Received: 5 March 2019 Revised: 9 October 2019 Accepted: 18 June 2020

DOI: 10.1002/ijfe.1851

RESEARCH ARTICLE

Market efficiency and volatility persistence

of cryptocurrency during pre- and post-crash periods
of Bitcoin: Evidence based on fractional integration

OlaOluwa S. Yaya1,2 | Ahamuefula E. Ogbonna1,2 | Robert Mudida3 |

Nuruddeen Abu4

1
Economic and Financial Statistics Unit,
Department of Statistics, University of Abstract
Ibadan, Ibadan, Nigeria This article investigates both market efficiency and volatility persistence in
2
Centre for Econometric and Allied 12 cryptocurrencies during pre-crash and post-crash periods. The article con-
Research, University of Ibadan, Ibadan,
tributes to the debate on the market efficiency of cryptocurrencies in the pres-
Nigeria
3
Institute for Public Policy and
ence of volatility, considering robust fractional integration methods in both
Governance, Strathmore University, linear and nonlinear setups. We find that markets of Bitcoin and most altcoins
Nairobi, Kenya considered in our study can be dubbed as efficient, and are also highly volatile,
4
Department of Economics, Umaru Musa
particularly, in the post-crash period that we are experiencing now. The
Yar'adua University, Katsina, Nigeria
volatilities are more likely to persist for a shorter period than volatilities in the
Correspondence pre-crash period. Our work, therefore, renders important information to cryp-
OlaOluwa S. Yaya, Department of
tocurrency market participants and portfolio managers.
Statistics, University of Ibadan, Ibadan,
Nigeria.
KEYWORDS
Email: [email protected]; o.s.olaoluwa@
gmail.com Bitcoin, cryptocurrency, fractional integration, market efficiency, virtual currency

JEL CLASSIFICATION
C22

1 | INTRODUCTION over 2000 cryptocurrency types are being traded 24/7 on

the internet.1 For the first time, having attained a steady
As a result of the 2007/2008 crash in the global financial and eventually astronomical increase in prices by late
market, individuals and traders have lost interest in 2017 to around $19,000 per coin, Bitcoin crashed and
investing in economic policy driven investments, such as caused prices of other alternative coins to fall (see
stocks, foreign exchange rates, oil and gold. Pricing of Cointelegraph, 2018). Other alternative coins (altcoins),
these assets is driven by global markets, which are which include Dash, Doge, Ethereum, Litecoin, Monero,
influenced considerably by United States economic and Ripple, Stellar, Vertcoin, and so on, were introduced due
political activities, and are thus traditional market sys- to the rising popularity of Bitcoin. Litecoin, for example,
tems. As a result, the global market is moving to a non- was introduced to conserve the computing power for
traditional monetary system, independent of government mining coin; Dash was introduced to render faster
policy and politics. Cryptocurrency, the digital currency, processing and improved privacy protection (Ciaian &
has become another investment source, accepted in the Rajcaniova, 2018). All other cryptocurrency types have
global market (Weber, 2016). Its market was popularized reasons for being in existence. Yet, none of the other
with the introduction of the first digital coin, the Bitcoin cryptocurrency types have overtaken Bitcoin's price.
in 2009, priced as low as $5 per coin, and since then, well While the process to accept or reject cryptocurrency by

Int J Fin Econ. 2020;1–18. wileyonlinelibrary.com/journal/ijfe © 2020 John Wiley & Sons, Ltd. 1
2 YAYA ET AL.

TABLE 1 A cursory review of the literature on efficient market efficiency of cryptocurrency

Authors Objectives Methodology Data structure Findings

Bariviera, 2017; Market efficiency of Runs test, Ljung–Box, Daily closing prices for Informational inefficiency in
Nadarajah & Bitcoin automatic variance Bitcoin in USD from the Bitcoin return series,
Chu, 2017; ratio test, Bartels 1 August 2010 to 31 which was dependent on the
Urquhart, 2016 test, BDS and R/S July 2016 sample period considered.
Hurst The informational efficiency
was observed in periods
covering recent years
Cheah, Mishra, Parhi, & Cross-market Bitcoin FCVAR model Daily closing prices for Fractional co-integration in
Zhang, 2018 prices as long memory Bitcoin in USD from cross-country Bitcoin prices,
processes and study 27 November 2011 to while disequilibrium errors
dynamic 17 March 2017, for were found to adjust slowly
interdependence five developed over the long run.
countries—Europe, Heterogeneous degree of
United States, inefficiency was observed in
Australia, Canada the Bitcoin markets
and United Kingdom
Bariviera, 2017; Tiwari, Market efficiency via time- DFA and MF-DCCA 2011–2017 The R/S method detected long
Jana, Das, & varying behaviour of memory behaviour in the
Roubaud, 2018; long memory of returns cryptocurrency, whereas
Zhang, Wang, Li, & on Bitcoin and volatility DFA method discriminated
Shen, 2018 informational efficiency
across time. There was
persistence of daily returns
in the first half of the
sample, whereas its
behaviour was more
informational efficient since
2014
Asanidze, 2018 Efficiency of Bitcoin Three different Daily data in the A changing level of returns,
market according to variance ratio tests, period from finding both statistically
adaptive market Kim's wild September 2010 to significant and non-
hypothesis, to check for bootstrapping based May 2017 significant values. His results
linear independency in on Lo and McKinlay were in accordance with the
returns (1988), Wright's adaptive market hypothesis,
(2000) Signs and confirming evolving
Rank tests predictability, and thus, the
efficiency of Bitcoin
Caporale & Price over reactions in The t test, ANOVA and Both categories of tests
Plastun, 2018 BitCoin, LiteCoin, ripple regression analysis confirmed the presence of
and dash by means of with dummy price patterns in
some parametric and variables, while the cryptocurrency, implying
non-parametric tests non-parametric test dependence of prices in these
was the Mann– markets. These market
Whitney U test reactions detected in the
cryptocurrency market
further indicated evidence
that abnormal profit by
traders was impossible and
this supported the EMH
Balcilar, Bouri, Gupta, & Market efficiency and Unit roots, Different time samples Evidence of long-range
Roubaud, 2017; Bouri long-range dependence autocorrelation, dependence and nonlinearity
et al., 2018; Yaya, in returns nonlinearity and in some cryptocurrencies
Ogbonna, & long-range
Olubusoye, 2019 dependence
YAYA ET AL. 3

TABLE 1 (Continued)

Authors Objectives Methodology Data structure Findings

Wei, 2018 Predictability of returns Ljung–Box, Bartels, lo 456 different Bitcoin's market return was
and MacKinlays cryptocurrencies found to be efficient, while
variance ratio, wild- other cryptocurrencies still
bootstrapped AVR, indicated signs of
the non-parametric autocorrelations
BDS and the R/S
Hurst exponent tests
Caporale, Gil-Alana, & Market efficiency in some Fractional persistence 2013–2017 Markets of those
Plastun, 2018 cryptocurrencies, such approach cryptocurrencies were
as BitCoin, LiteCoin, inefficient during the period;
ripple, dash, etc. though there was evidence of
correlations among those
returns time series
Caporale & Day of the week effect in Regression analysis 1 January 2013 to 31 Inconclusive evidence that the
Plastun, 2018 the cryptocurrency and simulation December 2017 cryptocurrency market is not
market approach efficient
Caporale et al., 2018 Persistence in the R/S analysis and 2013–2017 Market exhibits persistence
cryptocurrency market fractional integration and the degree changes over
time, implying some
evidence of market
inefficiency
Caporale, Plastun, & The role of the frequency Parametric and non- 1 May 2013 to 31 May The frequency of price over-
Oliinyk, 2019 of price over-reactions parametric methods, 2018 reactions of Bitcoin was able
in the cryptocurrency such as the ADF to provide useful information
Market in the case of tests, Granger necessary in the prediction of
BitCoin causality tests, price movements in
correlation analysis, cryptocurrency market. Also,
regression analysis no evidence of seasonality
with dummy was observed
variables, ARIMA
and ARMAX (with
exogenous variable
X) models, neural
net models, and
VAR models
Sigaki, Perc, & Informational market Permutation entropy Daily closing prices Evidence of informational
Ribeiro, 2019 efficiency of and statistical and the market market efficiency in some
cryptocurrencies complexity and capitalization of cryptocurrencies
hierarchical 1,509
clustering procedure cryptocurrencies
Vidal-Tomas, Ibadnez, Cap-weighted and Daily closing prices Cryptocurrency market is
& Farinos, 2019 equally weighted from 2015 to 2017 found to be weak-form
market portfolios inefficient due to the
behaviour of all the altcoins,
it is more inefficient over
time, especially in 2017
Charfeddine & Efficient market GHP, GSP, LW and 2013–2018 Evidence of inefficiency in
Maouchi, 2019 hypothesis of top four ELW most markets
cryptocurrencies
Grobys & Sapkota, 2019 Using Fama and 1 January 2014 to 31
French's (2008) December 2018
portfolio approach
(Continues)
4 YAYA ET AL.

TABLE 1 (Continued)

Authors Objectives Methodology Data structure Findings

Chu, Zhang, & AMH for two largest Regression analysis Hourly price from 1 Market efficiency varies over
Chan, 2019 cryptocurrencies July 2017 to 1 time, but sentiment and
September 2018 types of news and events do
not significantly influence
the efficiency of the market
of cryptocurrencies
Kristoufek & Efficient market Efficiency index and 1 January 2015 to 30 Cryptocurrencies were
Vosvrda, 2019 hypothesis bootstrapping June 2018 historically inefficient, but
most of these currencies
were efficient between July
2017 and June 2018

Note: Compiled by the Authors.

Abbreviations: ADF, augmented Dickey–Fuller; AMH, adaptive market hypothesis; ARIMA, autoregressive integrated moving average;
AVR, automatic variance; DFA, detrended fluctuation analysis; ELW, exact local whittle; EMH, efficient market hypothesis; FCVAR, frac-
tionally co-integrated vector autoregressive; GHP, Geweke and porter-Hudak; GSP, Gaussian semi-parametric; LW, local whittle; MF-DCCA,
multifractal detrended cross-correlation analysis; VAR, vector autoregressive.

economists and regulators is ongoing, the awareness con- returns are predictable, thus making it possible for inves-
tinues to grow. Cryptocurrency is also known for its huge tors to make abnormal returns. The random walk
volatility, although there are very few papers considering hypothesis implies market efficiency, since market
this. The predictability of the price, that is, efficiency of returns are unpredictable in such a time process. Thus, as
the market, could also be of interest to portfolio man- investors try to beat the market more, the market
agers and traders. Meanwhile, various academic papers becomes more efficient. The form of market efficiency
from the economic and finance perspectives have concen- defined in Fama (1970) is the weak form efficiency,
trated on the influence of other assets, such as gold and where current asset prices are expected to reflect all infor-
stocks, on cryptocurrency (see Barber, Boyen, Shi, & mation in the market transactional data, and no technical
Uzun, 2012; Bouri et al., 2018; Corbet, Meegan, Larkin, data analysis could help in realizing abnormal returns
Lucey, & Yarovaya, 2018; Dyhrberg, 2016a, 2016b; Gla- from such dataset.
ser, Zimmermann, Haferkorn, Weber, & Siering, 2014). Efficiency and volatility are inseparable, since effi-
The level of market efficiency is useful in the evalua- ciency is a function of market returns, while volatility is a
tion of the investment environment, in the description of function of variation from such returns, and persistence
the financial market and in knowing how to develop the is the time it takes to fizzle out. These variations are
market further. The efficiency or inefficiency of crypto- proxied by absolute or squared returns. Investigating effi-
currency market will render useful information to market ciency and volatility persistence in cryptocurrency mar-
players. Efficiency in a market posits that nothing but kets will, therefore, interest readers. In an efficient
own past information predicts the future dynamics of market, short period of time is expected for the effect of
market prices, that is, other influences, such as domestic price shocks/volatility to fizzle out.
and macroeconomic policy, do not influence price in this In the present article, we investigate market efficiency
context. The standard definition of efficient market in and volatility persistence in some highly priced and capi-
Fama (1965) says: ‘In an efficient market, at any point in talized cryptocurrencies based on daily data from
time, the actual price of a security will be a good estimate 7 August 2015 to 28 November 2018. We considered sam-
of its intrinsic value’. Fama (1970) then developed the ples, up to late 2017 cryptocurrency crash, and samples
efficiency market hypothesis (EMH), which states that after the crash, in order to remove the influence of struc-
prices of assets already contain past information, and in tural breaks in the market returns. We used fractional
the event of new information, there is a quick adjustment integration techniques on the returns to test the hypothe-
to the price such that the security is valued correctly. sis of market efficiency; while the squared returns were
That is, are returns of cryptocurrency predictable? used as proxy for volatility, where long memory evidence
Returns are expected to be unpredictable for EMH to in squared returns indicates the extent of volatility persis-
hold, while in an inefficient market, returns are predict- tence. Our approach of fractional integration estimation
able (see, Lim & Brooks, 2011). In an inefficient market, is robust, as it allows for both nonlinearity and possible
YAYA ET AL. 5

TABLE 2 A cursory review of the literature on volatility persistence of cryptocurrency

Authors Objectives Methodology Data structure Findings

Dyhrberg, 2016a,b Cryptocurrencies as GARCH modelling 19 July 2010 to 22 May Cryptocurrencies as just diversifier
alternative hedging 2015 devices rather than hedge
options instruments
Katsiampa, 2017 Specification of GARCH modelling 18 July 2010 to 1 CGARCH was found as an
optimal volatility October 2016 optimal model
model for Bitcoin
pricing among many
asymmetric GARCH
variants
Canh, Wongchoti, Volatility spillovers Dynamic conditional Daily closing prices Existence of volatility spillovers
Thanh, & and structural breaks correlation- from 5 August 2014 with strong positive correlations
Thong, 2019 in seven largest multivariate GARCH to 31 December 2018 between cryptocurrencies and
cryptocurrencies presence of structural breaks in
some
Bouri, Gil-Alana, Persistence in the level Parametric and semi- Two daily series of There was evidence of a long
Gupta, & and volatility of parametric Bitcoin price for two memory volatility of Bitcoin
Roubaud, 2019 Bitcoin price techniques different periods— price
August 2011 to April
2016, and July 2010
to December 2015
Bouri, Lau, Lucey, & The role of trading CGCD Seven-day week daily Trading volume predicts extreme
Roubaud, 2019 volume in the price returns of positive and negative returns of
predictability of seven large all cryptocurrencies, while it
return and volatility cryptocurrencies predicts volatility for only three
in the from April 2013 to cryptocurrencies (Litecoin, Nem,
cryptocurrency December 2017 and dash) particularly when
market volatility is less
Kumar & Dynamics of volatility GARCH and wavelet 15 August 2015 to 18 Existence of moderate volatility
Anandarao, 2019 spillover across four techniques January 2018 spillover is confirmed, and the
major volatility spillover is caused by
cryptocurrency the shocks in Bitcoin prices and
returns other exogenous factors
Ardia, Bluteau, & Presence of regime Markov-switching 19 August 2011 to 2 Bitcoin daily returns exhibit
Rüede, 2019 changes in the GARCH models March 2018 regime changes in their
GARCH volatility volatility dynamics
dynamics of Bitcoin
returns
Chaim & Dynamics of Bitcoin Markov chain Monte April 2013 to May 2018 Jumps to returns are
Laurini, 2018 daily volatility and Carlo procedures contemporaneous, while jumps
returns and threshold to volatility are permanent
sampling scheme
Chaim & Returns and volatility Mixed Markov chain 16 August 2015 to 31 Episodes of high volatility in early
Laurini, 2019 dynamics of major Monte Carlo October 2018 2017 and in early 2018
cryptocurrencies procedure and
Geweke and porter-
Hudak method
Charles & Volatility of four GARCH, APARCH, June 2014 to Cryptocurrency returns are
Darné, 2019 cryptocurrencies IGARCH and November 2018 influenced by infinite
FIGARCH persistence, and the jumps and
structural breaks are important
in modelling volatility of
cryptocurrencies
(Continues)
6 YAYA ET AL.

TABLE 2 (Continued)

Authors Objectives Methodology Data structure Findings

Cheikh, Zaied, & Presence of STGARCH 28 April 2013 to 1 The presence of a positive return–
Chevallier, 2019 asymmetric volatility December 2018 volatility relationship supports
dynamics in the safe-haven hypothesis in
cryptocurrencies cryptocurrencies
Fakhfekh & Optimum models for Five types of GARCH 7 August 7 2017 to 12 TGARCH model turned out to be
Jeribi, 2019 modelling most model December 2018 the most effective in the
popular modelling of most
cryptocurrencies cryptocurrencies
associated volatility

Note: Compiled by the Authors.

Abbreviations: APARCH, Asymmetric Power Autoregressive Conditional Heteroscedasticity; CGARCH, component generalized autoregressive
conditional heteroscedasticity; CGCD, Copula-granger-causality in distribution; FIGARCH, Fractionally Integrated Generalized Autoregressive
Conditional Heteroscedasticity; GARCH, generalized autoregressive conditional heteroscedasticity; IGARCH, Integrated Generalized Auto-
regressive Conditional Heteroscedasticity; STGARCH, smooth transitiion generalized autoregressive conditional heteroscedasticity.

smooth breaks in the returns and squared returns of market efficiency in Bitcoin using a different perspective.
cryptocurrency series, noting that Perron (2006) unit The authors considered daily closing prices of Bitcoin in
root-break test gives different break dates when one is United States dollars in five countries—Europe, United
not too sure of the specification of constant and trend States, Australia, Canada and United Kingdom—and found
in the testing regression. This is unlike the unit root/inte- evidence of co-integration with long-range dependence. The
gration testing framework by Augmented Dickey–Fuller authors further stated that the presence of co-integration
(ADF), Phillips–Perron (PP), Ng–Perron tests, etc., which informed market efficiency of Bitcoin in those countries.
restrictively test series between non-stationarity I(1) and sta-
tionary I(0) hypotheses. Even the random-walk hypothesis
needs sophisticated statistical tests, other than the conven- 2.2 | Modelling volatility and persistence
tional unit root tests, which have been found to lack power of cryptocurrency
under different alternative tests, particularly the fractional
unit root alternatives as noted in Diebold and Rud- Table 2 presents the literature on volatility modelling and
ebusch (1991), Hassler and Wolters (1994), Lee and persistence of cryptocurrencies using univariate and multi-
Schmidt (1996), among others. variate GARCH variants. Other authors applied Markov-
The rest of the article is structured as follows: Section 2 switching GARCH models, Markov Chain Monte Carlo and
presents a survey of the literature on market efficiency and wavelets techniques to study price variations in Bitcoin and
volatility of Bitcoin and cryptocurrency. Section 3 presents other cryptocurrency types.
the time series analysis approach used in this article. As Our approach of market efficiency and volatility per-
mentioned in the introduction, this is the fractional integra- sistence of Bitcoin and alternative cryptocurrencies is dif-
tion methodology in the linear and nonlinear case, for ferent from that applied so far in the literature, since it is
returns and squared returns of cryptocurrency series. In Sec- based on fractionally integrated time-series techniques,
tion 4, we present the data, some pre-tests and empirical and relates market efficiency with volatility persistence
results. Section 5 renders the concluding remarks. in the context of pre and post-crash periods of Bitcoin,
and provides robustness checks based on nonlinearity.
The alternative literature considered Bitcoin or other
2 | REVIEW OF LITERATURE cryptocurrency in relationship with other asset prices,
trying to find possible co-integration relationships, while
2.1 | On EMH of cryptocurrency others use different approaches.

Several prominent papers have examined the market ineffi-

ciency of the cryptocurrency, with specific focus on the most 3 | TIME SERIES ANALYSIS
valuable of all, the Bitcoin. This literature is summarized in APPROACH
Table 1. Most of the papers focused on Bitcoin prices in
United States dollars, while others considered Bitcoin with The Fractional integration framework is used throughout
other cryptocurrencies. Cheah et al. (2018) investigated to investigate both market efficiency and volatility
YAYA ET AL. 7

.10 .20
FIGURE 1 Plots of returns, absolute Bitcoin Bitcoinr Bitcoinsqr Dash Dashr Dashsqr .15
.05
and squared returns [Colour figure can be .10
.05
.00
viewed at wileyonlinelibrary.com] 20,000
.00
-.05
1,600
-.05
-.10
15,000 1,200 -.15
-.10

10,000 800

5,000 400

0 0
III IV I II III IV I II III IV I II III IV III IV I II III IV I II III IV I II III IV
2015 2016 2017 2018 2015 2016 2017 2018

.6 .3
Digibyte Digibyter Digibytesqr Doge Doger Dogesqr
.2
.4
.1
.2 .020 .0
.16
.0 .016 -.1

.12 -.2
-.2 .012

.08 -.3
.008

.04 .004

.00 .000
III IV I II III IV I II III IV I II III IV III IV I II III IV I II III IV I II III IV
2015 2016 2017 2018 2015 2016 2017 2018

.4 .3
Ethereum Ethereumr Ethereumsqr Litecoin Litecoinr Litecoinsqr
.2 .2

1,500 .0 .1

1,250 -.2 400 .0

1,000 -.4 -.1

300
750 -.6 -.2
200
500 -.8
100
250

0 0
III IV I II III IV I II III IV I II III IV III IV I II III IV I II III IV I II III IV
2015 2016 2017 2018 2015 2016 2017 2018

.2 .3
Maidsafecoin Maidsafecoinr Maidsafecoinsqr Monero Moneror Monerosqr
.1 .2

.1
.0
.0
-.1 500
1.2 -.1
1.0 -.2 400
-.2
0.8 300
0.6
200
0.4
100
0.2
0.0 0
III IV I II III IV I II III IV I II III IV III IV I II III IV I II III IV I II III IV
2015 2016 2017 2018 2015 2016 2017 2018

.6 .6
Nem Nemr Nemsqr Ripple Rippler Ripplesqr
.4 .4

.2
.2

4 .0
2.0 .0
-.2
3
1.5 -.2
-.4
2
1.0

0.5 1

0.0 0
III IV I II III IV I II III IV I II III IV III IV I II III IV I II III IV I II III IV
2015 2016 2017 2018 2015 2016 2017 2018

.4
.4
Vertcoin Vertcoinr Vertcoinsqr
Stellar Stellarr Stellarsqr .3
.2
.2

.1 .0

.0 10
1.0 -.2
-.1 8
0.8
-.2 -.4
6
0.6
4
0.4
2
0.2
0
0.0
III IV I II III IV I II III IV I II III IV
III IV I II III IV I II III IV I II III IV
2015 2016 2017 2018
2015 2016 2017 2018

persistence in this article. Recall that the unit integration the same spirit, Robinson (1994) sets up a testing frame-
by Dickey and Fuller (1979) (the Augmented Dickey– work for the three testing regression models. The
Fuller) test is based on three regression models of no con- approach uses the Lagrange Multiplier (LM) test, with
stant, constant only and linear trend with constant. In the model,
 8

TABLE 3 Descriptive statistics

Bitcoin Dash Digibyte Doge Ethereum Litecoin Maidsafecoin Monero Nem Ripple Stellar Vertcoin
Pane A: Prices
Full sample
Mean 3,596.646 177.316 0.013 0.002 213.463 47.745 0.234 73.391 0.143 0.256 0.093 0.996
Maximum 19,475.8 1,555.59 0.127 0.017 1,397.48 359.13 1.18 470.29 1.840 3.360 0.892 9.460
Minimum 210.07 2.08 0.000 0.000 0.432 2.64 0.012 0.369 0.000 0.004 0.001 0.018
SD 3,961.201 250.49 0.018 0.002 278.344 65.205 0.209 98.401 0.245 0.408 0.145 1.74
Jarque–Bera 403.8** 2,328.5** 3,042.7** 2,329.2** 626.3** 1,302.3** 634.8** 741.3** 12,569.1** 12,851.3** 962.2** 3,372.4**
Pre-crash sample
Mean 1,715.545 87.842 0.004 0.001 85.793 18.379 0.172 27.167 0.061 0.068 0.011 0.52
Maximum 19,475.8 1,014.51 0.06 0.006 700.59 315.36 0.716 328.06 0.684 0.862 0.235 9.46
Minimum 210.07 2.08 0.000 0.000 0.432 2.64 0.012 0.369 0.000 0.004 0.001 0.018
SD 2,582.042 156.871 0.008 0.001 136.454 32.481 0.173 48.94 0.105 0.108 0.023 1.39
Jarque–Bera 9,832.8** 3,663.8** 4,426.6** 1,071.1** 520.6** 42,763.3** 187.4** 6,163.8** 1,298.5** 2076.9** 38,921.2** 12,245.9**
Post-crash sample
Mean 8,293.962 400.74 0.035 0.005 532.267 121.076 0.389 188.818 0.347 0.725 0.298 2.183
Maximum 19,118.3 1,555.59 0.127 0.017 1,397.48 359.13 1.18 470.29 1.84 3.36 0.892 9.44
Minimum 3,765.95 88.68 0.011 0.002 107.91 29.23 0.153 53.12 0.068 0.263 0.144 0.28
SD 2,705.482 297.007 0.019 0.002 286.66 68.613 0.21 96.06 0.353 0.493 0.117 1.95
Jarque–Bera 260.2** 186.4** 1,002.9** 642.8** 24** 47.4** 384.3** 50.3** 485.6** 1,622.3** 374.3** 211.9**
Pane B: Log-returns
Full sample
Mean 0.0009 0.0012 0.0019 0.0009 0.0013 0.0007 0.0006 0.0016 0.0023 0.0014 0.0015 0.0005
Maximum 0.0971 0.1664 0.5004 0.2263 0.174 0.2252 0.1476 0.2465 0.464 0.4391 0.3058 0.3758
Minimum −0.0878 −0.1057 −0.1745 −0.2107 −0.5943 −0.1698 −0.1746 −0.1267 −0.1871 −0.2613 −0.1448 −0.2667
SD 0.0171 0.0257 0.045 0.03 0.0341 0.025 0.0298 0.0309 0.0396 0.0331 0.0364 0.0429
Jarque–Bera 1,149.8** 1,487.3** 28,667.9** 6,904.8** 312,176.0** 9,166.6** 520.1** 2,875.3** 22,151.8** 71,054.0** 12,543.5** 8,385.1**
Pre-crash sample
Mean 0.0021 0.0029 0.0031 0.0018 0.0028 0.0022 0.0016 0.0031 0.0043 0.0023 0.0023 0.0024
Maximum 0.097 0.166 0.5 0.226 0.174 0.225 0.148 0.247 0.464 0.439 0.306 0.376
Minimum −0.088 −0.106 −0.175 −0.211 −0.594 −0.17 −0.119 −0.127 −0.187 −0.261 −0.145 −0.267
YAYA ET AL.
YAYA ET AL. 9

yt = α + βt + x t ; t = 1, 2, :…, ð1Þ
Vertcoin

5,817.4**

−0.0042

−0.142

74.4**
0.046

0.228

0.035
with,

ð1 −LÞd x t = ut ; t = 0,  1,  2,:…, ð2Þ

10,592.1**

−0.0008
Stellar

712.3**
−0.133
0.038

0.032
0.2
where the linear model in (1) applies to the three stan-
dard cases of (a) no constant (i.e. α = β = 0); (b) an inter-
86,303.6**

cept (α unknown and β = 0 a priori); and (c) a linear

−0.0009
Ripple

−0.153

892**
0.033

0.184

0.033
time trend (i.e. α and β unknown). The regressor xt is the
time series under investigation, to be fractionally
differenced with exponent d. Robinson (1994) tests the
17,099.9**

−0.0028

null hypothesis,
−0.15

577**
Nem
0.042

0.204

0.032

H 0 : d = d0 , ð3Þ
2,833.2**
Monero

−0.0023

−0.115

where d0 is any real value. The time-series, yt, is the

37.9**
0.032

0.098

0.028

resulting covariance stationary process obtained after the

integration process has been carried out on the series, xt.
Since this estimation approach is parametric, it is, there-
Maidsafecoin

fore, imperative to specify a particular functional form

−0.0019

for the I(0) error term, ut. Although there are different
392.6**

150.8**
−0.175
0.029

0.099

0.032

functional forms, say Autoregressive [AR(1)] or seasonal

ARMA (Autoregressive Moving Average), as the dataset
may permit. Our datasets are in daily frequency, thus sea-
10,698.8**
Litecoin

−0.0029

sonal AR is not expected. We, therefore, consider only

−0.088

111**
0.025

0.124

0.024

the white noise (uncorrelated error) case. Equations (1)

and (2) can be combined as one equation as:
218,287.7**
Ethereum

y*t = α0 1*t + β0 t*t + ut , t = 1, 2,…, ð4Þ

−0.0024

−0.092

43.4**
0.037

0.064

0.024

where y*t = ð1 − LÞd0 yt , 1*t = ð1 −LÞd0 1t ; t *t = ð1 −LÞd0 tt .

9,735.4**

−0.0013

Since ut is I(0), then it is straightforward to estimate α0

257.1**
−0.161
Doge
0.029

0.169

0.033

and β0 in (4), by ordinary least squares (OLS) methods.

The LM statistics for Robinson (1994) test has the func-
tional form given by:
23,058.9**
Digibyte

−0.0014

343.6**
−0.162
0.048

0.222

0.037

T −1
R^ = 4 a^0 A^ a,^ ð5Þ
σ^
1,562.2**

−0.0032

Note: ** denotes 5% level of significance.

118.1**
−0.093

where T is the sample size, and

Dash
0.025

0.118

0.026

−2π X
*    −1  
a^ = ψ λ j gu λ j ; τ^ I λ j ; σ^2
1,507.9**
(Continued)

Bitcoin

T
−0.002

−0.083

j
74.1**
0.016

0.061

0.019

2π X
T −1  −1  
Post-crash sample

= σ 2 ðτ^Þ = g λ j ; τ^ I λ j ,
T j=1 u
Jarque–Bera

Jarque–Bera
Maximum
TABLE 3

Minimum

! −1 !
Mean

2 X
*    0 X *    0 X *    0 X
*    0
A^ =
SD

SD

ψ λj ψ λj − ψ λ j ε^ λ j ε^ λ j ε^ λ j ε^ λ j ψ λ j ;
T j j j j
10 YAYA ET AL.

TABLE 4 Estimates of d for the log-returns in the whole sample

No regressors An intercept A linear time trend

Bitcoin 0.0280 (0.0221) 0.0281 (0.0221) 0.0252 (0.0222)
Dash 0.0050 (0.0202) 0.0050 (0.0202) −9.78E-05 (0.0204)
Digibyte 0.0370 (0.0220) 0.0370 (0.0220) 0.0359 (0.0221)
Doge 0.0635 (0.0227) 0.0635 (0.0228) 0.0634 (0.0227)
Ethereum 0.0435 (0.0227) 0.0437 (0.0227) 0.0407 (0.0229)
Litecoin 0.0257 (0.0214) 0.0258 (0.0214) 0.0251 (0.0214)
Maidsafecoin −0.0430 (0.0209) −0.0431 (0.0209) −0.0489 (0.0213)
Monero −0.0008 (0.0210) −0.0008 (0.0211) −0.0043 (0.0213)
Nem −0.0231 (0.0196) −0.0216 (0.0197) −0.0332 (0.0202)
Ripple 0.0563 (0.0209) 0.0564 (0.0209) 0.0563 (0.0209)
Stellar 0.0622 (0.0227) 0.0622 (0.0227) 0.0619 (0.0228)
Vertcoin −0.0368 (0.0217) −0.0370 (0.0217) −0.0373 (0.0217)

Note: Significant parameters, d, are in bold.

TABLE 5 Estimates of d for the log-returns in the pre-crash sample

No regressors An intercept A linear time trend

Bitcoin 0.0240 (0.0266) 0.0241 (0.0266) 0.0004 (0.0279)
Dash −0.0149 (0.0249) −0.0146 (0.0249) −0.0318 (0.0259)
Digibyte 0.0330 (0.0267) 0.0331 (0.0267) 0.0319 (0.0268)
Doge 0.0492 (0.0278) 0.0493 (0.0278) 0.0439 (0.0279)
Ethereum 0.0341 (0.0280) 0.0342 (0.0280) 0.0316 (0.0281)
Litecoin 0.0381 (0.0260) 0.0381 (0.0260) 0.0152 (0.0273)
Maidsafecoin −0.0449 (0.0257) −0.0449 (0.0257) −0.0452 (0.0257)
Monero −0.0004 (0.0260) −0.0004 (0.0260) −0.0078 (0.0264)
Nem −0.0534 (0.0243) −0.0534 (0.0243) −0.0542 (0.0243)
Ripple 0.0468 (0.0244) 0.0468 (0.0244) 0.0373 (0.0250)
Stellar 0.0641 (0.0274) 0.0642 (0.0274) 0.0530 (0.0281)
Vertcoin −0.0392 (0.0267) −0.0392 (0.0267) −0.0763 (0.0294)

Note: Significant parameters, d, are in bold.

  0. Significance of this parameter then implies autocor-

   λ j   ∂  
ψ λ j = log 2 sin ; ε^ λ j = loggu λ j ; τ^ , relations of lagged observations to the current time
2 ∂τ
series, hence the returns seem to be predictable, and
consequently, implying market inefficiency. The
squared log-returns act as a proxy for volatility persis-
where I(λj) is the periodogram of u^t , and tence whenever a series-based approach to volatility
τ^ = arg minτ∈T * σ 2 ðτÞ, with T* as a suitable subset of the persistence is considered.2 The transformed series is
Rq Euclidean space, and λj = 2πj/T, with the summations expected to contain long memory, a case where obser-
in * above equations that are bounded for all frequency vations distant in time lag correlate with current time
in the spectrum. observation.
To test for the hypothesis of market efficiency, we For robustness, we investigate a possible structural
expect to find randomness in the returns of crypto- break in the data sample by means of a Fourier smooth
currency. Thus, fractional integrated parameters, d, function that induces nonlinearity. This function allows
are expected to be insignificantly different from for non-linear smooth break in the time series of interest,
YAYA ET AL. 11

TABLE 6 Estimates of d for the log-returns in the post-crash sample

No regressors An intercept A linear time trend

Bitcoin −0.0094 (0.0430) −0.0094 (0.0432) −0.0106 (0.0434)
Dash −0.0693 (0.0433) −0.0693 (0.0434) −0.0700 (0.0435)
Digibyte −0.0033 (0.0428) −0.0034 (0.0428) −0.0064 (0.0429)
Doge 0.0813 (0.0448) 0.0813 (0.0448) 0.0811 (0.0447)
Ethereum 0.0179 (0.0420) 0.0180 (0.0420) 0.0050 (0.0429)
Litecoin −0.0860 (0.0432) −0.0860 (0.0433) −0.0860 (0.0433)
Maidsafecoin −0.0935 (0.0405) −0.0933 (0.0405) −0.0934 (0.0406)
Monero −0.0991 (0.0430) −0.0992 (0.0431) −0.0995 (0.0430)
Nem −0.0424 (0.0420) −0.0427 (0.0421) −0.0430 (0.0421)
Ripple 0.0657 (0.0438) 0.0658 (0.0438) 0.0661 (0.0437)
Stellar 0.0369 (0.0435) 0.0369 (0.0435) 0.0341 (0.0434)
Vertcoin −0.1226 (0.0411) −0.1229 (0.0413) −0.1241 (0.0415)

Note: Significant parameters, d, are in bold.

TABLE 7 Estimates of d for log-returns for nonlinear Fourier function case in whole sample

d Intercept, α λ γ
Bitcoin 0.0151 (0.0228) −9.13E-06 (0.0005) 5.58E-05 (0.0008) −0.0018 (0.0007)
Dash −0.0217 (0.0216) 3.00E-05 (0.0006) 1.10E-03 (0.0009) −0.0033 (0.0009)
Digibyte 0.0333 (0.0222) 3.12E-05 (0.0016) −3.64E-04 (0.0022) −0.0027 (0.0022)
Doge 0.0598 (0.0230) −4.58E-05 (0.0013) −4.21E-04 (0.0017) −0.0021 (0.0016)
Ethereum 0.0329 (0.0234) −7.01E-05 (0.0012) 1.20E-03 (0.0017) −0.0036 (0.0016)
Litecoin 0.0056 (0.0225) −1.79E-05 (0.0007) −3.60E-04 (0.0010) −0.0032 (0.0010)
Maidsafecoin −0.0537 (0.0216) 4.59E-05 (0.0006) 1.36E-03 (0.0009) −0.0015 (0.0009)
Monero −0.0196 (0.0221) 8.74E-06 (0.0008) 1.42E-03 (0.0011) −0.0032 (0.0011)
Nem −0.0386 (0.0204) 6.85E-06 (0.0009) 2.40E-03 (0.0013) −0.0029 (0.0013)
Ripple 0.0477 (0.0214) −8.02E-06 (0.0013) −6.80E-04 (0.0017) −0.0034 (0.0017)
Stellar 0.0542 (0.0232) −6.84E-06 (0.0015) −1.79E-03 (0.0020) −0.0030 (0.0019)
Vertcoin −0.0645 (0.0232) 2.80E-05 (0.0008) −3.31E-04 (0.0012) −0.0045 (0.0013)

Note: In bold, significant estimates. SE of intercept and slope, as well as that of Fourier function parameters, are given in parentheses.

particularly, when the form of the break and the break

y*t = α0 1*t + β0 t *t + λ1 sin*1,t + γ 1 cos*1,t + x t , t = 1, 2, …, ð6Þ
date are unknown. Fourier function is first considered for
unit root case in Becker, Enders, and Lee (2006) and   2πt
Enders and Lee (2012a, 2012b). Moreover, we were moti- where sin*1,t = ð1 −LÞd0 sin 2πt * d0
T and cos1,t = ð1 −LÞ sin T
vated by authors, such as Dolado, Gonzalo, and May- . Significance of at least one of the Fourier parameters, λ1
oral (2002), who proposed the first Dickey–Fuller and γ 1, implies nonlinearity in the time series.
fractional integration model. Our proposed test is an
extension of the linear model by Robinson (1994), to the
nonlinear case, which uses flexible Fourier form (FFF) to 4 | DATA AND EMPIRICAL
mimic non-linearities in the process (see, Gil-Alana & RESULTS
Yaya, 2018). Technically, by extending (4) to the
nonlinear case proposed in Gil-Alana and Yaya (2018), The dataset are the daily prices of cryptocurrencies from
we write, 7 August 2015 and 28 November 2018. Based on high
12 YAYA ET AL.

TABLE 8 Estimates of d for log-returns for nonlinear Fourier function case in pre-crash sample

d Intercept, α λ γ
Bitcoin 0.0072 (0.0279) 4.53E-06 (0.0006) −1.13E-03 (0.0008) 0.0013 (0.0008)
Dash −0.0252 (0.0258) 2.90E-05 (0.0007) −1.85E-03 (0.0011) 0.0000 (0.0011)
Digibyte 0.0281 (0.0271) 7.48E-05 (0.0019) −2.78E-03 (0.0026) 0.0016 (0.0026)
Doge 0.0456 (0.0281) 5.71E-05 (0.0013) −1.30E-03 (0.0017) 0.0010 (0.0017)
Ethereum 0.0336 (0.0281) −7.40E-05 (0.0016) −7.28E-04 (0.0021) −0.0004 (0.0021)
Litecoin 0.0227 (0.0273) −3.74E-06 (0.0010) −2.46E-03 (0.0014) 0.0013 (0.0013)
Maidsafecoin −0.0456 (0.0257) 4.27E-05 (0.0007) 2.86E-06 (0.0011) −0.0006 (0.0011)
Monero −0.0025 (0.0263) −6.00E-06 (0.0011) −5.17E-04 (0.0015) −0.0010 (0.0015)
Nem −0.0538 (0.0244) −6.84E-06 (0.0010) −4.25E-04 (0.0015) 0.0002 (0.0016)
Ripple 0.0373 (0.0252) 3.20E-05 (0.0014) −3.27E-03 (0.0019) 0.0003 (0.0019)
Stellar 0.0581 (0.0280) 1.07E-04 (0.0019) −2.45E-03 (0.0024) 0.0017 (0.0024)
Vertcoin −0.0650 (0.0288) 2.00E-05 (0.0010) −3.96E-03 (0.0016) 0.0015 (0.0016)

Note: In bold, significant estimates. SE of intercept and slope, as well as that of Fourier function parameters, are given in parentheses.

TABLE 9 Estimates of d for log-returns for nonlinear Fourier function case in post-crash sample

d Intercept, α λ γ
Bitcoin −0.0294 (0.0447) 2.15E-05 (0.0009) −7.71E-04 (0.0013) −0.0023 (0.0013)
Dash −0.0768 (0.0439) −1.47E-05 (0.0009) −1.00E-03 (0.0014) −0.0013 (0.0015)
Digibyte −0.0070 (0.0432) −4.94E-05 (0.0019) −2.56E-04 (0.0027) −0.0017 (0.0027)
Doge 0.0766 (0.0452) −1.21E-04 (0.0027) −2.61E-03 (0.0034) −0.0015 (0.0033)
Ethereum 0.0148 (0.0424) −1.00E-04 (0.0014) 1.03E-03 (0.0020) −0.0005 (0.0019)
Litecoin −0.0884 (0.0434) −1.98E-05 (0.0008) 1.22E-04 (0.0013) −0.0011 (0.0013)
Maidsafecoin −0.1025 (0.0412) −1.00E-04 (0.0010) −3.04E-04 (0.0015) −0.0021 (0.0016)
Monero −0.1003 (0.0431) 1.23E-05 (0.0009) −5.87E-04 (0.0014) −0.0006 (0.0015)
Nem −0.0448 (0.0422) −2.00E-04 (0.0013) −1.12E-03 (0.0020) 0.0005 (0.0021)
Ripple 0.0620 (0.0441) −1.36E-05 (0.0025) −1.87E-03 (0.0032) 0.0020 (0.0032)
Stellar 0.0366 (0.0435) −3.66E-05 (0.0021) −8.41E-04 (0.0028) 0.0002 (0.0028)
Vertcoin −0.1301 (0.0417) 3.74E-05 (0.0009) −8.71E-04 (0.0015) −0.0017 (0.0016)

Note: In bold, significant estimates. SE of intercept and slope, as well as that of Fourier function parameters, are given in parentheses.

price and market capitalizations, we have only included valuable cryptocurrency (Yaya et al., 2019). Conse-
12 cryptocurrencies in our analysis.3 The cryptos are: quently, the chosen break date was 17 December 2017,
Bitcoin, Dash, Digibyte, Doge, Ethereum, Litecoin, when the cryptocurrency (particularly, the Bitcoin)
Maidsafecoin, Monero, Nem, Ripple, Stellar and Ver- prices were observed to crash, thus, trending down-
tcoin. We obtained log-transformed prices for returns wards.4 Bai and Perron (2003) breakpoint test was con-
and squared returns. The Plots of these are given in ducted to determine the break date and 10 October
Figure 1. The price series in each plot is observed to 2017 was detected as the break date. Since this day was
depart from its original trends, which may be an indi- farther away from the peak point of Bitcoin in the
cation of the presence of structural breaks. Therefore, late 2017, we further carried out Perron's (2006)
in addition to the full sample period considered, the ADF-structural break test with the assumption of
series was further sub-divided into different two sam- innovational outlier for constant and trend model spec-
ple periods—the pre-crash and post-crash periods. The ification, with maximum trend break t-statistic. This
adopted point for the sub-division was the peak point test gave a break date of 20 December 2017, which is
with respect to the Bitcoin prices, being the most consistent with our chosen crash date of Bitcoin.5
YAYA ET AL. 13

TABLE 10 Estimates of d for the squared returns in the whole sample

No regressors An intercept A linear time trend

Bitcoin 0.1683 (0.0208) 0.1683 (0.0207) 0.1683 (0.0207)
Dash 0.1253 (0.0213) 0.1254 (0.0213) 0.1220 (0.0215)
Digibyte 0.0684 (0.0208) 0.0684 (0.0208) 0.0676 (0.0208)
Doge 0.1718 (0.0221) 0.1719 (0.0221) 0.1684 (0.0224)
Ethereum 0.0324 (0.0214) 0.0328 (0.0215) 0.0275 (0.0217)
Litecoin 0.1304 (0.0211) 0.1304 (0.0211) 0.1256 (0.0215)
Maidsafecoin 0.1103 (0.0214) 0.1104 (0.0214) 0.1102 (0.0214)
Monero 0.1191 (0.0221) 0.1191 (0.0221) 0.1190 (0.0221)
Nem 0.1200 (0.0238) 0.1200 (0.0238) 0.1195 (0.0238)
Ripple 0.2264 (0.0243) 0.2264 (0.0243) 0.2257 (0.0244)
Stellar 0.3132 (0.0263) 0.3131 (0.0262) 0.3126 (0.0263)
Vertcoin 0.4183 (0.0306) 0.4183 (0.0306) 0.4182 (0.0306)

Note: Significant parameters, d, are in bold.

TABLE 11 Estimates of d for the squared returns in the pre-crash sample

No regressors An intercept A linear time trend

Bitcoin 0.1977 (0.0274) 0.1976 (0.0274) 0.1801 (0.0286)
Dash 0.1361 (0.0260) 0.1361 (0.0260) 0.1218 (0.0270)
Digibyte 0.0617 (0.0250) 0.0617 (0.0250) 0.0608 (0.0251)
Doge 0.1541 (0.0273) 0.1541 (0.0273) 0.1409 (0.0280)
Ethereum 0.0306 (0.0254) 0.0311 (0.0256) 0.0264 (0.0257)
Litecoin 0.1367 (0.0256) 0.1367 (0.0256) 0.1117 (0.0272)
Maidsafecoin 0.1175 (0.0272) 0.1176 (0.0272) 0.1174 (0.0271)
Monero 0.1190 (0.0269) 0.1190 (0.0269) 0.1170 (0.0270)
Nem 0.1212 (0.0288) 0.1212 (0.0288) 0.1198 (0.0287)
Ripple 0.2412 (0.0293) 0.2412 (0.0293) 0.2374 (0.0296)
Stellar 0.3364 (0.0320) 0.3365 (0.0320) 0.3311 (0.0326)
Vertcoin 0.4502 (0.0370) 0.4502 (0.0370) 0.4501 (0.0371)

Note: Significant parameter, d, are in bold.

Each plot in Figure 1 contains the actual, returns and sizes 1,210, 864 and 346, respectively. The results are
squared returns series of each cryptocurrency. Crypto- presented in three different panes, each displaying the
currency returns are characterized by volatility clustering statistical properties of the cryptocurrency prices and the
and notable jumps at different time periods, particularly, log-returns (see Table 3). Pane A and Pane B show
in the post-crash periods. This is quite noticeable in the the summary statistics for the cryptocurrency prices and
log-returns for Bitcoin, Dash, Digibyte, Doge, Litecoin, the log-return series under the three sample periods.
Maidsafecoin, Monero, Ripple, Stellar and Vertcoin. Bitcoin is the highest cryptocurrency and is averagely
Descriptive statistics, presented in Table 3, gives the priced at 3596.6 USD, 1495 USD and 8,194 USD in the
statistical distribution of the cryptocurrencies, as it gives full sample, pre-crash and post-crash samples, respec-
more reliable results than the graphical approach in tively. On the other hand, Doge, the least valuable crypto-
Figure 1. We consider the actual price, log-returns and currency among the considered cryptocurrencies, is
squared returns of the cryptocurrency series under three averagely priced at 0.001 USD, 0.002 USD and 0.005 USD
different sample periods: the full sample, the pre-crash in the full, pre-crash and post-crash sample periods,
sample and the post-crash sample periods, with sample respectively. Regardless of the cryptocurrency considered,
14 YAYA ET AL.

TABLE 12 Estimates of d for the squared returns in the post-crash sample

No regressors An intercept A linear time trend

Bitcoin 0.1234 (0.0332) 0.1246 (0.0334) 0.0373 (0.0375)
Dash 0.0836 (0.0376) 0.0844 (0.0378) 0.0436 (0.0401)
Digibyte 0.1537 (0.0403) 0.1560 (0.0407) 0.0938 (0.0441)
Doge 0.3795 (0.0543) 0.3815 (0.0543) 0.3677 (0.0552)
Ethereum 0.0607 (0.0392) 0.0608 (0.0392) 0.0326 (0.0416)
Litecoin 0.0866 (0.0351) 0.0867 (0.0351) 0.0202 (0.0394)
Maidsafecoin 0.0905 (0.0363) 0.0906 (0.0364) 0.0185 (0.0405)
Monero 0.1107 (0.0344) 0.1119 (0.0346) 0.0374 (0.0386)
Nem 0.0878 (0.0357) 0.0878 (0.0357) 0.0044 (0.0401)
Ripple 0.0897 (0.0367) 0.0905 (0.0369) 0.0545 (0.0376)
Stellar 0.1557 (0.0375) 0.1565 (0.0376) 0.1170 (0.0401)
Vertcoin 0.1300 (0.0462) 0.1302 (0.0462) 0.1229 (0.0471)

Note: Significant parameters, d, are in bold.

TABLE 13 Estimates of d for squared returns for nonlinear Fourier function case in the whole sample

d Intercept, α λ γ
Bitcoin 0.1395 (0.0224) 2.60E-06 (0.0001) −2.22E-04 (0.0001) −0.0001 (0.0001)
Dash 0.1059 (0.0225) −1.73E-06 (0.0001) −2.92E-04 (0.0001) −0.0002 (0.0001)
Digibyte 0.0646 (0.0210) 6.10E-06 (0.0005) −5.39E-04 (0.0006) −0.0006 (0.0006)
Doge 0.1581 (0.0231) −7.25E-06 (0.0003) −6.70E-04 (0.0003) −0.0003 (0.0003)
Ethereum 0.0310 (0.0218) 5.03E-05 (0.0004) 1.00E-04 (0.0005) 0.0008 (0.0005)
Litecoin 0.1008 (0.0230) 5.66E-06 (0.0001) −4.94E-04 (0.0002) −0.0004 (0.0002)
Maidsafecoin 0.1064 (0.0216) −1.01E-05 (0.0001) −1.90E-04 (0.0001) 0.0000 (0.0001)
Monero 0.1136 (0.0225) 2.65E-06 (0.0002) −5.63E-06 (0.0002) −0.0003 (0.0002)
Nem 0.1189 (0.0238) −1.08E-05 (0.0005) −2.83E-04 (0.0006) −0.0002 (0.0005)
Ripple 0.2211 (0.0248) 2.53E-05 (0.0008) −7.21E-04 (0.0009) −0.0008 (0.0008)
Stellar 0.3065 (0.0269) 6.70E-05 (0.0011) −9.38E-04 (0.0011) −0.0010 (0.0010)
Vertcoin 0.4180 (0.0306) 8.65E-05 (0.0028) −3.94E-04 (0.0025) −0.0003 (0.0021)

Note: In bold, significant estimates. SE of intercept and slope, as well as that of Fourier function parameters, are given in parentheses.

prices seemed to be more volatile in post-crash period Stellar had the highest log-returns of −0.0008 in the post-
than in the pre-crash sample period, since the observed crash sample period. The least log-returns in the full, pre-
standard deviation values are higher in the former than crash and post-crash sample periods were observed for
in the latter sample period. This may not be completely Maidsafecoin (0.0006, 0.0016) and Vertcoin (−0.0042),
disconnected from the speculation induced by the crypto- respectively. The log-returns were found to be highly vol-
currency crash of 17 December 2017. Moreover, the cryp- atile, with the SD that were twice the means in most
tocurrency prices are not normally distributed, given the cases, and were also not normally distributed.
statistically significant Jarque–Bera statistics, which for- Since market efficiency is a function of returns (price
mally combines the skewness and kurtosis. changes), volatility comes into play (high, medium or low)
On the statistical distribution of the log-returns of and the function of time the volatility will fizzle out is the
cryptocurrency (see, Pane B in Table 3), Nem had the volatility persistence, which is an autocorrelation issue in
highest log-returns of 0.0023 and 0.0043 in the full sam- the absolute or squared returns, or both as proxies.
ple and the pre-crash sample periods, respectively, while Starting with the case of market efficiency, investigated for
YAYA ET AL. 15

TABLE 14 Estimates of d for squared returns for nonlinear Fourier function case in pre-crash sample

d Intercept, α λ γ
Bitcoin 0.1786 (0.0290) 2.12E-06 (0.0001) −1.00E-04 (0.0001) 0.0002 (0.0001)
Dash 0.1191 (0.0274) −8.13E-06 (0.0001) −3.16E-04 (0.0002) 0.0002 (0.0002)
Digibyte 0.0487 (0.0258) −7.14E-06 (0.0005) −8.98E-04 (0.0007) 0.0012 (0.0007)
Doge 0.1375 (0.0285) 6.35E-06 (0.0003) −4.18E-04 (0.0003) 0.0006 (0.0003)
Ethereum 0.0263 (0.0261) 5.20E-05 (0.0005) 9.69E-05 (0.0007) 0.0012 (0.0007)
Litecoin 0.1095 (0.0277) 1.06E-05 (0.0002) −5.45E-04 (0.0002) 0.0004 (0.0002)
Maidsafecoin 0.1120 (0.0276) −1.30E-05 (0.0001) 1.25E-04 (0.0002) 0.0001 (0.0002)
Monero 0.1179 (0.0270) 2.42E-06 (0.0002) −6.30E-05 (0.0003) −0.0001 (0.0003)
Nem 0.1160 (0.0291) 3.16E-05 (0.0006) 3.25E-04 (0.0007) 0.0009 (0.0007)
Ripple 0.2363 (0.0298) 3.21E-05 (0.0011) −1.15E-03 (0.0012) 0.0003 (0.0011)
Stellar 0.3321 (0.0326) −0.0002 (0.0016) −1.12E-03 (0.0015) 0.0006 (0.0013)
Vertcoin 0.4496 (0.0371) −0.0005 (0.0039) −3.26E-04 (0.0034) 0.0008 (0.0029)

Note: In bold, significant estimates. SE of intercept and slope, as well as that of Fourier function parameters, are given in parentheses.

TABLE 15 Estimates of d for squared returns for nonlinear Fourier function case in post-crash sample

d Intercept, α λ γ
Bitcoin 0.0268 (0.0370) 3.39E-06 (0.0000) 2.60E-04 (0.0001) 0.0002 (0.0001)
Dash 0.0506 (0.0401) 8.42E-06 (0.0001) 1.89E-04 (0.0001) 0.0003 (0.0001)
Digibyte 0.1269 (0.0429) 1.00E-04 (0.0004) 5.51E-04 (0.0004) 0.0008 (0.0004)
Doge 0.3736 (0.0551) 4.20E-04 (0.0010) −1.45E-04 (0.0009) 0.0007 (0.0008)
Ethereum 0.0201 (0.0423) −2.69E-06 (0.0001) 1.36E-04 (0.0001) 0.0002 (0.0001)
Litecoin 0.0079 (0.0398) −5.32E-06 (0.0001) 3.23E-04 (0.0001) 0.0003 (0.0001)
Maidsafecoin 0.0399 (0.0396) −1.21E-05 (0.0001) 4.45E-04 (0.0002) 0.0004 (0.0002)
Monero 0.0444 (0.0381) 6.51E-06 (0.0001) 3.87E-04 (0.0001) 0.0004 (0.0001)
Nem 0.0255 (0.0395) −2.30E-05 (0.0002) 6.32E-04 (0.0002) 0.0006 (0.0002)
Ripple 0.0458 (0.0390) 1.34E-05 (0.0002) 2.03E-04 (0.0003) 0.0009 (0.0003)
Stellar 0.1286 (0.0395) 4.49E-05 (0.0003) 4.44E-04 (0.0004) 0.0007 (0.0004)
Vertcoin 0.1270 (0.0465) 1.90E-05 (0.0004) 1.74E-04 (0.0005) 0.0002 (0.0004)

Note: In bold, significant estimates. Standard errors of intercept and slope, as well as that of Fourier function parameters, are given in
parentheses.

the whole sample period (Table 4), pre-crash sample Litecoin, Maidsafecoin, Monero and Vertcoin, in the three
(Table 5) and post-crash sample (Table 6), we found evi- test regression specifications.
dences to support market efficiency of some The results presented in Tables 4–6 are based on the
cryptocurrencies. The null hypothesis of random walk for assumption of linearity in log-returns of cryptocurrencies.
market efficiency was not rejected in the case of Bitcoin, We then checked for robustness by means of nonlinear
Dash, Digibyte and Ethereum markets in the full sample, fractional integration, using flexible Fourier function
pre-crash and after crash samples (see Bartos, 2015). Dur- described in the methodology part.6 Noting that Fourier
ing the pre-crash sample, evidence to support market inef- function allows one to model remaining structural breaks
ficiency was found in Nem and Stellar, since random walk in returns, as smooth breaks rather than instantaneous
was evident, while the linear time trend specification breaks. In Tables 7–9, it is interesting to find nonlinearity
points to non-rejection of market efficiency in Stellar. In in the full returns sample, as expected, at least in the
the post-crash sample, market becomes more inefficient, form of a break during the crash for some crypto-
as we observe rejection of random walk in returns of currency. The coefficient of cosine part of the nonlinear
16 YAYA ET AL.

function, γ, is significant in Bitcoin, Dash, Ethereum, return of Bitcoin and other eight altcoins. Meanwhile,
Litecoin, Monero, Nem, Ripple and Vertcoin, implying significant d values were found in Digibyte, Doge, Stellar
the relevance of nonlinear fractional integration results and Vertcoin.
here. Now, looking at the estimated d parameter for
returns, we find, in addition to four cases of market ineffi-
ciency (Doge, Maidsafecoin, Ripple and Stellar), Vertcoin 5 | CONCLUSION
indicating rejection of the null hypothesis of random walk
of market hypothesis. Similar stances were found in the In this article, we have considered the market efficiency,
cases of pre-crash and post-crash samples. volatility and persistence in 12 cryptocurrencies, with
Next, we consider the issue of volatility persistence in data sampled from 7 August 2015, to 28 November 2018.
cryptocurrency, analysing fractional integration parame- The considered cryptocurrencies were the Bitcoin, Dash,
ter in the squared returns series, although we found vola- Digibyte, Doge, Ethereum, Litecoin, Maidsafecoin, Mon-
tility in the post-crash sample to be higher than that of ero, Nem, Ripple, Stellar and Vertcoin. The findings
the pre-crash sample. This is actually the expectation of obtained about the level of market efficiency of crypto-
many researchers, since they have found lesser volatility currency indicated evidence of random walk in the
during bear periods compared to bull periods (Gomez & returns of most cryptocurrencies including Bitcoin,
Biscarri, 2004; Gonzalez, Powell, Shi, & Wilson, 2005). which is contrary to what was published by some
The results of volatility persistence for the linear case are authors, who found inefficiency in Bitcoin. Thus, future
presented in Tables 10–12 for full sample, pre-crash Bitcoin values are unpredictable. Our approach of investi-
and post-crash samples. We only found evidence of gating market efficiency is novel, and robustness checks
no persistence of volatility in the case of Ethereum were based on nonlinearity and fractional integration
across the three sampled periods, while the remaining techniques, which have rarely been applied in the market
11 cryptocurrencies indicated evidence of significant vol- efficiency literature so far. We closely followed the defini-
atility persistence. In the full sample, Vertcoin had the tion of EMH by Fama (1970) on returns unpredictability
highest volatility persistence, and similarly in the pre- or being a random-walk process to imply market effi-
crash sample. Next to this is Stellar. Lowest significant ciency. Although volatility was found to be higher during
volatility was found for Maidsafecoin. Results of volatility the post-crash period, this was likely to persist for a
persistence observed during post-crash periods were not shorter period, compared to pre-crash period. As a result
consistent with that of pre-crash, since the highest persis- of this unpredictability of returns in Bitcoin, particularly,
tence of volatility was found in Doge, across the three traders cannot boast of making abnormal profits in the
testing regression model, while low volatility persistence cryptocurrency markets, as revealed in the findings.
was found in Dash, Nem, Ripple and Maidsafecoin. In Many works on cryptocurrency have shown evidence
the post-crash sample, volatility persistence was lower of market inefficiency, while others concluded that there
than that in the pre-crash sample, as the observed volatil- would be a possibility of approaching the market effi-
ity took a shorter time, than in the pre-crash period, to ciency state. Meanwhile, different data points have been
fizzle out. used by these authors and their results were very sensi-
As we did earlier, by introducing non-linear smooth tive to the time range. Our approach of analysis in this
break function to capture possible structural breaks in article followed Gil-Alana, Gupta, Shittu, and Yaya (2018)
the time series at hand, we similarly apply this to the on efficiency and volatility persistence of Baltic stock
squared returns. The results are presented in markets, which has better power compared to Dickey–
Tables 13–15. We first observed some non-linearities in Fuller-like unit root testing framework of market effi-
the squared returns of Bitcoin, Dash, Doge, Litecoin ciency, such as applied by those authors. This work will,
(in the full sample); Bitcoin and Litecoin (in the pre- therefore, interest market participants, portfolio man-
crash sample); and Bitcoin, Dash, Ethereum, Litecoin, agers and policy makers in a number of ways: it can
Maidsafecoin, Monero, Nem, Ripple and Stellar (in the inform both short and long-term investment strategies,
post-crash sample). By comparing the d estimates in the and it has provided compelling evidence that even in the
linear and nonlinear cases, we still find only Ethereum presence of volatility in cryptocurrency markets, markets
with no-significant volatility persistence, for the three cannot be exploited to make abnormal profits, by design-
testing regression specifications. In the pre-crash sample, ing appropriate trend trading strategies.
Digibyte was found to be insignificant, due to low power
and increase in the degree of freedom of the test statistic. ORCID
The results computed for the case of post-crash showed OlaOluwa S. Yaya https://orcid.org/0000-0002-7948-
more non-linearities, with nonlinearity found for squared 011X
YAYA ET AL. 17

E N D N O T ES Becker, R., Enders, W., & Lee, J. (2006). A stationary test with an
1
Bitcoin was introduced by Satoshi Nakamoto in 2009 in a white unknown number of smooth breaks. Journal of Time Series
paper published online. Cryptocurrency generally is a new solu- Analysis, 27, 381–409.
tion to some internet enthusiasts since the introduction of inter- Beer, C., & Weber, B. (2015). Bitcoin-the promise and limits of pri-
net: as a form of digital cash that is ‘peer-to-peer’ and open vate innovation in monetary and payment systems. Monetary
sourced-based (Beer & Weber, 2015). The offer by Nakamoto Policy and the Economy, Q4/14, 53–66.
allows a platform to create more private currency and ‘buy and Bollerslev, T. (1986). Generalized autoregressive conditional hetero-
sell’ takes place without recourse to central banks and commer- skedasticity. Journal of Econometrics, 31, 307–327.
cial banks. Bouri, E., Gil-Alana, L. A., Gupta, R., & Roubaud, D. (2019). Model-
2 ling long memory volatility in the Bitcoin market: Evidence of
Model-based volatility persistence uses the Generalized Auto-
persistence and structural breaks. International Journal of
regressive Conditional Heteroscedastic (GARCH) model of
Finance & Economics, 24(1), 412–426.
Bollerslev (1986).
Bouri, E., Gupta, R., Lahiani, A., & Shahbaz, M. (2018). Testing for
3
For dataset downloads, visit https://coinmetrics.io/data-downloads/. Asymmetric Nonlinear Short-and Long-Run Relationships
4
Note that it took some altcoins about 30–40 days to react to price between Bitcoin, Aggregate Commodity and Gold Prices.
crash induced from Bitcoin market while some altcoins reacted Resources Policy, 57, 224–235.
within the first week of price crash. Bouri, E., Lau, C. K. M., Lucey, B., & Roubaud, D. (2019). Trading
5
The 3 days (20 December 2017) after the crash date of volume and the predictability of return and volatility in the
17 December 2017, in the daily Bitcoin prices will in no way bias cryptocurrency market. Finance Research Letters, 29, 340–346.
our results. Canh, N. P., Wongchoti, U., Thanh, S. D., & Thong, N. T. (2019).
6
Systematic risk in cryptocurrency market: Evidence from DCC-
The nonlinear fractional integration framework considers only
MGARCH model. Finance Research Letters, 29, 90–100.
the case of intercept, since time trend coefficients are insignificant
Caporale, G. M., Gil-Alana, L., & Plastun, A. (2018). Persistence in
in the linear cases.
the cryptocurrency market. Research in International Business
and Finance, 46, 141–148.
Caporale, G. M., & Plastun, A. (2018). The day of the week effect in
DATA AVAILABILITY STATEMENT the cryptocurrency market. Finance Research Letters. https://
Data sharing is not applicable to this article as no new doi.org/10.1016/j.frl.2018.11.012
data were created or analyzed in this study. Caporale, G. M., Plastun, A., & Oliinyk, V. (2019). Bitcoin fluctua-
tions and the frequency of price overreactions. Financial Mar-
kets and Portfolio Management, 33, 109–131.
ORCID Chaim, P., & Laurini, M. P. (2018). Volatility and return jumps in
OlaOluwa S. Yaya https://orcid.org/0000-0002-7948- Bitcoin. Economics Letters, 173, 158–163.
Chaim, P., & Laurini, M. P. (2019). Nonlinear dependence in cryp-
011X
tocurrency markets. North American Journal of Economics and
Finance, 48, 32–47.
R EF E RE N C E S Charfeddine, L., & Maouchi, Y. (2019). Are shocks on the returns
Ardia, D., Bluteau, K., & Rüede, M. (2019). Regime changes in and volatility of cryptocurrencies really persistent? Finance
Bitcoin GARCH volatility dynamics. Finance Research Letters, Research Letters, 28, 423–430.
29, 266–271. Charles, A., & Darné, O. (2019). Volatility estimation for
Asanidze, Levan. (2018). Adaptive efficiency of Bitcoin market. Bach- cryptocurrencies: Further evidence with jumps and structural
elor Thesis, Poland: Lazarski University. https://doi.org/10. breaks. Economics Bulletin, 39(2), 954–968.
13140/RG.2.2.13796.22402. Cheah, E.-T., Mishra, T., Parhi, M., & Zhang, Z. (2018). Long mem-
Bai, J., & Perron, P. (2003). Computation and analysis of multiple ory interdependency and inefficiency in Bitcoin markets. Eco-
structural change models. Journal of Applied Economics, nomics Letters, 167, 18–25.
18, 1–22. Cheikh, N. B., Zaied, Y. B., & Chevallier, J. (2019). Asymmetric vol-
Balcilar, M., Bouri, E., Gupta, R., & Roubaud, D. (2017). Can vol- atility in cryptocurrency markets: New evidence from smooth
ume predict Bitcoin returns and volatility? A quantiles-based transition GARCH models. Finance Research Letters. https://
approach. Economic Modelling, 64, 74–81. doi.org/10.1016/j.frl.2019.09.008
Barber, S., Boyen, X., Shi, E., & Uzun, E. (2012). Bitter to better- Chu, J., Zhang, Y., & Chan, S. (2019). The adaptive market hypoth-
how to make BitCoin a better currency. In A. D. Keromytis esis in the high frequency cryptocurrency market. International
(Ed.), Financial cryptography and data security. Volume 7397 of Review of Financial Analysis. 64(C), 221–231.
lecture notes in computer science (pp. 399–414). Berlin, Ger- Ciaian, P., & Rajcaniova, M. (2018). Virtual relationships:
many: Springer. Short-and long-run evidence from Bitcoin and altcoin mar-
Bariviera, A. F. (2017). The inefficiency of Bitcoin revisited: A kets. Journal of International Financial Markets, Institu-
dynamic approach. Economic Letters, 161, 1–4. tions and Money, 52, 173–195.
Bartos, J. (2015). Does Bitcoin follow the hypothesis of efficient mar- Cointelegraph. (2018). CoinMarketCap News. Retrieved from
ket? International Journal of Economic Sciences, 4(2), 10–23. https://cointelegraph.com/tags/coinmarketcap
18 YAYA ET AL.

Corbet, S., Meegan, A., Larkin, C., Lucey, B., & Yarovaya, L. (2018). Kumar, A. S., & Anandarao, S. (2019). Volatility spillover in crypto-
Exploring the dynamic relationships between cryptocurrencies currency markets: Some evidences from GARCH and wavelet
and other financial assets. Economics Letters, 165, 28–34. analysis. Physica A: Statistical Mechanics and its Applications,
Dickey, D. A., & Fuller, W. A. (1979). Distribution of the estimators 524, 448–458.
for autoregressive time series with a unit root. Journal of the Lee, D., & Schmidt, P. (1996). On the power of the KPSS test of
American Statistical Association, 74, 427–431. stationarity against fractionally integrated alternatives. Journal
Diebold, F. X., & Rudebusch, G. D. (1991). On the power of Dickey- of Econometrics, 73, 285–302.
Fuller test against fractional alternatives. Economics Letters, 35, Lim, K. P., & Brooks, R. (2011). The evolution of stock market effi-
155–160. ciency over time: A survey of the empirical literature. Journal
Dolado, J. J., Gonzalo, J., & Mayoral, L. (2002). A fractional Dickey- of Economic Surveys, 25(1), 69–108.
Fuller test for unit roots. Econometrica, 70(5), 1963–2006. Lo, A. W., & MacKinlay, A. C. (1988). Stock market prices do not
Dyhrberg, A. H. (2016a). Bitcoin, gold and the dollar–a GARCH follow random walks: Evidence from a single specification test.
volatility analysis. Finance Research Letters, 16, 85–92. Review of Financial Studies, 1(1), 41–66.
Dyhrberg, A. H. (2016b). Hedging capabilities of Bitcoin. Is it the Nadarajah, S., & Chu, J. (2017). On the inefficiency of Bitcoin. Eco-
virtual gold. Finance Research Letters, 16, 139–144. nomics Letters, 150, 6–9.
Enders, W., & Lee, J. (2012a). A unit root test using a Fourier series Perron, P. (2006). Dealing with structural breaks. Palgrave Hand-
to approximate smooth breaks. Oxford Bulleting of Economics book of Econometrics, 1, 278–352.
and Statistics, 74, 574–599. Robinson, P. M. (1994). Efficient tests of nonstationary hypotheses.
Enders, W., & Lee, J. (2012b). The flexible Fourier form and Dickey- Journal of the American Statistical Association, 89, 1420–1437.
Fuller-type unit root tests. Economic Letters, 117, 196–199. Sigaki, H. Y. D., Perc, M., & Ribeiro, H. V. (2019). Clustering pat-
Fakhfekh, M., & Jeribi, A. (2019). Volatility dynamics of crypto-cur- terns in efficiency and the coming-of-age of the cryptocurrency
rencies' returns: Evidence from asymmetric and long memory market. Scientific Reports, 9, 1440. https://doi.org/10.1038/
GARCH models. Research in International Business and s41598-018-37773-3
Finance, 101075. https://doi.org/10.1016/j.ribaf.2019.101075 Tiwari, A. K., Jana, R. K., Das, D., & Roubaud, D. (2018). Informa-
Fama, E. F. (1965). Random walks in stock market prices. Financial tional efficiency of Bitcoin: An extension. Economics Letters,
Analysis Journal, 21, 55–59. 163, 106–109.
Fama, E. F. (1970). Efficient capital markets: A review of theory Urquhart, A. (2016). The inefficiency of bitcoin. Economic Letters,
and empirical work. Journal of Finance, 25(2), 383–417. 148, 80–82.
Fama, E. F., & French, K. R. (2008). Dissecting Anomalies. The Vidal-Tomas, D., Ibadnez, A. M., & Farinos, J. E. (2019). Weak effi-
cross-section of expected stock returns, The Journal of Finance, ciency of the cryptocurrency market: A market portfolio
63, 1653–1678. approach. Applied Economics Letters, 26(19), 1627–1633.
Gil-Alana, L. A., Gupta, R., Shittu, O. I., & Yaya, O. S. (2018). Market Weber, B. (2016). Bitcoin and the legitimacy crisis of money. Cam-
efficiency of Baltic stock markets: A fractional integration approach. bridge Journal of Economics, 40(1), 17–41.
Physica A, Statistical Mechanics and its Applications, 511, 251–262. Wei, W. (2018). Liquidity and market efficiency in cryptocurrencies.
Gil-Alana, L. A., & Yaya, O. S. (2018). Testing fractional unit roots Economics Letters, 168, 21–24.
with non-linear smooth break approximations using Fourier Wright, J. H. (2000). Alternative variance-ratio tests using ranks
functions (MPRA Paper No. 90516). Retrieved from https:// and signs. Journal of Business and Economic Statistics, 18(1),
mpra.ub.uni-muenchen.de/90516/ 1–9.
Glaser, F., Zimmermann, K., Haferkorn, M., Weber, M. C., & Yaya, O. S., Ogbonna, E. A., & Olubusoye, O. E. (2019). How persis-
Siering, M., (2014, April 15). Bitcoin – Asset or currency? tent and dynamic inter-dependent are pricing Bitcoin to other
Revealing users' hidden intentions. Cryptocurrencies before and after 2017/18 crash? Physica A,
Gomez, J., & Biscarri, F. P. d. G. (2004). Stock market cycles and Statistical Mechanics and its Applications, 531, 121732. https://
stock market development in Spain. Spanish Economic Review, doi.org/10.1016/j.physa.2019.121732
6, 127–151. Zhang, W., Wang, P., Li, X., & Shen, D. (2018). The inefficiency of
Gonzalez, L., Powell, J. G., Shi, J., & Wilson, A. (2005). Two centu- cryptocurrency and its cross-correlation with Dow Jones indus-
ries of bull and bear market cycles. International Review of Eco- trial average. Physica A, 510, 658–670.
nomic and Finance, 14, 469–486.
Grobys, K., & Sapkota, N. (2019). Cryptocurrencies and momen-
tum. Economics Letters, 180, 6–10. How to cite this article: Yaya OS, Ogbonna AE,
Hassler, U., & Wolters, J. (1994). On the power of unit root tests
Mudida R, Abu N. Market efficiency and volatility
against fractional alternatives. Economic Letters, 45, 1–5.
Katsiampa, P. (2017). Volatility estimation for Bitcoin: A compari-
persistence of cryptocurrency during pre- and post-
son of GARCH models. Economics Letters, 158, 3–6. crash periods of Bitcoin: Evidence based on
Kristoufek, L., & Vosvrda, M. (2019). Cryptocurrencies market effi- fractional integration. Int J Fin Econ. 2020;1–18.
ciency ranking: Not so straightforward. Physica A: Statistical https://doi.org/10.1002/ijfe.1851
Mechanics and its Applications, 531, 120853.