive power of the detected in a series of 12 cryptocurrency returns, and signif-
most frequent edges of the transaction network (a public icant jumping activity was found in all cases. More results
ledger that records all Bitcoin transactions) on the future underscore the importance of the jump in trading volume for
price of Bitcoin; and, (ii) to provide an efficient technique the formation of cryptocurrency leapfrogging.
for applying this untapped dataset in day trading. The re- Some researchers explored the relationship between cryp-
search found a significantly high accuracy (60.05%) for the tocurrency and different factors, including futures, gold, etc.
price movement classifications base on information that can Hale et al. [123] suggested that Bitcoin prices rise and fall
be obtained using a small subset of edges (approximately rapidly after CME issues futures consistent with pricing dy-
0.45% of all unique edges). It is worth noting that, Kondor namics. Specifically, the authors pointed out that the rapid
et al. [157, 159] firstly published some papers giving analy- rise and subsequent decline in prices after the introduction
sis on transaction networks on cryptocurrency markets and of futures is consistent with trading behaviour in the cryp-
applied related research in identifying Bitcoin users [139]. tocurrency market. Werner et al. [161] focused on the asym-
Abay et al. [2] attempted to understand the network dynam- metric interrelationships between major currencies and cryp-
ics behind the Blockchain graphs using topological features. tocurrencies. The results of multiple fractal asymmetric de-
The results showed that standard graph features such as the trending cross-correlation analysis show evidence of signif-
degree distribution of transaction graphs may not be suffi- icant persistence and asymmetric multiplicity in the cross-
cient to capture network dynamics and their potential im- correlation between most cryptocurrency pairs and ETF pairs.
pact on Bitcoin price fluctuations. Maurice et al [202] ap- Bai et al. [14] studied a trading algorithm for foreign ex-
plied wavelet time-scale persistence in analysing returns and change on a cryptocurrency Market using the Automated
volatility in cryptocurrency markets. The experiment exam- Triangular Arbitrage method. Implementing a pricing strat-
ined the long-memory and market efficiency characteristics egy, implementing trading algorithms and developing a given
in cryptocurrency markets using daily data for more than trading simulation are three problems solved by this research.
two years. The authors employed a log-periodogram re- Kang et al. [146] examined the hedging and diversification
gression method in researching stationarity in the cryptocur- properties of gold futures versus Bitcoin prices by using
rency market and used ARFIMA-FIGARCH class of mod- dynamic conditional correlations (DCCs) and wavelet co-
els in examining long-memory behaviour of cryptocurren- herence. DCC-GARCH model [95] is used to estimate the
cies across time and scale. In general, experiments indicated time-varying correlation between Bitcoin and gold futures
that heterogeneous memory behaviour existed in eight cryp- by modeling the variance and the co-variance but also this
tocurrency markets using daily data over the full-time period two flexibility. Wavelet coherence method focused more on
and across scales (August 25, 2015 to March 13, 2018). co-movement between Bitcoin and gold futures. From ex-
periments, the wavelet coherence results indicated volatility
persistence, causality and phase difference between Bitcoin
and gold. Dyhrberg et al [92] applied the GARCH model
and the exponential GARCH model in analysing similarities

between Bitcoin, gold and the US dollar. The experiments portfolios. Castro et al. [56] developed a portfolio optimi-
showed that Bitcoin, gold and the US dollar have similari- sation model based on the Omega measure which is more
ties with the variables of the GARCH model, have similar comprehensive than the Markowitz model and applied this
hedging capabilities and react symmetrically to good and to four crypto-asset investment portfolios by means of a nu-
bad news. The authors observed that Bitcoin can combine merical application. Experiments showed crypto-assets im-
some advantages of commodities and currencies in finan- proves the return of the portfolios, but on the other hand,
cial markets to be a tool for portfolio management. Baur et also increase the risk exposure.
al. [20] extended the research of Dyhrberg et al.; the same Bedi et al. [22] examined diversification capabilities of
data and sample periods are tested [92] with GARCH and Bitcoin for a global portfolio spread across six asset classes
EGARCH-(1,1) models but the experiments reached differ- from the standpoint of investors dealing in five major fiat
ent conclusions. Baur et al. found that Bitcoin has unique currencies, namely US Dollar, Great Britain Pound, Euro,
risk-return characteristics compared with other assets. They Japanese Yen and Chinese Yuan. They employed modified
noticed that Bitcoin excess returns and volatility resemble a Conditional Value-at-Risk and standard deviation as mea-
rather highly speculative asset with respect to gold or the sures of risk to perform portfolio optimisations across three
US dollar. Bouri et al. [40] studied the relationship be- asset allocation strategies and provided insights into the sharp
tween Bitcoin and energy commodities by applying DCCs disparity in Bitcoin trading volumes across national curren-
and GARCH (1,1) models. In particular, the results showed cies from a portfolio theory perspective. Similar research
that Bitcoin is a strong hedge and safe haven for energy has been done by Antipova et al. [7], which explored the
commodities. Kakushadze [142] proposed factor models for possibility of establishing and optimizing a global portfolio
the cross-section of daily cryptoasset returns and provided by diversifying investments using one or more cryptocur-
source code for data downloads, computing risk factors and rencies, and assessing returns to investors in terms of risks
backtesting for all cryptocurrencies and a host of various and returns. Fantazzini et al. [102] proposed a set of models
other digital assets. The results showed that cross-sectional that can be used to estimate the market risk for a portfo-
statistical arbitrage trading may be possible for cryptoas- lio of crypto-currencies, and simultaneously estimate their
sets subject to efficient executions and shorting. Beneki et credit risk using the Zero Price Probability (ZPP) model.
al. [25] tested hedging abilities between Bitcoin and Ethereum The results revealed the superiority of the t-copula/skewed-t
by a multivariate BEKK-GARCH methodology and impulse GARCH model for market risk, and the ZPP-based models
response analysis within VAR model. The results indicated for credit risk. Qiang et al. [134] examined the common
a volatility transaction from Ethereum to Bitcoin, which im- dynamics of bitcoin exchanges. Using a connectivity met-
plied possible profitable trading strategies on the cryptocur- ric based on the actual daily volatility of the bitcoin price,
rency derivatives market. Guglielmo et al. [54] examined they found that Coinbase is undoubtedly the market leader,
the week effect in cryptocurrency markets and explored the while Binance performance is surprisingly weak. The re-
feasibility of this indicator in trading practice. Student 𝑡-test, sults also suggested that safer asset extraction is more im-
ANOVA, Kruskal–Wallis and Mann–Whitney tests were car- portant for volatility linkages between Bitcoin exchanges
ried out for cryptocurrency data in order to compare time relative to trading volumes.
periods that may be characterised by anomalies with other Trucios et al. [240] proposed a methodology based on
time periods. When an anomaly is detected, an algorithm vine copulas and robust volatility models to estimate the
was established to exploit profit opportunities (MetaTrader Value-at-Risk (VaR) and Expected Shortfall (ES) of cryp-
terminal in MQL4 is mentioned in this research). The re- tocurrency portfolios. The proposed algorithm displayed
sults showed evidence of anomaly (abnormal positive re- good performance in estimating both VaR and ES. Hrytsiuk
turns on Mondays) in the Bitcoin market by backtesting in et al. [129] showed that the cryptocurrency returns can be
2013-2016. described by the Cauchy distribution and obtained the an-
alytical expressions for VaR risk measures and performed
8.2. Crypto-asset Portfolio Research calculations accordingly. As a result of the optimisation,
Some researchers applied portfolio theory for crypto as- the sets of optimal cryptocurrency portfolios were built in
sets. Corbet et al. [83] gave a systematic analysis of cryp- their experiments.
tocurrencies as financial assets. Brauneis et al. [46] ap- Jiang et al. [136] proposed a two-hidden-layer CNN that
plied the Markowitz mean-variance framework in order to takes the historical price of a group of cryptocurrency assets
assess the risk-return benefits of cryptocurrency portfolios. as an input and outputs the weight of the group of cryp-
In an out-of-sample analysis accounting for transaction cost, tocurrency assets. This research focused on portfolio re-
they found that combining cryptocurrencies enriches the set search in cryptocurrency assets using emerging technolo-
of ‘low’-risk cryptocurrency investment opportunities. In gies like CNN. Training is conducted in an intensive man-
terms of the Sharpe ratio and certainty equivalent returns, ner to maximise cumulative returns, which is considered a
the 1∕𝑁-portfolio (i.e., “naive” strategies, such as equally reward function of the CNN network. The performance of
dividing amongst asset classes) outperformed single cryp- the CNN strategy is compared with the three benchmarks
tocurrencies and more than 75% in terms of the Sharpe ratio and the other three portfolio management algorithms (buy
and certainty equivalent returns of mean-variance optimal and hold strategy, Uniform Constant Rebalanced Portfolio

and Universal Portfolio with Online Newton Step and Pas- evidence that bubbles mirror the social epidemic-like spread
sive Aggressive Mean Reversion); the results are positive of an investment idea. Guglielmo et al. [53] examined the
in that the model is only second to the Passive Aggressive price overreactions in the case of cryptocurrency trading.
Mean Reversion algorithm (PAMR). Estalayo et al. [99] re- Some parametric and non-parametric tests confirmed the pres-
ported initial findings around the combination of DL mod- ence of price patterns after overreactions, which identified
els and Multi-Objective Evolutionary Algorithms (MOEAs) that the next-day price changes in both directions are bigger
for allocating cryptocurrency portfolios. Technical rationale than after “normal” days. The results also showed that the
and details were given on the design of a stacked DL recur- overreaction detected in the cryptocurrency market would
rent neural network, and how its predictive power can be ex- not give available profit opportunities (possibly due to trans-
ploited for yielding accurate ex-ante estimates of the return action costs) that cannot be considered as evidence of the
and risk of the portfolio. Results obtained for a set of exper- EMH. Chaim et al. [62] analysed the high unconditional
iments carried out with real cryptocurrency data have veri- volatility of cryptocurrency from a standard log-normal stochas-
fied the superior performance of their designed deep learn- tic volatility model to discontinuous jumps of volatility and
ing model with respect to other regression techniques. returns. The experiment indicated the importance of in-
corporating permanent jumps to volatility in cryptocurrency
markets.
9. Market Condition Research
9.1. Bubbles and Crash Analysis 9.2. Extreme condition
Phillips and Yu proposed a methodology to test for the Differently from traditional fiat currencies, cryptocur-
presence of cryptocurrency bubble [68], which is extended rencies are risky and exhibit heavier tail behaviour. Paraskevi
by Shaen et al. [84]. The method is based on supremum et al. [151] found extreme dependence between returns and
Augmented Dickey–Fuller (SADF) to test for the bubble trading volumes. Evidence of asymmetric return-volume re-
through the inclusion of a sequence of forwarding recur- lationship in the cryptocurrency market was also found by
sive right-tailed ADF unit root tests. An extended method- the experiment, as a result of discrepancies in the correlation
ology generalised SADF (GSAFD), is also tested for bub- between positive and negative return exceedances across all
bles within cryptocurrency data. The research concluded the cryptocurrencies.
that there is no clear evidence of a persistent bubble in cryp- There has been a price crash in late 2017 to early 2018 in
tocurrency markets including Bitcoin or Ethereum. Bouri cryptocurrency [253]. Yaya et al. [253] researched the per-
et al. [44] date-stamped price explosiveness in seven large sistence and dependence of Bitcoin on other popular alter-
cryptocurrencies and revealed evidence of multiple periods native coins before and after the 2017/18 crash in cryptocur-
of explosivity in all cases. GSADF is used to identify mul- rency markets. The result showed that higher persistence of
tiple explosiveness periods and logistic regression is em- shocks is expected after the crash due to speculations in the
ployed to uncover evidence of co-explosivity across cryp- mind of cryptocurrency traders, and more evidence of non-
tocurrencies. The results showed that the likelihood of ex- mean reversions, implying chances of further price fall in
plosive periods in one cryptocurrency generally depends on cryptocurrencies.
the presence of explosivity in other cryptocurrencies and
points toward a contemporaneous co-explosivity that does 10. Others related to Cryptocurrency Trading
not necessarily depend on the size of each cryptocurrency.
Extended research by Phillips et al. [208, 209] (who ap- Some other research papers related to cryptocurrency
plied a recursive augmented Dickey-Fuller algorithm, which trading treat distributed in market behaviour, regulatory mech-
is called PSY test) and Landsnes et al. [97] studied pos- anisms and benchmarks.
sible predictors of bubble periods of certain cryptocurren- Krafft et al. [160] and Yang [252] analysed market dy-
cies. The evaluation includes multiple bubble periods in all namics and behavioural anomalies respectively to under-
cryptocurrencies. The result shows that higher volatility and stand effects of market behaviour in the cryptocurrency mar-
trading volume is positively associated with the presence of ket. Krafft et al. discussed potential ultimate causes, poten-
bubbles across cryptocurrencies. In terms of bubble predic- tial behavioural mechanisms and potential moderating con-
tion, the authors found the probit model to perform better textual factors to enumerate possible influence of GUI and
than the linear models. API on cryptocurrency markets. Then they highlighted the
Phillips et al. [210] used Hidden Markov Model (HMM) potential social and economic impact of human-computer
and Superiority and Inferiority Ranking (SIR) method to interaction in digital agency design. Yang, on the other
identify bubble-like behaviour in cryptocurrency time se- hand, applied behavioural theories of asset pricing anoma-
ries. Considering HMM and SIR method, an epidemic de- lies in testing 20 market anomalies using cryptocurrency
tection mechanism is used in social media to predict cryp- trading data. The results showed that anomaly research fo-
tocurrency price bubbles, which classify bubbles through cused more on the role of speculators, which gave a new
epidemic and non-epidemic labels. Experiments have demon- idea to research the momentum and reversal in the cryp-
strated a strong relationship between Reddit usage and cryptocurrency market. Cocco et al. [75] implemented a mech-
tocurrency prices. This work also provides some empirical anism to form a Bitcoin price and specific behaviour for
each type of trader including the initial wealth distribution

following Pareto’s law, order-based transaction and price pliance with efficient market assumptions. Cocco et al. [74]
settlement mechanism. Specifically, the model reproduced described an agent-based artificial cryptocurrency market in
the unit root attributes of the price series, the fat tail phe- which heterogeneous agents buy or sell cryptocurrencies.
nomenon, the volatility clustering of price returns, the gen- The proposed simulator is able to reproduce some real sta-
eration of Bitcoins, hashing power and power consumption. tistical properties of price returns observed in the Bitcoin
Leclair [169] and Vidal-Thomás et al. [241] analysed the real market. Marko [200] considered the future use of cryp-
existence of herding in the cryptocurrency market. Leclair tocurrencies as money based on the long-term value of cryp-
applied herding methods of Huang and Salmon [133] in esti- tocurrencies. Neil et al. [112] analysed the influence of net-
mating the market herd dynamics in the CAPM framework. work effect on the competition of new cryptocurrency mar-
Vidal-Thomás et al. analyse the existence of herds in the kets. Bariviera and Merediz-Sola [17] gave a survey based
cryptocurrency market by returning the cross-sectional stan- on hybrid analysis, which proposed a methodological hybrid
dard (absolute) deviations. Both their findings showed sig- method for a comprehensive literature review and provided
nificant evidence of market herding in the cryptocurrency the latest technology in the cryptocurrency economics liter-
market. Makarov et al. [180] studied price impact and arbi- ature.
trage dynamics in the cryptocurrency market and found that There also exists some research and papers introducing
85% of the variations in bitcoin returns and the idiosyncratic the basic process and rules of cryptocurrency trading in-
components of order flow play an important role in explain- cluding findings of Hansel et al. [124], Kate [148], Garza
ing the size of the arbitrage spreads between exchanges. et al. [114], Ward et al. [248] and Fantazzini et al. [101].
In November 2019, Griffin et al. put forward a paper Hansel et al. [124] introduced the basics of cryptocurrency,
on the thesis of unsupported digital money inflating cryp- Bitcoin and Blockchain, ways to identify the profitable trends
tocurrency prices [119], which caused a great stir in the aca- in the market, ways to use Altcoin trading platforms such
demic circle and public opinion. Using algorithms to anal- as GDAX and Coinbase, methods of using a crypto wal-
yse Blockchain data, they found that purchases with Tether let to store and protect the coins in their book. Kate et
are timed following market downturns and result in sizeable al. [148] set six steps to show how to start an investment
increases in Bitcoin prices. By mapping the blockchains of without any technical skills in the cryptocurrency market.
Bitcoin and Tether, they were able to establish that one large This book is an entry-level trading manual for starters learn-
player on Bitfinex uses Tether to purchase large amounts of ing cryptocurrency trading. Garza et al. [114] simulated
Bitcoin when prices are falling and following the prod of an automatic cryptocurrency trading system, which helps
Tether. investors limit systemic risks and improve market returns.
More researches involved benchmark and development This paper is an example to start designing an automatic
in cryptocurrency market [127, 259], regulatory framework cryptocurrency trading system. Ward et al. [248] discussed
analysis [220], data mining technology in cryptocurrency algorithmic cryptocurrency trading using several general al-
trading [204], application of efficient market hypothesis in gorithms, and modifications thereof including adjusting the
the cryptocurrency market [223] and artificial financial mar- parameters used in each strategy, as well as mixing mul-
kets for studying a cryptocurrency market [74]. Hileman tiple strategies or dynamically changing between strategies.
et al. [127] segmented the cryptocurrency industry into four This paper is an example to start algorithmic trading in cryp-
key sectors: exchanges, wallets, payments and mining. They tocurrency market. Fantazzini et al. [101] introduced the
gave a benchmarking study of individuals, data, regulation, R packages Bitcoin-Finance and bubble, including financial
compliance practices, costs of firms and a global map of analysis of cryptocurrency markets including Bitcoin.
mining in the cryptocurrency market in 2017. Zhou et al. [259] A community resource, that is, a platform for scholarly
discussed the status and future of computer trading in the communication, about cryptocurrencies and Blockchains is
largest group of Asia-Pacific economies and then consid- “Blockchain research network", see [197].
ered algorithmic and high frequency trading in cryptocur-
rency markets as well. Shanaev et al. [220] used data on
11. Summary Analysis of Literature Review
120 regulatory events to study the implications of cryptocur-
rency regulation and the results showed that stricter regula- This section analyses the timeline, the research distribu-
tion of cryptocurrency is not desirable. Akhilesh et al. [204] tion among technology and methods, the research distribu-
used the average absolute error calculated between the action among properties. It also summarises the datasets that
tual and predicted values of the market sentiment of differ- have been used in cryptocurrency trading research.
ent cryptocurrencies on that day as a method for quantifying
the uncertainty. They used the comparison of uncertainty 11.1. Timeline
quantification methods and opinion mining to analyse cur- Figure 8 shows several major events in cryptocurrency
rent market conditions. Sigaki et al. [223] used permutation trading. The timeline contains milestone events in cryp-
entropy and statistical complexity on the sliding time win- tocurrency trading and important scientific breakthroughs in
dow returned by the price log to quantify the dynamic ef- this area.
ficiency of more than four hundred cryptocurrencies. As a As early as 2009, Satoshi Nakamoto proposed and in-
result, the cryptocurrency market showed significant com- vented the first decentralised cryptocurrency, Bitcoin [192].
It is considered to be the start of cryptocurrency. In 2010,

Figure 8: Timeline of cryptocurrency trading research
the first cryptocurrency exchange was founded, which means

cryptocurrency would not be an OTC market but traded on
exchanges based on an auction market system.
Figure 9: Research distribution among cryptocurrency trad-
In 2013, Kristoufek [162] concluded that there is a strong
ing properties
correlation between Bitcoin price and the frequency of “Bit-
coin” search queries in Google Trends and Wikipedia. In
2014, Lee and Yang [170] firstly proposed to check causal- Table 7
ity from copula-based causality in the quantile method from Search hits of research distribution in all trading areas
trading volumes of seven major cryptocurrencies to returns Technology Category Google Scholar hits Google hits Arxiv hits
and volatility. Statistical methods 1.22M 62M 1240
Machine learning methods 483K 150M 520
In 2015, Cheah et al. [66] discussed the bubble and spec-
ulation of Bitcoin and cryptocurrencies. In 2016, Dyhrberg
explored Bitcoin volatility using GARCH models combined
pers basically research technical-level cryptocurrency trad-
with gold and US dollars [92].
ing including mathematical modeling and statistics. Other
From late 2016 to 2017, machine learning and deep learn-
papers related to trading systems on pure technical indica-
ing technology were applied in the prediction of cryptocur-
tors and introducing the industry and its history are not in-
rency return. In 2016, McNally et al. [184] predicted Bit-
cluded in this analysis. Among all 87 papers, 75 papers
coin price using the LSTM algorithm. Bell and Zbikowski
(86.21%) present statistical methods and technologies in cryp-
et al. [23, 255] applied SVM algorithm to predict trends of
tocurrency trading research and 13.8% papers research ma-
cryptocurrency price. In 2017, Jiang et al. [136] used double
chine learning applied to cryptocurrency trading (cf. Fig-
Q-network and pre-trained it using DBM for the prediction
ure 10). It is interesting to mention that, there are 16 pa-
of cryptocurrencies portfolio weights.
pers (18.4%) applying and comparing more than one tech-
In recent years, several research directions including cross
nique in cryptocurrency trading. More specifically, Bach et
asset portfolios [22, 56, 46], transaction network applica-
al. [12], Alessandretti et al. [5], Vo et al. [243], Phaladis-
tions [163, 44], machine learning optimisation [214, 9, 258]
ailoed et al. [207], Siaminos [222], Rane et al. [214] used
have been considered in the cryptocurrency trading area.
both statistical methods and machine learning methods in
cryptocurrency trading.
11.2. Research Distribution among Properties
Table 7 shows the results of search hits in all trading ar-
We counted the number of papers covering different as-
eas (not limited to cryptocurrencies). From the table, we can
pects of cryptocurrency trading. Figure 9 shows the result.
see that most research findings focused on statistical meth-
The attributes in the legend are ranked according to the num-
ods in trading, which means most of the research on tradi-
ber of papers that specifically test the attribute.
tional markets still focused on using statistical methods for
Over one-third (38.10%) of the papers research predic-
trading. But we observed that machine learning in trading
tion of returns. Another one-third of papers focus on re-
had a higher degree of attention. It might because the tra-
searching bubbles and extreme conditions and the relation-
ditional technical and fundamental have been arbitraged, so
ship between pairs and portfolios in cryptocurrency trading.
the market has moved in recent years to find new anomalies
The remaining researching topics (prediction of volatility,
to exploit. Meanwhile, the results also showed there exist
trading system, technical trading and others) have roughly
many opportunities for research in the widely studied areas
one-third share.
of machine learning applied to trade in cryptocurrency mar-
11.3. Research Distribution among Categories and kets (cf. Section 12).
Technologies 11.3.1. Research Distribution among Statistical
This section introduces and compares categories and tech- methods
nologies in cryptocurrency trading. When papers cover mul- As from Figure 10, we further classified the papers us-
tiple technologies or compare different methods, we draw ing statistical methods into 6 categories: (i) basic regres-
statistics from different technical perspectives. sion methods; (ii) linear classifiers and clustering; (iii) time-
Among all the 126 papers, 87 papers (69.05%) cover sta- series analysis; (iv) decision trees and probabilistic classi-
tistical methods and machine learning categories. These pa- fiers; (v) modern portfolios theory; and, (vi) Others.

ing volume, order-level information, collected from cryp-

tocurrency exchanges. Table 9 shows the sentiment-based
data. Most of the datasets in this table contain market data
and media/Internet data with emotional or statistical labels.
Table 10 gives two examples of datasets used in the col-
lected papers that are not covered in the first two tables.
The column “Currency” shows the types of cryptocur-
rencies included; this shows that Bitcoin is the most com-
monly used currency for cryptocurrency researches. The
column “Description” shows a general description and types
of datasets. The column “Data Resolution” means latency
of the data (e.g., used in the backtest) – this is useful to dis-
tinguish between high-frequency trading and low-frequency
trading. The column “Time range” shows the time span of
datasets used in experiments; this is convenient to distin-
Figure 10: Research distribution among cryptocurrency guish between the current performance in a specific time
trading technologies and methods interval and the long-term effect. We also present how the
dataset has been used (i.e., the task), cf. column “Usage”.
“Data Sources” gives details on where the data is retrieved
Basic regression methods include regression methods from, including cryptocurrency exchanges, aggregated cryp-
(Linear Regression), function estimation and CGCD method. tocurrency index and user forums (for sentiment analysis).
Linear Classifiers and Clustering include SVM and KNN Alexander et al. [6] made an investigation of cryptocur-
algorithm. Time-series analysis include GARCH model, rency data as well. They summarised data collected from
BEKK model, ARIMA model, Wavelet time-scale method. 152 published and SSRN discussion papers about cryptocur-
Decision Trees and probabilistic classifiers include Boost- rencies and analysed their data quality. They found that less
ing Tree, RF model. Modern portfolio theory include Value- than half the cryptocurrency papers published since January
at-Risk (VaR) theory, expected-shortfall (ES), Markowitz 2017 employ correct data.
mean-variance framework. Others include industry, market
data and research analysis in cryptocurrency market. 12. Opportunities in Cryptocurrency Trading
The figure shows that basic Regression methods and time-
This section discusses potential opportunities for future
series analysis are the most commonly used methods in this
research in cryptocurrency trading.
area.
Sentiment-based research. As discussed above, there
11.3.2. Research Distribution among Machine is a substantial body of work, which uses natural language
Learning Categories processing technology, for sentiment analysis with the ulti-
Papers using machine learning account for 22.78% (c.f mate goal of using news and media contents to improve the
Figure 10) of the total. We further classified these papers performance of cryptocurrency trading strategies.
into three categories: (vii) ANNs, (viii) LSTM/RNN/GRUs, Possible research directions may lie in a larger volume
and (ix) DL/RL. of media input (e.g., adding video sources) in sentiment
The figure also shows that methods based on LSTM, analysis; updating baseline natural language processing model
RNN and GRU are the most popular in this subfield. to perform more robust text preprocessing; applying neu-
ANNs contains papers researching ANN applications in ral networks in label training; extending samples in terms
cryptocurrency trading such as back propagation (BP) NN. of holding period; transaction-fees; and, user reputation re-
LSTM/RNN/GRUs include papers using neural networks search.
that exploit the temporal structure of the data, a technology Long-and-short term research. There are significant
especially suitable for time series prediction and financial differences between long and short time horizons in cryp-
trading. DL/RL includes papers using Multilayer Neural tocurrency trading. In long-term trading, investors might
Networks and Reinforcement Learning. The difference be- obtain greater profits but have more possibilities to control
tween ANN and DL is that generally, DL refers to an ANN risk when managing a position for weeks or months. It is
with multiple hidden layers while ANN refers to simple mandatory to control for risk on long term strategies due to
structure neural network contained input layer, hidden layer the increase in the holding period, directly proportional to
(one or multiple), and an output layer. the risk incurred by the trader. On the other hand, the longer
the horizon, the higher the risk and the most important the
11.4. Datasets used in Cryptocurrency Trading risk control. The shorter the horizon, the higher the cost and
Tables 8–10 show the details for some representative the lower the risk, so cost takes over the design of a strat-
datasets used in cryptocurrency trading research. Table 8 egy. In short-term trading, automated algorithmic trading
shows the market datasets. They mostly include price, trad- can be applied when holding periods are less than a week.

Table 8
Datasets (1/3):Market Data
Research Source Description Currency Data Resolution Time Range Usage Data Sources
Bouri et al. [41] price, Bitcoin, daily From: 2013/01/01 Prediction of volatility/return CoinMarketCap
volatility, Ethereum, To: 2017/12/31
detrended volume data 5 other cryptocurrencies
Nakano et al. [193] high frequency price, Bitcoin 15min From: 2016/07/31 Prediction of return Poloniex
volume data To: 2018/01/24
Bu et al. [49] three pieces time slice for Bitcoin and seven altcoins Not mentioned From: 2016/05/14 Maximum profit with DRL Not mentioned
different research objectives To: 2016/07/03
From: 2018/01/01
To: 2018/01/31
From: 2017/07/01
To: 2017/07/31
Sun et al. [229] price, volatility ETC-USDT, 1 minute, From: August 2017 Prediction of return Binance, Bitfinex
other 12 cryptocurrencies 5 minutes, To: December 2018
30 minutes,
one hour,
one day
Guo et al. [121] volatility, Bitcoin hourly volatility observations, From: September 2015 Prediction of volatility Not mentioned
order book data order book snapshots To: April 2017
Vo et al. [243] timestamps, Bitcoin 1minute From: Starting 2015 Prediction of return Bitstamp, Btce, Btcn,
the OHLC prices etc. To: End 2016 Coinbase, Coincheck, and Kraken
Ross et al. [210] price Bitcoin, daily From: April 2015 Predicting bubbles CryptoCompare
other 3 cryptocurrencies To: September 2016
Yaya et al. [253] price Bitcoin, daily From: 2015/08/07 Bubbles and crashes Coin Metrics
other 12 cryptocurrencies To: 2018/11/28
Brauneis et al. [46] individual price, 500 most capitalized daily From: 2015/01/01 Portfolios management CoinMarketCap
trading volume Cryptocurrencies To: 2017/12/31
Feng et al. [104] order-level USD/BTC Bitcoin order-level From: 2011/09/13 Trading strategy Bitstamp
trading data To: 2017/07/17
Table 9
Datasets (2/3):Sentiment-based data
Research Source Description Currency Time range Usage Data Sources
Kim et al. [156] Online cryptocurrency communities data Bitcoin,Ethereum, Ripple From: December 2013 Prediction of fluctuation Each community’s HTML page
and market data To: August, 2016 (Bitcoin)
From: August 2015
To: August, 2016 (Ethereum)
From: Creation
To: August, 2016 (Ripple)
Phillips et al. [212] Social media data and orice data Bitcoin and Ethereum From: 2016/08/30 Predict Mutual-Excitation of Reddit
To: 2017/08/30 Cryptocurrency Market Returns
Smtus [225] Hourly data on price and trading volume Bitcoin, Ethereum From: 2017/12/01 Prediction of price Google Trends, Telegram chat groups
and search terms from Google Trends and their respective pricedrivers To: 2018/06/30
Lamon et al. [167] Daily price data and cryptocurrency Bitcoin, Ethereum, Litecoin From: 2017/01/01 Prediction of price Kaggle, news headline
related news article headlines To: 2017/11/30
Phillips et al. [211] Price and social media factors from Reddit Bitcoin, Ethereum, Monero From: 2010/09/10 Waveletcoherence analysis of price BraveNewCoin
To: 2017/05/31 (Bitcoin)
Others can reference the paper
Kang et al. [145] Market data and posts and comments Bitcoin From: 2009/11/22 Relationships Between Bitcoin Bitcoin forum
including metadata To: 2018/02/02 Prices and User Groups in
Online Community
Researchers can differentiate between long-term and short- long-term research.

term trading in cryptocurrency trading by applying wavelet The existing work is mainly about showing the differ-
technology analysing bubble regimes [211] and consider- ences between long and short-term cryptocurrency trading.
ing price explosiveness [44] hypotheses for short-term and Long-term trading means less time would cost in trend trac-
Table 10
Datasets (3/3):Others
Research Source Description Time range Usage Data Sources
Kurbucz [163, 158] Raw and preprocessed data of all From: 2016/11/09 Predicting the price of Bitcoin Bitcoin network dataset [189]
Bitcoin transactions and daily returns To: 2018/02/05 with transaction network
Bedi et al. [22] A diversified portfolio including equity, From: July 2010 Cross-currency including cryptocurrency Portfolio sources [22]
fixedincome, real estate, alternative To: December 2018 researching portfolios
investments, commodities and money market

ing and simple technical indicators in market analysis. Short- markets from academic publications. Similarly, cryptocur-
term trading can limit overall risk because small positions rency market predictability could also be affected by re-
are used in every transaction. But market noise (interfer- search papers in the area. A possible attempt is to try new
ence) and short transaction time might cause some stress in pricing methods applying real-time market changes. Con-
short term trading. It might also be interesting to explore sidering the proportion of informed traders increasing in
the extraction of trading signals, time series research, appli- the cryptocurrency market in the pricing process is another
cation to portfolio management, the relationship between a breaking point (looking for a balance between alpha trading
huge market crash and small price drop, derivative pricing and trading research literature).
in cryptocurrency market, etc.
Correlation between cryptocurrency and others. By 13. Conclusions
the effects of monetary policy and business cycles that are
not controlled by the central bank, cryptocurrency is always We provided a comprehensive overview and analysis of
negatively correlated with overall financial market trends. the research work on cryptocurrency trading. This survey
There have been some studies discussing correlations be- presented a nomenclature of the definitions and current state
tween cryptocurrencies and other financial markets [146, of the art. The paper provides a comprehensive survey of
56], which can be used to predict the direction of the cryp- 126 cryptocurrency trading papers and analyses the research
tocurrency market. distribution that characterise the cryptocurrency trading lit-
Considering the characteristics of cryptocurrency, the erature. We further summarised the datasets used for exper-
correlation between cryptocurrency and other assets still re- iments and analysed the research trends and opportunities in
quires further research. Possible breakthroughs might be cryptocurrency trading.
achieved with principal component analysis, the relation- We expect this survey to be beneficial to academics (e.g.,
ship between cryptocurrency and other currencies in extreme finance researchers) and quantitative traders alike. The sur-
conditions (i.e., financial collapse). vey represents a quick way to get familiar with the litera-
Bubbles and crash research. To discuss the high volatil- ture on cryptocurrency trading and can motivate more re-
ity and return of cryptocurrencies, current research has fo- searchers to contribute to the pressing problems in the area,
cussed on bubbles of cryptocurrency markets [68], corre- for example along the lines we have identified.
lation between cryptocurrency bubbles and indicators like
volatility index (VIX) [97] (which is a “panic index” to mea- References
sure the implied volatility of S&P500 Index Options), spillover [1] 3commas, 2020. 3commas smart trading terminal and auto trading
effects in cryptocurrency market [178]. bots. https://3commas.io/. [Online, Accessed: January 26, 2020].
Additional research for bubbles and crashes in cryptocur- [2] Abay, N.C., Akcora, C.G., Gel, Y.R., Kantarcioglu, M., Islambekov,
rency trading could include a connection between the pro- U.D., Tian, Y., Thuraisingham, B., 2019. Chainnet: Learning on
blockchain graphs with topological features, in: 2019 IEEE Interna-
cess of bubble generation and financial collapse and con-
tional Conference on Data Mining (ICDM), IEEE. pp. 946–951.
ducting a coherent analysis (coherent process analysis from [3] Adeyanju, C., 2019. What crypto exchanges do to comply with
the formation of bubbles to aftermath analysis of bubble kyc, aml and cft regulations. https://cointelegraph.com/news/
burst), analysis of bubble theory by Microeconomics, trying what-crypto-exchanges-do-to-comply-with-kyc-aml-and-cft-
other physical or industrial models in analysing bubbles in regulations. [Online, Accessed January 11, 2020].
[4] Ahamad, S., Nair, M., Varghese, B., 2013. A survey on crypto cur-
cryptocurrency market (i.e., Omori law [249]), discussing
rencies, in: 4th International Conference on Advances in Computer
the supply and demand relationship of cryptocurrency in Science, AETACS, Citeseer. pp. 42–48.
bubble analysis (like using supply and demand graph to sim- [5] Alessandretti, L., ElBahrawy, A., Aiello, L.M., Baronchelli, A.,
ulate the generation of bubbles and simulate the bubble burst). 2018. Anticipating cryptocurrency prices using machine learning.
Game theory and agent-based analysis. Applying game Complexity 2018.
[6] Alexander, C., Dakos, M., 2020. A critical investigation of cryp-
theory or agent-based modelling in trading is a hot research
tocurrency data and analysis. Quantitative Finance 20, 173–188.
direction in the traditional financial market. It might also be [7] Antipova, V., 2019. Building and testing global investment portfo-
interesting to apply this method to trading in cryptocurrency lios using alternative asset classes. Master’s thesis. Vytautas Mag-
markets. nus University. Lithuania.
Public nature of Blockchain technology. Investiga- [8] Ardia, D., Bluteau, K., Rüede, M., 2019. Regime changes in bitcoin
garch volatility dynamics. Finance Research Letters 29, 266–271.
tions on the connections between the formation of a given
[9] Atsalakis, G.S., Atsalaki, I.G., Pasiouras, F., Zopounidis, C., 2019.
currency’s transaction network and its price has increased Bitcoin price forecasting with neuro-fuzzy techniques. European
rapidly in recent years; the growing attention on user iden- Journal of Operational Research 276, 770–780.
tification [139] also strongly supports this direction. With [10] Attanasio, G., Cagliero, L., Garza, P., Baralis, E., 2019. Quantita-
an in-depth understanding of these networks, we may iden- tive cryptocurrency trading: exploring the use of machine learning
techniques, in: Proceedings of the 5th Workshop on Data Science
tify new features in price prediction and may be closer to
for Macro-modeling with Financial and Economic Datasets, ACM.
understanding financial bubbles in cryptocurrency trading. p. 1.
Balance between the opening of trading research lit- [11] Authority, F.C., 2019. Contract for differences. https:
erature and the fading of alphas. Mclean et al. [183] //www.fca.org.uk/firms/contracts-for-difference, https://www:fca:
pointed out that investors learn about mispricing in stock org:uk/rms/contracts-for-dierence. [Online, Accessed January 29,
2020].

[12] Bach, W.G., Kasper, L., 2018. On Machine Learning Based Cryp- use/sa. [Online, Accessed February 11, 2020].
tocurrency Trading. Master’s thesis. Aalborg University. Denmark. [36] Blackbird, 2020. Blackbird bitcoin arbitrage: a long/short market-
[13] Badenhorst, J.J., et al., 2019. Effect of Bitcoin spot and derivative neutral strategy. https://github.com/butor/blackbird. [Online, Ac-
trading volumes on price volatility. Ph.D. thesis. University of Pre- cessed: January 26, 2020].
toria. [37] Bloomberg, 2020. Coinbase inc. https://www.bloomberg.com/
[14] Bai, S., Robinson, F., 2019. Automated Triangular Arbitrage:: A profile/company/0776164D:US. [Online, Accessed: February 11,
Trading Algorithm for Foreign Exchange on a Cryptocurrency Mar- 2020].
ket. Bachelor’s thesis. KTH Royal Institute of Technology. Sweden. [38] Bodart, V., Candelon, B., 2009. Evidence of interdependence and
[15] BAKKT, 2020a. Bakkt markets. https://www.bakkt.com/index. contagion using a frequency domain framework. Emerging markets
[Online, Accessed February 11, 2020]. review 10, 140–150.
[16] BAKKT, 2020b. Bakkt terms of use. https://www.bakkt.com/terms- [39] Bouri, E., Gupta, R., 2019. Predicting bitcoin returns: Comparing
of-use. [Online, Accessed: February 11, 2020]. the roles of newspaper-and internet search-based measures of uncer-
[17] Bariviera, A.F., Merediz-Sola, I., 2020. Where do we stand in cryp- tainty. Finance Research Letters , 101398.
tocurrencies economic research? a survey based on hybrid analysis. [40] Bouri, E., Jalkh, N., Molnár, P., Roubaud, D., 2017. Bitcoin for
arXiv preprint arXiv:2003.09723 . energy commodities before and after the december 2013 crash: di-
[18] Barnwal, A., Bharti, H., Ali, A., Singh, V., 2019. Stacking versifier, hedge or safe haven? Applied Economics 49, 5063–5073.
with neural network for cryptocurrency investment. arXiv preprint [41] Bouri, E., Lau, C.K.M., Lucey, B., Roubaud, D., 2019a. Trading
arXiv:1902.07855 . volume and the predictability of return and volatility in the cryp-
[19] Bartolucci, S., Destefanis, G., Ortu, M., Uras, N., Marchesi, M., tocurrency market. Finance Research Letters 29, 340–346.
Tonelli, R., 2019. The butterfly affect: Impact of development prac- [42] Bouri, E., Lucey, B., Roubaud, D., 2020a. The volatility surprise
tices on cryptocurrency prices . of leading cryptocurrencies: Transitory and permanent linkages. Fi-
[20] Baur, D.G., Dimpfl, T., Kuck, K., 2018. Bitcoin, gold and the us nance Research Letters 33, 101188.
dollar–a replication and extension. Finance Research Letters 25, [43] Bouri, E., Roubaud, D., Shahzad, S.J.H., 2020b. Do bitcoin and
103–110. other cryptocurrencies jump together? The Quarterly Review of
[21] Bauriya, A., Tikone, A., Nandgaonkar, P., Sakure, K.S., 2019. Real- Economics and Finance 76, 396–409.
time cryptocurrency trading system. International Research Journal [44] Bouri, E., Shahzad, S.J.H., Roubaud, D., 2019b. Co-explosivity in
of Engineering and Technology (IRJET) 06. the cryptocurrency market. Finance Research Letters 29, 178–183.
[22] Bedi, P., Nashier, T., 2020. On the investment credentials of bitcoin: [45] Bouri, E., Vo, X.V., Saeed, T., 2020c. Return equicorrelation in
A cross-currency perspective. Research in International Business the cryptocurrency market: Analysis and determinants. Finance Re-
and Finance 51, 101087. search Letters , 101497.
[23] Bell, T., 2016. Bitcoin trading agents. University of Southampton . [46] Brauneis, A., Mestel, R., 2019. Cryptocurrency-portfolios in a
[24] Ben-Akiva, M., McFadden, D., Train, K., Walker, J., Bhat, C., Bier- mean-variance framework. Finance Research Letters 28, 259–264.
laire, M., Bolduc, D., Boersch-Supan, A., Brownstone, D., Bunch, [47] van den Broek, L., Sharif, Z., 2018. Cointegration-based pairs trad-
D.S., et al., 2002. Hybrid choice models: progress and challenges. ing framework with application to the cryptocurrency market .
Marketing Letters 13, 163–175. [48] Brunnermeier, M.K., Oehmke, M., 2013. Bubbles, financial crises,
[25] Beneki, C., Koulis, A., Kyriazis, N.A., Papadamou, S., 2019. In- and systemic risk, in: Handbook of the Economics of Finance. El-
vestigating volatility transmission and hedging properties between sevier. volume 2, pp. 1221–1288.
bitcoin and ethereum. Research in International Business and Fi- [49] Bu, S.J., Cho, S.B., 2018. Learning optimal q-function using deep
nance 48, 219–227. boltzmann machine for reliable trading of cryptocurrency, in: Inter-
[26] Binance, 2020a. Binance partners with coinfirm to protect the global national Conference on Intelligent Data Engineering and Automated
cryptocurrency economy and ensure compliance with fatf aml rules. Learning, Springer. pp. 468–480.
https://www.binance.com/en/blog/386484403820867584/Binance- [50] Calo, B., Johnson, W., 2002. Global trading system. US Patent App.
Partners-with-Coinfirm-to-Protect-the-Global-Cryptocurrency- 09/769,036.
Economy-and-Ensure-Compliance-with-FATF-AML-Rules. [51] Capfolio, 2020. Capfolio cryptocurrency trading platform. https:
[Online, Accessed February 11, 2020]. //www.capfol.io/. [Online, Accessed: January 26, 2020].
[27] Binance, 2020b. Binance review 2020: Pros, cons, fees, features, [52] Caporale, G.M., Gil-Alana, L., Plastun, A., 2018. Persistence in
and safety. https://insidebitcoins.com/cryptocurrency-exchanges/ the cryptocurrency market. Research in International Business and
binance-review/. [Online, Accessed: February 11, 2020]. Finance 46, 141–148.
[28] Bitfinex, 2020a. Bitfinex markets. https://www.bitfinex.com/. [On- [53] Caporale, G.M., Plastun, A., 2018. Price Overreactions in the Cryp-
line, Accessed: February 11, 2020]. tocurrency Market. CESifo Working Paper 6861. Munich. URL:
[29] Bitfinex, 2020b. Bitfinex terms of service. https: http://hdl.handle.net/10419/174984.
//www.bitfinex.com/legal/terms. [Online, Accessed: February [54] Caporale, G.M., Plastun, A., 2019. The day of the week effect in
11, 2020]. the cryptocurrency market. Finance Research Letters 31.
[30] Bitfinex, 2020c. New york court rules that state attorney has ju- [55] Caporin, M., McAleer, M., 2012. Do we really need both bekk and
risdiction over bitfinex. https://cointelegraph.com/news/new-york- dcc? a tale of two multivariate garch models. Journal of Economic
court-rules-that-state-attorney-has-jurisdiction-over-bitfinex. [On- Surveys 26, 736–751.
line, Accessed February 11, 2020]. [56] Castro, J.G., Tito, E.A.H., Brandão, L.E.T., Gomes, L.L., 2019.
[31] Bitmex, 2020a. Beginner’s guide to bitmex: Complete review. Crypto-assets portfolio optimization under the omega measure. The
https://blockonomi.com/bitmex-review/. [Online, Accessed: Febru- Engineering Economist , 1–21.
ary 11, 2020]. [57] Catalyst, 2020. An algorithmic trading library for crypto-assets in
[32] Bitmex, 2020b. Bitmex. https://www.bitmex.com/register. [Online, python. https://github.com/enigmampc/catalyst. [Online, Accessed:
Accessed: February 11, 2020]. January 26, 2020].
[33] Bitstamp, 2020a. Bitstamp review 2020. https: [58] CBOE, 2020a. Cboe history. http://www.cboe.com/aboutcboe/
//www.fxempire.com/crypto/exchange/bitstamp/review. [On- history. [Online, Accessed: February 11, 2020].
line, Accessed: February 11, 2020]. [59] CBOE, 2020b. Cboe products. https://www.cboe.com. [Online,
[34] Bitstamp, 2020b. Bitstamp who we are. https://www.bitstamp.net/ Accessed: February 11, 2020].
about-us/. [Online, Accessed: February 11, 2020]. [60] CBOE, 2020c. Cfe regulation. https://www.cboe.com/aboutcboe/
[35] Bitstamp, 2020c. Terms of use. https://www.bitstamp.net/terms-of- legal-regulatory/departmental-overviews/cfe-regulation. [Online,

Accessed February 11, 2020]. [85] Corbet, S., Meegan, A., Larkin, C., Lucey, B., Yarovaya, L., 2018b.
[61] Ccxt, 2020. Ccxt – cryptocurrency exchange trading library. https: Exploring the dynamic relationships between cryptocurrencies and
//github.com/ccxt/ccxt. [Online, Accessed: January 26, 2020]. other financial assets. Economics Letters 165, 28–34.
[62] Chaim, P., Laurini, M.P., 2018. Volatility and return jumps in bit- [86] Cryptosignal, 2020. Automated crypto trading and technical analy-
coin. Economics Letters 173, 158–163. sis (ta) bot. https://github.com/CryptoSignal/crypto-signal. [Online,
[63] Chaim, P., Laurini, M.P., 2019. Nonlinear dependence in cryptocur- Accessed: January 26, 2020].
rency markets. The North American Journal of Economics and Fi- [87] Ctubio, 2020. Ctubio - cryptocurrency trading bot. https://
nance 48, 32–47. github.com/ctubio/Krypto-trading-bot. [Online, Accessed: January
[64] Chang, C.C., Hsieh, P.F., Wang, Y.H., 2015. Sophistication, senti- 26, 2020].
ment, and misreaction. Journal of Financial and Quantitative Anal- [88] Diebold, F.X., Yılmaz, K., 2014. On the network topology of
ysis 50, 903–928. variance decompositions: Measuring the connectedness of financial
[65] Charles, A., Darné, O., et al., 2019. Volatility estimation for cryp- firms. Journal of Econometrics 182, 119–134.
tocurrencies: Further evidence with jumps and structural breaks. [89] Domínguez, M.A., Lobato, I.N., 2003. Testing the martingale dif-
Economics Bulletin 39, 954–968. ference hypothesis. Econometric Reviews 22, 351–377.
[66] Cheah, E.T., Fry, J., 2015. Speculative bubbles in bitcoin markets? [90] Doran, M.D., 2014. A forensic look at bitcoin cryptocurrency. Ph.D.
an empirical investigation into the fundamental value of bitcoin. thesis. Utica College.
Economics Letters 130, 32–36. [91] Dutta, A., Kumar, S., Basu, M., 2020. A gated recurrent unit ap-
[67] Cheng, J., Dong, L., Lapata, M., 2016. Long short-term memory- proach to bitcoin price prediction. Journal of Risk and Financial
networks for machine reading. arXiv preprint arXiv:1601.06733 . Management 13, 23.
[68] Cheung, A., Roca, E., Su, J.J., 2015. Crypto-currency bubbles: an [92] Dyhrberg, A.H., 2016. Bitcoin, gold and the dollar–a garch volatil-
application of the phillips–shi–yu (2013) methodology on mt. gox ity analysis. Finance Research Letters 16, 85–92.
bitcoin prices. Applied Economics 47, 2348–2358. [93] Easley, D., Engle, R.F., O’Hara, M., Wu, L., 2008. Time-varying
[69] Choi, B., 2012. ARMA model identification. Springer Science & arrival rates of informed and uninformed trades. Journal of Financial
Business Media. Econometrics 6, 171–207.
[70] Chung, J., Gulcehre, C., Cho, K., Bengio, Y., 2014. Empirical eval- [94] Elliott, R.J., Van Der Hoek*, J., Malcolm, W.P., 2005. Pairs trading.
uation of gated recurrent neural networks on sequence modeling. Quantitative Finance 5, 271–276.
arXiv preprint arXiv:1412.3555 . [95] Engle, R., 2002. Dynamic conditional correlation: A simple class of
[71] CME, 2020a. Cme cryptocurrency products. https:// multivariate generalized autoregressive conditional heteroskedastic-
www.cmegroup.com/trading/cryptocurrency-indices.html. [Online, ity models. Journal of Business & Economic Statistics 20, 339–350.
Accessed: February 11, 2020]. [96] Engle, R.F., Kroner, K.F., 1995. Multivariate simultaneous general-
[72] CME, 2020b. Cme group rules and regulation overview. ized arch. Econometric theory 11, 122–150.
https://www.cmegroup.com/education/courses/market-regulation/ [97] Enoksen, F.A., Landsnes, C.J., 2019. What Can Predict Bubbles in
overview/cme-group-rules-and-regulation-overview.html. [Online, Cryptocurrency Prices? Master’s thesis. University of Stavanger,
Accessed February 11, 2020]. Norway.
[73] CME, 2020c. Cme groups overview. https://www.cmegroup.com/ [98] Escanciano, J.C., Velasco, C., 2006. Generalized spectral tests for
company/history/. [Online, Accessed: February 11, 2020]. the martingale difference hypothesis. Journal of Econometrics 134,
[74] Cocco, L., Concas, G., Marchesi, M., 2017. Using an artificial fi- 151–185.
nancial market for studying a cryptocurrency market. Journal of [99] Estalayo, I., Del Ser, J., Osaba, E., Bilbao, M.N., Muhammad, K.,
Economic Interaction and Coordination 12, 345–365. Gálvez, A., Iglesias, A., 2019. Return, diversification and risk in
[75] Cocco, L., Marchesi, M., 2016. Modeling and simulation of the cryptocurrency portfolios using deep recurrent neural networks and
economics of mining in the bitcoin market. PloS one 11. multi-objective evolutionary algorithms, in: 2019 IEEE Congress
[76] Coinbase, 2020a. Coinbase supported cryptocurrencies. on Evolutionary Computation (CEC), IEEE. pp. 755–761.
https://help.coinbase.com/en/coinbase/getting-started/general- [100] Fang, F., Chung, W., Ventre, C., Basios, M., Kanthan, L., Li, L., Wu,
crypto-education/supported-cryptocurrencies.html. [Online, F., 2019. Ascertaining price formation in cryptocurrency markets
Accessed: February 11, 2020]. with deeplearning. arXiv preprint arXiv:2003.00803 .
[77] Coinbase, 2020b. Our path to listing sec-regulated crypto secu- [101] Fantazzini, D., 2019. Quantitative finance with r and cryptocurren-
rities. https://blog.coinbase.com/our-path-to-listing-sec-regulated- cies. Amazon KDP, ISBN-13 , 978–1090685315.
crypto-securities-a1724e13bb5a. [Online, Accessed February 11, [102] Fantazzini, D., Zimin, S., 2020. A multivariate approach for the
2020]. simultaneous modelling of market risk and credit risk for cryptocur-
[78] CoinMaketCap, 2019. Top 100 cryptocurrencies by market capital- rencies. Journal of Industrial and Business Economics 47, 19–69.
ization URL: https://coinmarketcap.com/. [103] Farell, R., 2015. An analysis of the cryptocurrency industry .
[79] Coinmarketcap, 2020. Percentage of total market capitalization. [104] Feng, W., Wang, Y., Zhang, Z., 2018. Informed trading in the bitcoin
https://coinmarketcap.com/charts/}dominance-percentage. [Online, market. Finance Research Letters 26, 63–70.
Accessed January 11, 2020]. [105] Fil, M., 2019. Pairs trading in cryptocurrency markets. Univerzita
[80] Colianni, S., Rosales, S., Signorotti, M., 2015. Algorithmic trad- Karlova, Fakulta sociálních věd .
ing of cryptocurrency based on twitter sentiment analysis. CS229 [106] Flood, R., Hodrick, R.J., Kaplan, P., 1986. An evaluation of recent
Project , 1–5. evidence on stock market bubbles. National Bureau of Economic
[81] Conrad, C., Custovic, A., Ghysels, E., 2018. Long-and short- Research Cambridge, Mass., USA .
term cryptocurrency volatility components: A garch-midas analysis. [107] Flori, A., 2019. News and subjective beliefs: A bayesian approach
Journal of Risk and Financial Management 11, 23. to bitcoin investments. Research in International Business and Fi-
[82] Corbet, S., Eraslan, V., Lucey, B., Sensoy, A., 2019a. The effective- nance 50, 336–356.
ness of technical trading rules in cryptocurrency markets. Finance [108] Fretrade, 2020. Freqtrade. https://github.com/freqtrade/freqtrade.
Research Letters 31, 32–37. [Online, Accessed: January 26, 2020].
[83] Corbet, S., Lucey, B., Urquhart, A., Yarovaya, L., 2019b. Cryp- [109] Friedl, M.A., Brodley, C.E., 1997. Decision tree classification of
tocurrencies as a financial asset: A systematic analysis. Interna- land cover from remotely sensed data. Remote sensing of environ-
tional Review of Financial Analysis 62, 182–199. ment 61, 399–409.
[84] Corbet, S., Lucey, B., Yarovaya, L., 2018a. Datestamping the bit- [110] Friedman, J., Tibshirani, R., 1984. The monotone smoothing of
coin and ethereum bubbles. Finance Research Letters 26, 81–88. scatterplots. Technometrics 26, 243–250.

[111] Friedman, J.H., 2001. Greedy function approximation: a gradient with deep reinforcement learning, in: 2017 Intelligent Systems Con-
boosting machine. Annals of statistics , 1189–1232. ference (IntelliSys), IEEE. pp. 905–913.
[112] Gandal, N., Halaburda, H., 2014. Competition in the cryptocurrency [137] Jianliang, M., Haikun, S., Ling, B., 2009. The application on intru-
market. CEPR Discussion Paper No. DP10157. sion detection based on k-means cluster algorithm, in: 2009 Inter-
[113] Garcia, D., Schweitzer, F., 2015. Social signals and algorithmic national Forum on Information Technology and Applications, IEEE.
trading of bitcoin. Royal Society open science 2, 150288. pp. 150–152.
[114] Garza, P., 2019. Formal automatic trading in the cryptocurrency era. [138] Juchli, M., 2018. Limit order placement optimization with Deep
Ph.D. thesis. Politecnico di Torino. Reinforcement Learning: Learning from patterns in cryptocurrency
[115] Gatev, E., Goetzmann, W.N., Rouwenhorst, K.G., 2006. Pairs trad- market data. Master’s thesis. TU Delft Electrical Engineering.
ing: Performance of a relative-value arbitrage rule. The Review of Netherlands.
Financial Studies 19, 797–827. [139] Juhász, P.L., Stéger, J., Kondor, D., Vattay, G., 2018. A bayesian
[116] Gerritsen, D.F., Bouri, E., Ramezanifar, E., Roubaud, D., 2019. The approach to identify bitcoin users. PloS one 13.
profitability of technical trading rules in the bitcoin market. Finance [140] Kajtazi, A., Moro, A., 2019. The role of bitcoin in well diversified
Research Letters . portfolios: A comparative global study. International Review of
[117] Golang, 2020. A golang implementation of a console-based trad- Financial Analysis 61, 143–157.
ing bot for cryptocurrency exchanges. https://github.com/saniales/ [141] Kakushadze, Z., 2016. 101 formulaic alphas. Wilmott 2016, 72–81.
golang-crypto-trading-bot. [Online, Accessed: January 26, 2020]. [142] Kakushadze, Z., 2018. Cryptoasset factor models. Algorithmic Fi-
[118] Grayblock, 2018. Elliptic-curve cryptography. https://medium.com/ nance , 1–18.
coinmonks/elliptic-curve-cryptography-6de8fc748b8b. [Online, [143] Kalchbrenner, N., Grefenstette, E., Blunsom, P., 2014. A convo-
Accessed December 29, 2019]. lutional neural network for modelling sentences. arXiv preprint
[119] Griffin, J.M., Shams, A., 2019. Is bitcoin really un-tethered? Avail- arXiv:1404.2188 .
able at SSRN 3195066 . [144] Kamrat, S., Suesangiamsakul, N., Marukatat, R., 2018. Technical
[120] Gu, S., Lillicrap, T., Sutskever, I., Levine, S., 2016. Continuous analysis for cryptocurrency trading on mobile phones, in: 2018 3rd
deep q-learning with model-based acceleration, in: International Technology Innovation Management and Engineering Science In-
Conference on Machine Learning, pp. 2829–2838. ternational Conference (TIMES-iCON), IEEE. pp. 1–4.
[121] Guo, T., Bifet, A., Antulov-Fantulin, N., 2018. Bitcoin volatility [145] Kang, K., Choo, J., Kim, Y., 2019a. Whose opinion mat-
forecasting with a glimpse into buy and sell orders, in: 2018 IEEE ters? analyzing relationships between bitcoin prices and user
International Conference on Data Mining (ICDM), IEEE. pp. 989– groups in online community. Social Science Computer Review ,
994. 0894439319840716.
[122] Ha, S., Moon, B.R., 2018. Finding attractive technical patterns in [146] Kang, S.H., McIver, R.P., Hernandez, J.A., 2019b. Co-movements
cryptocurrency markets. Memetic Computing 10, 301–306. between bitcoin and gold: A wavelet coherence analysis. Physica
[123] Hale, G., Krishnamurthy, A., Kudlyak, M., Shultz, P., et al., 2018. A: Statistical Mechanics and its Applications , 120888.
How futures trading changed bitcoin prices. FRBSF Economic Let- [147] Kat, H.M., Heynen, R.C., 1994. Volatility prediction: A comparison
ter 12. of the stochastic volatility, garch (1, 1) and egarch (1, 1) models.
[124] Hansel, D., 2018. Cryptocurrency trading: How to make money by Journal of Derivatives 2.
trading bitcoin and other cryptocurrency (volume 2) . [148] Kate, C., 2018. Cryptocurrency trading for beginners: 6-steps action
[125] Harwick, C., 2016. Cryptocurrency and the problem of intermedia- plan to your first investment .
tion. The Independent Review 20, 569–588. [149] Katsiampa, P., 2019. An empirical investigation of volatility dynam-
[126] Henderson, P., Islam, R., Bachman, P., Pineau, J., Precup, D., ics in the cryptocurrency market. Research in International Business
Meger, D., 2018. Deep reinforcement learning that matters, in: and Finance .
Thirty-Second AAAI Conference on Artificial Intelligence. [150] Katsiampa, P., Corbet, S., Lucey, B.M., 2018a. Volatility spillover
[127] Hileman, G., Rauchs, M., 2017. Global cryptocurrency benchmark- effects in leading cryptocurrencies: A bekk-mgarch analysis. Avail-
ing study. Cambridge Centre for Alternative Finance 33. able at SSRN 3232912 .
[128] Holmes, G., Donkin, A., Witten, I.H., 1994. Weka: A machine [151] Katsiampa, P., Gkillas, K., Longin, F., 2018b. Cryptocur-
learning workbench, in: Proceedings of ANZIIS’94-Australian New rency market activity during extremely volatile periods URL:
Zealnd Intelligent Information Systems Conference, IEEE. pp. 357– https://ssrn.com/abstract=3220781, doi:https://dx.doi.org/
361. 10.2139/ssrn.3220781.
[129] Hrytsiuk, P., Babych, T., Bachyshyna, L., 2019. Cryptocur- [152] Kaufman, P.J., 2013. Trading Systems and Methods,+ Website. vol-
rency portfolio optimization using value-at-risk measure, in: Strate- ume 591. John Wiley & Sons.
gies, Models and Technologies of Economic Systems Management [153] Keerthi, S.S., Shevade, S.K., Bhattacharyya, C., Murthy, K.R.K.,
(SMTESM 2019), Atlantis Press. 2001. Improvements to platt’s smo algorithm for svm classifier de-
[130] Hudson, R., Urquhart, A., 2019. Technical analysis and cryptocur- sign. Neural computation 13, 637–649.
rencies. Available at SSRN 3387950 . [154] Khuntia, S., Pattanayak, J., 2018. Adaptive market hypothesis and
[131] Hultman, H., 2018. Volatility Forecasting An Empirical Study on evolving predictability of bitcoin. Economics Letters 167, 26–28.
Bitcoin Using Garch and Stochastic Volatility models. Master’s the- [155] Kim, K., Kim, J., Rinaldo, A., 2018. Time series featurization via
sis. Lund University. Sweden. topological data analysis: an application to cryptocurrency trend
[132] Hutto, C.J., Gilbert, E., 2014. Vader: A parsimonious rule-based forecasting. arXiv preprint arXiv:1812.02987 .
model for sentiment analysis of social media text, in: Eighth inter- [156] Kim, Y.B., Kim, J.G., Kim, W., Im, J.H., Kim, T.H., Kang, S.J.,
national AAAI conference on weblogs and social media. Kim, C.H., 2016. Predicting fluctuations in cryptocurrency transac-
[133] Hwang, S., Salmon, M., 2004. Market stress and herding. Journal tions based on user comments and replies. PloS one 11, e0161197.
of Empirical Finance 11, 585–616. [157] Kondor, D., Csabai, I., Szüle, J., Pósfai, M., Vattay, G., 2014a. In-
[134] Ji, Q., Bouri, E., Kristoufek, L., Lucey, B., 2019a. Realised volatil- ferring the interplay between network structure and market effects
ity connectedness among bitcoin exchange markets. Finance Re- in bitcoin. New Journal of Physics 16, 125003.
search Letters , 101391. [158] Kondor, D., Csabai, I., Szule, J., Psfai, M., Vattay, G., 2014b. Infer-
[135] Ji, Q., Bouri, E., Lau, C.K.M., Roubaud, D., 2019b. Dynamic con- ring the interplay between network structure and market effects in
nectedness and integration in cryptocurrency markets. International bitcoin. New Journal of Physics 16, 125003. URL: https://doi.org/
Review of Financial Analysis 63, 257–272. 10.1088%2F1367-2630%2F16%2F12%2F125003, doi:10.1088/
[136] Jiang, Z., Liang, J., 2017. Cryptocurrency portfolio management 1367-2630/16/12/125003.

[159] Kondor, D., Pósfai, M., Csabai, I., Vattay, G., 2014c. Do the rich why_world_leader_crypto_exchange_binance_moved_to_malta}.XlKZ8Gj7Q2x.
get richer? an empirical analysis of the bitcoin transaction network. [Online, Accessed: February 11, 2020].
PloS one 9. [182] Markowitz, H., 1952. Portfolio selection. The Journal of Finance 7,
[160] Krafft, P.M., Della Penna, N., Pentland, A.S., 2018. An experimen- 77–91.
tal study of cryptocurrency market dynamics, in: Proceedings of the [183] McLean, R.D., Pontiff, J., 2016. Does academic research destroy
2018 CHI Conference on Human Factors in Computing Systems, stock return predictability? The Journal of Finance 71, 5–32.
ACM. p. 605. [184] McNally, S., 2016. Predicting the price of Bitcoin using Machine
[161] Kristjanpoller, W., Bouri, E., Takaishi, T., 2020. Cryptocurrencies Learning. Ph.D. thesis. Dublin, National College of Ireland.
and equity funds: Evidence from an asymmetric multifractal anal- [185] McNally, S., Roche, J., Caton, S., 2018. Predicting the price of bit-
ysis. Physica A: Statistical Mechanics and its Applications 545, coin using machine learning, in: 2018 26th Euromicro International
123711. Conference on Parallel, Distributed and Network-based Processing
[162] Kristoufek, L., 2013. Bitcoin meets google trends and wikipedia: (PDP), IEEE. pp. 339–343.
Quantifying the relationship between phenomena of the internet era. [186] Merediz-Solà, I., Bariviera, A.F., 2019. A bibliometric analysis of
Scientific reports 3, 3415. bitcoin scientific production. Research in International Business and
[163] Kurbucz, M.T., 2019. Predicting the price of bitcoin by the most Finance 50, 294–305.
frequent edges of its transaction network. Economics Letters 184, [187] Meunier, S., 2018. Blockchain 101: What is blockchain and how
108655. does this revolutionary technology work?, in: Transforming Climate
[164] Kutner, M.H., Nachtsheim, C.J., Neter, J., Li, W., et al., 2005. Ap- Finance and Green Investment with Blockchains. Elsevier, pp. 23–
plied linear statistical models. volume 5. McGraw-Hill Irwin New 34.
York. [188] Mikolov, T., Kombrink, S., Burget, L., Černockỳ, J., Khudanpur, S.,
[165] Kwon, D.H., Kim, J.B., Heo, J.S., Kim, C.M., Han, Y.H., 2019. 2011. Extensions of recurrent neural network language model, in:
Time series classification of cryptocurrency price trend based on a 2011 IEEE international conference on acoustics, speech and signal
recurrent lstm neural network. Journal of Information Processing processing (ICASSP), IEEE. pp. 5528–5531.
Systems 15. [189] MIT, 2015. Bitcoin network dataset. https://senseable2015-
[166] Kyriazis, N.A., 2019. A survey on efficiency and profitable trad- 6.mit.edu/bitcoin. [Online, Accessed January 11, 2020].
ing opportunities in cryptocurrency markets. Journal of Risk and [190] Molina, J., 2019. Develop your crypto-trading system using plain
Financial Management 12, 67. logic, part 1. URL: https://medium.com/swlh/develop-your-crypto-
[167] Lamon, C., Nielsen, E., Redondo, E., 2017. Cryptocurrency price trading-system-using-plain-logic-part-1-caac02f0a37d.
prediction using news and social media sentiment. SMU Data Sci. [191] Mukhopadhyay, U., Skjellum, A., Hambolu, O., Oakley, J., Yu, L.,
Rev 1, 1–22. Brooks, R., 2016. A brief survey of cryptocurrency systems, in:
[168] Lawrence, S., Giles, C.L., Tsoi, A.C., Back, A.D., 1997. Face 2016 14th annual conference on privacy, security and trust (PST),
recognition: A convolutional neural-network approach. IEEE trans- IEEE. pp. 745–752.
actions on neural networks 8, 98–113. [192] Nakamoto, S., 2009. Bitcoin open source implementation of p2p
[169] Leclair, E.M., 2018. Herding in the cryptocurrency market. Econ currency. P2P foundation 18.
5029 final research. Carleton University. Canada. [193] Nakano, M., Takahashi, A., Takahashi, S., 2018. Bitcoin technical
[170] Lee, T.H., Yang, W., 2014. Granger-causality in quantiles between trading with artificial neural network. Physica A: Statistical Me-
financial markets: Using copula approach. International Review of chanics and its Applications 510, 587–609.
Financial Analysis 33, 70–78. [194] Narayanan, A., Bonneau, J., Felten, E., Miller, A., Goldfeder, S.,
[171] Li, T.N., Tourin, A., 2016. Optimal pairs trading with time- 2016. Bitcoin and cryptocurrency technologies: a comprehensive
varying volatility. International Journal of Financial Engineering introduction. Princeton University Press.
3, 1650023. [195] Nasir, M.A., Huynh, T.L.D., Nguyen, S.P., Duong, D., 2019. Fore-
[172] Li, T.R., Chamrajnagar, A., Fong, X., Rizik, N., Fu, F., 2019. casting cryptocurrency returns and volume using search engines. Fi-
Sentiment-based prediction of alternative cryptocurrency price fluc- nancial Innovation 5, 2.
tuations using gradient boosting tree model. Frontiers in Physics 7, [196] Neter, J., Kutner, M.H., Nachtsheim, C.J., Wasserman, W., 1996.
98. Applied linear statistical models. volume 4. Irwin Chicago.
[173] Liaw, A., Wiener, M., et al., 2002. Classification and regression by [197] research network, B., 2020. Platform for scholarly com-
randomforest. R news 2, 18–22. munication about cryptocurrencies and blockchains. https://
[174] Lintilhac, P.S., Tourin, A., 2017. Model-based pairs trading in the www.blockchainresearchnetwork.org/. [Online, Accessed: April
bitcoin markets. Quantitative Finance 17, 703–716. 17, 2020].
[175] Liu, W., 2019. Portfolio diversification across cryptocurrencies. Fi- [198] Ng, A.Y., Jordan, M.I., 2002. On discriminative vs. generative clas-
nance Research Letters 29, 200–205. sifiers: A comparison of logistic regression and naive bayes, in: Ad-
[176] Localbtc, 2020. Localbitcoins purchasing online. https:// vances in neural information processing systems, pp. 841–848.
localbitcoins.com. [Online, Accessed: January 11, 2020]. [199] Nomics, 2020. Top cryptocurrency exchanges list. https://
[177] Lucarelli, G., Borrotti, M., 2019. A deep reinforcement learning ap- nomics.com/exchanges. [Online, Accessed: January 11, 2020].
proach for automated cryptocurrency trading, in: IFIP International [200] Ogorevc, M., 2019. Cryptocurrency as money: A trading strategy
Conference on Artificial Intelligence Applications and Innovations, solution. Available at SSRN 3436041 .
Springer. pp. 247–258. [201] Omane-Adjepong, M., Alagidede, I.P., 2019. Multiresolution anal-
[178] Luu Duc Huynh, T., 2019. Spillover risks on cryptocurrency mar- ysis and spillovers of major cryptocurrency markets. Research in
kets: A look from var-svar granger causality and student copulas. International Business and Finance 49, 191–206.
Journal of Risk and Financial Management 12, 52. [202] Omane-Adjepong, M., Alagidede, P., Akosah, N.K., 2019. Wavelet
[179] Madan, I., Saluja, S., Zhao, A., 2015. Automated bitcoin trad- time-scale persistence analysis of cryptocurrency market returns and
ing via machine learning algorithms. URL: http://cs229. stanford. volatility. Physica A: Statistical Mechanics and its Applications 514,
edu/proj2014/Isaac% 20Madan 20. 105–120.
[180] Makarov, I., Schoar, A., 2020. Trading and arbitrage in cryptocur- [203] Pascanu, R., Mikolov, T., Bengio, Y., 2013. On the difficulty of
rency markets. Journal of Financial Economics 135, 293–319. training recurrent neural networks, in: International conference on
[181] Maltatoday, 2020. Why world leader crypto machine learning, pp. 1310–1318.
exchange binance moved to malta. https:// [204] Patil, A.P., Akarsh, T., Parkavi, A., 2018. A study of opinion mining
www.maltatoday.com.mt/business/business_news/93170/ and data mining techniques to analyse the cryptocurrency market,

in: 2018 3rd International Conference on Computational Systems Performance Evaluation Review 46, 131–134.
and Information Technology for Sustainable Solutions (CSITSS), [226] Sriram, A., Jun, H., Satheesh, S., Coates, A., 2017. Cold fusion:
IEEE. pp. 198–203. Training seq2seq models together with language models. arXiv
[205] Păuna, C., 2018. Arbitrage trading systems for cryptocurrencies. preprint arXiv:1708.06426 .
design principles and server architecture. Informatica Economica [227] Stocksharp, 2020. Stocksharp - trading platform. https://
22, 35–42. github.com/StockSharp/StockSharp. [Online, Accessed: January
[206] Persson, S., Slottje, A., Shaw, I., 2018. Hybrid Autoregressive- 26, 2020].
Recurrent Neural Network Architecture for Algorithmic Trading of [228] Stuerner, P., 2019. Algorithmic Cryptocurrency Trading. Ph.D. the-
Cryptocurrencies. Cs230 deep learning thesis. Stanford University. sis. Ulm University.
[207] Phaladisailoed, T., Numnonda, T., 2018. Machine learning models [229] Sun, J., Zhou, Y., Lin, J., 2019. Using machine learning for cryp-
comparison for bitcoin price prediction, in: 2018 10th International tocurrency trading, in: 2019 IEEE International Conference on In-
Conference on Information Technology and Electrical Engineering dustrial Cyber Physical Systems (ICPS), IEEE. pp. 647–652.
(ICITEE), IEEE. pp. 506–511. [230] Sutton, R.S., Barto, A.G., et al., 1998. Introduction to reinforcement
[208] Phillips, P.C., Shi, S., Yu, J., 2015a. Testing for multiple bubbles: learning. volume 135. MIT press Cambridge.
Historical episodes of exuberance and collapse in the s&p 500. In- [231] Sze, V., Chen, Y.H., Yang, T.J., Emer, J.S., 2017. Efficient process-
ternational economic review 56, 1043–1078. ing of deep neural networks: A tutorial and survey. Proceedings of
[209] Phillips, P.C., Shi, S., Yu, J., 2015b. Testing for multiple bubbles: the IEEE 105, 2295–2329.
Limit theory of real-time detectors. International Economic Review [232] Tapscott, D., Tapscott, A., 2016. Blockchain revolution: how the
56, 1079–1134. technology behind bitcoin is changing money, business, and the
[210] Phillips, R.C., Gorse, D., 2017. Predicting cryptocurrency price world. Penguin.
bubbles using social media data and epidemic modelling, in: 2017 [233] TradingstrategyGuides, 2019a. Eos cryptocurrency trading
IEEE Symposium Series on Computational Intelligence (SSCI), strategy–turtle soup pattern. https://tradingstrategyguides.com/eos-
IEEE. pp. 1–7. cryptocurrency-trading-strategy/. [Online, Accessed January 29,
[211] Phillips, R.C., Gorse, D., 2018a. Cryptocurrency price drivers: 2020].
Wavelet coherence analysis revisited. PloS one 13, e0195200. [234] TradingstrategyGuides, 2019b. Free omni cryptocurrency strategy–
[212] Phillips, R.C., Gorse, D., 2018b. Mutual-excitation of cryptocur- amazing gann box. https://tradingstrategyguides.com/free-omni-
rency market returns and social media topics, in: Proceedings of the cryptocurrency-strategy/. [Online, Accessed January 29, 2020].
4th International Conference on Frontiers of Educational Technolo- [235] TradingstrategyGuides, 2019c. Iota cryptocurrency strategy–
gies, ACM. pp. 80–86. busted double top pattern. https://tradingstrategyguides.com/iota-
[213] Poloniex, 2020. Poloniex markets. https://poloniex.com/. [Online, cryptocurrency-strategy/. [Online, Accessed January 29, 2020].
Accessed: February 11, 2020]. [236] TradingstrategyGuides, 2019d. Nem (xem) cryp-
[214] Rane, P.V., Dhage, S.N., 2019. Systematic erudition of bitcoin price tocurrency strategy–momentum pinball setup. https:
prediction using machine learning techniques, in: 2019 5th Interna- //tradingstrategyguides.com/nem-xem-cryptocurrency-strategy/.
tional Conference on Advanced Computing & Communication Sys- [Online, Accessed January 29, 2020].
tems (ICACCS), IEEE. pp. 594–598. [237] TradingstrategyGuides, 2019e. Tether trading strategy–bottom
[215] Rebane, J., Karlsson, I., Denic, S., Papapetrou, P., 2018. Seq2seq rotation trading. https://tradingstrategyguides.com/tether-trading-
rnns and arima models for cryptocurrency prediction: A compara- strategy/. [Online, Accessed January 29, 2020].
tive study. SIGKDD Fintech 18. [238] TradingView, 2020. Total crypto market capitalization and
[216] Rish, I., et al., 2001. An empirical study of the naive bayes clas- volume. https://www.tradingview.com/markets/cryptocurrencies/
sifier, in: IJCAI 2001 workshop on empirical methods in artificial global-charts/. [Online, Accessed February 23, 2020].
intelligence, pp. 41–46. [239] Troster, V., Tiwari, A.K., Shahbaz, M., Macedo, D.N., 2019. Bitcoin
[217] Rose, C., 2015. The evolution of digital currencies: Bitcoin, a cryp- returns and risk: A general garch and gas analysis. Finance Research
tocurrency causing a monetary revolution. International Business & Letters 30, 187–193.
Economics Research Journal (IBER) 14, 617–622. [240] Trucíos, C., Tiwari, A.K., Alqahtani, F., 2019. Value-at-risk and ex-
[218] S., J., 2018. How do blockchain mining and transactions work pected shortfall in cryptocurrencies’ portfolio: A vine copula-based
explained in 7 simple steps. https://blog.goodaudience.com/ approach. Available at SSRN 3441892 .
how-a-miner-adds-transactions-to-the-blockchain-in-seven-steps- [241] Vidal-Tomás, D., Ibáñez, A.M., Farinós, J.E., 2019. Herding in the
856053271476. [Online, Accessed: January 26, 2020]. cryptocurrency market: Cssd and csad approaches. Finance Re-
[219] Salakhutdinov, R., Hinton, G., 2009. Deep boltzmann machines, in: search Letters 30, 181–186.
Artificial intelligence and statistics, pp. 448–455. [242] Virk, D.S., 2017. Prediction of Bitcoin Price using Data Mining.
[220] Shanaev, S., Sharma, S., Ghimire, B., Shuraeva, A., 2020. Tam- Master’s thesis. National College of Ireland.
ing the blockchain beast? regulatory implications for the cryptocur- [243] Vo, A., Yost-Bremm, C., 2018. A high-frequency algorithmic trad-
rency market. Research in International Business and Finance 51, ing strategy for cryptocurrency. Journal of Computer Information
101080. Systems , 1–14.
[221] Sharma, S., Krishma, N., Raina, E., 2017. Survey paper on cryp- [244] Vogelvang, B., 2005. Econometrics: theory and applications with
tocurrency. International Journal of Scientific Research in Computer Eviews. Pearson Education.
Science, Engineering and Information Technology Vol. 2 Issue 3, [245] Wagstaff, K., Cardie, C., Rogers, S., Schrödl, S., et al., 2001. Con-
307–310. strained k-means clustering with background knowledge, in: Icml,
[222] Siaminos, G., 2019. Predicting the value of cryptocurrencies using pp. 577–584.
machine learning time series analysis time series analysis time . [246] Wang, H., He, D., Ji, Y., 2017. Designated-verifier proof of assets
[223] Sigaki, H.Y., Perc, M., Ribeiro, H.V., 2019. Clustering patterns for bitcoin exchange using elliptic curve cryptography. Future Gen-
in efficiency and the coming-of-age of the cryptocurrency market. eration Computer Systems .
Scientific reports 9, 1440. [247] Wang, L., 2005. Support vector machines: theory and applications.
[224] Sirignano, J., Cont, R., 2019. Universal features of price formation volume 177. Springer Science & Business Media.
in financial markets: perspectives from deep learning. Quantitative [248] Ward, M., 2018. Algorithmic Trading for Cryptocurrencies. De-
Finance 19, 1449–1459. partmental honors thesis. UtahState University. United States.
[225] Smuts, N., 2019. What drives cryptocurrency prices?: An investiga- [249] Weber, P., Wang, F., Vodenska-Chitkushev, I., Havlin, S., Stanley,
tion of google trends and telegram sentiment. ACM SIGMETRICS H.E., 2007. Relation between volatility correlations in financial

markets and omori processes occurring on all scales. Physical Re-

view E 76, 016109.
[250] Wohlin, C., 2014. Guidelines for snowballing in systematic litera-
ture studies and a replication in software engineering, in: Proceed-
ings of the 18th international conference on evaluation and assess-
ment in software engineering, pp. 1–10.
[251] Xu, Z., Wang, S., Zhu, F., Huang, J., 2017. Seq2seq fingerprint:
An unsupervised deep molecular embedding for drug discovery, in:
Proceedings of the 8th ACM International Conference on Bioinfor-
matics, Computational Biology, and Health Informatics, pp. 285–
294.
[252] Yang, H., 2018. Behavioral anomalies in cryptocur-
rency markets. Available at SSRN 3174421 URL:
https://ssrn.com/abstract=3174421, doi:https://dx.doi.org/
10.2139/ssrn.3174421.
[253] Yaya, O.S., Ogbonna, E.A., Olubusoye, O.E., 2018. How persistent
and dependent are pricing of bitcoin to other cryptocurrencies before
and after 2017/18 crash? .
[254] Zamuda, A., Crescimanna, V., Burguillo, J.C., Dias, J.M., Wegrzyn-
Wolska, K., Rached, I., Gonzlez, H., Senkerik, R., Pop, C., Cioara,
T., et al., 2019. Forecasting cryptocurrency value by sentiment anal-
ysis: An hpc-oriented survey of the state-of-the-art in the cloud era,
in: High-Performance Modelling and Simulation for Big Data Ap-
plications. Springer, pp. 325–349.
[255] Żbikowski, K., 2016. Application of machine learning algorithms
for bitcoin automated trading, in: Machine Intelligence and Big
Data in Industry. Springer, pp. 161–168.
[256] Zemmal, N., Azizi, N., Dey, N., Sellami, M., 2016. Adaptive
semi supervised support vector machine semi supervised learning
with features cooperation for breast cancer classification. Journal of
Medical Imaging and Health Informatics 6, 53–62.
[257] Zhang, S., Yao, L., Sun, A., Tay, Y., 2019. Deep learning based rec-
ommender system: A survey and new perspectives. ACM Comput.
Surv. 52. URL: https://doi.org/10.1145/3285029, doi:10.1145/
3285029.
[258] Zhengyang, W., Xingzhou, L., Jinjin, R., Jiaqing, K., 2019. Predic-
tion of cryptocurrency price dynamics with multiple machine learn-
ing techniques, in: Proceedings of the 2019 4th International Con-
ference on Machine Learning Technologies, ACM. pp. 15–19.
[259] Zhou, H., Kalev, P.S., 2019. Algorithmic and high frequency trading
in asia-pacific, now and the future. Pacific-Basin Finance Journal
53, 186–207.
[260] Ślepaczuk, R., Zenkova, M., 2018. Robustness of support vector
machines in algorithmic trading on cryptocurrency market. Central
European Economic Journal 5, 186 – 205.
View publication stats

Cryptocurrency Trading: A Comprehensive Survey: March 2020

Uploaded by

Copyright:

Available Formats

Cryptocurrency Trading: A Comprehensive Survey: March 2020

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Cryptocurrency Trading: A Comprehensive Survey: March 2020

Uploaded by

Copyright:

Available Formats

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

Cryptocurrency Trading: A Comprehensive Survey

Preprint · March 2020

Fang Fan Carmine Ventre

SEE PROFILE SEE PROFILE

DASSE project View project

The user has requested enhancement of the downloaded file.

ARTICLE INFO ABSTRACT

First Author et al. Page 1 of 30

ogy or technical methods in researching cryptocurrency trad-

2. Cryptocurrency Trading 2.1.2. From Blockchain to Cryptocurrencies

First Author et al. Page 2 of 30

the current block. Miners in Blockchain accept transactions,

First Author et al. Page 3 of 30

a set of predefined rules to buy and sell on cryptocurrency

2.3.2. Advantages of Trading Cryptocurrency

First Author et al. Page 4 of 30

First Author et al. Page 5 of 30

ally, crypto asset denotes a digital asset (i.e., cryptocurren-

First Author et al. Page 6 of 30

First Author et al. Page 7 of 30

5. Cryptocurrency Trading Software Systems

First Author et al. Page 8 of 30

First Author et al. Page 9 of 30

First Author et al. Page 10 of 30

6.2. Pairs Trading

First Author et al. Page 11 of 30

First Author et al. Page 12 of 30

First Author et al. Page 13 of 30

First Author et al. Page 14 of 30

First Author et al. Page 15 of 30

First Author et al. Page 16 of 30

7.3. Others 8. Portfolio and Cryptocurrency Assets

First Author et al. Page 17 of 30

First Author et al. Page 18 of 30

First Author et al. Page 19 of 30

First Author et al. Page 20 of 30

Figure 8: Timeline of cryptocurrency trading research

the first cryptocurrency exchange was founded, which means

First Author et al. Page 21 of 30

ing volume, order-level information, collected from cryp-

First Author et al. Page 22 of 30

Researchers can differentiate between long-term and short- long-term research.

First Author et al. Page 23 of 30

First Author et al. Page 24 of 30

First Author et al. Page 25 of 30

First Author et al. Page 26 of 30

First Author et al. Page 27 of 30

First Author et al. Page 28 of 30

First Author et al. Page 29 of 30

markets and omori processes occurring on all scales. Physical Re-

First Author et al. Page 30 of 30

View publication stats

You might also like