Rise of the Machines? Intraday High-Frequency Trading Patterns of Cryptocurrencies
Alla A. Petukhina, Raphael C. G. Reule, Wolfgang Karl Härdle
RRise of the Machines?
Intraday High-Frequency Trading Patterns of Cryptocurrencies
Alla A. Petukhina
Humboldt-Universit¨at zu Berlin.Firamis GmbH, Germany.alla.petukhina[at]wiwi.hu-berlin.de
Raphael C. G. Reule
Humboldt-Universit¨at zu Berlin.irtg1792.wiwi[at]wiwi.hu-berlin.de
Wolfgang Karl H¨ardle
Humboldt-Universit¨at zu Berlin, IRTG 1792, Dorotheenstr. 1, 10117 Berlin, Germany School ofBusiness, Singapore Management University, 50 Stamford Road, Singapore 178899 Faculty ofMathematics and Physics, Charles University, Ke Karlovu 3, 121 16 Prague, Czech RepublicDepartment of Information Management and Finance, National Chiao Tung University, Taiwan, ROChaerdle[at]wiwi.hu-berlin.de
September 10, 2020
Abstract
This research analyses high-frequency data of the cryptocurrency market in re-gards to intraday trading patterns related to algorithmic trading and its impacton the European cryptocurrency market. We study trading quantitatives such asreturns, traded volumes, volatility periodicity, and provide summary statistics ofreturn correlations to CRIX (CRyptocurrency IndeX), as well as respective over-all high-frequency based market statistics with respect to temporal aspects. Ourresults provide mandatory insight into a market, where the grand scale employ-ment of automated trading algorithms and the extremely rapid execution of tradesmight seem to be a standard based on media reports. Our findings on intradaymomentum of trading patterns lead to a new quantitative view on approaching thepredictability of economic value in this new digital market.
JEL Classification: G02, G11, G12, G14, G15, G23.Keywords: Cryptocurrency, High-Frequency Trading, Algorithmic Trading, Liquidity,Volatility, Price Impact, FinTech, CRIX.
The financial support of the Czech Science Foundation under grant no. 19-28231X, theFiramis GmbH, Robert-Kempner-Ring 27, 61440 Oberursel (Taunus), the Yushan Scholar a r X i v : . [ q -f i n . T R ] S e p rogram, and European Unions Horizon 2020 FIN-TECH Project, under the grant no.825215 (Topic ICT-35-2018, Type of actions: CSA), as well as data support by dyossolutions GmbH, Oberwallstr. 8, 10117 Berlin, are greatly acknowledged. This is a post-peer-review, pre-copyedit version of an article published in The EuropeanJournal of Finance. The final authenticated version is available online at: https://doi.org/10.1080/1351847X.2020.1789684 Motivation
High-frequency trading takes advantage of the incredible rise of computing power pro-vided by the steady development of ever more capable structures. Algorithms are alreadymajor players in a variety of financial applications, and have proven to be more efficientthan their human counterparts. By employing these so-called “algos”, positive effects canbe exploited to their maximum and market inefficiencies can potentially be eliminated(Hrdle et al., 2020). However, just like for every coin, there is a flipside, such as thenegative impact on capital markets caused by technological inefficiencies (Emem, 2018).One of the most noted events of an early point of attack for these algorithms was theFlash Crash of 2010.No matter what, the machines are here to stay and their influence, possibly poweredby “learning” algorithms, will certainly increase even more with time - especially inregards to new emerging markets such as cryptocurrencies. The rising popularity andacceptance of this alternative asset, as it has yet to be understood as an alternativeto fiat currency, requires for specialised strategies to maximise the potential return ofinvestments (Akhtaruzzaman et al., 2019; Platanakis and Urquhart, 2019; Trimborn etal., 2018; Petukhina et al., 2020, 2019).
Figure 1:
CRIX Time Series.3et, did the quantlets or algorithms really venture in the realm of autonomous ma-chines, the digital world, or are they still with the world of the humans, the world ofmanual labour oil and baby nutrition companies?This is especially of interest since the cryptocurrency market has significantly maturedin recent years and has attracted enormous investments, not only by major players butespecially by individuals. Especially
F inT ech Startups are of high interest, as absurdamounts of financial backing was (and is still to some extent) being generated by justpresenting a briefly written
W hitepaper -PDF marketing outlet (Zetzsche et al., 2019).The early cryptocurrency market kick-off starting in late 2017 is evidently presentingsuch happenings. The discrepancy between sentiment and tone generated by marketingversus the delivered performance is fascinating (with further references Hrdle et al., 2020;Chen et al., 2019a; Qian et al., 2019).In this research, we are analysing high-frequency data (5-minute intervals) gainedfrom the cryptocurrency market and see, if there is really 24/7 algorithmic trading, orif there are still people sitting behind their computers creating and executing orders byhand after they have returned from their daily jobs.Previous research outputs on this theme, such as Zhang et al. (2019), have used timespans ranging from 1 hour to 12 hours. Their methods yielded results, which lead todifferent conclusions, yet opened up further thoughts towards factors such as tradingpatterns, variations in returns, volatility and trading volume. Zhang et al. (2018) arealso looking at the same aspects as the previous research, with the additional findingof a power-law correlation between price and volume. Rschli et al. (2017) respectivelybuild a uni- and multivariate analysis of quantitative facts to show off stylized facts ofcryptocurrencies. Schnaubelt et al. (2019) analyzed limit order data from cryptocurrencyexchanges. Besides their recovery of common qualitative facts, they find that these dataexhibit many of the properties found for classic limit order exchanges, such as a symmet-ric average limit order book, the autocorrelation of returns only at the tick level and thetiming of large trades. Yet they find that cryptocurrency exchanges exhibit a relativelyshallow limit order book with quickly rising liquidity costs for larger volumes, many smalltrades and an extended distribution of limit order volume far beyond the current mid-price.Given the search for the most efficient trading strategies, Caporale and Plastun (2019)provide a range of historic scientific works on the time of day effects to reap abnormalprofits. In contrast to their work, we aim at identifying the market drivers, which areresponsible for how this new emerging market, which is still full of conundrums for many,4ehaves - i.e. do market movements fit into human activity patterns or are these inde-pendent from time.Preliminary research has therefore not touched the highly topical question of humanimpact in the wake of digital systems. There are many papers with interesting approachesand solutions, but only for problems that are already known and have been rebrewn forsome time now. Yet, with the advent and popular discussion of the employment of LongShort Term Memory Neural Networks (LSTM) and hence deep learning for finance, AIadvisory, essentially based on the human factor of sentiment in the realm of cryptocur-rencies (Chen et al., 2019a), will play a major role in especially this completely digitalmarket. This, as a circular argument, brings us once again to the fundamental idea ofenforcing the understanding of market behaviour based on the time of the day and theagents acting in these markets that are predestined to be ruled by the machines.As a polemic term, we are using
P roof − Of − Human ( P oH ; derived from
P roof − Of − W ork , P roof − Of − Stake et cetera consensus algorithms) to underline the hy-potheses that not algorithms are the major players in this market, but humans. Humansdon’t act as programmed like algorithms - they act based on biological and psychologicalinput, such as hunger or fatigue. The majority of humans will have certain times at whichthey are active, and at which they rest and are therefore inactive. Alternatively spoken,algorithms need humans to start and then exacerbate a price trend - the question is,therefore if the cryptocurrency market is dominated by human or algorithmic behaviour.Eventually, we can differentiate algorithmic and human trading patterns expressed withinthe market (with further references Caporale et al., 2016).The paper is structured by giving a brief general introduction and data source disclo-sure and methodology section, followed by a respective intraday data analysis, which isconcluded by a section on Time-Of-Day effects and the Proof-Of-Human.All presented graphical and numerical examples shown are reproducible and can befound on (Borke and H¨ardle, 2018) and are indicated as CCID.
To understand the dynamics of this new high-frequency market, it is mandatory to in-vestigate the statistical properties of various high-frequency variables, for example, trad-ing volume or volatility, to find respective answers to questions like option pricing andforecasting. Preliminary research to visualize the cryptocurrency market was done byTrimborn and Hrdle (2018) with the CRyptocurrency IndeX, CRIX (crix.berlin), in5rder to represent the performance of the cryptocurrency market with the help of themost mature and accepted cryptocurrencies, such as Bitcoin (BTC), Ethereum (ETH),or Ripple (XRP) - see appendix section 5.1 for further used abbreviations. As the CRIXindex family covers a range of cryptocurrencies based on different liquidity rules andvarious model selection criteria, we have chosen this as the main data source. CRIXrepresents the cryptocurrency market, but by its very nature is dominated by a few mainplayers with BTC being the absolute market driver over time.Furthermore, we used data provided by dyos solutions GmbH (dyos.io) compiled fromvarious exchanges’ data, to ensure that our findings are coherent with other data avail-able. It is important to keep in mind, that the 5-minute data analysed in this researchis gained from sources located in the
European markets (+1h GMT) and therefore thetime-of-day effects may look different for markets from the Americas or Asia. We willmake an exegesis on this important point in subsection 3.3.In addition, the analysed data sample belongs to the time period after the cryptocur-rency market heated up immensely around the end of 2017, followed by a sharp cooldownat the beginning of 2018. By that time a plethora of euphoric media outlets was praisingthe endless possibilities which the blockchain technology may provide - and what eventu-ally also lead to quite a lot of ICO scams (Zetzsche et al., 2019). At that time, algorithmictrading in cryptocurrency markets was not seen as being a mere idea, but reality by moreor less promising FinTech startups. These emerging enterprises are offering a wide varietyof blockchain-related services, such as trading, asset management, or technical support.Especially FinTech startups related to the financial sector, in contrast to for examplesupply chain oriented ventures, are heavily interested in ArtificialIntelligence (learningalgorithms) and are marketing their individual related products as groundbreaking andready-to-use. Given the chosen typical vacation period, July and August, one shouldhence expect a less pronounced human, but algorithmic driven market behavior to con-tradict the hypotheses of the PoH concept - more on that as well in subsection 3.3.Regarding data handling, we are coherent with previous research on high-frequencydata based on traditional data sources, such as the NYSE, which has underlined datapreparation issues and the specific statistical properties of various high-frequency vari-ables (Hautsch, 2011). As we are dealing with a subject, where individuals can actdirectly with the market without involving a middle-man, the characteristics of our dataobserved on transaction level, therefore, are especially irregularly spaced in time andwithout interruption - see section 3. 6
Intraday Data Analysis
In the following chapter, we provide an overview of the methods employed to analyzeour high-frequency data at hand with further statistical intraday cryptocurrency marketobservations.
This paper undertakes a fresh empirical investigation of key financial variables of cryp-tocurrency market, such as volatility, returns and trading volume. Following, for example,Hussein (2011), intraday return volatility is calculated as absolute log-returns as definedin (2). As we are looking at high-frequency data, there is no need to use measures like,for example, the compounded annual growth rate (CAGR) instead of absolute returns,which is used to get the per-annum returns and does not support the analysis in this case.The simple return
Ret t is defined as Ret t = P t − P t − P t − , (1)where P t und P t − are prices of coins at time points t and t − ret t is defined as ret t = log P t P t − = log(1 + Ret t ) . (2)In order to expressively visualize some features of our high dimensional and nonsta-tionary time series gained from our large high-frequency dataset of the specifically chosenperiod of time, a Generalized Additive M odel (GAM) is best suited. A GAM is a gen-eralized linear model (GLM), where the nonlinear predictor is given by a specified sumof smooth functions of the covariates, as well as a conventional parametric componentof the linear predictor (Hrdle, 1990). The basic advantage of GAM is the possibility tomodel highly complex nonlinear relationships given a large number of potential predic-tors. In particular, recent computational developments in GAM fitting methods, such asWood et al. (2015), Wand (2017), and Wood (2017), have made it possible to use thesemodels to explore very large datasets. Moreover, in the last two decades, GAM methodshave intensively developed in terms of the range of models that can be fitted. All theseadvantages make GAMs a feasible tool to investigate intraday seasonality patterns withhigh-frequency trading data. In general, the model has a structure something like: g { E ( y i ) } = β + f ( x i ) + · · · + f p ( x ip ) (3)7here y = ( y , . . . , y n ) (cid:62) observation of a response variable Y , g is a link function(identical, logarithmic or inverse, etc.), x . . . x p are independent variables, β is an inter-cept, f ( x i ) . . . f p ( x ip ) are unknown nonparametric smooth functions, and ε i is an i.i.d.random error. In our application we use the identity link function, since the LHS of ourequations are features/variables observed or measured on a continuous scale, to fit thefollowing statistical model: y i = f ( x ,i ) + f ( x ,i ) + . . . + f p ( x p,i ) + ε i (4)Here y i will be a trading volume, volatility, or returns as defined in (2), x q,i will bethe daily and weekly effects. The nonlinear function f q is a smooth function, composedby sum of basis functions b qj (for example B-splines, P-splines or cubic splines) and theircorresponding regression coefficients β q,j . Thus, each function f q is expressed as: f q ( x ) = k q (cid:88) j =1 β q,j b qj ( x ) (5)where k q is the dimension of the spline basis.The smooth function m ( x , ..., x p ) = (cid:80) pq =1 f q ( x q ) is estimated by penalized regression: n (cid:88) i =1 (cid:32) y i − p (cid:88) q =1 f q ( x i ) (cid:33) + p (cid:88) q =1 λ q (cid:90) (cid:13)(cid:13) f (cid:48)(cid:48) q ( x ) (cid:13)(cid:13) dx (6)where the penalty parameter Λ = ( λ , . . . , λ p ) is a smoothing parameter controllingthe fitsmoothness tradeoff for f q and can be selected by minimization of the GeneralizedCross Validation (GCV) score, see (Wood, 2004) and (Wood, 2011). Denoting B thematrix formed by concatenation of the b qj , one has to solve the following problem: (cid:98) β = arg min λ,β (cid:40) (cid:107) Y − Bβ (cid:107) + p (cid:88) q =1 λ ˙ q β (cid:62) S q β (cid:41) (7)where β = ( β , . . . , β p ) (cid:62) is the vector of the unknown regression parameters, S q is amatrix of known coefficients (a smoothing matrix) and depends on the spline basis. Thus,given λ , expression (7) may readily be minimized to yield the coefficient estimates ˆ β λ .The method of obtaining the estimate of the β is called Penalized Iteratively Re-weightedLeast Squares (P-IRLS) which is implemented in the mgcv R package, see (Wood, 2019). As an introduction to the data analyzed in this brief research, we are providing sum-mary statistics regarding its statistical properties to form a basic understanding of themarket at hand. Firstly, the trading data density of cryptocurrencies against the normal8istribution of BTC is far from normally distributed, see figure 2. Hence the behaviourof agents in this market is far from what we would see in classic markets. This implies,that new rules are being employed, and therefore we have to rethink our common wayon how to approach the quantitative analysis of markets in general. We will start ourdiscussion on the specific research question by first providing a general overview of thecryptocurrency market with increasingly narrowed focus and attention to detail regardingspecific timeframes and parameters for individual crypto-assets.
Density of cryptos against normal distribution D en s i t y −0.005 0.000 0.005 Figure 2:
Density of intraday CCs returns. 01. July 2018 - 31. August 2018. Theprobability density functions of the distributions of daily returns for the analized cryp-tocurrencies with the following colour code: BCH, BTC, DASH, ETC, ETH, LTC, REP,STR, XMR, XRP, ZEC. A normal distribution with the same mean and standard devi-ation as the returns on BTC is displayed as a histogram in the backgroundSecondly, using GAM, we gain interesting insights into the trading activities in this24/7 market. Cryptocurrencies are being traded without any forced break, as we know itfrom classic markets, for example, if the stock exchange closes for the night or especiallyfor weekends. In addition to this fact, we have to consider, that there is no centralizedtrading in the act, but a plethora of service providers, so-called cryptocurrency exchanges.As we disclose the origin of our data, we underline, that caused by this very decentralizednature of cryptocurrency genesis and their respective trading, partially greatly divergingprice data is available for each individual cryptocurrency. Again, this is caused by the de-9entralized root of individual, unsupervised and unregulated, places for exchange. Thereis no fixed price for BTC contrary to, for example, for exchange rates of USD-EUR. - -
01 2018 - -
06 2018 - -
11 2018 - -
16 2018 - -
21 2018 - -
26 2018 - -
01 2018 - -
06 2018 - -
11 2018 - -
16 2018 - -
21 2018 - -
26 2018 - -
31 2018 - -
05 2018 - -
10 2018 - -
15 2018 - -
20 2018 - -
25 2018 - -
30 2018 - -
04 2018 - -
09 2018 - -
14 2018 - -
19 2018 - -
24 2018 - - Date14000160001800020000220002400026000 P r i c e Figure 3:
Candlestick chart of CRIX. 01. July 2018 - 29. September 2018. - -
01 2018 - -
08 2018 - -
15 2018 - -
22 2018 - -
29 2018 - -
05 2018 - -
12 2018 - -
19 2018 - - Date600065007000750080008500 B T C P r i c e (a) BTC - -
01 2018 - -
08 2018 - -
15 2018 - -
22 2018 - -
29 2018 - -
05 2018 - -
12 2018 - -
19 2018 - - Date250300350400450500 E T H P r i c e (b) ETH - -
01 2018 - -
08 2018 - -
15 2018 - -
22 2018 - -
29 2018 - -
05 2018 - -
12 2018 - -
19 2018 - - Date0.250.300.350.400.450.50 X R P P r i c e (c) XRP Figure 4:
Chandlestick charts for individual price movements. 01. July 2018 - 31.August 2018.In contrast to the CRIX candlestick chart presented in figure 3, where five minutehigh-frequency data is aggregated to 60 minutes, we present respective individual plotsfor each examined cryptocurrency, as shown in figure 4 to give an easier entry to under-stand this volatile market. Consistency between Figure 1, 3, and 4 can be seen in thecontext of the findings in Chen et al. (2019a), where the impact of sentiment on cryp-tocurrency prices is evident (with further references Chen et al., 2019c; Qian et al., 2019).Furthermore, when recurring to the observable price structure of the Flash Crash of 2010as well, we can see quite many jumps in these figures - a phenomenon also described in10hen et al. (2019b) and Qian et al. (2019).Figure 5, shows the intraday 5-minutes returns for the period from the 01. July 2018to the 31. August 2018. As indicated, overall returns across the board are very extreme- a phenomenon generally unknown to classic financial markets. In addition, we canobserve an extreme activity cluster around the second half of August. We can link thisactivity to increased media outlets regarding cryptocurrencies: the more investors floodedinto this market, the higher the trading activity, fueled by sentiment, became - leadingto partially absurd returns; positive as well as negative.
Jul Aug Sep − . . . . Index B T C (a) BTC Jul Aug Sep − . − . − . . . . . Index E T H (b) ETH Jul Aug Sep − . − . . . . Index X R P (c) XRP Figure 5:
Intraday Returns (5 minutes). 01. July 2018 - 31. August 2018.
Jul Aug Sep . . . . . Index B T C (a) BTC Jul Aug Sep . . . . . . . . Index E T H (b) ETH Jul Aug Sep . . . . . Index X R P (c) XRP Figure 6:
Intraday Volatility. 01. July 2018 - 31. August 2018.Figure 6 adds to this finding, presenting the overall volatility from the beforehandstated period. As we can see, the return activity cluster in August from figure 5 is mir-rored in the volatility activity cluster in figure 4. Hence, we proof the beforehand statedclaim of cryptocurrency activity being fueled by media outlets as well as sentiment, asbeing attested. 11able 1 displays the estimated values of selected parameters for the cryptocurrencyintraday trading for the given period of the 01. July 2018 to the 31. August 2018. Thelargest autocorrelation is for DASH (0.01), the smallest autocorrelation is for STR (-0.09).
Table 1:
Estimated first-order autocorrelation of the returns, (cid:98) ρ ( ret t ), the squaredreturns, (cid:98) ρ ( ret t ), and the absolute returns, (cid:98) ρ ( | ret t | ), as well as the estimated skewness, (cid:98) S , the estimated excess kurtosis, (cid:92) e.Kurt , and the Jarque-Bera test statistic, JB, with therespective, obviously very small, p-value for the overall summed intraday high-frequencydata from the 01. July 2018 to the 31. August 2018. (cid:98) ρ ( ret t ) (cid:98) ρ ( ret t ) (cid:98) ρ ( | ret t | ) (cid:98) S (cid:92) e.Kurt JB JB p-valueBCH -0.01 0.12 0.20 0.49 13.69 140148.24 0.00BTC -0.05 0.13 0.24 1.30 49.44 1823779.80 0.00DASH 0.01 0.17 0.20 0.73 28.98 626596.64 0.00ETC -0.06 0.26 0.26 0.70 26.07 507374.39 0.00ETH -0.01 0.18 0.27 0.17 16.34 198777.58 0.00LTC -0.01 0.11 0.19 0.44 14.91 166121.81 0.00REP -0.08 0.22 0.19 0.35 21.89 356937.91 0.00STR -0.09 0.12 0.18 0.28 8.12 49354.96 0.00XMR -0.07 0.13 0.14 0.03 10.51 82241.48 0.00XRP -0.05 0.17 0.25 0.11 11.44 97390.58 0.00ZEC -0.07 0.25 0.22 1.30 26.66 534032.89 0.00While the first-order autocorrelation of the returns of all cryptocurrencies is all closeto zero and mostly negative, the autocorrelations of the squared and absolute returns ofall cryptocurrencies are positive and significantly larger than zero. Obviously, there is alinear relationship in the absolute and squared values of the chronologically sequentialreturns. Since the autocorrelation is positive, it can be concluded, that small absolutereturns are followed sequentially by small absolute returns and large absolute returns arefollowed by large ones again. This means, that there are quiet periods with small pricechanges and dynamic periods with large oscillations.Furthermore, whereas the estimate for skewness is mostly close to zero, except forBTC and ZEC, the estimate for excess kurtosis is in every case significantly larger than3. The smallest estimated excess kurtosis is by STR (yet with an expressive (cid:92) e.Kurt of8.12), and the largest by BTC ( (cid:92) e.Kurt = 49.44). These values show, that the tested con-stituents are far from normally distributed. Negative skewness signals about increasingthe downside risk and is a consequence of asymmetric volatility models. Positively skewed12istributions have a longer right tail, meaning for investors a greater chance of extremelypositive outcomes. A well-known stylized fact about returns distributions highlights theirleptokurtic nature: they have more mass around the centre and in the tails than a nor-mal distribution. For example, Hussein (2011) reports relatively high levels of kurtosis instock data from the United States of America. This phenomenon is known as kurtosis risk.The combined test of the normal distribution from Jarque and Bera (JB) can be de-rived as asymptotically χ distribution with two degrees of freedom. The last column intable 1 shows, that in all cases the normal distribution hypothesis is clearly rejected. Thisis above all caused by the value of kurtosis, which is significantly larger than 3, caused bya very frequent appearance of outliers in this new market. The higher kurtosis, comparedto a normal distribution, proves that these extreme points result in leptokurtic distribu-tions and are evidence of fat tails relative to the normal distribution’s tail. However, asthis asymmetry is common to financial markets, it is especially strong in the cryptocur-rency markets with potentially extreme returns and a very pronounced volatility.The following tables respectively show the individual correlation to CRIX, if the mar-ket is acting positively, table 2, or negatively, table 3. Extensive care should be put onour main actors - BTC, ETH and XRP - when studying these. As these enjoy a largemarket acceptance and hence are long-term drivers of the cryptocurrency market, we canonce again, underline our findings given beforehand.On a side note, tables 2 and 3 show that among the top 11 cryptocurrencies, most pairsexhibit low return correlations, what suggest strong diversification benefits in a portfolio,especially outside the major cryptocurrencies presented, see also (Petukhina et al., 2020).We can observe, that the correlation to CRIX in both tables presents itself as clusteredaround well-known cryptocurrencies, namely BTC, ETH, XRP, as well as BCH, and ETC.Therefore, this activity can be interpreted in a way, which indicates these constituentsas the market drivers. This finding also correlates with the long term trading activityregistered on many online sources for these coins. We should note, without going intodetail, that LTC and BCH are closely related to BTC, and that ETC is closely tied tothe history of ETH. XRP itself was able to carve out its very specific niche early enoughfor certain applications, especially in the banking sector - in contrast, BTC can be seenas the genesis of digital currency without any intrinsic value, whereas the ETH systemenables many different applications, majorly through so-called “smart contracts”.13 able 2: Pairwise crypto-currency correlations of returns for positive market-movementdays, as defined by returns on CRIX. 01. July 2018 - 31. August 2018. UP BCH BTC DASH ETC ETH LTC REP STR XMR XRP ZECBCH 0.50 0.23 0.33 0.47 0.46 0.13 0.29 0.25 0.37 0.23BTC 0.50 0.27 0.36 0.55 0.49 0.18 0.34 0.30 0.40 0.27DASH 0.23 0.27 0.17 0.22 0.22 0.10 0.17 0.17 0.22 0.14ETC 0.33 0.36 0.17 0.37 0.31 0.11 0.21 0.17 0.28 0.14ETH 0.47 0.55 0.22 0.37 0.47 0.16 0.30 0.27 0.42 0.22LTC 0.46 0.49 0.22 0.31 0.47 0.17 0.26 0.25 0.39 0.23REP 0.13 0.18 0.10 0.11 0.16 0.17 0.12 0.11 0.11 0.11STR 0.29 0.34 0.17 0.21 0.30 0.26 0.12 0.18 0.27 0.19XMR 0.25 0.30 0.17 0.17 0.27 0.25 0.11 0.18 0.20 0.15XRP 0.37 0.40 0.22 0.28 0.42 0.39 0.11 0.27 0.20 0.19ZEC 0.23 0.27 0.14 0.14 0.22 0.23 0.11 0.19 0.15 0.19
Table 3:
Pairwise crypto-currency correlations of returns for negative market-movementdays, as defined by returns on CRIX. 01. July 2018 - 31. August 2018.
DOWN
BCH BTC DASH ETC ETH LTC REP STR XMR XRP ZECBCH 0.48 0.21 0.32 0.47 0.43 0.15 0.27 0.23 0.37 0.22BTC 0.48 0.26 0.36 0.52 0.45 0.19 0.33 0.30 0.41 0.24DASH 0.21 0.26 0.15 0.22 0.21 0.11 0.16 0.18 0.18 0.14ETC 0.32 0.36 0.15 0.36 0.30 0.14 0.21 0.18 0.30 0.16ETH 0.47 0.52 0.22 0.36 0.42 0.16 0.29 0.23 0.40 0.21LTC 0.43 0.45 0.21 0.30 0.42 0.16 0.26 0.24 0.35 0.19REP 0.15 0.19 0.11 0.14 0.16 0.16 0.11 0.12 0.13 0.08STR 0.27 0.33 0.16 0.21 0.29 0.26 0.11 0.16 0.26 0.16XMR 0.23 0.30 0.18 0.18 0.23 0.24 0.12 0.16 0.20 0.15XRP 0.37 0.41 0.18 0.30 0.40 0.35 0.13 0.26 0.20 0.17ZEC 0.22 0.24 0.14 0.16 0.21 0.19 0.08 0.16 0.15 0.1714 .3 Time-Of-Day Effects and Proof-Of-Human
To support our hypothesis of mostly dealing with human agent initiated trades, whichwe coin as PoH, we present our findings regarding the time-of-day trading in this section.Additional material on information arrival, news sentiment, volatilities and jumps of in-traday returns can also be taken from Qian et al. (2019).Cryptocurrency exchanges, as introduced in section 2, are often designed to servea certain target group, for example by emphasizing compliance with national regula-tory frameworks. By plotting the trade volume against the timestamps, we can alsoobserve certain properties of market activity and draw coherent conclusions to the ori-gin of the market participants: are these mostly human, who are doing trades by hand,or are we looking at a well oiled automatic machinery full of algorithms - just as com-monly portrayed. Keep in mind, as mentioned in section 2, that our data is gained fromEurope-based sources, and taken from periods that are overwhelmingly identifiable bycorporate staff vacations. One should hence expect a less pronounced human, but algo-rithmic driven market behaviour to contradict our hypotheses.To underline this argument, it is useful to imagine a transitional system, whereashuman interference is completely removed or not relevant to a market system (e.g. Ca-porale et al., 2016), and where the trading pattern will, therefore, be independent of thetime-of-day effects: human + human + human (cid:98) = human driven networkhuman + algorithm + human (cid:98) = predominantly human driven networkhuman + algorithm + algorithm (cid:98) = predominantly machine driven networkalgorithm + algorithm + algorithm (cid:98) = algorithmic driven network With increasing market participation of algorithms, we expect, for example, nighttimeto have a negligible impact on the market activity. In contrast, we expect nighttime tohave an impact on market activity if the market is dominated by human interaction.The following figures employ GAM to observe daily and weekly patterns for intradayvolatility and trading volume. For daily seasonality cubic regression splines, for weeklyseasonality P -splines are used, and a number of knots are logically set to the number ofunique values, i.e 62 for daily patterns and 7 for weekly. The summary statistics of GAMfor all cryptocurrencies demonstrate a high significance of smooth terms combined with aquite low explanatory power (coefficients of determination are around 1%). Nevertheless,we can observe distinct intraday seasonality patterns.15 − − − + − − − − BTC
Time A b s . r e t u r n s :
05 02 :
05 04 :
05 06 :
05 08 :
05 10 :
05 12 :
05 14 :
05 16 :
05 18 :
05 20 :
05 22 : (a) BTC − − − − + − − − ETH
Time A b s . r e t u r n s :
05 02 :
05 04 :
05 06 :
05 08 :
05 10 :
05 12 :
05 14 :
05 16 :
05 18 :
05 20 :
05 22 : (b) ETH − − + − − − XRP
Time A b s . r e t u r n s :
05 02 :
05 04 :
05 06 :
05 08 :
05 10 :
05 12 :
05 14 :
05 16 :
05 18 :
05 20 :
05 22 : (c) XRP Figure 7:
Daily seasonality: fit of Generalized Additive Model (5 min nodes) withcubic regression splines for absolute returns of cryptocurrencies (shaded regions representconfidence bands for smooths), 01. July 2018 - 31. August 2018. − − − − + − Time A b s . r e t u r n s (a) BTC − − − − + − − Time A b s . r e t u r n s (b) ETH − − − − − − + − − − Time A b s . r e t u r n s (c) XRP Figure 8:
Weekly seasonality: fit of Generalized Additive Model with p-splines for abso-lute returns of cryptocurrencies (shaded regions represent confidence bands for smooths),01. July 2018 - 31. August 2018.Assuming that the majority of employed persons do work from 09:00 to 17:00 o’clockin Europe, figures 7 and 8 (data time is +1 GMT) present us with a very clear pictureof returns and volume. Characteristic human activity curves are presented by figure7 showing the daily seasonality - a curve driven by algorithms as the main actor, or
Artif icial Intelligence in a FinTech startup buzzword context, should not present sucha comparatively extreme low around a typical time for the majority of humans to beasleep. Following that point, the curves expresse a significant growth, only to flat outagain around lunch break time. Most figures present a peak between 17:00 and 20:00o’clock, just when most people finish their daily routine jobs, followed by an expressivedecline of the curves. This is surprising, as media outlets and startup marketing generallypraise the non-stop availability and easy access to cryptocurrency exchanges, and hencewe would presume to see a curve different to that of a “routine”-job. Further adding16o this argument of trading being mostly done by humans organized in cooperations(regarding figure 8 with the seven numbers indicating the days of the week), is researchon anomalies such as the “Monday Effect” applied to our findings (e.g. Cross, 1973;Basher and Sadorsky, 2006). By applying both parametric and nonparametric methods,Caporale and Plastun (2019) find abnormal returns for no other cryptocurrency thanBTC, and that only on Mondays - yet, in figure 8 we can observe that weekly absolutereturns across cryptocurrencies reach their peak only in the period from Tuesdays toaround Thursdays, with a steep decline in activity during the weekends. D a il y
50 100 150 200 250 W ee k l y Trading Volume (a) BTC D a il y
50 100 150 200 250 W ee k l y Trading Volume (b) ETH D a il y
50 100 150 200 250 W ee k l y Trading Volume (c) XRP
Figure 9:
Daily and weekly seasonality: fit of Generalized Additive Model with cubicand p-splines for trading volume of cryptocurrencies (5 min nodes), 01. July 2018 - 31.August 2018. 01. July 2018 - 31. August 2018.Figure 9 presents us a respective lower trading volume during the weekends, comparedto for example Thursdays or especially Fridays. Similar results can be seen in figure 10,presenting us with low volatility on the cryptocurrency market at said times - one as-sumption from this could be taken from the immense influx of financially potent startupsorganized as cooperations in this emerging market (c.f. Benedetti and Kostovetsky, 2018).Yet, we can see that human interaction is shaping how the market behaves during thegiven time frames. Trade limited to regular working hours and days in Europe leads tothe conclusion, that the majority of trades are not done by algorithms, which are active24/7, but by human agents themselves making transactions and orders individually andby hand. This is especially obvious through figure 8, which is presenting a much loweractivity pattern observable during the weekends. Should algorithms really be the driversin this, technically predestined, fully digitized market, then this curve should not dropoff as observable on Saturdays and Sundays. These findings are similar across the board(see appendix sections 5.2 - 5.4). While there is a plethora of well working, open-sourcetrading bots available for these markets, for example via Github (Nevskii, 2019), as wellas an abundance of commercially available trading bots (Norry, 2020), the trust in these- or the knowledge of how to employ them in this emerging market - is certainly low.17his is especially surprising, as the possibility for arbitrage or mean reversion is obviouswith multiple exchanges trading the same assets each with individually different prices,see section 3. The inherent possibility to take advantage of this inefficiency of the dis-tributed trading, with near-simultaneous transactions, leads to great opportunities fortraders unseen in most traditional markets for most assets. Hence we can assume, asalgorithms need humans to get deployed and take action, like reacting to price changes,that the overall impact of these is not significant, if not negligible at all. D a il y
50 100 150 200 250 W ee k l y Volatility (a) BTC D a il y
50 100 150 200 250 W ee k l y Volatility (b) ETH D a il y
50 100 150 200 250 W ee k l y Volatility (c) XRP
Figure 10:
Daily and weekly seasonality: fit of Generalized Additive Model with cubicand p-splines for volatility of cryptocurrencies (5 min nodes), 01. July 2018 - 31. August2018. 01. July 2018 - 31. August 2018.In total we can observe, that the activity patterns displayed in this market not onlytend to express human interaction but also corporate structures as well, as most trad-ing is done Mondays to Fridays, with the weekends expressing a low intensity of tradestaking place. The previously mentioned immense increase of financially potent FinTechentities have attracted absurd amounts of financial backing compared to the output de-livered via initial coin offerings, ICOs for short (c.f. Benedetti and Kostovetsky, 2018;Zetzsche et al., 2019). To enable new industries using the blockchain technology, star-tups and commercial companies have been launching ICOs, similar to the initial publicofferings (IPOs) of companies, to sell tokens in a transparent and decentralized mannerand therefore creating a new method of raising funds without intermediaries, like tradi-tional financial institutes. Some of these tokens are pegged to other (monetary) systemsor even cryptocurrency constructions directly, as these have already gained a high marketacceptance - especially the Ethereum ecosystem is facilitating this by providing excessivetools and documentaries, paired with a focused and growing community of developers, tocreate what they coined as “coloured coins” in order to expand the utility of the existingblockchain (Walters, 2018). Besides the fact, that the legality of ICOs is disputed andpotential responses from regulatory agencies are growing to be imminent, ICOs enableanyone within the community to participate in the investment, providing opportunities18or small-scale investors. Hence the assumption would be, that especially these special-ized corporate startups are working on their backend and maintain their ecosystem, whilstbeing active drivers of trading in this market - yet predominantly human ones.Coming back to the 2010 Flash Crash mentioned in the introduction of this paper,one could argue, that such a flash crash is not possible due to the delay that is inherentto blockchains - the so-called blocktime (Hrdle et al., 2020). However, as research hasshown, it is easily possible to derive sentiment and therefore market reactions from Twit-ter, Facebook, Stocktwits, or similar public forums. As most activity can be seen on therespective cryptocurrency exchanges, where the order books are not handled on-chain,but necessarily off-chain to quickly process the exchange users trading requests (Chen etal., 2019c,a; Qian et al., 2019). Therefore a crash related to certain cryptocurrency pricesmay be seen only after the respective information has been seeded into the network andaccepted as new information to the individual blockchain. This is creating an inherentrisk, as market behaviour can not be seen by only relying on on-chain data to predict cer-tain price movements. The previously mentioned “‘learning” algorithms could thereforebe, if they are employed in this manner, be dangerous if a “false-postive” is identifiedand results in a respective process leading to dumping a certain asset, which in turncould then generate a waterfall when other algorithms, that respond to blockchain pricedata, reply to this movement (Zinovyeva et al., 2019). Hence, a grand scale applicationof algorithms needs to be finely tuned in order to avoid any humanely unforeseeable, buttechnically feasible, consequences.With the cryptocurrency market being easy to join and to actively participate in,financial traders are becoming redundant - unless they provide specialized services. Mak-ing many transactions doesn’t cost time to interact with a trader and money to pay thisperson, as one can do that by hand at home with very low transaction costs. This said,there is a big competition going on between the exchanges, who themselves may act astraders or brokers. The future has to tell if through this competition the rise of the ma-chines and the respective mass employment of algorithmic trading in this digital realmwill become reality.
We have shown, that meanwhile there are certainly grand-scale employers of algorith-mic trading around in this new emerging market of cryptocurrencies, yet, based on thetime-of-day effects and the evidence gained, we can conclude, that the impact of 24/7algorithmic trading is rather negligible given the empirical facts we have at hand. Thisleads us to the conclusion, that even though this new digital market appears predestined19o be ruled by algorithms and specialised AI advisors, the digital realm of cryptocurren-cies has yet to be conquered by the machines and is still firmly in the hands of humansor generally driven by respective startup’s.Further research should certainly step into this breach, that we have proven to beexistent, and create means on how to best exploit this open ground on a market-orientedbasis, as well as on an individual level, say in regards to the exchanges. Necessarily, suchresearch not only needs to be of quantitative or technical origin, but also needs to includea regulatory point of view, as especially this field on blockchain research is more andmore characterized by its evident interdisciplinary nature.
References
Akhtaruzzaman, M., A. Sensoy, and S. Corbet , “The influence of Bitcoin on portfolio diversifi-cation and design,”
Finance Research Letters , 2019, p. 101344.
Basher, S.A. and P. Sadorsky , “Day-of-the-week effects in emerging stock markets,”
Applied Eco-nomics Letters , 2006, (10), 621–628. Benedetti, H. and L. Kostovetsky , “Digital Tulips? Returns to Investors in Initial Coin Offer-ings,” 2018. SSRN. Retrieved on the 12.02.2020 from https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3182169 . Borke, L. and W. H¨ardle , “Q3-D3-LSA: D3.js and Generalized Vector Space Models for StatisticalComputing,” in Lu H. Shen X. H¨ardle, W., ed.,
Handbook of Big Data Analytics , Springer-Verlag BerlinHeidelberg, ISBN 978-3-319-18284-1, DOI: 10.1007/978-3-319-18284-1, 2018, pp. 377–424.
Caporale, G.M. and A. Plastun , “The day of the week effect in the cryptocurrency market,”
Finance Research Letters , 2019, , L. Gil-Alana, A. Plastun, and I. Makarenko , “Intraday Anomalies and Market Efficiency: ATrading Robot Analysis,”
Computational Economics , 2016, , 275295. Chen, C. YH., R. Despres, L. Guo, and T. Renault , “What makes cryptocurrencies spe-cial? Investor sentiment and return predictability during the bubble,” 2019. IRTG 1792 Discus-sion Paper. ISSN: 2568-5619. Forthcoming. . , W. Kuo, L. Guo, and W.K. Hrdle , “Risk of Bitcoin Market: Volatility, Jumps, andForecasts,” 2019. IRTG 1792 Discussion Paper. ISSN: 2568-5619. Forthcoming. . , W.K. Hrdle, and Y. Klochkov , “SONIC: SOcial Network with Influencers and Communities,”2019. IRTG 1792 Discussion Paper. ISSN: 2568-5619. Forthcoming. . Cross, F. , “The Behavior of Stock Prices on Fridays and Mondays,”
Financial Analysts Journal , 1973, (6). mem, M. , “Wall Street Shenanigans: Rampant Trading Bots are Exploiting Crypto Exchanges,”2018. CNN. Retrieved on the 28.06.2019 from . Hautsch, N. , Econometrics of financial high-frequency data , ISBN 978-3642219245.: Springer Science& Business Media, 2011.
Hussein, S. , “The Intraday Behaviour of Bid-Ask Spreads, Trading Volume and Return Volatility:Evidence from DAX30,”
International Journal of Economics and Finance , 2011, (1, February). Hrdle, W. , Applied Nonparametric Regression , Cambridge, New York, Melbourne: Cambridge Uni-versity Press, 1990.
Hrdle, W.K., C.R. Harvey, and R.C.G. Reule , “Understanding Cryptocurrencies,”
Journal ofFinancial Econometrics , 2020.
Nevskii , “Crypto trading bots review 2019,” 2019. Medium. Retrieved on the 25.06.2019 from https://medium.com/@vladnevskii/crypto-trading-bots-review-2018-312cf3e8145c . Norry, A. , “Beginners Guide to Bitcoin & Crypto Trading Bots,” 2020. Blockonomi. Retrieved onthe 12.02.2020 from https://blockonomi.com/bitcoin-trading-bots/ . Petukhina, A., S. Trimborn, W. Hrdle, and H. Elendner , “Investing With CryptocurrenciesEvaluating the Potential of Portfolio Allocation Strategies,” 2020. Quantitative Finance, submitted,RR 20200320.
Platanakis, E. and A. Urquhart , “Should investors include bitcoin in their portfolios? a portfoliotheory approach,”
The British Accounting Review , 2019, p. 100837.
Qian, Y., J. Tu, and W. Hrdle , “Information Arrival, News Sentiment, Volatili-ties and Jumps of Intraday Returns,” 2019. IRTG 1792 Discussion Paper. ISSN: 2568-5619. Forthcoming. . Rschli, K., L. Vetter, W. Breymann, J. Osterrieder, J. Chu, S. Nadarajah, and S. Chan ,“Stylized Facts of Cryptourrencies,” Technical Report, Zurich University of Applied Sciences/ZHAW2017.
Schnaubelt, M., J. Rende, and C. Krauss , “Testing Stylized Facts of Bitcoin Limit Order Books,”
Journal of Risk and Financial Management , 2019, (25, doi:10.3390/jrfm12010025). Trimborn, S. and W. Hrdle , “CRIX an Index for cryptocurrencies,”
Journal of Empirical Finance ,2018,
49, December , 107–122, https://doi.org/10.1016/j.jempfin.2018.08.004. , M. Li, and W.K Hrdle , “Investing with cryptocurrencies-A liquidity constrained investmentapproach,”
Published version: Trimborn, S., Li, M. and WK Hrdle (2019)” Investing with Cryp-tocurrenciesa Liquidity Constrained Investment Approach” Journal of Financial Econometrics, doi.org/10.1093/jjfinec/nbz016 , 2018.
Walters, S. , “What are Colored Coins? The Ultimate Guide,” 2018. unblock. Retrieved on the13.02.2020 from https://unblock.net/what-are-colored-coins/ . Wand, Matt P , “Fast approximate inference for arbitrarily large semiparametric regression modelsvia message passing,”
Journal of the American Statistical Association , 2017, (517), 137–168.
Wood, S. , “Stable and efficient multiple smoothing parameter estimation for generalized additivemodels,”
Journal of the American Statistical Association , 2004, (467), 673–686. “Fast stable restricted maximum likelihood and marginal likelihood estimation of semiparametricgeneralized linear models,” Journal of the Royal Statistical Society: Series B (Statistical Methodology) ,2011, (1), 3–36., “gam.models,” 2019. RDocumentation. Retrieved on the 17.02.2020 from . Wood, Simon N , Generalized additive models: an introduction with R , CRC press, 2017. , Yannig Goude, and Simon Shaw , “Generalized additive models for large data sets,”
Journal ofthe Royal Statistical Society: Series C (Applied Statistics) , 2015, (1), 139–155. Zetzsche, D.A., R.P. Buckley, D. Arner, and L. F¨ohr , “The ICO Gold Rush: It’s a Scam, It’sa Bubble, It’s a Super Challenge for Regulators,”
Harvard International Law Journal , 2019, (2),275295. Zhang, W., P. Wang, X. Li, and D. Shen , “Some stylized facts of the cryptocurrency market,”
Applied Economics , 2018, (55), 5950–5965. Zhang, Y., S. Chan, J. Chu, and S. Nadarajah , “Stylised facts for high frequency cryptocurrencydata,”
Physica A , 2019, , 598–612.
Zinovyeva, E., W.K. Hrdle, and S. Lessmann , “Antisocial Online Behavior De-tection Using Deep Learning,” 2019. IRTG 1792 Discussion Paper. ISSN: 2568-5619. Forthcoming. . Appendix
Abbrev. CC WebsiteBCH Bitcoin Cash bitcoincash.org
BTC (XBT) Bitcoin bitcoin.com , bitcoin.org DASH Dash dash.org
ETC Ethereum Classic ethereumclassic.github.io
ETH Ethereum ethereum.org
LTC Litecoin litecoin.com , litecoin.org REP Augur augur.net
STR Stalker staker.network
XMR Monero getmonero.org
XRP Ripple ripple.com
ZEC Zcash z.cash - -
01 2018 - -
08 2018 - -
15 2018 - -
22 2018 - -
29 2018 - -
05 2018 - -
12 2018 - -
19 2018 - - Date500600700800900 B C H P r i c e (a) BCH - -
01 2018 - -
08 2018 - -
15 2018 - -
22 2018 - -
29 2018 - -
05 2018 - -
12 2018 - -
19 2018 - - Date101214161820 E T C P r i c e (b) ETC - -
01 2018 - -
08 2018 - -
15 2018 - -
22 2018 - -
29 2018 - -
05 2018 - -
12 2018 - -
19 2018 - - Date5060708090 L T C P r i c e (c) LTC Figure 11:
Candlestick charts for individual price movements. 01. July 2018 - 31.August 2018. 23 ul Aug Sep − . − . . . . . . Index B CH (a) BCH Jul Aug Sep − . − . . . . . . Index E T C (b) ETC Jul Aug Sep − . . . . Index L T C (c) LTC Figure 12:
Intraday 5-minutes log-returns. 01. July 2018 - 31. August 2018.
Jul Aug Sep . . . . . Index B CH (a) BCH Jul Aug Sep . . . . Index E T C (b) ETC Jul Aug Sep . . . . . Index L T C (c) LTC Figure 13:
Intraday volatility (absolute values of 5-minutes log-returns) . 01. July 2018- 31. August 2018. − − − − + − − − − BCH
Time A b s . r e t u r n s :
05 02 :
05 04 :
05 06 :
05 08 :
05 10 :
05 12 :
05 14 :
05 16 :
05 18 :
05 20 :
05 22 : (a) BCH − − + − − ETC
Time A b s . r e t u r n s :
05 02 :
05 04 :
05 06 :
05 08 :
05 10 :
05 12 :
05 14 :
05 16 :
05 18 :
05 20 :
05 22 : (b) ETC − − + − − LTC
Time A b s . r e t u r n s :
05 02 :
05 04 :
05 06 :
05 08 :
05 10 :
05 12 :
05 14 :
05 16 :
05 18 :
05 20 :
05 22 : (c) LTC Figure 14:
Generalized additive model of volatility. 01. July 2018 - 31. August 2018.24 − BCH
Time T r ad i ng v o l u m e 00 :
00 02 :
00 04 :
00 06 :
00 08 :
00 10 :
00 12 :
00 14 :
00 16 :
00 18 :
00 20 :
00 22 : (a) BCH − − ETC
Time T r ad i ng v o l u m e 00 :
00 02 :
00 04 :
00 06 :
00 08 :
00 10 :
00 12 :
00 14 :
00 16 :
00 18 :
00 20 :
00 22 : (b) ETC − LTC
Time T r ad i ng v o l u m e 00 :
00 02 :
00 04 :
00 06 :
00 08 :
00 10 :
00 12 :
00 14 :
00 16 :
00 18 :
00 20 :
00 22 : (c) LTC Figure 15:
Generalized Additive Model of trading volume of cryptocurrencies. 01. July2018 - 31. August 2018. D a il y
50 100 150 200 250 W ee k l y Trading Volume (a) BCH D a il y
50 100 150 200 250 W ee k l y Trading Volume (b) ETC D a il y
50 100 150 200 250 W ee k l y Trading Volume (c) LTC
Figure 16:
Daily and weekly seasonality: fit of Generalized Additive Model with cubicand p-splines for trading volume of cryptocurrencies (5 min nodes), 01. July 2018 - 31.August 2018. 01. July 2018 - 31. August 2018.25 a il y
50 100 150 200 250 W ee k l y Volatility (a) BCH D a il y
50 100 150 200 250 W ee k l y Volatility (b) ETC D a il y
50 100 150 200 250 W ee k l y Volatility (c) LTC
Figure 17:
Daily and weekly seasonality: fit of Generalized Additive Model with cubicand p-splines for volatility of cryptocurrencies (5 min nodes), 01. July 2018 - 31. August2018. 01. July 2018 - 31. August 2018. - -
01 2018 - -
08 2018 - -
15 2018 - -
22 2018 - -
29 2018 - -
05 2018 - -
12 2018 - -
19 2018 - - Date140160180200220240260280 D A S H P r i c e (a) DASH - -
01 2018 - -
08 2018 - -
15 2018 - -
22 2018 - -
29 2018 - -
05 2018 - -
12 2018 - -
19 2018 - - Date152025303540 R E P P r i c e (b) REP - -
01 2018 - -
08 2018 - -
15 2018 - -
22 2018 - -
29 2018 - -
05 2018 - -
12 2018 - -
19 2018 - - Date0.1750.2000.2250.2500.2750.3000.3250.350 S T R P r i c e (c) STR Figure 18:
Candlestick charts for individual price movements (60-minutes intervals).01. July 2018 - 31. August 2018.
Jul Aug Sep − . − . − . . . . . Index D AS H (a) DASH Jul Aug Sep − . . . . Index R EP (b) REP Jul Aug Sep − . − . . . . . Index S T R (c) STR Figure 19:
Intraday log-returns (5-minutes). 01. July 2018 - 31. August 2018.26 ul Aug Sep . . . . . . . . Index D AS H (a) DASH Jul Aug Sep . . . . . . Index R EP (b) REP Jul Aug Sep . . . . . . . Index S T R (c) STR Figure 20:
Intraday volatility (absolute 5-minutes log-returns). 01. July 2018 - 31.August 2018. − − + − − DASH
Time A b s . r e t u r n s :
05 02 :
05 04 :
05 06 :
05 08 :
05 10 :
05 12 :
05 14 :
05 16 :
05 18 :
05 20 :
05 22 : (a) DASH − − + − − REP
Time A b s . r e t u r n s :
05 02 :
05 04 :
05 06 :
05 08 :
05 10 :
05 12 :
05 14 :
05 16 :
05 18 :
05 20 :
05 22 : (b) REP − − + − STR
Time A b s . r e t u r n s :
05 02 :
05 04 :
05 06 :
05 08 :
05 10 :
05 12 :
05 14 :
05 16 :
05 18 :
05 20 :
05 22 : (c) STR Figure 21:
Generalized Additive Model of volatility of cryptocurrencies. 01. July 2018- 31. August 2018. − − DASH
Time T r ad i ng v o l u m e 00 :
00 02 :
00 04 :
00 06 :
00 08 :
00 10 :
00 12 :
00 14 :
00 16 :
00 18 :
00 20 :
00 22 : (a) DASH − − REP
Time T r ad i ng v o l u m e 00 :
00 02 :
00 04 :
00 06 :
00 08 :
00 10 :
00 12 :
00 14 :
00 16 :
00 18 :
00 20 :
00 22 : (b) REP − STR
Time T r ad i ng v o l u m e 00 :
00 02 :
00 04 :
00 06 :
00 08 :
00 10 :
00 12 :
00 14 :
00 16 :
00 18 :
00 20 :
00 22 : (c) STR Figure 22:
Generalized Additive Model of intraday trading volume of cryptocurrencies.01. July 2018 - 31. August 2018. 27 a il y
50 100 150 200 250 W ee k l y Trading Volume (a) DASH D a il y
50 100 150 200 250 W ee k l y Trading Volume (b) REP D a il y
50 100 150 200 250 W ee k l y Trading Volume (c) STR
Figure 23:
Daily and weekly seasonality: fit of Generalized Additive Model with cubicand p-splines for trading volume of cryptocurrencies (5 min nodes), 01. July 2018 - 31.August 2018. 01. July 2018 - 31. August 2018. D a il y
50 100 150 200 250 W ee k l y Volatility (a) DASH D a il y
50 100 150 200 250 W ee k l y Volatility (b) REP D a il y
50 100 150 200 250 W ee k l y Volatility (c) STR
Figure 24:
Daily and weekly seasonality: fit of Generalized Additive Model with cubicand p-splines for volatility of cryptocurrencies (5 min nodes), 01. July 2018 - 31. August2018. 01. July 2018 - 31. August 2018. - -
01 2018 - -
08 2018 - -
15 2018 - -
22 2018 - -
29 2018 - -
05 2018 - -
12 2018 - -
19 2018 - - Date8090100110120130140150 X M R P r i c e (a) XMR - -
01 2018 - -
08 2018 - -
15 2018 - -
22 2018 - -
29 2018 - -
05 2018 - -
12 2018 - -
19 2018 - - Date120140160180200220 Z E C P r i c e (b) ZEC Figure 25:
Chandlestick charts for individual price movements. 01. July 2018 - 31.August 2018. 28 ul Aug Sep − . − . . . . Index X M R (a) XMR Jul Aug Sep − . − . . . . . . . Index Z E C (b) ZEC Figure 26:
Intraday 5-minutes log-returns. 01. July 2018 - 31. August 2018.
Jul Aug Sep . . . . . Index X M R (a) XMR Jul Aug Sep . . . . . . Index Z E C (b) ZEC Figure 27:
Intraday Volatility. 01. July 2018 - 31. August 2018. − − − − + − − − XMR
Time A b s . r e t u r n s :
05 02 :
05 04 :
05 06 :
05 08 :
05 10 :
05 12 :
05 14 :
05 16 :
05 18 :
05 20 :
05 22 : (a) XMR − − − − + − − ZEC
Time A b s . r e t u r n s :
05 02 :
05 04 :
05 06 :
05 08 :
05 10 :
05 12 :
05 14 :
05 16 :
05 18 :
05 20 :
05 22 : (b) ZEC Figure 28:
Generalized Additive Model of volatility of cryptocurrencies. 01. July 2018- 31. August 2018. 29
XMR
Time T r ad i ng v o l u m e 00 :
00 02 :
00 04 :
00 06 :
00 08 :
00 10 :
00 12 :
00 14 :
00 16 :
00 18 :
00 20 :
00 22 : (a) XMR − ZEC
Time T r ad i ng v o l u m e 00 :
00 02 :
00 04 :
00 06 :
00 08 :
00 10 :
00 12 :
00 14 :
00 16 :
00 18 :
00 20 :
00 22 : (b) ZEC Figure 29:
Generalized Additive Model of the 62 intraday trading volume of cryptocur-rencies. 01. July 2018 - 31. August 2018. D a il y
50 100 150 200 250 W ee k l y Trading Volume (a) XMR D a il y
50 100 150 200 250 W ee k l y Trading Volume (b) ZEC
Figure 30:
Daily and weekly seasonality: fit of Generalized Additive Model with cubicand p-splines for trading volume of cryptocurrencies (5 min nodes), 01. July 2018 - 31.August 2018. 01. July 2018 - 31. August 2018.30 a il y
50 100 150 200 250 W ee k l y Volatility (a) XMR D a il y
50 100 150 200 250 W ee k l y Volatility (b) ZEC