Cryptocurrency Trading: A Comprehensive Survey
Fan Fang, Carmine Ventre, Michail Basios, Hoiliong Kong, Leslie Kanthan, Lingbo Li, David Martinez-Regoband, Fan Wu
CCryptocurrency Trading: A Comprehensive Survey
Fan Fang a , ∗ , Carmine Ventre a , Michail Basios b , Hoiliong Kong b , Leslie Kanthan b ,David Martinez-Rego b , Fan Wu b and Lingbo Li b , ∗ a King’s College London, UK b Turing Intelligence Technology Limited, UK
A R T I C L E I N F O
Keywords :trading, cryptocurrency, machine learn-ing, econometrics
A B S T R A C T
In recent years, the tendency of the number of financial institutions including cryptocurrencies intheir portfolios has accelerated. Cryptocurrencies are the first pure digital assets to be included byasset managers. Even though they share some commonalities with more traditional assets, they have aseparate nature of its own and their behaviour as an asset is still under the process of being understood.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 126research papers on various aspects of cryptocurrency trading ( e . g ., cryptocurrency trading systems,bubble and extreme condition, prediction of volatility and return, crypto-assets portfolio constructionand crypto-assets, technical trading and others). This paper also analyses datasets, research trends anddistribution among research objects (contents/properties) and technologies, concluding with somepromising 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 andtrading strategies. The academic community has similarlyspent considerable efforts in researching cryptocurrency trad-ing. This paper seeks to provide a comprehensive survey ofthe research on cryptocurrency trading, by which we meanany study aimed at facilitating and building strategies totrade cryptocurrencies.As an emerging market and research direction, cryp-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 papershave appeared since 2018, demonstrating the emergence ofcryptocurrency trading as a new research area in financialtrading.The literature is organised according tosix distinct as-pects of cryptocurrency trading:• Cryptocurrency trading software systems (i.e., real-time trading systems, turtle trading systems, arbitragetrading systems);• Systematic trading including technical analysis, pairstrading and other systematic trading methods; [email protected] ( Fan Fang); [email protected] ( Fan Fang); [email protected] ( Carmine Ventre); [email protected] ( Michail Basios); [email protected] ( Hoiliong Kong); [email protected] ( Leslie Kanthan); [email protected] ( David Martinez-Rego); [email protected] ( Fan Wu); [email protected] . cvr . cc , cvrŽsayahna . org ( Fan Fang) ORCID (s): ( Fan Fang); ( Lingbo Li)
Figure 1:
Cryptocurrency Trading Publications (cumulative)during 2013-2019 • Emergent trading technologies including economet-ric methods, machine learning technology and otheremergent trading methods;• Portfolio and cryptocurrency assets including researchamong cryptocurrency co-movements and crypto-assetportfolio research;• Market condition research including bubbles [106] orcrash analysis and extreme conditions;• Other Miscellaneous cryptocurrency trading research.In this survey we aim at compiling the most relevant re-search in these areas and extract a set of descriptive indica-tors that can give an idea of the level of maturity research inthis area has achieved.We also summarise research distribution (among researchproperties, categories and research technologies). The dis-tribution among properties and categories identifies classifi-cations of research objectives and contents. The distributionamong technologies identifies classifications of methodol-
First Author et al.
Page 1 of 30 a r X i v : . [ q -f i n . T R ] J a n ryptocurrency Trading: A Comprehensive Survey ogy or technical methods in researching cryptocurrency trad-ing. Specifically, we subdivide research distribution amongcategories and technologies into statistical methods and ma-chine learning technologies. Moreover, We identify datasetsand opportunities (potential research directions) that haveappeared in the cryptocurrency trading area. To ensure thatour survey is self-contained, we aim to provide sufficientmaterial to adequately guide financial trading researcherswho are interested in cryptocurrency trading.There has been related work that discussed or partiallysurveyed the literature related to cryptocurrency trading. Kyr-iazis et al. [166] surveyed efficiency and profitable tradingopportunities in cryptocurrency markets. Ahamad et al. [4]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 bibliometricanalysis of bitcoin literature. These relate work outcomesfocused 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-tocurrency trading.In summary, the paper makes the following contribu-tions: Definition.
This paper defines cryptocurrency trading andcategorises it into: cryptocurrency markets, cryptocur-rency trading models, and cryptocurrency trading strate-gies. The core content of this survey is trading strate-gies for cryptocurrencies while we cover all aspectsof it.
Multidisciplinary Survey.
The paper provides a compre-hensive survey of 126 cryptocurrency trading papers,across different academic disciplines such as financeand economics, artificial intelligence and computerscience. Some papers may cover multiple aspects andwill be surveyed for each category.
Analysis.
The paper analyses the research distribution, datasetsand trends that characterise the cryptocurrency trad-ing literature.
Horizons.
The paper identifies challenges, promising re-search directions in cryptocurrency trading, aimed topromote and facilitate further research.Figure 2 depicts the paper structure, which is informedby the review schema adopted. More details about this canbe found in Section 4.
2. Cryptocurrency Trading
This section provides an introduction to cryptocurrencytrading. We will discuss
Blockchain , as the enabling tech-nology, cryptocurrency markets and cryptocurrency trad-ing strategies . Figure 2:
Tree structure of the contents in this paper
Figure 3:
Workflow of Blockchain transaction
Blockchain is a digital ledger of economic transactionsthat can be used to record not just financial transactions, butany object with an intrinsic value. [232]. In its simplestform, a Blockchain is a series of immutable data recordswith timestamps,which are managed by a cluster of ma-chines that do not belong to any single entity. Each of thesedata block s is protected by cryptographic principle and boundto each other in a chain (cf. Figure 3 for the workflow).Cryptocurrencies like Bitcoin are made on a peer-to-peer network structure. Each peer has a complete historyof all transactions, thus recording the balance of each ac-count. For example, a transaction is a file that says “A paysX Bitcoins to B” that is signed by A using its private key.This is basic public-key cryptography, but also the build-ing block on which cryptocurrencies are based. After beingsigned, the transaction is broadcast on the network. When apeer discovers a new transaction, it checks to make sure thatthe signature is valid (this amounts to use the signer’s publickey, denoted as the algorithm in Figure 3). If the verificationis valid then the block is added to the chain; all other blocksadded after it will “confirm” that transaction. For example,if a transaction is contained in block 502 and the length ofthe blockchain is 507 blocks, it means that the transactionhas 5 confirmations (507-502) [218].
Confirmation is a critical concept in cryptocurrencies;only miners can confirm transactions. Miners add blocksto the Blockchain; they retrieve transactions in the previ-ous block and combine it with the hash of the precedingblock to obtain its hash, and then store the derived hash into
First Author et al. Page 2 of 30ryptocurrency Trading: A Comprehensive Survey 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 nodemust add it to its database. In layman terms, it has becomepart of the Blockchain and miners undertake this work toobtain cryptocurrency tokens, such as Bitcoin. In contrastto Blockchain, cryptocurrencies are related to the use of to-kens based on distributed ledger technology. Any transac-tion involving purchase, sale, investment, etc. involves aBlockchain native token or sub-token. Blockchain is a plat-form that drives cryptocurrency and is a technology that actsas a distributed ledger for the network. The network createsa means of transaction and enables the transfer of value andinformation. Cryptocurrencies are the tokens used in thesenetworks to send value and pay for these transactions. Theycan be thought of as tools on the Blockchain, and in somecases can also function as resources or utilities. In otherinstances, they are used to digitise the value of assets. Insummary, Cryptocurrencies are part of an ecosystem-basedon Blockchain technology.
Cryptocurrency is a decentralised medium of exchangewhich uses cryptographic functions to conduct financial trans-actions [90]. Cryptocurrencies leverage the Blockchain tech-nology to gain decentralisation, transparency, and immutabil-ity [187]. In the above, we have discussed how Blockchaintechnology is implemented for cryptocurrencies.In general, the security of cryptocurrencies is built oncryptography, neither by people nor on trust [194]. For ex-ample, Bitcoin uses a method called ”Elliptic Curve Cryp-tography” to ensure that transactions involving Bitcoin aresecure [246]. Elliptic curve cryptography is a type of public-key cryptography that relies on mathematics to ensure thesecurity of transactions. When someone attempts to circum-vent the aforesaid encryption scheme by brute force, it takesthem one-tenth the age of the universe to find a value matchwhen trying 250 billion possibilities every second [118].Regarding its use as a currency, cryptocurrency has the sameproperties as money. It has a controlled supply. Most cryp-tocurrencies limit the supply of tokens. E.g. for Bitcoin,the supply will decrease over time and will reach its finalquantity sometime around 2,140. All cryptocurrencies con-trol the supply of tokens through a timetable encoded in theBlockchain.One of the most important features of cryptocurrenciesis the exclusion of financial institution intermediaries [125].The absence of a “middleman” lowers transaction costs fortraders. For comparison, if a bank’s database is hacked ordamaged, the bank will rely entirely on its backup to recoverany information that is lost or compromised. With cryp-tocurrencies, even if part of the network is compromised,the rest will continue to be able to verify transactions cor-rectly. Cryptocurrencies also have the important feature ofnot being controlled by any central authority [217]: the de-centralised nature of the Blockchain ensures cryptocurren-
Figure 4:
Total Market Capitalization and Volume of cryp-tocurrency market, USD [238] cies are theoretically immune to government control and in-terference.As of December 20, 2019, there exist 4,950 cryptocur-rencies and 20,325 cryptocurrency markets; the market capis around 190 billion dollars [78]. Figure 4 shows histor-ical data on global market capitalisation and 24-hour trad-ing volume [238]. The total market cap is calculated byaggregating the dollar market cap of all cryptocurrencies.From the figure, we can observe how cryptocurrencies expe-rience exponential growth in 2017 and a large bubble burstin early 2018. But in recent years, cryptocurrencies haveshown signs of stabilisation.There are three mainstream cryptocurrencies: Bitcoin(BTC), Ethereum (ETH), and Litecoin (LTC). Bitcoin wascreated in 2009 and garnered massive popularity. On Oc-tober 31, 2008, an individual or group of individuals oper-ating under the pseudonym Satoshi Nakamoto released theBitcoin white paper and described it as: ”A pure peer-to-peer version of electronic cash that can be sent online forpayment from one party to another without going througha counterparty, ie. a financial institution.” [193] Launchedby Vitalik Buterin in 2015, Ethereum is a special Blockchainwith a special token called Ether (ETH symbol in exchanges).A very important feature of Ethereum is the ability to createnew tokens on the Ethereum Blockchain. The Ethereum net-work went live on July 30, 2015, and pre-mined 72 millionEthereum. Litecoin is a peer-to-peer cryptocurrency createdby Charlie Lee. It was created according to the Bitcoin pro-tocol, but it uses a different hashing algorithm. Litecoinuses a memory-intensive proof-of-work algorithm, Scrypt.ScryptFigure 5 shows percentages of total cryptocurrency mar-ket capitalisation; Bitcoin and Ethereum occupy the vastmajority of the total market capitalisation (data collected on8 Jan 2020).
A cryptocurrency exchange or digital currency exchange(DCE) is a business that allows customers to trade cryp-tocurrencies. Cryptocurrency exchanges can be market mak-ers, usually using the bid-ask spread as a commission forservices, or as a matching platform, by simply charging fees.Table 1 shows the top or classical cryptocurrency ex-
First Author et al. Page 3 of 30ryptocurrency Trading: A Comprehensive Survey
Table 1
Cryptocurrency exchanges Lists
Exchanges Category Supported currencies Fiat Currency Registration country Regulatory authorityCME 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 > 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] - Figure 5:
Percentage of Total Market Capitalisation [79] changes according to the rank list, by volume, compiledon “nomics” website [199]. Chicago Mercantile Exchange(CME), Chicago Board Options Exchange (CBOE) as wellas BAKKT (backed by New York Stock Exchange) are reg-ulated cryptocurrency exchanges. Fiat currency data alsocomes from “nomics” website [199]. Regulatory authorityand supported currencies of listed exchanges are collectedfrom official websites or blogs.
Firstly we give a definition of cryptocurrency trading . Definition 1.
Cryptocurrency trading is the act of buyingand selling of cryptocurrencies with the intention of makinga profit.
The definition of cryptocurrency trading can be broken downinto three aspects: object, operation mode and trading strat-egy. The object of cryptocurrency trading is the asset beingtraded, which is “cryptocurrency”. The operation mode ofcryptocurrency trading depends on the means of transactionin the cryptocurrency market, which can be classified into“trading of cryptocurrency Contract for Differences (CFD)”(The contract between the two parties, often referred to asthe “buyer” and “seller”, stipulates that the buyer will paythe seller the difference between themselves when the po-sition closes [11]) and “buying and selling cryptocurrenciesvia an exchange”. A trading strategy in cryptocurrency trad-ing, formulated by an investor, is an algorithm that defines a set of predefined rules to buy and sell on cryptocurrencymarkets.
The benefits of cryptocurrency trading include:
Drastic fluctuations.
The volatility of cryptocurrencies areoften likely to attract speculative interest and investors.The rapid fluctuations of intraday prices can providetraders with great money-earning opportunities, but italso includes more risk.
The cryptocurrency market is available24 hours a day, 7 days a week because it is a de-centralised market. Unlike buying and selling stocksand commodities, the cryptocurrency market is nottraded physically from a single location. Cryptocur-rency transactions can take place between individuals,in different venues across the world.
Near Anonymity.
Buying goods and services using cryp-tocurrencies is done online and does not require tomake one’s own identity public. With increasing con-cerns over identity theft and privacy, cryptocurren-cies can thus provide users with some advantages re-garding privacy. Different exchanges have specificKnow-Your-Customer (KYC) measures used to iden-tify users or customers [3]. The KYC undertook inthe exchanges allows financial institutions to reducethe financial risk while maximising the wallet owner’sanonymity.
Peer-to-peer transactions.
One of the biggest benefits ofcryptocurrencies is that they do not involve financialinstitution intermediaries. As mentioned above, thiscan reduce transaction costs. Moreover, this featuremight appeal to users who distrust traditional systems.Over-the-counter (OTC) cryptocurrency markets of-fer, in this context, peer-to-peer transactions on theBlockchain. The most famous cryptocurrency OTCmarket is “LocalBitcoin [176]”.
Programmable “smart” capabilities.
Some cryptocurren-cies can bring other benefits to holders, including lim-ited ownership and voting rights. Cryptocurrenciesmay also include a partial ownership interest in phys-ical assets such as artwork or real estate.
First Author et al. Page 4 of 30ryptocurrency Trading: A Comprehensive Survey
3. Cryptocurrency Trading Strategy
Cryptocurrency trading strategy is the main focus of thissurvey. There are many trading strategies, which can bebroadly divided into two main categories: technical and fun-damental. They are similar in the sense that they both relyon quantifiable information that can be backtested againsthistorical data to verify their performance. In recent years,a third kind of trading strategy, which we call quantitative,has received increasing attention. Such a trading strategy issimilar to a technical trading strategy because it uses trad-ing activity information on the exchange to make buying orselling decisions. Quantitative traders build trading strate-gies with quantitative data, which is mainly derived fromprice, volume, technical indicators or ratios to take advan-tage of inefficiencies in the market and are executed auto-matically by trading software. Cryptocurrency market is dif-ferent from traditional markets as there are more arbitrageopportunities, higher fluctuation and transparency. Due tothese characteristics, most traders and analysts prefer usingquantitative trading strategies in cryptocurrency markets.
Software trading systems allow international transactions,process customer accounts and information, and accept andexecute transaction orders [50]. A cryptocurrency tradingsystem is a set of principles and procedures that are pre-programmed to allow trade between cryptocurrencies andbetween fiat currencies and cryptocurrencies. Cryptocur-rency trading systems are built to overcome price manip-ulation, cybercriminal activities and transaction delays [21].When developing a cryptocurrency trading system, we mustconsider the capital market, base asset, investment plan andstrategies [190]. Strategies are the most important part of aneffective cryptocurrency trading system and they will be in-troduced below. There exist several cryptocurrency tradingsystems that are available commercially, for example, Cap-folio, 3Commas, CCXT, Freqtrade and Ctubio. From thesecryptocurrency trading systems, investors can obtain pro-fessional trading strategy support, fairness and transparencyfrom the professional third-party consulting companies andfast customer services.
Systematic Trading is a way to define trading goals,risk controls and rules. In general, systematic trading in-cludes high frequency trading and slower investment typeslike systematic trend tracking. In this survey, we dividesystematic cryptocurrency trading into technical analysis,pairs trading and others. Technical analysis in cryptocur-rency trading is the act of using historical patterns of trans-action data to assist a trader in assessing current and pro-jecting future market conditions for the purpose of makingprofitable trades. Price and volume charts summarise alltrading activity made by market participants in an exchangeand affect their decisions. Some experiments showed thatthe use of specific technical trading rules allows generat-ing excess returns, which is useful to cryptocurrency traders and investors in making optimal trading and investment de-cisions [116]. Pairs trading is a systematic trading strat-egy that considers two similar assets with slightly differentspreads. If the spread widens, short the high stocks and buythe low stocks. When the spread narrows again to a certainequilibrium value, a profit is generated [94]. Papers shownin this section involve the analysis and comparison of tech-nical indicators, pairs and informed trading, amongst otherstrategies.
Emergent trading strategies for cryptocurrency includestrategies that are based on econometrics and machine learn-ing technologies.
Econometric methods apply a combination of statisticaland economic theories to estimate economic variables andpredict their values [244].
Statistical models use mathe-matical equations to encode information extracted from thedata [152]. In some cases, statistical modeling techniquescan quickly provide sufficiently accurate models [24]. Othermethods might be used, such as sentiment-based predictionand long-and-short-term volatility classification based pre-diction [64]. The prediction of volatility can be used tojudge the price fluctuation of cryptocurrencies, which is alsovaluable for the pricing of cryptocurrency-related deriva-tives [147].When studying cryptocurrency trading using economet-rics, researchers apply statistical models on time-series datalike generalised autoregressive conditional heteroskedastic-ity (GARCH) and BEKK (named after Baba, Engle, Kraftand Kroner, 1995 [96]) models to evaluate the fluctuationof cryptocurrencies [55]. A linear statistical model is amethod to evaluate the linear relationship between pricesand an explanatory variable [196]. When there exists morethan one explanatory variable, we can model the linear re-lationship between explanatory (independent) and response(dependent) variables with multiple linear models. The com-mon linear statistical model used in the time-series analysisis the autoregressive moving average (ARMA) model [69].
Machine learning is an efficient tool for developing Bit-coin and other cryptocurrency trading strategies [185] be-cause it can infer data relationships that are often not di-rectly observable by humans. From the most basic perspec-tive, Machine Learning relies on the definition of two maincomponents: input features and objective function. The def-inition of Input Features (data sources) is where knowledgeof fundamental and technical analysis comes into play. Wemay divide the input into several groups of features, for ex-ample, those based on Economic indicators (such as, grossdomestic product indicator, interest rates, etc.), Social indi-cators (Google Trends, Twitter, etc.), Technical indicators(price, volume, etc.) and other Seasonal indicators (time ofday, day of the week, etc.). The objective function definesthe fitness criteria one uses to judge if the Machine Learn-
First Author et al. Page 5 of 30ryptocurrency Trading: A Comprehensive Survey
Figure 6:
Process of machine learning in predicting cryp-tocurrency ing model has learnt the task at hand. Typical predictivemodels try to anticipate numeric (e.g., price) or categorical(e.g., trend) unseen outcomes. The machine learning modelis trained by using historic input data (sometimes calledin-sample) to generalise patterns therein to unseen (out-of-sample) data to (approximately) achieve the goal defined bythe objective function. Clearly, in the case of trading, thegoal is to infer trading signals from market indicators whichhelp to anticipate asset future returns.Generalisation error is a pervasive concern in the appli-cation of Machine Learning to real applications, and of ut-most importance in Financial applications. We need to usestatistical approaches, such as cross validation, to validatethe model before we actually use it to make predictions.In machine learning, this is typically called “validation”.The process of using machine learning technology to pre-dict cryptocurrency is shown in Figure 6.Depending on the formulation of the main learning loop,we can classify Machine Learning approaches into threecategories: Supervised learning, Unsupervised learning andReinforcement learning.
Supervised learning is used to de-rive a predictive function from labeled training data. La-beled training data means that each training instance in-cludes inputs and expected outputs. Usually, these expectedoutputs are produced by a supervisor and represent the ex-pected behaviour of the model. The most used labels intrading are derived from in sample future returns of assets.
Unsupervised learning tries to infer structure from unla-beled training data and it can be used during exploratorydata analysis to discover hidden patterns or to group dataaccording to any pre-defined similarity metrics.
Reinforce-ment learning utilises software agents trained to maximisea utility function, which defines their objective; this is flex-ible enough to allow agents to exchange short term returnsfor future ones. In the financial sector, some trading chal-lenges can be expressed as a game in which an agent aimsat maximising the return at the end of the period.The use of machine learning in cryptocurrency tradingresearch encompasses the connection between data sources’understanding and machine learning model research. Fur-ther concrete examples are shown in a later section.
Portfolio theory advocates diversification of investmentsto maximize returns for a given level of risk by allocatingassets strategically. The celebrated mean-variance optimisa-tion is a prominent example of this approach [182]. Gener- ally, crypto asset denotes a digital asset (i.e., cryptocurren-cies and derivatives). There are some common ways to builda diversified portfolio in crypto assets. The first method isto diversify across markets, which is to mix a wide vari-ety of investments within a portfolio of the cryptocurrencymarket. The second method is to consider the industry sec-tor, which is to avoid investing too much money in any onecategory. Diversified investment of portfolio in the cryp-tocurrency market includes portfolio across cryptocurren-cies [175] and portfolio across the global market includingstocks and futures [140].
Market condition research appears especially importantfor cryptocurrencies. A financial bubble is a significant in-crease in the price of an asset without changes in its intrinsicvalue [48]. Many experts pinpoint a cryptocurrency bubblein 2017 when the prices of cryptocurrencies grew by 900 % .In 2018, Bitcoin faced a collapse in its value. This signif-icant fluctuation inspired researchers to study bubbles andextreme conditions in cryptocurrency trading.
4. Paper Collection and Review Schema
The section introduces the scope and approach of ourpaper collection, a basic analysis, and the structure of oursurvey.
We adopt a bottom-up approach to the research in cryp-tocurrency trading, starting from the systems up to risk man-agement techniques. For the underlying trading system, thefocus is on the optimisation of trading platforms structureand improvements of computer science technologies.At a higher level, researchers focus on the design ofmodels to predict return or volatility in cryptocurrency mar-kets. These techniques become useful to the generation oftrading signals. on the next level above predictive mod-els, researchers discuss technical trading methods to tradein real cryptocurrency markets. Bubbles and extreme condi-tions are hot topics in cryptocurrency trading because, asdiscussed above, these markets have shown to be highlyvolatile (whilst volatility went down after crashes). Portfo-lio and cryptocurrency asset management are effective meth-ods to control risk. We group these two areas in risk man-agement research. Other papers included in this survey in-clude topics like pricing rules, dynamic market analysis,regulatory implications, and so on. Table 2 shows the gen-eral scope of cryptocurrency trading included in this survey.Since many trading strategies and methods in cryptocur-rency trading are closely related to stock trading, some re-searchers migrate or use the research results for the latterto the former. When conducting this research, we only con-sider those papers whose research focuses on cryptocurrencymarkets or a comparison of trading in those and other finan-cial markets.Specifically, we apply the following criteria when col-lecting papers related to cryptocurrency trading:
First Author et al. Page 6 of 30ryptocurrency Trading: A Comprehensive Survey
Table 2
Survey scope table
Trading (bottom up) Trading SystemPrediction (return)Prediction (volatility)Technical trading methodsRisk management Bubble and extreme conditionPorfolio and Cryptocurrency assetOthers
1. The paper introduces or discusses the general idea ofcryptocurrency trading or one of the related aspects ofcryptocurrency trading.2. The paper proposes an approach, study or frameworkthat targets optimised efficiency or accuracy of cryp-tocurrency trading.3. The paper compares different approaches or perspec-tives in trading cryptocurrency.By “cryptocurrency trading” here, we mean one of the termslisted in Table 2 and discussed above.Some researchers gave a brief survey of cryptocurrency [4,221], cryptocurrency systems [191] and cryptocurrency trad-ing opportunities [166]. These surveys are rather limited inscope as compared to ours, which also includes a discussionon the latest papers in the area; we want to remark that thisis a fast-moving research field.
To collect the papers in different areas or platforms, weused keyword searches on Google Scholar and arXiv, twoof the most popular scientific databases. We also chooseother public repositories like SSRN but we find that almostall academic papers in these platforms can also be retrievedvia Google Scholar; consequently, in our statistical analysis,we count those as Google Scholar hits. We choose arXiv asanother source since it allows this survey to be contempo-rary with all the most recent findings in the area. The in-terested reader is warned that these papers have not under-gone formal peer review. The keywords used for searchingand collecting are listed below. [Crypto] means the cryp-tocurrency market, which is our research interest becausemethods might be different among different markets. Weconducted 6 searches across the two repositories just beforeOctober 15, 2019.- [Crypto] + Trading- [Crypto] + Trading system- [Crypto] + Prediction- [Crypto] + Trading strategy- [Crypto] + Risk Management- [Crypto] + PortfolioTo ensure high coverage, we adopted the so-called snow-balling [250] method on each paper found through thesekeywords. We checked papers added from snowballing meth-ods that satisfy the criteria introduced above until we reachedclosure.
Table 3
Paper query results.
Key Words
Table 3 shows the details of the results from our papercollection. Keyword searches and snowballing resulted in126 papers across the six research areas of interest in Sec-tion 4.1.Figure 7 shows the distribution of papers published atdifferent research sites. Among all the papers, 45.24% pa-pers are published in Finance and Economics venues suchas Journal of Financial Economics (JFE), Cambridge Centrefor Alternative Finance (CCAF), Finance Research Letters,Centre for Economic Policy Research (CEPR) and Journalof Risk and Financial Management (JRFM); 4.76% papersare published in Science venues such as Public Library OfScience one (PLOS one), Royal Society open science andSAGE; 15.87% papers are published in Intelligent Engi-neering and Data Mining venues such as Symposium Se-ries on Computational Intelligence (SSCI), Intelligent Sys-tems Conference (IntelliSys), Intelligent Data Engineeringand Automated Learning (IDEAL) and International Con-ference on Data Mining (ICDM); 4.76% papers are pub-lished in Physics / Physicians venues (mostly in Physicsvenue) such as Physica A; 10.32% papers are published inAI and complex system venues such as Complexity and In-ternational Federation for Information Processing (IFIP); 17.46%papers are published in Others venues which contains inde-pendently published papers and dissertations; 1.59% papersare published on arXiv. The distribution of different venuesshows that cryptocurrency trading is mostly published in Fi-nance and Economics venues, but with a wide diversity oth-erwise.
We discuss the contributions of the collected papers anda statistical analysis of these papers in the remainder of thepaper, according to Table 4.The papers in our collection are organised and presentedfrom six angles. We introduce the work about several dif-ferent cryptocurrency trading software systems in Section5. Section 6 introduces systematic trading applied to cryp-tocurrency trading. In Section 7, we introduce some emer-gent trading technologies including econometrics on cryp-tocurrencies, machine learning technologies and other emer-
First Author et al. Page 7 of 30ryptocurrency Trading: A Comprehensive Survey
Table 4
Review Schema
Classification Sec TopicCryptocurrency Trading Software System 5.1 Trading Infrastructure System5.2 Real-time Cryptocurrency Trading System5.3 Turtle trading system in Cryptocurrency market5.4 Arbitrage Trading Systems for Cryptocurrencies5.5 Comparison of three cryptocurrency trading systemsSystematic Trading 6.1 Technical Analysis6.2 Pairs Trading6.3 OthersEmergent Trading Technologies 7.1 Econometrics on cryptocurrency7.2 Machine learning technology7.3 OthersPortfolio and Cryptocurrency Assets 8.1 Research among cryptocurrency pairs and related factors8.2 Crypto-asset portfolio researchMarket condition research 9.1 Bubbles and crash analysis9.2 Extreme conditionOthers 10 Others related to Cryptocurrency TradingSummary Analysis of Literature Review 11.1 Timeline11.2 Research distribution among properties11.3 Research distribution among categories and technologies11.4 Datasets used in cryptocurrency trading
Figure 7:
Publication Venue Distribution gent trading technologies in the cryptocurrency market. Sec-tion 8 introduces research on cryptocurrency pairs and re-lated factors and crypto-asset portfolios research. In Sec-tion 9 we discuss cryptocurrency market condition research,including bubbles, crash analysis, and extreme conditions.Section 10 introduces other research included in cryptocur-rency trading not covered above.We would like to emphasize that the six headings abovefocus on a particular aspect of cryptocurrency trading; wegive a complete organisation of the papers collected undereach heading. This implies that those papers covering morethan one aspect will be discussed in different sections, oncefrom each angle.We analyse and compare the number of research paperson different cryptocurrency trading properties and technolo-gies in Section 11, where we also summarise the datasetsand the timeline of research in cryptocurrency trading.We build upon this review to conclude in Section 12 withsome opportunities for future research.
5. Cryptocurrency Trading Software Systems
Following the development of computer science and cryp-tocurrency trading, many cryptocurrency trading systems/botshave been developed. Table 5 compares the cryptocurrencytrading systems existing in the market. The table is sortedbased on URL types (GitHub or Official website) and GitHubstars (if appropriate).
Capfolio is a proprietary payable cryptocurrency tradingsystem which is a professional analysis platform and has anadvanced backtesting engine [51]. It supports five differentcryptocurrency exchanges. is a proprietary payable cryptocurrency trad-ing system platform that can take profit and stop-loss ordersat the same time [1]. Twelve different cryptocurrency ex-changes are compatible with this system.
CCXT is a cryptocurrency trading system with a unifiedAPI out of the box and optional normalized data and sup-ports many Bitcoin / Ether / Altcoin exchange markets andmerchant APIs. Any trader or developer can create a tradingstrategy based on this data and access public transactionsthrough the APIs [61]. The CCXT library is used to con-nect and trade with cryptocurrency exchanges and paymentprocessing services worldwide. It provides quick access tomarket data for storage, analysis, visualisation, indicator de-velopment, algorithmic trading, strategy backtesting, auto-mated code generation and related software engineering. Itis designed for coders, skilled traders, data scientists and fi-nancial analysts to build trading algorithms. Current CCXTfeatures include:I. Support for many cryptocurrency exchanges;II. Fully implemented public and private APIs;III. Optional normalized data for cross-exchange analysisand arbitrage;
First Author et al. Page 8 of 30ryptocurrency Trading: A Comprehensive Survey
IV. Out-of-the-box unified API, very easy to integrate.
Blackbird
Bitcoin Arbitrage is a C++ trading systemthat automatically executes long / short arbitrage betweenBitcoin exchanges. It can generate market-neutral strate-gies that do not transfer funds between exchanges [36]. Themotivation behind Blackbird is to naturally profit from thesetemporary price differences between different exchanges whilebeing market neutral. Unlike other Bitcoin arbitrage sys-tems, Blackbird does not sell but actually short sells Bitcoinon the short exchange. This feature offers two importantadvantages. Firstly, the strategy is always market agnostic:fluctuations (rising or falling) in the Bitcoin market will notaffect the strategy returns. This eliminates the huge risks ofthis strategy. Secondly, this strategy does not require trans-ferring funds (USD or BTC) between Bitcoin exchanges.Buy and sell transactions are conducted in parallel on twodifferent exchanges. There is no need to deal with transmis-sion delays.
StockSharp is an open-source trading platform for trad-ing at any market of the world including 48 cryptocurrencyexchanges [227]. It has a free C
Freqtrade is a free and open-source cryptocurrency trad-ing robot system written in Python. It is designed to supportall major exchanges and is controlled by telegram. It con-tains backtesting, mapping and money management tools,and strategy optimization through machine learning [108].Freqtrade has the following features:I. Persistence: Persistence is achieved through SQLitetechnology;II. Strategy optimization through machine learning: Usemachine learning to optimize your trading strategy pa-rameters with real trading data;III. Marginal Position Size: Calculates winning rate, risk-return ratio, optimal stop loss and adjusts position size,and then trades positions for each specific market;IV. Telegram management: use telegram to manage therobot.V. Dry run: Run the robot without spending money;
CryptoSignal is a professional technical analysis cryp-tocurrency trading system [86]. Investors can track over 500coins of Bittrex, Bitfinex, GDAX, Gemini and more. Auto-mated technical analysis includes momentum, RSI, IchimokuCloud, MACD, etc. The system gives alerts including Email,Slack, Telegram, etc. CryptoSignal has two primary fea-tures. First of all, it offers modular code for easy implemen-tation of trading strategies; Secondly, it is easy to install withDocker.
Ctubio is a C++ based low latency (high frequency)cryptocurrency trading system [87]. This trading system canplace or cancel orders through supported cryptocurrency ex-changes in less than a few milliseconds. Moreover, it pro-vides a charting system that can visualise the trading ac-count status including trades completed, target position forfiat currency, etc.
Catalyst is an analysis and visualization of the cryp-tocurrency trading system [57]. It makes trading strategieseasy to express and backtest them on historical data (dailyand minute resolution), providing analysis and insights intothe performance of specific strategies. Catalyst allows usersto share and organise data and build profitable, data-driveninvestment strategies. Catalyst not only supports the tradingexecution but also offers historical price data of all cryptoassets (from minute to daily resolution). Catalyst also hasbacktesting and real-time trading capabilities, which enablesusers to seamlessly transit between the two different trad-ing modes. Lastly, Catalyst integrates statistics and machinelearning libraries (such as matplotlib, scipy, statsmodels andsklearn) to support the development, analysis and visualiza-tion of the latest trading systems.
Golang Crypto Trading Bot is a Go based cryptocur-rency trading system [117]. Users can test the strategy insandbox environment simulation. If simulation mode is en-abled, a fake balance for each coin must be specified foreach exchange.
Amit et al. [21] developed a real-time CryptocurrencyTrading System. A real-time cryptocurrency trading systemis composed of clients, servers and databases. Traders usea web-application to login to the server to buy/sell cryptoassets. The server collects cryptocurrency market data bycreating a script that uses the Coinmarket API. Finally, thedatabase collects balances, trades and order book informa-tion from the server. The authors tested the system with anexperiment that demonstrates user-friendly and secure expe-riences for traders in the cryptocurrency exchange platform.
The original Turtle Trading system is a trend followingtrading system developed in the 1970s. The idea is to gen-erate buy and sell signals on stock for short-term and long-term breakouts and its cut-loss condition which is measuredby Average true range (ATR) [144]. The trading system willadjust the size of assets based on their volatility. Essen-tially, if a turtle accumulates a position in a highly volatile
First Author et al. Page 9 of 30ryptocurrency Trading: A Comprehensive Survey
Table 5
Comparison of existing cryptocurrency trading systems.
Name Features market, it will be offset by a low volatility position. Ex-tended Turtle Trading system is improved with smaller timeinterval spans and introduces a new rule by using exponen-tial moving average (EMA). Three EMA values are used totrigger the “buy” signal: 30EMA (Fast), 60EMA (Slow),100EMA (Long). The author of [144] performed backtest-ing and comparing both trading systems (Original Turtle andExtended Turtle) on 8 prominent cryptocurrencies. Throughthe experiment, Original Turtle Trading System achieved an18.59% average net profit margin (percentage of net profitover total revenue) and 35.94% average profitability (per-centage of winning trades over total numbers of trades) in87 trades through nearly one year. Extended Turtle Trad-ing System achieved 114.41% average net profit margin and52.75% average profitability in 41 trades through the sametime interval. This research showed how Extended TurtleTrading System compared can improve over Original TurtleTrading System in trading cryptocurrencies.
Christian [205] introduced arbitrage trading systems forcryptocurrencies. Arbitrage trading aims to spot the differ-ences in price that can occur when there are discrepancies inthe levels of supply and demand across multiple exchanges.As a result, a trader could realise a quick and low-risk profitby buying from one exchange and selling at a higher priceon a different exchange. Arbitrage trading signals are caughtby automated trading software. The technical differencesbetween data sources impose a server process to be organ-ised for each data source. Relational databases and SQLare reliable solution due to the large amounts of relationaldata. The author used the system to catch arbitrage oppor-tunities on 25 May 2018 among 787 cryptocurrencies on 7different exchanges. The research paper [205] listed the bestten trading signals made by this system from 186 availablefound signals. The results showed that the system caughtthe trading signal of “BTG-BTC” to get a profit of up to 495.44% when arbitraging to buy in Cryptopia exchangeand sell in Binance exchange. Another three well-traded ar-bitrage signals (profit expectation around 20% mentioned bythe author) were found on 25 May 2018. Arbitrage TradingSoftware System introduced in that paper presented generalprinciples and implementation of arbitrage trading systemin the cryptocurrency market.
Real-time trading systems use real-time functions to col-lect data and generate trading algorithms. Turtle trading sys-tem and arbitrage trading system have shown a sharp con-trast in their profit and risk behaviour. Using Turtle tradingsystem in cryptocurrency markets got high returns with highrisk. Arbitrage trading system is inferior in terms of revenuebut also has a lower risk. One feature that turtle trading sys-tem and arbitrage trading system have in common is theyperformed well in capturing alpha.
6. Systematic Trading
Many researchers have focused on technical indicators(patterns) analysis for trading on cryptocurrency markets.Examples of studies with this approach include “Turtle Souppattern strategy” [233], “Nem (XEM) strategy” [236], “Amaz-ing Gann Box strategy” [234], “Busted Double Top Pat-tern strategy” [235], and “Bottom Rotation Trading strat-egy” [237]. Table 6 shows the comparison among thesefive classical technical trading strategies using technical in-dicators. “Turtle soup pattern strategy” [233] used a 2-daybreakout of price in predicting price trends of cryptocurren-cies. This strategy is a kind of chart trading pattern. “Nem(XEM) strategy” combined Rate of Change (ROC) indica-tor and Relative Strength Index (RSI) in predicting pricetrends [236]. “Amazing Gann Box” predicted exact pointsof increase and decrease in Gann Box which are used to
First Author et al. Page 10 of 30ryptocurrency Trading: A Comprehensive Survey
Table 6
Comparison among five classical technical trading strategies
Technical trading strategy Core Methods Tecchnical tools/patternsTurtle Soup pattern [233] 2-daybreakout of price Chart trading patternsNem (XEM) [236] Price trends combined ROC & RSI Rate of Change indictor (ROC)Relative strength index (RSI)Amazing Gann Box [234] Predict exact points of rises and fallsin Gann Box (catch explosive trends) Candlestick, boxcharts withFibonacci RetracementBusted Double Top Pattern [235] Bearish reversal trading pattern thatgenerates a sell signal Price chart patternBottom Rotation Trading [237] Pick the bottom before the reversalhappens Price chart pattern, box chart catch explosive trends of cryptocurrency price [234]. Tech-nical analysis tools such as candlestick and box charts withFibonacci Retracement based on golden ratio are used in thistechnical analysis. Fibonacci Retracement uses horizontallines to indicate where possible support and resistance lev-els are in the market. “Busted Double Top Pattern” used aBearish reversal trading pattern which generates a sell sig-nal to predict price trends [235]. “Bottom Rotation Trading”is a technical analysis method that picks the bottom beforethe reversal happens. This strategy used a price chart patternand box chart as technical analysis tools.Sungjoo et al. [122] investigated using genetic program-ming (GP) to find attractive technical patterns in the cryp-tocurrency market. Over 12 technical indicators includingMoving Average (MA) and Stochastic oscillator were usedin experiments; adjusted gain, match count, relative mar-ket pressure and diversity measures have been used to quan-tify the attractiveness of technical patterns. With extendedexperiments, the GP system is shown to find successfullyattractive technical patterns, which are useful for portfoliooptimization. Hudson et al. [130] applied almost , to technical trading rules (classified into MA rules, filterrules, support resistance rules, oscillator rules and channelbreakout rules). This comprehensive study found that tech-nical trading rules provide investors with significant pre-dictive power and profitability. Corbet et al. [82] analysedvarious technical trading rules in the form of the movingaverage-oscillator and trading range break-out strategies togenerate higher returns in cryptocurrency markets. By usingone-minute dollar-denominated Bitcoin close-price data, thebacktest showed variable-length moving average (VMA) ruleperforms best considering it generates the most useful sig-nals in high frequency trading. Pairs trading is a trading strategy that attempts to ex-ploit the mean-reversion between the prices of certain secu-rities. Miroslav [105] investigated the applicability of stan-dard pairs trading approaches on cryptocurrency data withthe benchmarks of Gatev et al. [115]. The pairs trading strat-egy is constructed in two steps. Firstly, suitable pairs witha stable long-run relationship are identified. Secondly, thelong-run equilibrium is calculated and pairs trading strategyis defined by the spread based on the values. The researchalso extended intra-day pairs trading using high frequency data. Overall, the model was able to achieve a 3% monthlyprofit in Miroslav’s experiments [105]. Broek [47] ap-plied pairs trading based on cointegration in cryptocurrencytrading and 31 pairs were found to be significantly cointe-grated (within sector and cross-sector). By selecting fourpairs and testing over a 60-day trading period, the pairs trad-ing strategy got its profitability from arbitrage opportunities,which rejected the Efficient-market hypothesis (EMH) forthe cryptocurrency market. Lintihac et al [174] proposed anoptimal dynamic pair trading strategy model for a portfolioof assets. The experiment used stochastic control techniquesto calculate optimal portfolio weights and correlated the re-sults with several other strategies commonly used by practi-tioners including static dual-threshold strategies. Thomas etal. [171] proposed a pairwise trading model incorporatingtime-varying volatility with constant elasticity of variancetype. The experiment calculated the best pair strategy byusing a finite difference method and estimated parametersby generalised moment method.
Other systematic trading methods in cryptocurrency trad-ing mainly include informed trading. Using USD / BTC ex-change rate trading data, Feng et al. [104] found evidenceof informed trading in the Bitcoin market in those quantilesof the order sizes of buyer-initiated (seller-initiated) ordersare abnormally high before large positive (negative) events,compared to the quantiles of seller-initiated (buyer-initiated)orders; this study adopts a new indicator inspired by the vol-ume imbalance indicator [93]. The evidence of informedtrading in the Bitcoin market suggests that investors profiton their private information when they get information be-fore it is widely available.
7. Emergent Trading Technologies
Copula-quantile causality analysis and Granger-causalityanalysis are methods to investigate causality in cryptocur-rency trading analysis. Bouri et al. [41] applied a copula-quantile causality approach on volatility in the cryptocur-rency market. The approach of the experiment extended theCopula-Granger-causality in distribution (CGCD) methodof Lee and Yang [170] in 2014. The experiment constructedtwo tests of CGCD using copula functions. The paramet-
First Author et al. Page 11 of 30ryptocurrency Trading: A Comprehensive Survey ric test employed six parametric copula functions to dis-cover dependency density between variables. The perfor-mance matrix of these functions varies with independentcopula density. Three distribution regions are the focus ofthis research: left tail (1%, 5%, 10% quantile), central re-gion (40%, 60% quantile and median) and right tail (90%,95%, 99% quantile). The study provided significant evi-dence of Granger causality from trading volume to the re-turns of seven large cryptocurrencies on both left and righttails. Elie et al. [42] examined the causal linkages amongthe volatility of leading cryptocurrencies via the frequency-domain test of Bodart and Candelon [38] and distinguishedbetween temporary and permanent causation. The resultsshowed that permanent shocks are more important in ex-plaining Granger causality whereas transient shocks dom-inate the causality of smaller cryptocurrencies in the longterm. Badenhorst [13] attempted to reveal whether spotand derivative market volumes affect Bitcoin price volatilitywith the Granger-causality method and ARCH (1,1). Theresult shows spot trading volumes have a significant posi-tive effect on price volatility while the relationship betweencryptocurrency volatility and the derivative market is uncer-tain. Elie et al. [45] used a dynamic equicorrelation (DECO)model and reported evidence that the average earnings equi-librium correlation changes over time between the 12 lead-ing cryptocurrencies. The results showed increased cryp-tocurrency market consolidation despite significant price de-clined in 2018. Furthermore, measurement of trading vol-ume and uncertainty are key determinants of integration.Several econometrics methods in time-series research,such as GARCH and BEKK, have been used in the liter-ature on cryptocurrency trading. Conrad et al. [81] usedthe GARCH-MIDAS model to extract long and short-termvolatility components of the Bitcoin market. The technicaldetails of this model decomposed the conditional varianceinto the low-frequency and high-frequency components. Theresults identified that S&P 500 realized volatility has a nega-tive and highly significant effect on long-term Bitcoin volatil-ity and S&P 500 volatility risk premium has a significantlypositive effect on long-term Bitcoin volatility. Ardia et al. [8]used the Markov Switching GARCH (MSGARCH) modelto test the existence of institutional changes in the GARCHvolatility dynamics of Bitcoin’s logarithmic returns. More-over, a Bayesian method was used for estimating model pa-rameters and calculating VaR prediction. The results showedthat MSGARCH models clearly outperform single-regimeGARCH for Value-at-Risk forecasting. Troster et al. [239]performed general GARCH and GAS (Generalized Auto-regressive Score) analysis to model and predict Bitcoin’s re-turns and risks. The experiment found that the GAS modelwith heavy-tailed distribution can provide the best out-of-sample prediction and goodness-of-fit attributes for Bitcoin’sreturn and risk modeling. The results also illustrated theimportance of modeling excess kurtosis for Bitcoin returns.Charles et al. [65] studied four cryptocurrency markets in-cluding Bitcoin, Dash, Litecoin and Ripple. Results showedcryptocurrency returns are strongly characterised by the pres- ence of jumps as well as structural breaks except the Dashmarket. Four GARCH-type models (i.e., GARCH, APARCH,IGARCH and FIGARCH) and three return types with struc-tural breaks (original returns, jump-filtered returns, and jump-filtered returns) are considered. The research indicated theimportance of jumps in cryptocurrency volatility and struc-tural breakthroughs.Some researchers focused on long memory methods forvolatility in cryptocurrency markets. Long memory meth-ods focused on long-range dependence and significant long-term correlations among fluctuations on markets. Chaim etal. [63] estimated a multivariate stochastic volatility modelwith discontinuous jumps in cryptocurrency markets. Theresults showed that permanent volatility appears to be drivenby major market developments and popular interest levels.Caporale et al. [52] examined persistence in the cryptocur-rency market by Rescaled range (R/S) analysis and frac-tional integration. The results of the study indicated thatthe market is persistent (there is a positive correlation be-tween its past and future values) and that its level changesover time. Khuntin et al. [154] applied the adaptive markethypothesis (AMH) in the predictability of Bitcoin evolvingreturns. The consistent test of Dominguez and Lobato [89],generalized spectral (GS) of Escanciano and Velasco [98]are applied in capturing time-varying linear and nonlineardependence in bitcoin returns. The results verified EvolvingEfficiency in Bitcoin price changes and evidence of dynamicefficiency in line with AMH’s claims.Katsiampa et al. [150] applied three pair-wise bivariateBEKK models to examine the conditional volatility dynam-ics along with interlinkages and conditional correlations be-tween three pairs of cryptocurrencies in 2018. More specifi-cally, the BEKK-MGARCH methodology also captured cross-market effects of shocks and volatility, which are also knownas shock transmission effects and volatility spillover effects.The experiment found evidence of bi-directional shock trans-mission effects between Bitcoin and both Ether and Lit-coin. In particular, bi-directional shock spillover effects areidentified between three pairs (Bitcoin, Ether and Litcoin)and time-varying conditional correlations exist with positivecorrelations mostly prevailing. In 2019, Katsiampa [149]further researched an asymmetric diagonal BEKK modelto examine conditional variances of five cryptocurrenciesthat are significantly affected by both previous squared er-rors and past conditional volatility. The experiment testedthe null hypothesis of the unit root against the stationar-ity hypothesis. Once stationarity is ensured, ARCH LMis tested for ARCH effects to examine the requirement ofvolatility modeling in return series. Moreover, volatilityco-movements among cryptocurrency pairs are also testedby the multivariate GARCH model. The results confirmedthe non-normality and heteroskedasticity of price returns incryptocurrency markets. The finding also identified the ef-fects of cryptocurrencies’ volatility dynamics due to majornews. Hultman [131] set out to examine GARCH (1,1),bivariate-BEKK (1,1) and a standard stochastic model toforecast the volatility of Bitcoin. A rolling window approach
First Author et al. Page 12 of 30ryptocurrency Trading: A Comprehensive Survey is used in these experiments. Mean absolute error (MAE),Mean squared error (MSE) and Root-mean-square deviation(RMSE) are three loss criteria adopted to evaluate the de-gree of error between predicted and true values. The re-sult shows the following rank of loss functions: GARCH(1,1) > bivariate-BEKK (1,1) > Standard stochastic for allthe three different loss criteria; in other words, GARCH(1,1)appeared best in predicting the volatility of Bitcoin. Wavelettime-scale persistence analysis is also applied in the predic-tion and research of volatility in cryptocurrency markets [202].The results showed that information efficiency (efficiency)and volatility persistence in the cryptocurrency market arehighly sensitive to time scales, measures of returns and volatil-ity, and institutional changes. Adjepong et al. [202] con-nected with similar research by Corbet et al. [85] and showedthat GARCH is quicker than BEKK to absorb new informa-tion regarding the data.
As we have previously stated, Machine learning technol-ogy constructs computer algorithms that automatically im-prove themselves by finding patterns in existing data with-out explicit instructions [128]. The rapid development ofmachine learning in recent years has promoted its applica-tion to cryptocurrency trading, especially in the predictionof cryptocurrency returns.
Several machine learning technologies are applied in cryp-tocurrency trading. We distinguish these by the objective setto the algorithm: classification, clustering, regression, rein-forcement learning. We have separated a section specificallyon deep learning due to its intrinsic variation of techniquesand wide adoption.
Classification Algorithms . Classification in machinelearning has the objective of categorising incoming objectsinto different categories as needed, where we can assignlabels to each category (e.g., up and down). Naive Bayes(NB) [216], Support Vector Machine (SVM) [247], K-NearestNeighbours (KNN) [247], Decision Tree (DT) [109], Ran-dom Forest (RF) [173] and Gradient Boosting (GB) [111]algorithms habe been used in cryptocurrency trading basedon papers we collected. NB is a probabilistic classifier basedon Bayes’ theorem with strong (naive) conditional indepen-dence assumptions between features [216]. SVM is a su-pervised learning model that aims at achieving high mar-gin classifiers connecting to learning bounds theory [256].SVMs assign new examples to one category or another, mak-ing it a non-probabilistic binary linear classifier [247], al-though some corrections can make a probabilistic interpre-tation of their output [153]. KNN is a memory-based orlazy learning algorithm, where the function is only approx-imated locally, and all calculations are being postponed toinference time [247]. DT is a decision support tool algo-rithm that uses a tree-like decision graph or model to seg-ment input patterns into regions to then assign an associ-ated label to each region [109]. RF is an ensemble learn- ing method. The algorithm operates by constructing a largenumber of decision trees during training and outputting theaverage consensus as predicted class in the case of classi-fication or mean prediction value in the case of regression[173]. GB produces a prediction model in the form of anensemble of weak prediction models [111].
Clustering Algorithms . Clustering is a machine learn-ing technique that involves grouping data points in a waythat each group shows some regularity [137]. K-Means is avector quantization used for clustering analysis in data min-ing. K-means stores the 𝑘 -centroids used to define the clus-ters; a point is considered to be in a particular cluster if it iscloser to the cluster’s centroid than any other centroid [245].K-Means is one of the most used clustering algorithms usedin cryptocurrency trading according to the papers we col-lected. Regression Algorithms . We have defined regressionas any statistical technique that aims at estimating a con-tinuous value [164]. Linear Regression (LR) and Scatter-plot Smoothing are common techniques used in solving re-gression problems in cryptocurrency trading. LR is a lin-ear method used to model the relationship between a scalarresponse (or dependent variable) and one or more explana-tory variables (or independent variables) [164]. ScatterplotSmoothing is a technology to fit functions through scatterplots to best represent relationships between variables [110].
Deep Learning Algorithms . Deep learning is a moderntake on artificial neural networks (ANNs) [257], made pos-sible by the advances in computational power. An ANN isa computational system inspired by the natural neural net-works that make up the animal’s brain. The system “learns”to perform tasks including the prediction by considering ex-amples. Deep learning’s superior accuracy comes from highcomputational complexity cost. Deep learning algorithmsare currently the basis for many modern artificial intelli-gence applications [231]. Convolutional neural networks(CNNs) [168], Recurrent neural networks (RNNs) [188],Gated recurrent units (GRUs) [70], Multilayer perceptron(MLP) and Long short-term memory (LSTM) [67] networksare the most common deep learning technologies used incryptocurrency trading. A CNN is a specific type of neu-ral network layer commonly used for supervised learning.CNNs have found their best success in image processingand natural language processing problems. An attempt touse CNNs in cryptocurrency can be shown in [143]. AnRNN is a type of artificial neural network in which con-nections between nodes form a directed graph with possi-ble loops. This structure of RNNs makes them suitable forprocessing time-series data [188] due to the introduction ofmemory in the recurrent connections. They face neverthe-less for the vanishing gradients problem [203] and so dif-ferent variations have been recently proposed. LSTM [67]is a particular RNN architecture widely used. LSTMs haveshown to be superior to nongated RNNs on financial time-series problems because they have the ability to selectivelyremember patterns for a long time. A GRU [70] is anothergated version of the standard RNN which has been used in
First Author et al. Page 13 of 30ryptocurrency Trading: A Comprehensive Survey crypto trading [91]. Another deep learning technology usedin cryptocurrency trading is Seq2seq, which is a specificimplementation of the Encoder–Decoder architecture [251].Seq2seq was first aimed at solving natural language process-ing problems but has been also applied it in cryptocurrencytrend predictions in [226].
Reinforcement Learning Algorithms . Reinforcementlearning (RL) is an area of machine learning leveraging theidea that software agents act in the environment to maximizea cumulative reward [230]. Deep Q-Learning (DQN) [120]and Deep Boltzmann Machine (DBM) [219] are commontechnologies used in cryptocurrency trading using RL. DeepQ learning uses neural networks to approximate Q-valuefunctions. A state is given as input, and Q values for all pos-sible actions are generated as outputs [120]. DBM is a typeof binary paired Markov random field (undirected probabil-ity graphical model) with multiple layers of hidden randomvariables [219]. It is a network of randomly coupled randombinary units.
In the development of machine learning trading signals,technical indicators have usually been used as input fea-tures. Nakano et al. [193] explored Bitcoin intraday tech-nical trading based on ANNs for return prediction. The ex-periment obtained medium frequency price and volume data(time interval of data is 15min) of Bitcoin from a cryptocur-rency exchange. An ANN predicts the price trends (up anddown) in the next period from the input data. Data is pre-processed to construct a training dataset that contains a ma-trix of technical patterns including EMA, Emerging MarketsSmall Cap (EMSD), relative strength index (RSI), etc. Theirnumerical experiments contain different research aspects in-cluding base ANN research, effects of different layers, ef-fects of different activation functions, different outputs, dif-ferent inputs and effects of additional technical indicators.The results have shown that the use of various technical in-dicators possibly prevents over-fitting in the classificationof non-stationary financial time-series data, which enhancestrading performance compared to the primitive technical trad-ing strategy. (Buy-and-Hold is the benchmark strategy inthis experiment.)Some classification and regression machine learning mod-els are applied in cryptocurrency trading by predicting pricetrends. Most researchers have focused on the comparisonof different classification and regression machine learningmethods. Sun et al. [229] used random forests (RFs) withfactors in Alpha01 [141] (capturing features from the his-tory of the cryptocurrency market) to build a prediction model.The experiment collected data from API in cryptocurrencyexchanges and selected 5-minute frequency data for back-testing. The results showed that the performances are pro-portional to the amount of data (more data, more accurate)and the factors used in the RF model appear to have differentimportance. For example, “Alpha024” and “Alpha032” fea-tures appeared as the most important in the model adopted.(The alpha features come from paper “101 Formulaic Al- phas" [141].) Vo et al. [243] applied RFs in High-Frequencycryptocurrency Trading (HFT) and compared it with deeplearning models. Minute-level data is collected when util-ising a forward fill imputation method to replace the NULLvalue (i.e., a missing value). Different periods and RF treesare tested in the experiments. The authors also compared F-1 precision and recall metrics between RF and Deep Learn-ing (DL). The results showed that RF is effective despitemulticollinearity occurring in ML features, the lack of modelidentification also potentially leading to model identifica-tion issues; this research also attempted to create an HFTstrategy for Bitcoin using RF. Maryna et al. [260] inves-tigated the profitability of an algorithmic trading strategybased on training an SVM model to identify cryptocurren-cies with high or low predicted returns. The results showedthat the performance of the SVM strategy was the fourth be-ing better only than S&P B&H strategy, which simply buys-and-hold the S&P index. (There are other 4 benchmarkstrategies in this research.)The authors observed that SVMneeds a large number of parameters and so is very proneto overfitting, which caused its bad performance. Barnwalet al. [18] used generative and discriminative classifiers tocreate a stacking model, particularly 3 generative and 6 dis-criminative classifiers combined by a one-layer Neural Net-work, to predict the direction of cryptocurrency price. Adiscriminative classifier directly models the relationship be-tween unknown and known data, while generative classifiersmodel the prediction indirectly through the data generationdistribution [198]. Technical indicators including trend, mo-mentum, volume and volatility, are collected as features ofthe model. The authors discussed how different classifiersand features affect the prediction. Attanasio et al. [10] com-pared a variety of classification algorithms including SVM,NB and RF in predicting next-day price trends of a givencryptocurrency. The results showed that due to the hetero-geneity and volatility of cryptocurrencies’ financial instru-ments, forecasting models based on a series of forecasts ap-peared better than a single classification technology in trad-ing cryptocurrencies. Madan et al. [179] modeled the Bit-coin price prediction problem as a binomial classificationtask, experimenting with a custom algorithm that leveragesboth random forests and generalized linear models. Dailydata, 10-minute data and 10-second data are used in theexperiments. The experiments showed that 10-minute datagave a better sensitivity and specificity ratio than 10-seconddata (10-second prediction achieved around 10% accuracy).Considering predictive trading, 10-minute data helped showclearer trends in the experiment compared to 10-second back-testing. Similarly, Virk [242] compared RF, SVM, GB andLR to predict the price of Bitcoin. The results showed thatSVM achieved the highest accuracy of 62.31% and preci-sion value 0.77 among binomial classification machine learn-ing algorithms.Different deep learning models have been used in find-ing patterns of price movements in cryptocurrency markets.Zhengy et al. [258] implemented two machine learning mod-els, fully-connected ANN and LSTM to predict cryptocur-
First Author et al. Page 14 of 30ryptocurrency Trading: A Comprehensive Survey rency price dynamics. The results showed that ANN, ingeneral, outperforms LSTM although theoretically, LSTMis more suitable than ANN in terms of modeling time seriesdynamics; the performance measures considered are MAEand RMSE in joint prediction (five cryptocurrencies dailyprices prediction). The findings show that the future state ofa time series for cryptocurrencies is highly dependent on itshistoric evolution. Kwon et al. [165] used an LSTM model,with a three-dimensional price tensor representing the pastprice changes of cryptocurrencies as input. This model out-performs the GB model in terms of F1-score. Specifically,it has a performance improvement of about over the GBmodel in 10-minute price prediction. In particular, the ex-periments showed that LSTM is more suitable when classi-fying cryptocurrency data with high volatility. Alessandrettiet al. [5] tested Gradient boosting decision trees (includ-ing single regression and XGBoost-augmented regression)and the LSTM model on forecasting daily cryptocurrencyprices. They found methods based on gradient boosting de-cision trees worked best when predictions were based onshort-term windows of 5/10 days while LSTM worked bestwhen predictions were based on 50 days of data. The rel-ative importance of the features in both models are com-pared and an optimised portfolio composition (based on ge-ometric mean return and Sharpe ratio) is discussed in thispaper. Phaladisailoed et al. [207] chose regression mod-els (Theil-Sen Regression and Huber Regression) and deeplearning-based models (LSTM and GRU) to compare theperformance of predicting the rise and fall of Bitcoin price.In terms of two common measure metrics, MSE and R-Square (R ), GRU shows the best accuracy. Fan et al. [100]applied an autoencoder-augmented LSTM structure in pre-dicting the mid-price of 8 cryptocurrency pairs. Level-2limit order book live data is collected and the experimentachieved 78% accuracy of price movements prediction inhigh frequency trading (tick level). This research improvedand verified the view of Sirignano et al. [224] that univer-sal models have better performance than currency-pair spe-cific models for cryptocurrency markets. Moreover, “Walk-through” (i.e., retrain the original deep learning model it-self when it appears to no longer be valid) is proposed asa method to optimise the training of a deep learning modeland shown to significantly improve the prediction accuracy.Researchers have also focused on comparing classicalstatistical models and machine/deep learning models. Raneet al. [214] described classical time series prediction meth-ods and machine learning algorithms used for predictingBitcoin price. Statistical models such as Autoregressive In-tegrated Moving Average models (ARIMA), Binomial Gen-eralized Linear Model and GARCH are compared with ma-chine learning models such as SVM, LSTM and Non-linearAuto-Regressive with Exogenous Input Model (NARX). Theobservation and results showed that the NARX model is thebest model with nearly 52% predicting accuracy based on10 seconds interval. Rebane et al. [215] compared tradi-tional models like ARIMA with a modern popular modellike seq2seq in predicting cryptocurrency returns. The re- sult showed that the seq2seq model exhibited demonstra-ble improvement over the ARIMA model for Bitcoin-USDprediction but the seq2seq model showed very poor perfor-mance in extreme cases. The authors proposed performingadditional investigations, such as the use of LSTM insteadof GRU units to improve the performance. Similar modelswere also compared by Stuerner et al. [228] who exploredthe superiority of automated investment approach in trendfollowing and technical analysis in cryptocurrency trading.Samuel et al. [206] explored the vector autoregressive model(VAR model), a more complex RNN, and a hybrid of the twoin residual recurrent neural networks (R2N2) in predictingcryptocurrency returns. The RNN with ten hidden layers isoptimised for the setting and the neural network augmentedby VAR allows the network to be shallower, quicker andto have a better prediction than an RNN. RNN, VAR andR2N2 models are compared. The results showed that theVAR model has phenomenal test period performance andthus props up the R2N2 model, while the RNN performspoorly. This research is an attempt at optimisation of modeldesign and applying to the prediction on cryptocurrency re-turns. Sentiment analysis, a popular research topic in the age ofsocial media, has also been adopted to improve predictionsfor cryptocurrency trading. This data source typically has tobe combined with Machine Learning for the generation oftrading signals.Lamon et al. [167] used daily news and social mediadata labeled on actual price changes, rather than on positiveand negative sentiment. By this approach, the predictionon price is replaced with positive and negative sentiment.The experiment acquired cryptocurrency-related news arti-cle headlines from the website like “cryptocoinsnews” andtwitter API. Weights are taken in positive and negative wordsin the cryptocurrency market. Authors compared LogisticRegression (LR), Linear Support Vector Machine (LSVM)and NB as classifiers and concluded that LR is the bestclassifier in daily price prediction with 43.9% of price in-creases correctly predicted and 61.9% of price decreasescorrectly forecasted. Smuts [225] conducted a similar bi-nary sentiment-based price prediction method with an LSTMmodel using Google Trends and Telegram sentiment. Indetail, the sentiment was extracted from Telegram by us-ing a novel measure called VADER [132]. The backtest-ing reached 76% accuracy on the test set during the firsthalf of 2018 in predicting hourly prices. Nasir et al. [195]researched the relationship between cryptocurrency returnsand search engines. The experiment employed a rich setof established empirical approaches including VAR frame-work, copulas approach and non-parametric drawings of timeseries. The results found that Google searches exert signif-icant influence on Bitcoin returns, especially in the short-term intervals. Kristoufek [162] discussed positive and neg-ative feedback on Google trends or daily views on Wikipedia.The author mentioned different methods including Cointe-
First Author et al. Page 15 of 30ryptocurrency Trading: A Comprehensive Survey gration, Vector autoregression and Vector error-correctionmodel to find causal relationships between prices and searchedterms in the cryptocurrency market. The results indicatedthat search trends and cryptocurrency prices are connected.There is also a clear asymmetry between the effects of in-creased interest in currencies above or below their trend val-ues from the experiment. Young et al. [156] analysed usercomments and replies in online communities and their con-nection with cryptocurrency volatility. After crawling com-ments and replies in online communities, authors tagged theextent of positive and negative topics. Then the relation-ship between price and the number of transactions of cryp-tocurrency is tested according to comments and replies toselected data. At last, a prediction model using machinelearning based on selected data is created to predict fluctu-ations in the cryptocurrency market. The results show theamount of accumulated data and animated community ac-tivities exerted a direct effect on fluctuation in the price andvolume of a cryptocurrency.Phillips et al. [212] applied dynamic topic modeling andHawkes model to decipher relationships between topics andcryptocurrency price movements. The authors used LatentDirichlet allocation (LDA) model for topic modeling, whichassumes each document contains multiple topics to differentextents. The experiment showed that particular topics tendto precede certain types of price movements in the cryp-tocurrency market and the authors proposed the relation-ships could be built into real-time cryptocurrency trading.Li et al. [172] analysed Twitter sentiment and trading vol-ume and an Extreme Gradient Boosting Regression TreeModel in the prediction of ZClassic (ZCL) cryptocurrencymarket. Sentiment analysis using natural language process-ing from the Python package “Textblob” assigns impactfulwords a polarity value. Values of weighted and unweightedsentiment indices are calculated on an hourly basis by sum-ming weights of coinciding tweets, which makes us com-pare this index to ZCL price data. The model achieveda Pearson correlation of 0.806 when applied to test data,yielding a statistical significance at the 𝑝 < . level.Flori [107] relied on a Bayesian framework that combinesmarket-neutral information with subjective beliefs to con-struct diversified investment strategies in the Bitcoin mar-ket. The result shows that news and media attention seem tocontribute to influence the demand for Bitcoin and enlargethe perimeter of the potential investors, probably stimulatingprice euphoria and upwards-downwards market dynamics.The authors’ research highlighted the importance of newsin guiding portfolio re-balancing. Elie et al. [39] comparedthe ability of newspaper-based metrics and internet search-based uncertainty metrics in predicting bitcoin returns. Thepredictive power of Internet-based economic uncertainty-related query indices is statistically stronger than that ofnewspapers in predicting bitcoin returns.Similarly, Colianni et al. [80], Garcia et al. [113], Za-muda et al. [254] et al. used sentiment analysis technol-ogy applying it in the cryptocurrency trading area and hadsimilar results. Colianni et al. [80] cleaned data and ap- plied supervised machine learning algorithms such as logis-tic regression, Naive Bayes and support vector machines,etc. on Twitter Sentiment Analysis for cryptocurrency trad-ing. Garcia et al. [113] applied multidimensional analy-sis and impulse analysis in social signals of sentiment ef-fects and algorithmic trading of Bitcoin. The results veri-fied the long-standing assumption that transaction-based so-cial media sentiment has the potential to generate a posi-tive return on investment. Zamuda et al. [254] adopted newsentiment analysis indicators and used multi-target portfo-lio selection to avoid risks in cryptocurrency trading. Theperspective is rationalized based on the elastic demand forcomputing resources of the cloud infrastructure. A gen-eral model evaluating the influence between user’s networkAction-Reaction-Influence-Model (ARIM) is mentioned inthis research. Bartolucci et al. [19] researched cryptocur-rency prices with the “Butterfly effect”, which means “is-sues” of the open-source project provides insights to im-prove prediction of cryptocurrency prices. Sentiment, po-liteness, emotions analysis of GitHub comments are appliedin Ethereum and Bitcoin markets. The results showed thatthese metrics have predictive power on cryptocurrency prices. Deep reinforcement algorithms bypass prediction andgo straight to market management actions to achieve highcumulated profit [126]. Bu et al. [49] proposed a combina-tion of double Q-network and unsupervised pre-training us-ing DBM to generate and enhance the optimal Q-function incryptocurrency trading. The trading model contains agentsin series in the form of two neural networks, unsupervisedlearning modules and environments. The input market stateconnects an encoding network which includes spectral fea-ture extraction (convolution-pooling module) and temporalfeature extraction (LSTM module). A double-Q networkfollows the encoding network and actions are generated fromthis network. Compared to existing deep learning models(LSTM, CNN, MLP, etc.), this model achieved the high-est profit even facing an extreme market situation (recorded24% of the profit while cryptocurrency market price dropsby -64%). Juchli [138] applied two implementations of rein-forcement learning agents, a Q-Learning agent, which servesas the learner when no market variables are provided, anda DQN agent which was developed to handle the featurespreviously mentioned. The DQN agent was backtested un-der the application of two different neural network architec-tures. The results showed that the DQN-CNN agent (convo-lutional neural network) is superior to the DQN-MLP agent(multilayer perceptron) in backtesting prediction. Lucarelliet al. [177] focused on improving automated cryptocurrencytrading with a deep reinforcement learning approach. Dou-ble and Dueling double deep Q-learning networks are com-pared for 4 years. By setting rewards functions as Sharperatio and profit, the double Q-learning method demonstratedto be the most profitable approach in trading cryptocurrency.
First Author et al. Page 16 of 30ryptocurrency Trading: A Comprehensive Survey
Atsalakis et al. [9] proposes a computational intelligencetechnique that uses a hybrid Neuro-Fuzzy controller, namelyPATSOS, to forecast the direction in the change of the dailyprice of Bitcoin. The proposed methodology outperformstwo other computational intelligence models, the first be-ing developed with a simpler neuro-fuzzy approach, and thesecond being developed with artificial neural networks. Ac-cording to the signals of the proposed model, the investmentreturn obtained through trading simulation is 71.21% higherthan the investment return obtained through a simple buyand hold strategy. This application is proposed for the firsttime in the forecasting of Bitcoin price movements. Topo-logical data analysis is applied to forecasting price trends ofcryptocurrency markets in [155]. The approach is to har-ness topological features of attractors of dynamical systemsfor arbitrary temporal data. The results showed that themethod can effectively separate important topological pat-terns and sampling noise (like bid–ask bounces, discrete-ness of price changes, differences in trade sizes or infor-mational content of price changes, etc.) by providing theo-retical results. Kurbucz [163] designed a complex methodconsisting of single-hidden layer feedforward neural net-works (SLFNs) to (i) determine the predictive power of themost frequent edges of the transaction network (a publicledger that records all Bitcoin transactions) on the futureprice of Bitcoin; and, (ii) to provide an efficient techniquefor applying this untapped dataset in day trading. The re-search found a significantly high accuracy (60.05%) for theprice movement classifications base on information that canbe obtained using a small subset of edges (approximately0.45% of all unique edges). It is worth noting that, Kondoret al. [157, 159] firstly published some papers giving analy-sis on transaction networks on cryptocurrency markets andapplied related research in identifying Bitcoin users [139].Abay et al. [2] attempted to understand the network dynam-ics behind the Blockchain graphs using topological features.The results showed that standard graph features such as thedegree distribution of transaction graphs may not be suffi-cient to capture network dynamics and their potential im-pact on Bitcoin price fluctuations. Maurice et al [202] ap-plied wavelet time-scale persistence in analysing returns andvolatility in cryptocurrency markets. The experiment exam-ined the long-memory and market efficiency characteristicsin cryptocurrency markets using daily data for more thantwo years. The authors employed a log-periodogram re-gression method in researching stationarity in the cryptocur-rency market and used ARFIMA-FIGARCH class of mod-els in examining long-memory behaviour of cryptocurren-cies across time and scale. In general, experiments indicatedthat heterogeneous memory behaviour existed in eight cryp-tocurrency markets using daily data over the full-time periodand across scales (August 25, 2015 to March 13, 2018).
8. Portfolio and Cryptocurrency Assets
Ji et al. [135] examined connectedness via return andvolatility spillovers across six large cryptocurrencies (col-lected from coinmarketcap lists from August 7 2015 to Febru-ary 22 2018) and found Litecoin and Bitcoin to have themost effect on other cryptocurrencies. The authors followedmethods of Diebold et al. [88] and built positive/negative re-turns and volatility connectedness networks. Furthermore,the regression model is used to identify drivers of variouscryptocurrency integration levels. Further analysis revealedthat the relationship between each cryptocurrency in termsof return and volatility is not necessarily due to its mar-ket size. Adjepong et al. [201] explored market coherenceand volatility causal linkages of seven leading cryptocurren-cies. Wavelet-based methods are used to examine marketconnectedness. Parametric and nonparametric tests are em-ployed to investigate directions of volatility spillovers of theassets. Experiments revealed from diversification benefits tolinkages of connectedness and volatility in cryptocurrencymarkets. Elie et al. [43] found the presence of jumps wasdetected in a series of 12 cryptocurrency returns, and signif-icant jumping activity was found in all cases. More resultsunderscore the importance of the jump in trading volume forthe formation of cryptocurrency leapfrogging.Some researchers explored the relationship between cryp-tocurrency and different factors, including futures, gold, etc.Hale et al. [123] suggested that Bitcoin prices rise and fallrapidly after CME issues futures consistent with pricing dy-namics. Specifically, the authors pointed out that the rapidrise and subsequent decline in prices after the introductionof futures is consistent with trading behaviour in the cryp-tocurrency market. Werner et al. [161] focused on the asym-metric interrelationships between major currencies and cryp-tocurrencies. The results of multiple fractal asymmetric de-trending cross-correlation analysis show evidence of signif-icant persistence and asymmetric multiplicity in the cross-correlation between most cryptocurrency pairs and ETF pairs.Bai et al. [14] studied a trading algorithm for foreign ex-change on a cryptocurrency Market using the AutomatedTriangular Arbitrage method. Implementing a pricing strat-egy, implementing trading algorithms and developing a giventrading simulation are three problems solved by this research.Kang et al. [146] examined the hedging and diversificationproperties of gold futures versus Bitcoin prices by usingdynamic conditional correlations (DCCs) and wavelet co-herence. DCC-GARCH model [95] is used to estimate thetime-varying correlation between Bitcoin and gold futuresby modeling the variance and the co-variance but also thistwo flexibility. Wavelet coherence method focused more onco-movement between Bitcoin and gold futures. From ex-periments, the wavelet coherence results indicated volatilitypersistence, causality and phase difference between Bitcoinand gold. Dyhrberg et al [92] applied the GARCH modeland the exponential GARCH model in analysing similarities
First Author et al. Page 17 of 30ryptocurrency Trading: A Comprehensive Survey between Bitcoin, gold and the US dollar. The experimentsshowed that Bitcoin, gold and the US dollar have similari-ties with the variables of the GARCH model, have similarhedging capabilities and react symmetrically to good andbad news. The authors observed that Bitcoin can combinesome advantages of commodities and currencies in finan-cial markets to be a tool for portfolio management. Baur etal. [20] extended the research of Dyhrberg et al.; the samedata and sample periods are tested [92] with GARCH andEGARCH-(1,1) models but the experiments reached differ-ent conclusions. Baur et al. found that Bitcoin has uniquerisk-return characteristics compared with other assets. Theynoticed that Bitcoin excess returns and volatility resemble arather highly speculative asset with respect to gold or theUS dollar. Bouri et al. [40] studied the relationship be-tween Bitcoin and energy commodities by applying DCCsand GARCH (1,1) models. In particular, the results showedthat Bitcoin is a strong hedge and safe haven for energycommodities. Kakushadze [142] proposed factor models forthe cross-section of daily cryptoasset returns and providedsource code for data downloads, computing risk factors andbacktesting for all cryptocurrencies and a host of variousother digital assets. The results showed that cross-sectionalstatistical arbitrage trading may be possible for cryptoas-sets subject to efficient executions and shorting. Beneki etal. [25] tested hedging abilities between Bitcoin and Ethereumby a multivariate BEKK-GARCH methodology and impulseresponse analysis within VAR model. The results indicateda volatility transaction from Ethereum to Bitcoin, which im-plied possible profitable trading strategies on the cryptocur-rency derivatives market. Guglielmo et al. [54] examinedthe week effect in cryptocurrency markets and explored thefeasibility of this indicator in trading practice. Student 𝑡 -test,ANOVA, Kruskal–Wallis and Mann–Whitney tests were car-ried out for cryptocurrency data in order to compare timeperiods that may be characterised by anomalies with othertime periods. When an anomaly is detected, an algorithmwas established to exploit profit opportunities (MetaTraderterminal in MQL4 is mentioned in this research). The re-sults showed evidence of anomaly (abnormal positive re-turns on Mondays) in the Bitcoin market by backtesting in2013-2016. Some researchers applied portfolio theory for crypto as-sets. Corbet et al. [83] gave a systematic analysis of cryp-tocurrencies as financial assets. Brauneis et al. [46] ap-plied the Markowitz mean-variance framework in order toassess the risk-return benefits of cryptocurrency portfolios.In an out-of-sample analysis accounting for transaction cost,they found that combining cryptocurrencies enriches the setof ‘low’-risk cryptocurrency investment opportunities. Interms of the Sharpe ratio and certainty equivalent returns,the 𝑁 -portfolio (i.e., “naive” strategies, such as equallydividing amongst asset classes) outperformed single cryp-tocurrencies and more than 75% in terms of the Sharpe ratioand certainty equivalent returns of mean-variance optimal portfolios. Castro et al. [56] developed a portfolio optimi-sation model based on the Omega measure which is morecomprehensive than the Markowitz model and applied thisto four crypto-asset investment portfolios by means of a nu-merical application. Experiments showed crypto-assets im-proves the return of the portfolios, but on the other hand,also increase the risk exposure.Bedi et al. [22] examined diversification capabilities ofBitcoin for a global portfolio spread across six asset classesfrom the standpoint of investors dealing in five major fiatcurrencies, namely US Dollar, Great Britain Pound, Euro,Japanese Yen and Chinese Yuan. They employed modifiedConditional Value-at-Risk and standard deviation as mea-sures of risk to perform portfolio optimisations across threeasset allocation strategies and provided insights into the sharpdisparity in Bitcoin trading volumes across national curren-cies from a portfolio theory perspective. Similar researchhas been done by Antipova et al. [7], which explored thepossibility of establishing and optimizing a global portfolioby diversifying investments using one or more cryptocur-rencies, and assessing returns to investors in terms of risksand returns. Fantazzini et al. [102] proposed a set of modelsthat can be used to estimate the market risk for a portfo-lio of crypto-currencies, and simultaneously estimate theircredit risk using the Zero Price Probability (ZPP) model.The results revealed the superiority of the t-copula/skewed-tGARCH model for market risk, and the ZPP-based modelsfor credit risk. Qiang et al. [134] examined the commondynamics of bitcoin exchanges. Using a connectivity met-ric based on the actual daily volatility of the bitcoin price,they found that Coinbase is undoubtedly the market leader,while Binance performance is surprisingly weak. The re-sults also suggested that safer asset extraction is more im-portant for volatility linkages between Bitcoin exchangesrelative to trading volumes.Trucios et al. [240] proposed a methodology based onvine copulas and robust volatility models to estimate theValue-at-Risk (VaR) and Expected Shortfall (ES) of cryp-tocurrency portfolios. The proposed algorithm displayedgood performance in estimating both VaR and ES. Hrytsiuket al. [129] showed that the cryptocurrency returns can bedescribed by the Cauchy distribution and obtained the an-alytical expressions for VaR risk measures and performedcalculations accordingly. As a result of the optimisation,the sets of optimal cryptocurrency portfolios were built intheir experiments.Jiang et al. [136] proposed a two-hidden-layer CNN thattakes the historical price of a group of cryptocurrency assetsas an input and outputs the weight of the group of cryp-tocurrency assets. This research focused on portfolio re-search in cryptocurrency assets using emerging technolo-gies like CNN. Training is conducted in an intensive man-ner to maximise cumulative returns, which is considered areward function of the CNN network. The performance ofthe CNN strategy is compared with the three benchmarksand the other three portfolio management algorithms (buyand hold strategy, Uniform Constant Rebalanced Portfolio First Author et al. Page 18 of 30ryptocurrency Trading: A Comprehensive Survey and Universal Portfolio with Online Newton Step and Pas-sive Aggressive Mean Reversion); the results are positivein that the model is only second to the Passive AggressiveMean Reversion algorithm (PAMR). Estalayo et al. [99] re-ported initial findings around the combination of DL mod-els and Multi-Objective Evolutionary Algorithms (MOEAs)for allocating cryptocurrency portfolios. Technical rationaleand details were given on the design of a stacked DL recur-rent neural network, and how its predictive power can be ex-ploited for yielding accurate ex-ante estimates of the returnand risk of the portfolio. Results obtained for a set of exper-iments carried out with real cryptocurrency data have veri-fied the superior performance of their designed deep learn-ing model with respect to other regression techniques.
9. Market Condition Research
Phillips and Yu proposed a methodology to test for thepresence of cryptocurrency bubble [68], which is extendedby Shaen et al. [84]. The method is based on supremumAugmented Dickey–Fuller (SADF) to test for the bubblethrough the inclusion of a sequence of forwarding recur-sive right-tailed ADF unit root tests. An extended method-ology generalised SADF (GSAFD), is also tested for bub-bles within cryptocurrency data. The research concludedthat there is no clear evidence of a persistent bubble in cryp-tocurrency markets including Bitcoin or Ethereum. Bouriet al. [44] date-stamped price explosiveness in seven largecryptocurrencies and revealed evidence of multiple periodsof explosivity in all cases. GSADF is used to identify mul-tiple explosiveness periods and logistic regression is em-ployed to uncover evidence of co-explosivity across cryp-tocurrencies. The results showed that the likelihood of ex-plosive periods in one cryptocurrency generally depends onthe presence of explosivity in other cryptocurrencies andpoints toward a contemporaneous co-explosivity that doesnot necessarily depend on the size of each cryptocurrency.Extended research by Phillips et al. [208, 209] (who ap-plied a recursive augmented Dickey-Fuller algorithm, whichis called PSY test) and Landsnes et al. [97] studied pos-sible predictors of bubble periods of certain cryptocurren-cies. The evaluation includes multiple bubble periods in allcryptocurrencies. The result shows that higher volatility andtrading volume is positively associated with the presence ofbubbles across cryptocurrencies. In terms of bubble predic-tion, the authors found the probit model to perform betterthan the linear models.Phillips et al. [210] used Hidden Markov Model (HMM)and Superiority and Inferiority Ranking (SIR) method toidentify bubble-like behaviour in cryptocurrency time se-ries. Considering HMM and SIR method, an epidemic de-tection mechanism is used in social media to predict cryp-tocurrency price bubbles, which classify bubbles throughepidemic and non-epidemic labels. Experiments have demon-strated a strong relationship between Reddit usage and cryp-tocurrency prices. This work also provides some empirical evidence that bubbles mirror the social epidemic-like spreadof an investment idea. Guglielmo et al. [53] examined theprice overreactions in the case of cryptocurrency trading.Some parametric and non-parametric tests confirmed the pres-ence of price patterns after overreactions, which identifiedthat the next-day price changes in both directions are biggerthan after “normal” days. The results also showed that theoverreaction detected in the cryptocurrency market wouldnot give available profit opportunities (possibly due to trans-action costs) that cannot be considered as evidence of theEMH. Chaim et al. [62] analysed the high unconditionalvolatility of cryptocurrency from a standard log-normal stochas-tic volatility model to discontinuous jumps of volatility andreturns. The experiment indicated the importance of in-corporating permanent jumps to volatility in cryptocurrencymarkets.
Differently from traditional fiat currencies, cryptocur-rencies are risky and exhibit heavier tail behaviour. Paraskeviet al. [151] found extreme dependence between returns andtrading volumes. Evidence of asymmetric return-volume re-lationship in the cryptocurrency market was also found bythe experiment, as a result of discrepancies in the correlationbetween positive and negative return exceedances across allthe cryptocurrencies.There has been a price crash in late 2017 to early 2018 incryptocurrency [253]. Yaya et al. [253] researched the per-sistence and dependence of Bitcoin on other popular alter-native coins before and after the 2017/18 crash in cryptocur-rency markets. The result showed that higher persistence ofshocks is expected after the crash due to speculations in themind of cryptocurrency traders, and more evidence of non-mean reversions, implying chances of further price fall incryptocurrencies.
10. Others related to Cryptocurrency Trading
Some other research papers related to cryptocurrencytrading treat distributed in market behaviour, regulatory mech-anisms and benchmarks.Krafft et al. [160] and Yang [252] analysed market dy-namics and behavioural anomalies respectively to under-stand effects of market behaviour in the cryptocurrency mar-ket. Krafft et al. discussed potential ultimate causes, poten-tial behavioural mechanisms and potential moderating con-textual factors to enumerate possible influence of GUI andAPI on cryptocurrency markets. Then they highlighted thepotential social and economic impact of human-computerinteraction in digital agency design. Yang, on the otherhand, applied behavioural theories of asset pricing anoma-lies in testing 20 market anomalies using cryptocurrencytrading data. The results showed that anomaly research fo-cused more on the role of speculators, which gave a newidea to research the momentum and reversal in the cryp-tocurrency market. Cocco et al. [75] implemented a mech-anism to form a Bitcoin price and specific behaviour foreach type of trader including the initial wealth distribution
First Author et al. Page 19 of 30ryptocurrency Trading: A Comprehensive Survey following Pareto’s law, order-based transaction and pricesettlement mechanism. Specifically, the model reproducedthe unit root attributes of the price series, the fat tail phe-nomenon, the volatility clustering of price returns, the gen-eration of Bitcoins, hashing power and power consumption.Leclair [169] and Vidal-Thomás et al. [241] analysed theexistence of herding in the cryptocurrency market. Leclairapplied herding methods of Huang and Salmon [133] in esti-mating the market herd dynamics in the CAPM framework.Vidal-Thomás et al. analyse the existence of herds in thecryptocurrency market by returning the cross-sectional stan-dard (absolute) deviations. Both their findings showed sig-nificant evidence of market herding in the cryptocurrencymarket. Makarov et al. [180] studied price impact and arbi-trage dynamics in the cryptocurrency market and found that85% of the variations in bitcoin returns and the idiosyncraticcomponents of order flow play an important role in explain-ing the size of the arbitrage spreads between exchanges.In November 2019, Griffin et al. put forward a paperon the thesis of unsupported digital money inflating cryp-tocurrency prices [119], which caused a great stir in the aca-demic circle and public opinion. Using algorithms to anal-yse Blockchain data, they found that purchases with Tetherare timed following market downturns and result in sizeableincreases in Bitcoin prices. By mapping the blockchains ofBitcoin and Tether, they were able to establish that one largeplayer on Bitfinex uses Tether to purchase large amounts ofBitcoin when prices are falling and following the prod ofTether.More researches involved benchmark and developmentin cryptocurrency market [127, 259], regulatory frameworkanalysis [220], data mining technology in cryptocurrencytrading [204], application of efficient market hypothesis inthe cryptocurrency market [223] and artificial financial mar-kets for studying a cryptocurrency market [74]. Hilemanet al. [127] segmented the cryptocurrency industry into fourkey sectors: exchanges, wallets, payments and mining. Theygave a benchmarking study of individuals, data, regulation,compliance practices, costs of firms and a global map ofmining in the cryptocurrency market in 2017. Zhou et al. [259]discussed the status and future of computer trading in thelargest group of Asia-Pacific economies and then consid-ered algorithmic and high frequency trading in cryptocur-rency markets as well. Shanaev et al. [220] used data on120 regulatory events to study the implications of cryptocur-rency regulation and the results showed that stricter regula-tion of cryptocurrency is not desirable. Akhilesh et al. [204]used the average absolute error calculated between the ac-tual and predicted values of the market sentiment of differ-ent cryptocurrencies on that day as a method for quantifyingthe uncertainty. They used the comparison of uncertaintyquantification methods and opinion mining to analyse cur-rent market conditions. Sigaki et al. [223] used permutationentropy and statistical complexity on the sliding time win-dow returned by the price log to quantify the dynamic ef-ficiency of more than four hundred cryptocurrencies. As aresult, the cryptocurrency market showed significant com- pliance with efficient market assumptions. Cocco et al. [74]described an agent-based artificial cryptocurrency market inwhich heterogeneous agents buy or sell cryptocurrencies.The proposed simulator is able to reproduce some real sta-tistical properties of price returns observed in the Bitcoinreal market. Marko [200] considered the future use of cryp-tocurrencies as money based on the long-term value of cryp-tocurrencies. Neil et al. [112] analysed the influence of net-work effect on the competition of new cryptocurrency mar-kets. Bariviera and Merediz-Sola [17] gave a survey basedon hybrid analysis, which proposed a methodological hybridmethod for a comprehensive literature review and providedthe latest technology in the cryptocurrency economics liter-ature.There also exists some research and papers introducingthe basic process and rules of cryptocurrency trading in-cluding findings of Hansel et al. [124], Kate [148], Garzaet al. [114], Ward et al. [248] and Fantazzini et al. [101].Hansel et al. [124] introduced the basics of cryptocurrency,Bitcoin and Blockchain, ways to identify the profitable trendsin the market, ways to use Altcoin trading platforms suchas GDAX and Coinbase, methods of using a crypto wal-let to store and protect the coins in their book. Kate etal. [148] set six steps to show how to start an investmentwithout any technical skills in the cryptocurrency market.This book is an entry-level trading manual for starters learn-ing cryptocurrency trading. Garza et al. [114] simulatedan automatic cryptocurrency trading system, which helpsinvestors limit systemic risks and improve market returns.This paper is an example to start designing an automaticcryptocurrency trading system. Ward et al. [248] discussedalgorithmic cryptocurrency trading using several general al-gorithms, and modifications thereof including adjusting theparameters used in each strategy, as well as mixing mul-tiple strategies or dynamically changing between strategies.This paper is an example to start algorithmic trading in cryp-tocurrency market. Fantazzini et al. [101] introduced theR packages Bitcoin-Finance and bubble, including financialanalysis of cryptocurrency markets including Bitcoin.A community resource, that is, a platform for scholarlycommunication, about cryptocurrencies and Blockchains is“Blockchain research network", see [197].
11. Summary Analysis of Literature Review
This section analyses the timeline, the research distribu-tion among technology and methods, the research distribu-tion among properties. It also summarises the datasets thathave been used in cryptocurrency trading research.
Figure 8 shows several major events in cryptocurrencytrading. The timeline contains milestone events in cryp-tocurrency trading and important scientific breakthroughs inthis area.As early as 2009, Satoshi Nakamoto proposed and in-vented the first decentralised cryptocurrency, Bitcoin [192].It is considered to be the start of cryptocurrency. In 2010,
First Author et al. Page 20 of 30ryptocurrency Trading: A Comprehensive Survey
Figure 8:
Timeline of cryptocurrency trading research the first cryptocurrency exchange was founded, which meanscryptocurrency would not be an OTC market but traded onexchanges based on an auction market system.In 2013, Kristoufek [162] concluded that there is a strongcorrelation between Bitcoin price and the frequency of “Bit-coin” search queries in Google Trends and Wikipedia. In2014, Lee and Yang [170] firstly proposed to check causal-ity from copula-based causality in the quantile method fromtrading volumes of seven major cryptocurrencies to returnsand volatility.In 2015, Cheah et al. [66] discussed the bubble and spec-ulation of Bitcoin and cryptocurrencies. In 2016, Dyhrbergexplored Bitcoin volatility using GARCH models combinedwith gold and US dollars [92].From late 2016 to 2017, machine learning and deep learn-ing technology were applied in the prediction of cryptocur-rency return. In 2016, McNally et al. [184] predicted Bit-coin price using the LSTM algorithm. Bell and Zbikowskiet al. [23, 255] applied SVM algorithm to predict trends ofcryptocurrency price. In 2017, Jiang et al. [136] used doubleQ-network and pre-trained it using DBM for the predictionof cryptocurrencies portfolio weights.In recent years, several research directions including crossasset portfolios [22, 56, 46], transaction network applica-tions [163, 44], machine learning optimisation [214, 9, 258]have been considered in the cryptocurrency trading area.
We counted the number of papers covering different as-pects of cryptocurrency trading. Figure 9 shows the result.The attributes in the legend are ranked according to the num-ber of papers that specifically test the attribute.Over one-third (38.10%) of the papers research predic-tion of returns. Another one-third of papers focus on re-searching bubbles and extreme conditions and the relation-ship between pairs and portfolios in cryptocurrency trading.The remaining researching topics (prediction of volatility,trading system, technical trading and others) have roughlyone-third share.
This section introduces and compares categories and tech-nologies in cryptocurrency trading. When papers cover mul-tiple technologies or compare different methods, we drawstatistics from different technical perspectives.Among all the 126 papers, 87 papers (69.05%) cover sta-tistical methods and machine learning categories. These pa-
Figure 9:
Research distribution among cryptocurrency trad-ing properties
Table 7
Search hits of research distribution in all trading areas
Technology Category Google Scholar hits Google hits Arxiv hitsStatistical methods 1.22M 62M 1240Machine learning methods 483K 150M 520 pers basically research technical-level cryptocurrency trad-ing including mathematical modeling and statistics. Otherpapers related to trading systems on pure technical indica-tors and introducing the industry and its history are not in-cluded in this analysis. Among all 87 papers, 75 papers(86.21%) present statistical methods and technologies in cryp-tocurrency trading research and 13.8% papers research ma-chine learning applied to cryptocurrency trading (cf. Fig-ure 10). It is interesting to mention that, there are 16 pa-pers (18.4%) applying and comparing more than one tech-nique in cryptocurrency trading. More specifically, Bach etal. [12], Alessandretti et al. [5], Vo et al. [243], Phaladis-ailoed et al. [207], Siaminos [222], Rane et al. [214] usedboth statistical methods and machine learning methods incryptocurrency trading.Table 7 shows the results of search hits in all trading ar-eas (not limited to cryptocurrencies). From the table, we cansee that most research findings focused on statistical meth-ods in trading, which means most of the research on tradi-tional markets still focused on using statistical methods fortrading. But we observed that machine learning in tradinghad a higher degree of attention. It might because the tra-ditional technical and fundamental have been arbitraged, sothe market has moved in recent years to find new anomaliesto exploit. Meanwhile, the results also showed there existmany opportunities for research in the widely studied areasof machine learning applied to trade in cryptocurrency mar-kets (cf. Section 12).
As from Figure 10, we further classified the papers us-ing statistical methods into 6 categories: (i) basic regres-sion methods; (ii) linear classifiers and clustering; (iii) time-series analysis; (iv) decision trees and probabilistic classi-fiers; (v) modern portfolios theory; and, (vi) Others.
First Author et al. Page 21 of 30ryptocurrency Trading: A Comprehensive Survey
Figure 10:
Research distribution among cryptocurrencytrading technologies and methods
Basic regression methods include regression methods(Linear Regression), function estimation and CGCD method.
Linear Classifiers and Clustering include SVM and KNNalgorithm.
Time-series analysis include GARCH model,BEKK model, ARIMA model, Wavelet time-scale method.
Decision Trees and probabilistic classifiers include Boost-ing Tree, RF model.
Modern portfolio theory include Value-at-Risk (VaR) theory, expected-shortfall (ES), Markowitzmean-variance framework.
Others include industry, marketdata and research analysis in cryptocurrency market.The figure shows that basic Regression methods and time-series analysis are the most commonly used methods in thisarea.
Papers using machine learning account for 22.78% (c.fFigure 10) of the total. We further classified these papersinto three categories: (vii) ANNs, (viii) LSTM/RNN/GRUs,and (ix) DL/RL.The figure also shows that methods based on LSTM,RNN and GRU are the most popular in this subfield.
ANNs contains papers researching ANN applications incryptocurrency trading such as back propagation (BP) NN.
LSTM/RNN/GRUs include papers using neural networksthat exploit the temporal structure of the data, a technologyespecially suitable for time series prediction and financialtrading.
DL/RL includes papers using Multilayer NeuralNetworks and Reinforcement Learning. The difference be-tween ANN and DL is that generally, DL refers to an ANNwith multiple hidden layers while ANN refers to simplestructure neural network contained input layer, hidden layer(one or multiple), and an output layer.
Tables 8–10 show the details for some representativedatasets used in cryptocurrency trading research. Table 8shows the market datasets. They mostly include price, trad- ing volume, order-level information, collected from cryp-tocurrency exchanges. Table 9 shows the sentiment-baseddata. Most of the datasets in this table contain market dataand 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. Thecolumn “Description” shows a general description and typesof datasets. The column “Data Resolution” means latencyof the data (e.g., used in the backtest) – this is useful to dis-tinguish between high-frequency trading and low-frequencytrading. The column “Time range” shows the time span ofdatasets used in experiments; this is convenient to distin-guish between the current performance in a specific timeinterval and the long-term effect. We also present how thedataset has been used (i.e., the task), cf. column “Usage”.“Data Sources” gives details on where the data is retrievedfrom, including cryptocurrency exchanges, aggregated cryp-tocurrency index and user forums (for sentiment analysis).Alexander et al. [6] made an investigation of cryptocur-rency data as well. They summarised data collected from152 published and SSRN discussion papers about cryptocur-rencies and analysed their data quality. They found that lessthan half the cryptocurrency papers published since January2017 employ correct data.
12. Opportunities in Cryptocurrency Trading
This section discusses potential opportunities for futureresearch in cryptocurrency trading.
Sentiment-based research . As discussed above, thereis a substantial body of work, which uses natural languageprocessing technology, for sentiment analysis with the ulti-mate goal of using news and media contents to improve theperformance of cryptocurrency trading strategies.Possible research directions may lie in a larger volumeof media input (e.g., adding video sources) in sentimentanalysis; updating baseline natural language processing modelto perform more robust text preprocessing; applying neu-ral networks in label training; extending samples in termsof holding period; transaction-fees; and, user reputation re-search.
Long-and-short term research . There are significantdifferences between long and short time horizons in cryp-tocurrency trading. In long-term trading, investors mightobtain greater profits but have more possibilities to controlrisk when managing a position for weeks or months. It ismandatory to control for risk on long term strategies due tothe increase in the holding period, directly proportional tothe risk incurred by the trader. On the other hand, the longerthe horizon, the higher the risk and the most important therisk control. The shorter the horizon, the higher the cost andthe lower the risk, so cost takes over the design of a strat-egy. In short-term trading, automated algorithmic tradingcan be applied when holding periods are less than a week.
First Author et al. Page 22 of 30ryptocurrency Trading: A Comprehensive Survey
Table 8
Datasets (1/3):Market Data
Research Source Description Currency Data Resolution Time Range Usage Data Sources
Bouri et al. [41] price,volatility,detrended volume data Bitcoin,Ethereum,5 other cryptocurrencies daily From: 2013/01/01To: 2017/12/31 Prediction of volatility/return CoinMarketCapNakano et al. [193] high frequency price,volume data Bitcoin 15min From: 2016/07/31To: 2018/01/24 Prediction of return PoloniexBu et al. [49] three pieces time slice fordifferent research objectives Bitcoin and seven altcoins Not mentioned From: 2016/05/14To: 2016/07/03From: 2018/01/01To: 2018/01/31From: 2017/07/01To: 2017/07/31 Maximum profit with DRL Not mentionedSun et al. [229] price, volatility ETC-USDT,other 12 cryptocurrencies 1 minute,5 minutes,30 minutes,one hour,one day From: August 2017To: December 2018 Prediction of return Binance, BitfinexGuo et al. [121] volatility,order book data Bitcoin hourly volatility observations,order book snapshots From: September 2015To: April 2017 Prediction of volatility Not mentionedVo et al. [243] timestamps,the OHLC prices etc. Bitcoin 1minute From: Starting 2015To: End 2016 Prediction of return Bitstamp, Btce, Btcn,Coinbase, Coincheck, and KrakenRoss et al. [210] price Bitcoin,other 3 cryptocurrencies daily From: April 2015To: September 2016 Predicting bubbles CryptoCompareYaya et al. [253] price Bitcoin,other 12 cryptocurrencies daily From: 2015/08/07To: 2018/11/28 Bubbles and crashes Coin MetricsBrauneis et al. [46] individual price,trading volume 500 most capitalizedCryptocurrencies daily From: 2015/01/01To: 2017/12/31 Portfolios management CoinMarketCapFeng et al. [104] order-level USD/BTCtrading data Bitcoin order-level From: 2011/09/13To: 2017/07/17 Trading strategy Bitstamp
Table 9
Datasets (2/3):Sentiment-based data
Research Source Description Currency Time range Usage Data Sources
Kim et al. [156] Online cryptocurrency communities dataand market data Bitcoin,Ethereum, Ripple From: December 2013To: August, 2016 (Bitcoin)From: August 2015To: August, 2016 (Ethereum)From: CreationTo: August, 2016 (Ripple) Prediction of fluctuation Each community’s HTML pagePhillips et al. [212] Social media data and orice data Bitcoin and Ethereum From: 2016/08/30To: 2017/08/30 Predict Mutual-Excitation ofCryptocurrency Market Returns RedditSmtus [225] Hourly data on price and trading volumeand search terms from Google Trends Bitcoin, Ethereumand their respective pricedrivers From: 2017/12/01To: 2018/06/30 Prediction of price Google Trends, Telegram chat groupsLamon et al. [167] Daily price data and cryptocurrencyrelated news article headlines Bitcoin, Ethereum, Litecoin From: 2017/01/01To: 2017/11/30 Prediction of price Kaggle, news headlinePhillips et al. [211] Price and social media factors from Reddit Bitcoin, Ethereum, Monero From: 2010/09/10To: 2017/05/31 (Bitcoin)Others can reference the paper Waveletcoherence analysis of price BraveNewCoinKang et al. [145] Market data and posts and commentsincluding metadata Bitcoin From: 2009/11/22To: 2018/02/02 Relationships Between BitcoinPrices and User Groups inOnline Community Bitcoin forum
Researchers can differentiate between long-term and short-term trading in cryptocurrency trading by applying wavelettechnology analysing bubble regimes [211] and consider-ing price explosiveness [44] hypotheses for short-term and long-term research.The existing work is mainly about showing the differ-ences between long and short-term cryptocurrency trading.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 allBitcoin transactions and daily returns From: 2016/11/09To: 2018/02/05 Predicting the price of Bitcoinwith transaction network Bitcoin network dataset [189]Bedi et al. [22] A diversified portfolio including equity,fixedincome, real estate, alternativeinvestments, commodities and money market From: July 2010To: December 2018 Cross-currency including cryptocurrencyresearching portfolios Portfolio sources [22]
First Author et al. Page 23 of 30ryptocurrency Trading: A Comprehensive Survey ing and simple technical indicators in market analysis. Short-term trading can limit overall risk because small positionsare used in every transaction. But market noise (interfer-ence) and short transaction time might cause some stress inshort term trading. It might also be interesting to explorethe extraction of trading signals, time series research, appli-cation to portfolio management, the relationship between ahuge market crash and small price drop, derivative pricingin cryptocurrency market, etc.
Correlation between cryptocurrency and others . Bythe effects of monetary policy and business cycles that arenot controlled by the central bank, cryptocurrency is alwaysnegatively correlated with overall financial market trends.There have been some studies discussing correlations be-tween cryptocurrencies and other financial markets [146,56], which can be used to predict the direction of the cryp-tocurrency market.Considering the characteristics of cryptocurrency, thecorrelation between cryptocurrency and other assets still re-quires further research. Possible breakthroughs might beachieved with principal component analysis, the relation-ship between cryptocurrency and other currencies in extremeconditions (i.e., financial collapse).
Bubbles and crash research . To discuss the high volatil-ity and return of cryptocurrencies, current research has fo-cussed on bubbles of cryptocurrency markets [68], corre-lation between cryptocurrency bubbles and indicators likevolatility index (VIX) [97] (which is a “panic index” to mea-sure the implied volatility of S&P500 Index Options), spillovereffects in cryptocurrency market [178].Additional research for bubbles and crashes in cryptocur-rency trading could include a connection between the pro-cess of bubble generation and financial collapse and con-ducting a coherent analysis (coherent process analysis fromthe formation of bubbles to aftermath analysis of bubbleburst), analysis of bubble theory by Microeconomics, tryingother physical or industrial models in analysing bubbles incryptocurrency market (i.e.,
Omori law [249]), discussingthe supply and demand relationship of cryptocurrency inbubble analysis (like using supply and demand graph to sim-ulate the generation of bubbles and simulate the bubble burst).
Game theory and agent-based analysis . Applying gametheory or agent-based modelling in trading is a hot researchdirection in the traditional financial market. It might also beinteresting to apply this method to trading in cryptocurrencymarkets.
Public nature of Blockchain technology . Investiga-tions on the connections between the formation of a givencurrency’s transaction network and its price has increasedrapidly in recent years; the growing attention on user iden-tification [139] also strongly supports this direction. Withan in-depth understanding of these networks, we may iden-tify new features in price prediction and may be closer tounderstanding financial bubbles in cryptocurrency trading.
Balance between the opening of trading research lit-erature and the fading of alphas . Mclean et al. [183]pointed out that investors learn about mispricing in stock markets from academic publications. Similarly, cryptocur-rency market predictability could also be affected by re-search papers in the area. A possible attempt is to try newpricing methods applying real-time market changes. Con-sidering the proportion of informed traders increasing inthe cryptocurrency market in the pricing process is anotherbreaking point (looking for a balance between alpha tradingand trading research literature).
13. Conclusions
We provided a comprehensive overview and analysis ofthe research work on cryptocurrency trading. This surveypresented a nomenclature of the definitions and current stateof the art. The paper provides a comprehensive survey of126 cryptocurrency trading papers and analyses the researchdistribution that characterise the cryptocurrency trading lit-erature. We further summarised the datasets used for exper-iments and analysed the research trends and opportunities incryptocurrency trading.We expect this survey to be beneficial to academics (e.g.,finance researchers) and quantitative traders alike. The sur-vey represents a quick way to get familiar with the litera-ture on cryptocurrency trading and can motivate more re-searchers to contribute to the pressing problems in the area,for example along the lines we have identified.
References [1] 3commas, 2020. 3commas smart trading terminal and auto tradingbots. https://3commas . io/. [Online, Accessed: January 26, 2020].[2] Abay, N.C., Akcora, C.G., Gel, Y.R., Kantarcioglu, M., Islambekov,U.D., Tian, Y., Thuraisingham, B., 2019. Chainnet: Learning onblockchain graphs with topological features, in: 2019 IEEE Interna-tional Conference on Data Mining (ICDM), IEEE. pp. 946–951.[3] Adeyanju, C., 2019. What crypto exchanges do to comply withkyc, aml and cft regulations. https://cointelegraph . . fca . org . First Author et al. Page 24 of 30ryptocurrency Trading: A Comprehensive Survey . bakkt . . bakkt . . binance . com/en/blog/386484403820867584/Binance-Partners-with-Coinfirm-to-Protect-the-Global-Cryptocurrency-Economy-and-Ensure-Compliance-with-FATF-AML-Rules.[Online, Accessed February 11, 2020].[27] Binance, 2020b. Binance review 2020: Pros, cons, fees, features,and safety. https://insidebitcoins . . bitfinex . . bitfinex . com/legal/terms. [Online, Accessed: February11, 2020].[30] Bitfinex, 2020c. New york court rules that state attorney has ju-risdiction over bitfinex. https://cointelegraph . com/news/new-york-court-rules-that-state-attorney-has-jurisdiction-over-bitfinex. [On-line, Accessed February 11, 2020].[31] Bitmex, 2020a. Beginner’s guide to bitmex: Complete review.https://blockonomi . . bitmex . . fxempire . . bitstamp . . bitstamp . net/terms-of- use/sa. [Online, Accessed February 11, 2020].[36] Blackbird, 2020. Blackbird bitcoin arbitrage: a long/short market-neutral strategy. https://github . . bloomberg . . capfol . io/. [Online, Accessed: January 26, 2020].[52] Caporale, G.M., Gil-Alana, L., Plastun, A., 2018. Persistence inthe cryptocurrency market. Research in International Business andFinance 46, 141–148.[53] Caporale, G.M., Plastun, A., 2018. Price Overreactions in the Cryp-tocurrency Market. CESifo Working Paper 6861. Munich. URL:http://hdl . handle . net/10419/174984.[54] Caporale, G.M., Plastun, A., 2019. The day of the week effect inthe cryptocurrency market. Finance Research Letters 31.[55] Caporin, M., McAleer, M., 2012. Do we really need both bekk anddcc? a tale of two multivariate garch models. Journal of EconomicSurveys 26, 736–751.[56] Castro, J.G., Tito, E.A.H., Brandão, L.E.T., Gomes, L.L., 2019.Crypto-assets portfolio optimization under the omega measure. TheEngineering Economist , 1–21.[57] Catalyst, 2020. An algorithmic trading library for crypto-assets inpython. https://github . . cboe . . cboe . . cboe . com/aboutcboe/legal-regulatory/departmental-overviews/cfe-regulation. [Online, First Author et al. Page 25 of 30ryptocurrency Trading: A Comprehensive Survey
Accessed February 11, 2020].[61] Ccxt, 2020. Ccxt – cryptocurrency exchange trading library. https://github . . cmegroup . com/trading/cryptocurrency-indices . . cmegroup . com/education/courses/market-regulation/overview/cme-group-rules-and-regulation-overview . . cmegroup . com/company/history/. [Online, Accessed: February 11, 2020].[74] Cocco, L., Concas, G., Marchesi, M., 2017. Using an artificial fi-nancial market for studying a cryptocurrency market. Journal ofEconomic Interaction and Coordination 12, 345–365.[75] Cocco, L., Marchesi, M., 2016. Modeling and simulation of theeconomics of mining in the bitcoin market. PloS one 11.[76] Coinbase, 2020a. Coinbase supported cryptocurrencies.https://help . coinbase . com/en/coinbase/getting-started/general-crypto-education/supported-cryptocurrencies . html. [Online,Accessed: February 11, 2020].[77] Coinbase, 2020b. Our path to listing sec-regulated crypto secu-rities. https://blog . coinbase . com/our-path-to-listing-sec-regulated-crypto-securities-a1724e13bb5a. [Online, Accessed February 11,2020].[78] CoinMaketCap, 2019. Top 100 cryptocurrencies by market capital-ization URL: https://coinmarketcap . com/.[79] Coinmarketcap, 2020. Percentage of total market capitalization.https://coinmarketcap . com/charts/}dominance-percentage. [Online,Accessed January 11, 2020].[80] Colianni, S., Rosales, S., Signorotti, M., 2015. Algorithmic trad-ing of cryptocurrency based on twitter sentiment analysis. CS229Project , 1–5.[81] Conrad, C., Custovic, A., Ghysels, E., 2018. Long-and short-term cryptocurrency volatility components: A garch-midas analysis.Journal of Risk and Financial Management 11, 23.[82] Corbet, S., Eraslan, V., Lucey, B., Sensoy, A., 2019a. The effective-ness of technical trading rules in cryptocurrency markets. FinanceResearch Letters 31, 32–37.[83] Corbet, S., Lucey, B., Urquhart, A., Yarovaya, L., 2019b. Cryp-tocurrencies as a financial asset: A systematic analysis. Interna-tional Review of Financial Analysis 62, 182–199.[84] Corbet, S., Lucey, B., Yarovaya, L., 2018a. Datestamping the bit-coin and ethereum bubbles. Finance Research Letters 26, 81–88. [85] Corbet, S., Meegan, A., Larkin, C., Lucey, B., Yarovaya, L., 2018b.Exploring the dynamic relationships between cryptocurrencies andother financial assets. Economics Letters 165, 28–34.[86] Cryptosignal, 2020. Automated crypto trading and technical analy-sis (ta) bot. https://github . com/CryptoSignal/crypto-signal. [Online,Accessed: January 26, 2020].[87] Ctubio, 2020. Ctubio - cryptocurrency trading bot. https://github . com/ctubio/Krypto-trading-bot. [Online, Accessed: January26, 2020].[88] Diebold, F.X., Yılmaz, K., 2014. On the network topology ofvariance decompositions: Measuring the connectedness of financialfirms. Journal of Econometrics 182, 119–134.[89] Domínguez, M.A., Lobato, I.N., 2003. Testing the martingale dif-ference hypothesis. Econometric Reviews 22, 351–377.[90] Doran, M.D., 2014. A forensic look at bitcoin cryptocurrency. Ph.D.thesis. Utica College.[91] Dutta, A., Kumar, S., Basu, M., 2020. A gated recurrent unit ap-proach to bitcoin price prediction. Journal of Risk and FinancialManagement 13, 23.[92] Dyhrberg, A.H., 2016. Bitcoin, gold and the dollar–a garch volatil-ity analysis. Finance Research Letters 16, 85–92.[93] Easley, D., Engle, R.F., O’Hara, M., Wu, L., 2008. Time-varyingarrival rates of informed and uninformed trades. Journal of FinancialEconometrics 6, 171–207.[94] Elliott, R.J., Van Der Hoek*, J., Malcolm, W.P., 2005. Pairs trading.Quantitative Finance 5, 271–276.[95] Engle, R., 2002. Dynamic conditional correlation: A simple class ofmultivariate generalized autoregressive conditional heteroskedastic-ity models. Journal of Business & Economic Statistics 20, 339–350.[96] Engle, R.F., Kroner, K.F., 1995. Multivariate simultaneous general-ized arch. Econometric theory 11, 122–150.[97] Enoksen, F.A., Landsnes, C.J., 2019. What Can Predict Bubbles inCryptocurrency Prices? Master’s thesis. University of Stavanger,Norway.[98] Escanciano, J.C., Velasco, C., 2006. Generalized spectral tests forthe martingale difference hypothesis. Journal of Econometrics 134,151–185.[99] Estalayo, I., Del Ser, J., Osaba, E., Bilbao, M.N., Muhammad, K.,Gálvez, A., Iglesias, A., 2019. Return, diversification and risk incryptocurrency portfolios using deep recurrent neural networks andmulti-objective evolutionary algorithms, in: 2019 IEEE Congresson Evolutionary Computation (CEC), IEEE. pp. 755–761.[100] Fang, F., Chung, W., Ventre, C., Basios, M., Kanthan, L., Li, L., Wu,F., 2019. Ascertaining price formation in cryptocurrency marketswith deeplearning. arXiv preprint arXiv:2003.00803 .[101] Fantazzini, D., 2019. Quantitative finance with r and cryptocurren-cies. Amazon KDP, ISBN-13 , 978–1090685315.[102] Fantazzini, D., Zimin, S., 2020. A multivariate approach for thesimultaneous modelling of market risk and credit risk for cryptocur-rencies. Journal of Industrial and Business Economics 47, 19–69.[103] Farell, R., 2015. An analysis of the cryptocurrency industry .[104] Feng, W., Wang, Y., Zhang, Z., 2018. Informed trading in the bitcoinmarket. Finance Research Letters 26, 63–70.[105] Fil, M., 2019. Pairs trading in cryptocurrency markets. UniverzitaKarlova, Fakulta sociálních vˇed .[106] Flood, R., Hodrick, R.J., Kaplan, P., 1986. An evaluation of recentevidence on stock market bubbles. National Bureau of EconomicResearch Cambridge, Mass., USA .[107] Flori, A., 2019. News and subjective beliefs: A bayesian approachto bitcoin investments. Research in International Business and Fi-nance 50, 336–356.[108] Fretrade, 2020. Freqtrade. https://github . com/freqtrade/freqtrade.[Online, Accessed: January 26, 2020].[109] Friedl, M.A., Brodley, C.E., 1997. Decision tree classification ofland cover from remotely sensed data. Remote sensing of environ-ment 61, 399–409.[110] Friedman, J., Tibshirani, R., 1984. The monotone smoothing ofscatterplots. Technometrics 26, 243–250. First Author et al. Page 26 of 30ryptocurrency Trading: A Comprehensive Survey [111] Friedman, J.H., 2001. Greedy function approximation: a gradientboosting machine. Annals of statistics , 1189–1232.[112] Gandal, N., Halaburda, H., 2014. Competition in the cryptocurrencymarket. CEPR Discussion Paper No. DP10157.[113] Garcia, D., Schweitzer, F., 2015. Social signals and algorithmictrading of bitcoin. Royal Society open science 2, 150288.[114] Garza, P., 2019. Formal automatic trading in the cryptocurrency era.Ph.D. thesis. Politecnico di Torino.[115] Gatev, E., Goetzmann, W.N., Rouwenhorst, K.G., 2006. Pairs trad-ing: Performance of a relative-value arbitrage rule. The Review ofFinancial Studies 19, 797–827.[116] Gerritsen, D.F., Bouri, E., Ramezanifar, E., Roubaud, D., 2019. Theprofitability of technical trading rules in the bitcoin market. FinanceResearch Letters .[117] Golang, 2020. A golang implementation of a console-based trad-ing bot for cryptocurrency exchanges. https://github . com/saniales/golang-crypto-trading-bot. [Online, Accessed: January 26, 2020].[118] Grayblock, 2018. Elliptic-curve cryptography. https://medium . com/coinmonks/elliptic-curve-cryptography-6de8fc748b8b. [Online,Accessed December 29, 2019].[119] Griffin, J.M., Shams, A., 2019. Is bitcoin really un-tethered? Avail-able at SSRN 3195066 .[120] Gu, S., Lillicrap, T., Sutskever, I., Levine, S., 2016. Continuousdeep q-learning with model-based acceleration, in: InternationalConference on Machine Learning, pp. 2829–2838.[121] Guo, T., Bifet, A., Antulov-Fantulin, N., 2018. Bitcoin volatilityforecasting with a glimpse into buy and sell orders, in: 2018 IEEEInternational Conference on Data Mining (ICDM), IEEE. pp. 989–994.[122] Ha, S., Moon, B.R., 2018. Finding attractive technical patterns incryptocurrency markets. Memetic Computing 10, 301–306.[123] Hale, G., Krishnamurthy, A., Kudlyak, M., Shultz, P., et al., 2018.How futures trading changed bitcoin prices. FRBSF Economic Let-ter 12.[124] Hansel, D., 2018. Cryptocurrency trading: How to make money bytrading bitcoin and other cryptocurrency (volume 2) .[125] Harwick, C., 2016. Cryptocurrency and the problem of intermedia-tion. The Independent Review 20, 569–588.[126] Henderson, P., Islam, R., Bachman, P., Pineau, J., Precup, D.,Meger, D., 2018. Deep reinforcement learning that matters, in:Thirty-Second AAAI Conference on Artificial Intelligence.[127] Hileman, G., Rauchs, M., 2017. Global cryptocurrency benchmark-ing study. Cambridge Centre for Alternative Finance 33.[128] Holmes, G., Donkin, A., Witten, I.H., 1994. Weka: A machinelearning workbench, in: Proceedings of ANZIIS’94-Australian NewZealnd Intelligent Information Systems Conference, IEEE. pp. 357–361.[129] Hrytsiuk, P., Babych, T., Bachyshyna, L., 2019. Cryptocur-rency portfolio optimization using value-at-risk measure, in: Strate-gies, Models and Technologies of Economic Systems Management(SMTESM 2019), Atlantis Press.[130] Hudson, R., Urquhart, A., 2019. Technical analysis and cryptocur-rencies. Available at SSRN 3387950 .[131] Hultman, H., 2018. Volatility Forecasting An Empirical Study onBitcoin Using Garch and Stochastic Volatility models. Master’s the-sis. Lund University. Sweden.[132] Hutto, C.J., Gilbert, E., 2014. Vader: A parsimonious rule-basedmodel for sentiment analysis of social media text, in: Eighth inter-national AAAI conference on weblogs and social media.[133] Hwang, S., Salmon, M., 2004. Market stress and herding. Journalof Empirical Finance 11, 585–616.[134] Ji, Q., Bouri, E., Kristoufek, L., Lucey, B., 2019a. Realised volatil-ity connectedness among bitcoin exchange markets. Finance Re-search Letters , 101391.[135] Ji, Q., Bouri, E., Lau, C.K.M., Roubaud, D., 2019b. Dynamic con-nectedness and integration in cryptocurrency markets. InternationalReview of Financial Analysis 63, 257–272.[136] Jiang, Z., Liang, J., 2017. Cryptocurrency portfolio management with deep reinforcement learning, in: 2017 Intelligent Systems Con-ference (IntelliSys), IEEE. pp. 905–913.[137] Jianliang, M., Haikun, S., Ling, B., 2009. The application on intru-sion detection based on k-means cluster algorithm, in: 2009 Inter-national Forum on Information Technology and Applications, IEEE.pp. 150–152.[138] Juchli, M., 2018. Limit order placement optimization with DeepReinforcement Learning: Learning from patterns in cryptocurrencymarket data. Master’s thesis. TU Delft Electrical Engineering.Netherlands.[139] Juhász, P.L., Stéger, J., Kondor, D., Vattay, G., 2018. A bayesianapproach to identify bitcoin users. PloS one 13.[140] Kajtazi, A., Moro, A., 2019. The role of bitcoin in well diversifiedportfolios: A comparative global study. International Review ofFinancial Analysis 61, 143–157.[141] Kakushadze, Z., 2016. 101 formulaic alphas. Wilmott 2016, 72–81.[142] Kakushadze, Z., 2018. Cryptoasset factor models. Algorithmic Fi-nance , 1–18.[143] Kalchbrenner, N., Grefenstette, E., Blunsom, P., 2014. A convo-lutional neural network for modelling sentences. arXiv preprintarXiv:1404.2188 .[144] Kamrat, S., Suesangiamsakul, N., Marukatat, R., 2018. Technicalanalysis for cryptocurrency trading on mobile phones, in: 2018 3rdTechnology Innovation Management and Engineering Science In-ternational Conference (TIMES-iCON), IEEE. pp. 1–4.[145] Kang, K., Choo, J., Kim, Y., 2019a. Whose opinion mat-ters? analyzing relationships between bitcoin prices and usergroups in online community. Social Science Computer Review ,0894439319840716.[146] Kang, S.H., McIver, R.P., Hernandez, J.A., 2019b. Co-movementsbetween bitcoin and gold: A wavelet coherence analysis. PhysicaA: Statistical Mechanics and its Applications , 120888.[147] Kat, H.M., Heynen, R.C., 1994. Volatility prediction: A comparisonof the stochastic volatility, garch (1, 1) and egarch (1, 1) models.Journal of Derivatives 2.[148] Kate, C., 2018. Cryptocurrency trading for beginners: 6-steps actionplan to your first investment .[149] Katsiampa, P., 2019. An empirical investigation of volatility dynam-ics in the cryptocurrency market. Research in International Businessand Finance .[150] Katsiampa, P., Corbet, S., Lucey, B.M., 2018a. Volatility spillovereffects in leading cryptocurrencies: A bekk-mgarch analysis. Avail-able at SSRN 3232912 .[151] Katsiampa, P., Gkillas, K., Longin, F., 2018b. Cryptocur-rency market activity during extremely volatile periods URL:https://ssrn . com/abstract=3220781, doi: https://dx . doi . org/10 . . .[152] Kaufman, P.J., 2013. Trading Systems and Methods,+ Website. vol-ume 591. John Wiley & Sons.[153] Keerthi, S.S., Shevade, S.K., Bhattacharyya, C., Murthy, K.R.K.,2001. Improvements to platt’s smo algorithm for svm classifier de-sign. Neural computation 13, 637–649.[154] Khuntia, S., Pattanayak, J., 2018. Adaptive market hypothesis andevolving predictability of bitcoin. Economics Letters 167, 26–28.[155] Kim, K., Kim, J., Rinaldo, A., 2018. Time series featurization viatopological data analysis: an application to cryptocurrency trendforecasting. arXiv preprint arXiv:1812.02987 .[156] Kim, Y.B., Kim, J.G., Kim, W., Im, J.H., Kim, T.H., Kang, S.J.,Kim, C.H., 2016. Predicting fluctuations in cryptocurrency transac-tions based on user comments and replies. PloS one 11, e0161197.[157] Kondor, D., Csabai, I., Szüle, J., Pósfai, M., Vattay, G., 2014a. In-ferring the interplay between network structure and market effectsin bitcoin. New Journal of Physics 16, 125003.[158] Kondor, D., Csabai, I., Szule, J., Psfai, M., Vattay, G., 2014b. Infer-ring the interplay between network structure and market effects inbitcoin. New Journal of Physics 16, 125003. URL: https://doi . org/10 . . . First Author et al. Page 27 of 30ryptocurrency Trading: A Comprehensive Survey [159] Kondor, D., Pósfai, M., Csabai, I., Vattay, G., 2014c. Do the richget richer? an empirical analysis of the bitcoin transaction network.PloS one 9.[160] Krafft, P.M., Della Penna, N., Pentland, A.S., 2018. An experimen-tal study of cryptocurrency market dynamics, in: Proceedings of the2018 CHI Conference on Human Factors in Computing Systems,ACM. p. 605.[161] Kristjanpoller, W., Bouri, E., Takaishi, T., 2020. Cryptocurrenciesand equity funds: Evidence from an asymmetric multifractal anal-ysis. Physica A: Statistical Mechanics and its Applications 545,123711.[162] Kristoufek, L., 2013. Bitcoin meets google trends and wikipedia:Quantifying the relationship between phenomena of the internet era.Scientific reports 3, 3415.[163] Kurbucz, M.T., 2019. Predicting the price of bitcoin by the mostfrequent edges of its transaction network. Economics Letters 184,108655.[164] Kutner, M.H., Nachtsheim, C.J., Neter, J., Li, W., et al., 2005. Ap-plied linear statistical models. volume 5. McGraw-Hill Irwin NewYork.[165] Kwon, D.H., Kim, J.B., Heo, J.S., Kim, C.M., Han, Y.H., 2019.Time series classification of cryptocurrency price trend based on arecurrent lstm neural network. Journal of Information ProcessingSystems 15.[166] Kyriazis, N.A., 2019. A survey on efficiency and profitable trad-ing opportunities in cryptocurrency markets. Journal of Risk andFinancial Management 12, 67.[167] Lamon, C., Nielsen, E., Redondo, E., 2017. Cryptocurrency priceprediction using news and social media sentiment. SMU Data Sci.Rev 1, 1–22.[168] Lawrence, S., Giles, C.L., Tsoi, A.C., Back, A.D., 1997. Facerecognition: A convolutional neural-network approach. IEEE trans-actions on neural networks 8, 98–113.[169] Leclair, E.M., 2018. Herding in the cryptocurrency market. Econ5029 final research. Carleton University. Canada.[170] Lee, T.H., Yang, W., 2014. Granger-causality in quantiles betweenfinancial markets: Using copula approach. International Review ofFinancial Analysis 33, 70–78.[171] Li, T.N., Tourin, A., 2016. Optimal pairs trading with time-varying volatility. International Journal of Financial Engineering3, 1650023.[172] Li, T.R., Chamrajnagar, A., Fong, X., Rizik, N., Fu, F., 2019.Sentiment-based prediction of alternative cryptocurrency price fluc-tuations using gradient boosting tree model. Frontiers in Physics 7,98.[173] Liaw, A., Wiener, M., et al., 2002. Classification and regression byrandomforest. R news 2, 18–22.[174] Lintilhac, P.S., Tourin, A., 2017. Model-based pairs trading in thebitcoin markets. Quantitative Finance 17, 703–716.[175] Liu, W., 2019. Portfolio diversification across cryptocurrencies. Fi-nance Research Letters 29, 200–205.[176] Localbtc, 2020. Localbitcoins purchasing online. https://localbitcoins . . maltatoday . com . mt/business/business_news/93170/ why_world_leader_crypto_exchange_binance_moved_to_malta} . XlKZ8Gj7Q2x.[Online, Accessed: February 11, 2020].[182] Markowitz, H., 1952. Portfolio selection. The Journal of Finance 7,77–91.[183] McLean, R.D., Pontiff, J., 2016. Does academic research destroystock return predictability? The Journal of Finance 71, 5–32.[184] McNally, S., 2016. Predicting the price of Bitcoin using MachineLearning. Ph.D. thesis. Dublin, National College of Ireland.[185] McNally, S., Roche, J., Caton, S., 2018. Predicting the price of bit-coin using machine learning, in: 2018 26th Euromicro InternationalConference on Parallel, Distributed and Network-based Processing(PDP), IEEE. pp. 339–343.[186] Merediz-Solà, I., Bariviera, A.F., 2019. A bibliometric analysis ofbitcoin scientific production. Research in International Business andFinance 50, 294–305.[187] Meunier, S., 2018. Blockchain 101: What is blockchain and howdoes this revolutionary technology work?, in: Transforming ClimateFinance and Green Investment with Blockchains. Elsevier, pp. 23–34.[188] Mikolov, T., Kombrink, S., Burget, L., ˇCernock`y, J., Khudanpur, S.,2011. Extensions of recurrent neural network language model, in:2011 IEEE international conference on acoustics, speech and signalprocessing (ICASSP), IEEE. pp. 5528–5531.[189] MIT, 2015. Bitcoin network dataset. https://senseable2015-6 . mit . edu/bitcoin. [Online, Accessed January 11, 2020].[190] Molina, J., 2019. Develop your crypto-trading system using plainlogic, part 1. URL: https://medium . . blockchainresearchnetwork . org/. [Online, Accessed: April17, 2020].[198] Ng, A.Y., Jordan, M.I., 2002. On discriminative vs. generative clas-sifiers: A comparison of logistic regression and naive bayes, in: Ad-vances in neural information processing systems, pp. 841–848.[199] Nomics, 2020. Top cryptocurrency exchanges list. https://nomics . com/exchanges. [Online, Accessed: January 11, 2020].[200] Ogorevc, M., 2019. Cryptocurrency as money: A trading strategysolution. Available at SSRN 3436041 .[201] Omane-Adjepong, M., Alagidede, I.P., 2019. Multiresolution anal-ysis and spillovers of major cryptocurrency markets. Research inInternational Business and Finance 49, 191–206.[202] Omane-Adjepong, M., Alagidede, P., Akosah, N.K., 2019. Wavelettime-scale persistence analysis of cryptocurrency market returns andvolatility. Physica A: Statistical Mechanics and its Applications 514,105–120.[203] Pascanu, R., Mikolov, T., Bengio, Y., 2013. On the difficulty oftraining recurrent neural networks, in: International conference onmachine learning, pp. 1310–1318.[204] Patil, A.P., Akarsh, T., Parkavi, A., 2018. A study of opinion miningand data mining techniques to analyse the cryptocurrency market, First Author et al. Page 28 of 30ryptocurrency Trading: A Comprehensive Survey in: 2018 3rd International Conference on Computational Systemsand Information Technology for Sustainable Solutions (CSITSS),IEEE. pp. 198–203.[205] P˘auna, C., 2018. Arbitrage trading systems for cryptocurrencies.design principles and server architecture. Informatica Economica22, 35–42.[206] Persson, S., Slottje, A., Shaw, I., 2018. Hybrid Autoregressive-Recurrent Neural Network Architecture for Algorithmic Trading ofCryptocurrencies. Cs230 deep learning thesis. Stanford University.[207] Phaladisailoed, T., Numnonda, T., 2018. Machine learning modelscomparison for bitcoin price prediction, in: 2018 10th InternationalConference on Information Technology and Electrical Engineering(ICITEE), IEEE. pp. 506–511.[208] Phillips, P.C., Shi, S., Yu, J., 2015a. Testing for multiple bubbles:Historical episodes of exuberance and collapse in the s&p 500. In-ternational economic review 56, 1043–1078.[209] Phillips, P.C., Shi, S., Yu, J., 2015b. Testing for multiple bubbles:Limit theory of real-time detectors. International Economic Review56, 1079–1134.[210] Phillips, R.C., Gorse, D., 2017. Predicting cryptocurrency pricebubbles using social media data and epidemic modelling, in: 2017IEEE Symposium Series on Computational Intelligence (SSCI),IEEE. pp. 1–7.[211] Phillips, R.C., Gorse, D., 2018a. Cryptocurrency price drivers:Wavelet coherence analysis revisited. PloS one 13, e0195200.[212] Phillips, R.C., Gorse, D., 2018b. Mutual-excitation of cryptocur-rency market returns and social media topics, in: Proceedings of the4th International Conference on Frontiers of Educational Technolo-gies, ACM. pp. 80–86.[213] Poloniex, 2020. Poloniex markets. https://poloniex . com/. [Online,Accessed: February 11, 2020].[214] Rane, P.V., Dhage, S.N., 2019. Systematic erudition of bitcoin priceprediction using machine learning techniques, in: 2019 5th Interna-tional Conference on Advanced Computing & Communication Sys-tems (ICACCS), IEEE. pp. 594–598.[215] Rebane, J., Karlsson, I., Denic, S., Papapetrou, P., 2018. Seq2seqrnns and arima models for cryptocurrency prediction: A compara-tive study. SIGKDD Fintech 18.[216] Rish, I., et al., 2001. An empirical study of the naive bayes clas-sifier, in: IJCAI 2001 workshop on empirical methods in artificialintelligence, pp. 41–46.[217] Rose, C., 2015. The evolution of digital currencies: Bitcoin, a cryp-tocurrency causing a monetary revolution. International Business &Economics Research Journal (IBER) 14, 617–622.[218] S., J., 2018. How do blockchain mining and transactions workexplained in 7 simple steps. https://blog . goodaudience . com/how-a-miner-adds-transactions-to-the-blockchain-in-seven-steps-856053271476. [Online, Accessed: January 26, 2020].[219] Salakhutdinov, R., Hinton, G., 2009. Deep boltzmann machines, in:Artificial intelligence and statistics, pp. 448–455.[220] Shanaev, S., Sharma, S., Ghimire, B., Shuraeva, A., 2020. Tam-ing the blockchain beast? regulatory implications for the cryptocur-rency market. Research in International Business and Finance 51,101080.[221] Sharma, S., Krishma, N., Raina, E., 2017. Survey paper on cryp-tocurrency. International Journal of Scientific Research in ComputerScience, Engineering and Information Technology Vol. 2 Issue 3,307–310.[222] Siaminos, G., 2019. Predicting the value of cryptocurrencies usingmachine learning time series analysis time series analysis time .[223] Sigaki, H.Y., Perc, M., Ribeiro, H.V., 2019. Clustering patternsin efficiency and the coming-of-age of the cryptocurrency market.Scientific reports 9, 1440.[224] Sirignano, J., Cont, R., 2019. Universal features of price formationin financial markets: perspectives from deep learning. QuantitativeFinance 19, 1449–1459.[225] Smuts, N., 2019. What drives cryptocurrency prices?: An investiga-tion of google trends and telegram sentiment. ACM SIGMETRICS Performance Evaluation Review 46, 131–134.[226] Sriram, A., Jun, H., Satheesh, S., Coates, A., 2017. Cold fusion:Training seq2seq models together with language models. arXivpreprint arXiv:1708.06426 .[227] Stocksharp, 2020. Stocksharp - trading platform. https://github . com/StockSharp/StockSharp. [Online, Accessed: January26, 2020].[228] Stuerner, P., 2019. Algorithmic Cryptocurrency Trading. Ph.D. the-sis. Ulm University.[229] Sun, J., Zhou, Y., Lin, J., 2019. Using machine learning for cryp-tocurrency trading, in: 2019 IEEE International Conference on In-dustrial Cyber Physical Systems (ICPS), IEEE. pp. 647–652.[230] Sutton, R.S., Barto, A.G., et al., 1998. Introduction to reinforcementlearning. volume 135. MIT press Cambridge.[231] Sze, V., Chen, Y.H., Yang, T.J., Emer, J.S., 2017. Efficient process-ing of deep neural networks: A tutorial and survey. Proceedings ofthe IEEE 105, 2295–2329.[232] Tapscott, D., Tapscott, A., 2016. Blockchain revolution: how thetechnology behind bitcoin is changing money, business, and theworld. Penguin.[233] TradingstrategyGuides, 2019a. Eos cryptocurrency tradingstrategy–turtle soup pattern. https://tradingstrategyguides . com/eos-cryptocurrency-trading-strategy/. [Online, Accessed January 29,2020].[234] TradingstrategyGuides, 2019b. Free omni cryptocurrency strategy–amazing gann box. https://tradingstrategyguides . com/free-omni-cryptocurrency-strategy/. [Online, Accessed January 29, 2020].[235] TradingstrategyGuides, 2019c. Iota cryptocurrency strategy–busted double top pattern. https://tradingstrategyguides . com/iota-cryptocurrency-strategy/. [Online, Accessed January 29, 2020].[236] TradingstrategyGuides, 2019d. Nem (xem) cryp-tocurrency strategy–momentum pinball setup. https://tradingstrategyguides . com/nem-xem-cryptocurrency-strategy/.[Online, Accessed January 29, 2020].[237] TradingstrategyGuides, 2019e. Tether trading strategy–bottomrotation trading. https://tradingstrategyguides . . tradingview . com/markets/cryptocurrencies/global-charts/. [Online, Accessed February 23, 2020].[239] Troster, V., Tiwari, A.K., Shahbaz, M., Macedo, D.N., 2019. Bitcoinreturns and risk: A general garch and gas analysis. Finance ResearchLetters 30, 187–193.[240] Trucíos, C., Tiwari, A.K., Alqahtani, F., 2019. Value-at-risk and ex-pected shortfall in cryptocurrencies’ portfolio: A vine copula-basedapproach. Available at SSRN 3441892 .[241] Vidal-Tomás, D., Ibáñez, A.M., Farinós, J.E., 2019. Herding in thecryptocurrency market: Cssd and csad approaches. Finance Re-search Letters 30, 181–186.[242] Virk, D.S., 2017. Prediction of Bitcoin Price using Data Mining.Master’s thesis. National College of Ireland.[243] Vo, A., Yost-Bremm, C., 2018. A high-frequency algorithmic trad-ing strategy for cryptocurrency. Journal of Computer InformationSystems , 1–14.[244] Vogelvang, B., 2005. Econometrics: theory and applications withEviews. Pearson Education.[245] Wagstaff, K., Cardie, C., Rogers, S., Schrödl, S., et al., 2001. Con-strained k-means clustering with background knowledge, in: Icml,pp. 577–584.[246] Wang, H., He, D., Ji, Y., 2017. Designated-verifier proof of assetsfor bitcoin exchange using elliptic curve cryptography. Future Gen-eration Computer Systems .[247] Wang, L., 2005. Support vector machines: theory and applications.volume 177. Springer Science & Business Media.[248] Ward, M., 2018. Algorithmic Trading for Cryptocurrencies. De-partmental honors thesis. UtahState University. United States.[249] Weber, P., Wang, F., Vodenska-Chitkushev, I., Havlin, S., Stanley,H.E., 2007. Relation between volatility correlations in financial First Author et al. Page 29 of 30ryptocurrency Trading: A Comprehensive Survey 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 . . .[253] Yaya, O.S., Ogbonna, E.A., Olubusoye, O.E., 2018. How persistentand dependent are pricing of bitcoin to other cryptocurrencies beforeand 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] ˙Zbikowski, K., 2016. Application of machine learning algorithmsfor bitcoin automated trading, in: Machine Intelligence and BigData in Industry. Springer, pp. 161–168.[256] Zemmal, N., Azizi, N., Dey, N., Sellami, M., 2016. Adaptivesemi supervised support vector machine semi supervised learningwith features cooperation for breast cancer classification. Journal ofMedical 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 . . .[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 tradingin asia-pacific, now and the future. Pacific-Basin Finance Journal53, 186–207.[260] ´Slepaczuk, R., Zenkova, M., 2018. Robustness of support vectormachines in algorithmic trading on cryptocurrency market. CentralEuropean Economic Journal 5, 186 – 205..[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 tradingin asia-pacific, now and the future. Pacific-Basin Finance Journal53, 186–207.[260] ´Slepaczuk, R., Zenkova, M., 2018. Robustness of support vectormachines in algorithmic trading on cryptocurrency market. CentralEuropean Economic Journal 5, 186 – 205.