Liquidity Commonality in The Cryptocurrency Market
Liquidity Commonality in The Cryptocurrency Market
Liquidity Commonality in The Cryptocurrency Market
To cite this article: Abhinava Tripathi, Alok Dixit & Vipul (2022) Liquidity commonality
in the cryptocurrency market, Applied Economics, 54:15, 1727-1741, DOI:
10.1080/00036846.2021.1982128
ABSTRACT KEYWORDS
Motivated by the unique transaction cost structure of the cryptocurrency (CC) market1, this study Liquidity commonality;
investigates the phenomenon of liquidity commonality across a sample of 53 CCs. The study employs Cryptocurrency; Principal
the google search volume index (GSVI) measure to capture the retail investor’s attention towards the component analysis;
Quantile regression;
CC market. Using the quantile regression method, we document the liquidity dynamics of CCs that is
Transaction costs
contrasting to other asset classes, and is ascribed to its unique transaction cost structure. In view of
the relatively high liquidity commonality levels found in the CC market, this paper sounds a note of JEL CLASSIFICATION
caution to retail investors on episodic non-availability of liquidity in CCs. G12; G14; C33
CONTACT Abhinava Tripathi [email protected] Department of Management Studies, Indian Institute of Technology Roorkee, Roorkee,
India
2
As per the https://coinmarketcap.com/ (accessed on 21-11-2019).
3
Bitcoin alone is traded on more than 350 markets across the world, in various currencies (Wei 2018).
© 2021 Informa UK Limited, trading as Taylor & Francis Group
1728 A. TRIPATHI ET AL.
drive-up the transaction fees in CC markets (Easley, the liquidity of securities. More recent literature
O’Hara, and Basu 2019). Because CC users are has employed Google Search Volume Index
required to pay transaction fees to get their transac (GSVI) as a measure of investor attention (Da,
tions posted on blockchain. As the transaction Engelberg, and Gao 2011, 2014; Ding and Hou
volumes rise and transaction processing capacities 2015). We contribute to this literature by adding
are fully utilized, the waiting times for users increase a novel and less explored asset class, that is CCs.
substantially. These users compete for priority to The study makes three important contributions
reduce the waiting time, and in turn, drive-up the to the literature. It is one of the first studies that
transaction fees. The transaction processing priority examines the phenomenon of liquidity common
is dynamically auctioned by the miners with auction ality in the CC market, employing 53 CCs. The
price being the transaction fee. Higher fees secure study also investigates the effect of investor atten
higher priority and less waiting time. tion on liquidity dynamics of CCs, using the daily
These unique features of the liquidity in CC GSVI measure. Finally, the study examines com
markets motivate us to examine the presence of monality across the conditional quantiles of liquid
a marketwide systematic factor that has been ity, employing the quantile regression framework
widely acknowledged to affect the liquidity of tra (Koenker and Bassett 1978; Koenker and Hallock
ditional securities; i.e. the phenomenon of liquidity 2001). The findings of this study suggest
commonality (Chordia, Roll, and Subrahmanyam a significant presence of commonality that is higher
2000). The literature on liquidity commonality has than other asset classes. The results show that
covered a diverse set of markets. These include investor attention has a positive relation with
equity markets (cash and derivatives, developed volume as well as transaction costs. These findings
and emerging markets), fixed income securities are in contrast to those pertaining to other asset
markets, currency markets, and commodity mar classes, and may be ascribed to the unique transac
kets. A summary of some of the most important tion cost structure that is integral to the CC market.
studies on commonality, covering these diverse That is, higher volumes (beyond certain threshold
markets, is provided in Table A1 of Appendix A. levels) drive-up the transaction fees and vice-versa.
These studies document the evidence of liquidity We further show that this property of the CC
commonality and its adverse implications for market attenuates the escalating commonality
investors. The literature on commonality supports levels around the extreme quantiles of liquidity.
the argument that the benefits of portfolio diversi The remaining study is organized into four sec
fication and various other trading strategies disap tions. Section 2 describes the data, sample, and
pear when the transaction costs are accounted for, variables, and Section 3 discusses the methodology.
especially when liquidity commonality is high. Section 4 presents the empirical findings, and
However, the liquidity dynamics of CCs remains Section 5 concludes.
less well understood, and in particular, the area of
liquidity commonality of CC markets has not been
II. Data, sample, and variables
explored (Marshall, Nguyen, and Visaltanachoti
2019). Hence, the results from this study are parti Daily price and volume data of cryptocurrencies
cularly important to investors that are primarily are provided by the Coinmarketcap website (www.
driven towards CCs by their portfolio concerns. coinmarketcap.com). Coinmarketcap aggregates
This study provides evidence on the phenomenon the trading activity of more than 2,200 cryptocur
of liquidity commonality in the context of the CC rencies across about 8,900 exchanges. It computes
market, using 53 cryptocurrencies. the volume-weighted average price of cryptocur
Another strand of literature examines the role of rency, converted into USD. We select the 53 cryp
investor attention on security prices and liquidity tocurrencies among the top-500 for which the
(Barber and Odean 2007; Ding and Hou 2015; trading history is available for the period spanning
Hirshleifer, Lim, and Teoh 2011). The literature 1 January 2017–31 October 2019. These currencies
suggests that increasing investor attention comprise about 85% of the total cryptocurrency
decreases information asymmetry and improves market capitalization (Table 1). Out of these 53
APPLIED ECONOMICS 1729
CCs, there are 49 coins and 4 tokens (Tether, their symbols (e.g. BTC) for all the CCs in our
Augur, Maidsafe, and Golem).4 The descriptive sample set. Prior studies, however, have relied on
statistics of the top-5 currencies are provided in weekly and monthly search volume index data to
Table 2. proxy retail investor attention.
For measuring the liquidity of CCs, we use The descriptive statistics for these liquidity mea
widely employed proxies of bid-ask spread that sures (for the top-5 currencies) are provided in
can be calculated using the daily price and Table 3. The average-correlations (averaged over
volume5 data. These are Amihud measure the 53 CC’s) across these measures are reported in
(Amihud 2002; ‘Amihud’), high-low price measure Table 4. Furthermore, average-correlations of these
(Corwin and Schultz 2012; ‘CS’), high-low-close measures with GSVIs (search terms:
range measure (Będowska-Sójka 2018; ‘RHL’), ‘Cryptocurrency’ and ‘Bitcoin’) are also shown.
high-low range measure (Chung and Zhang 2014; GSVIs exhibit a positive correlation with volume
‘HLR’), and volatility over volume measure (Fong, measure and various spread measures. The positive
Holden, and Tobek 2018; ‘VOV’). The formulae for correlation with spread measures are contradictory
these measures are as follows: to those observed in cases of other asset classes. It
jrt j
(a) Amihud t ¼ logðvolumet Þ suggests that high investor attention is related to
2ðeαt 1Þ
(b) CSt 1þeαt ; increasing volumes along with increasing transac
where, tion costs. Since the keyword ‘Cryptocurrency’ is
pffiffiffiffiffiffi pffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi � � �� 2 a broad keyword associated with the asset class, we
2βt βt γt Ht may expect it to exhibit a higher correlation than
αt ¼ pffiffi pffiffi ; βt ¼ In
3 2 2 3 2 2 Lt the keyword ‘Bitcoin’, which is related to an indi
� � �� 2 � � vidual (though major) currency.
Ht þ 1 maxfHt ; Htþ1 g
þ In and γt ¼ In
Lt þ 1 minfLt ; Ltþ1 g
III. Methodology
The study follows two approaches to measure com
(c) RHLt ¼ HCtt L1t monality: (a) Principal component and canonical
Ht Lt
(d) HLRt ¼ 0:5ðH t þLt Þ correlation analyses approach of Hasbrouck and
�
ln L t
H
Seppi (2001), using the liquidity and GSVI mea
(e) VOVt ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
t
sures in normalized form.6 (b) Market model of
logðvolumet Þ
CRS (2000) shown in Equation (1), which is further
Here, Ht , Lt ,Ct , and volumet are high, low, clos augmented [Equation (2)] with control variables
ing prices, and volume on day t; rt is the price [GSVI for CC full name/symbol, and return and
return on day t based on closing prices. We also volatility7 �of individual CC and the market]. Here,
employ the daily Google Search Volume Index �
Li;t
(GSVI) for these CCs as a proxy of investor atten DLi;t ¼ ln Li;t 1 denotes the daily changes in the
tion (Da, Engelberg, and Gao 2011, 2014). Given liquidity of the i-th Cryptocurrency (CC) from day
the dominance of retail investors in CC market, the t-1 to day t, and is the liquidity variable of interest
GSVI measure of investor attention appears to be (Chordia, Roll, and Subrahmanyam 2000; Karolyi,
a reasonable choice. We use daily GSVIs corre Lee, and Van Dijk 2012; Marshall, Nguyen, and
sponding to full currency names (e.g. Bitcoin) and Visaltanachoti 2013). Li;t is the liquidity measure
4
While the term ‘token’ is often employed interchangeably with ‘coin’, there are important differences (Howell, Niessner, and Yermack 2018; Tasca 2019).
Tokens (e.g. Tether) are CCs that do not have their own blockchain networks. Unlike coins, tokens do not act as a store of value, as they derive value from
some other asset (similar to derivatives). That is, they digitally represent some other asset (such as gold, U.S. Dollar). Moreover, unlike coins, tokens are not
mined. For example, Tether operates on the Ethereum network. It digitally represents the U.S. Dollar, and therefore, derives its value from the U.S. Dollar. The
inclusion of tokens in the study is supported on two grounds. (a) Tokens operate on the same blockchain network as coins (predominantly Ethereum and
Binance Coin) and trade on the CC markets. (b) The investor profile remains similar for both the CCs, i.e. tokens and coins. Therefore, if observed, the presence
of commonality can be largely ascribed to the CC market microstructure and similar investor profile.
5
The volume measure is computed as the natural log of the daily dollar trading volume.
6
The normalization procedure involves scaling the difference between the observation and time-series mean by the standard deviation of the series.
7
Return-squares are employed to proxy volatility. The results remain qualitatively similar with the application of conditional volatility obtained from GARCH
models (not reported here for brevity).
1730 A. TRIPATHI ET AL.
8
Prior literature has suggesed the inclusion of lead and lag terms in such market model regressions (Dang et al. 2015; Galariotis and Giouvris 2007; Karolyi, Lee,
and Van Dijk 2012; Kempf and Mayston 2008; Marshall, Nguyen, and Visaltanachoti 2013; Moshirian et al. 2017). For example, Chordia, Roll, and
Subrahmanyam (2000) state that ‘the leads and lags are designed to capture any lagged adjustment in commonality.’ Infrequent trading introduces bias
in the market model regressions of the kind employed in this study (Dimson 1979). The recorded price and volume information across securities (CCs in this
study) may be non-synchronous, and therefore, any market index constructed from such data is an average of temporally ordered values of the securities.
Thus, the covariance of less frequently traded securities with the market is underestimated and vice-versa. To account for this bias, Dimson (1979) proposed
the aggregated coefficients (AC) method that employs lead, lag, and contemporaneous terms of the market variable. Dimson (1979) argued that infrequently
traded securities exhibit higher covariance with the lagged market term. In contrast, heavily traded securities show higher covariance with the leading market
term. Therefore, the non-inclusion of lead and lag terms may vitiate the estimation.
APPLIED ECONOMICS 1731
Table 2. Descriptive statistics for price returns and unit root tests.
BTC ETH XRP USDT LTC
Panel A: Descriptive statistics for return series
Observations 1,034 1,034 1,034 1,034 1,034
Mean 0.22% 0.30% 0.37% 0.00% 0.25%
Median 0.29% 0.03% −0.29% 0.00% −0.08%
Maximum 22.51% 29.01% 102.74% 5.72% 51.03%
Minimum −20.75% −31.55% −61.63% −4.74% −39.52%
Std. Dev. 4.39% 5.90% 8.10% 0.72% 6.44%
Skewness −0.07 0.23 2.81 0.21 1.13
Kurtosis 6.30 6.96 37.65 12.85 12.11
JB# 470.66*** 685.08*** 53,101.42*** 4,188.24*** 3,794.48***
ARCH(1)+ 40.20*** 49.39*** 87.53*** 166.52*** 18.23***
ARCH(5) 59.00*** 94.02*** 97.81*** 172.70*** 29.29***
Panel B: Unit root test statistics
ADF@ −9.02*** −8.85*** −8.26*** −11.44*** −9.31***
PP& −1061.66*** −1128.32*** −1229.57*** −1269.36*** −1064.73***
The table presents descriptive statistics for the top-5 currencies (by market-capitalization) in the sample, over the study period
(1 January 2017 to 31 October 2019).
Notes:
#
Jarque Bera statistic for the significance of skewness and kurtosis differing from a normal distribution
+
Engle’s ARCH Lagrange Multiplier (LM) test
@
Augmented Dicky-Fuller test
&
Phillips-Perron test
*, ** and *** indicate significance at the 10%, 5% and 1% levels, respectively.
Table 3. Descriptive statistics for daily liquidity measures. the retail investor attention across CCs is driven by
BTC ETH XRP USDT LTC a set of similar factors, and may contribute to the
Observations 1,031 1,031 1,031 1,031 1,031
Panel A: Volume measure [log(volume)] commonality in liquidity.
Mean 22.11 18.76 19.43 20.91 19.85 Next, we follow the market model of CRS (2000)
Median 22.35 19.07 19.86 21.69 19.83
Maximum 24.53 21.27 22.93 24.70 22.66 to examine the commonality. The regressions are
Minimum 17.92 12.64 12.35 13.43 14.78 performed in the fixed-effect panel data frame
Std. Dev. 1.38 1.64 2.01 2.52 1.71
Panel B: Amihud measure (%)
work, and t-statistics are computed using robust
Mean 0.14% 0.24% 0.23% 0.03% 0.21% standard errors.9 The results are shown in Table 6.
Median 0.09% 0.15% 0.12% 0.02% 0.14%
Maximum 1.00% 2.34% 5.34% 0.27% 2.68% Panels A and B present the results with and without
Minimum 0.00% 0.00% 0.00% 0.00% 0.00% control variables. For brevity, we show the details
Std. Dev. 0.14% 0.27% 0.34% 0.03% 0.24%
only for the contemporaneous market liquidity and
Panel C: CS measure (%)
Mean 1.65% 2.59% 2.73% 1.05% 2.32% DGSVI measures. The contemporaneous market
Median 1.33% 2.10% 2.15% 0.97% 1.91%
Maximum 18.98% 25.18% 58.20% 7.88% 16.45%
liquidity coefficient is significantly positive for all
Minimum 0.00% 0.00% 0.00% 0.00% 0.00% the liquidity measures and different model specifi
Std. Dev. 1.42% 2.31% 3.44% 0.74% 1.73%
cations. This suggests that, indeed, there is
Panel D: RHL measure (%)
Mean 5.39% 8.35% 8.54% 1.97% 7.78% a significant presence of liquidity commonality in
Median 4.19% 6.22% 5.58% 1.66% 6.06% the CC market. The results from the pool regres
Maximum 27.75% 78.73% 227.23% 20.19% 68.97%
Minimum 0.39% 0.93% 0.92% 0.00% 0.52% sion model are qualitatively similar, and therefore,
Std. Dev. 4.25% 6.97% 11.47% 1.61% 6.64%
not reported for brevity. These results [shown in
Panel E: HLR measure (%)
Mean 1.35% 2.08% 2.07% 0.49% 1.93% Panel C (Table 6)] suggest that the majority of the
Median 1.05% 1.54% 1.40% 0.41% 1.51% CCs exhibit a positive and significant coefficient for
Maximum 7.59% 14.24% 26.71% 4.75% 13.08%
Minimum 0.10% 0.23% 0.23% 0.00% 0.13% the contemporaneous market liquidity measures.
Std. Dev. 1.08% 1.72% 2.26% 0.40% 1.57% To examine the robustness of these results, the
Panel F: VOV measure (%)
Mean 1.15% 1.92% 1.88% 0.43% 1.73% CRS (2000) model regressions [Equations (1) and
Median 0.89% 1.43% 1.28% 0.36% 1.35% (2)] are estimated separately across individual CCs.
Maximum 6.29% 13.16% 27.16% 4.37% 12.10%
Minimum 0.09% 0.22% 0.21% 0.00% 0.13% The regression coefficients of DGSVI measure
Std. Dev. 0.91% 1.56% 2.07% 0.36% 1.40% (Panel B, Table 6) are consistently positive and
The table provides summary statistics for various daily liquidity measures for significant for most of the liquidity measures,
the top five CCs (by market-capitalization), for the study period
(1 January 2017 to 31 October 2019). which is contradictory to the results of earlier stu
dies. Previous studies on conventional asset classes
suggest that this measure has a negative relation
with spread measures. These studies argue that
Canonical correlations indicate that liquidity mea increasing investor attention leads to higher trad
sures and GSVIs share a significant positive common ing volumes and less information asymmetry, and
variation. Overall, the results from CCA suggest that hence, an improvement in liquidity (Adachi,
9
Heteroscedasticity and autocorrelation consistent (HAC) robust standard errors.
APPLIED ECONOMICS 1733
Table 5. Commonality in the CCs using Principal Component and mean may not fully capture the liquidity dynamics
Canonical Correlation analyses. in CCs. To capture this aspect of commonality, we
PC1 PC2 PC3
employ the quantile regression methodology. The
Panel A: Common variation in liquidity measures explained by the first
three principal components (PC’s) results from CRS (2000) model [Equation (1)]10 in
Volume 43.66% 20.90% 4.24% a quantile regression framework are summarized in
Amihud 26.44% 3.68% 3.27%
CS 16.91% 3.82% 3.12% Figure 1 (detailed results are provided in Table B1,
RHL 27.05% 3.84% 3.59% Appendix B). We show the t-statistic for the coeffi
HLR 36.74% 3.69% 3.15%
VOV 34.69% 4.21% 3.15% cient of contemporaneous market liquidity and
Panel B: Common variation in GSVI measures explained by the first three
goodness-of-fit measure for all the liquidity mea
principal components (PC’s) sures, in Panels A and B.11 These results suggest
Full name of CC that the liquidity behaviour for the extreme quan
Top-10 CCs 43.09% 19.70% 13.49% tiles (i.e. large positive and negative liquidity
All 53 CCs 30.78% 9.13% 7.35%
Symbol of CC changes) differs considerably from that during nor
Top-10 CCs 46.19% 17.11% 10.95%
All 53 CCs 24.04% 9.54% 9.26%
mal liquidity conditions. Relatively lower com
Panel C: First two canonical correlations (CVs) between all the liquidity
monality is observed for the top and bottom
measures and GSVIs deciles than that for the other quantiles. The plau
Coefficient (%) Chi-square sible reasons for this behaviour are as follows.
Top-10 CCs During the times of extreme positive volume
CV1 18.41*** 405.29
CV2 6.95*** 49.92 changes in CCs, driven by marketwide systematic
All (53) CCs shocks, the transaction costs also increase. The
CV1 18.38*** 1,841.50
CV2 5.01*** 125.56 excessive increase in transaction costs impedes
*, ** and *** indicate significance at the 10%, 5% and 1% levels, respectively. the trading activity in CC market, and weakens
Panels A and B report the highest common variation explained by the first
three principal components (PCs), for the volume (natural log of dollar the effect of marketwide systematic factor.
trading volume), Amihud, CS, RHL, HLR, and VOV measures, and GSVIs. For Conversely, during the extreme negative volume
GSVIs, we report the PCAs for the top-10 and all the (53) CCs. The GSVI
keyword search involves two sets, namely, (a) full currency name (e.g. changes, the transaction costs may come down.
Bitcoin) and (b) symbol (e.g. BTC). Panel C presents the first two canonical This may boost the trading activity in CC market,
correlations (CCAs: CV1 and CV2) and corresponding Chi-square statistics
between the set of two covariates comprising liquidity measures (Volume, and in turn, may neutralize the effect of negative
Amihud, CS, RHL, HLR, and VOV) and GSVI measures (full name and
symbol). The CCAs are presented for the top-10 and all the CCs.
marketwide liquidity shocks. This causes the beha
viour of liquidity commonality in CC markets to be
Masuda, and Takeda 2017; Bank, Larch, and Peter different from that for conventional asset classes.
2011; Ding and Hou 2015). Our results indicate The levels of commonality are higher than those
that increasing investor attention, though posi found for other asset classes. However, in contrast
tively related to volume, also has a positive rela to conventional assets, it decreases for the periods
tionship with bid-ask spreads. This contrasting of liquidity shocks.
result for the CC market can be ascribed to the In summary, we ascribe the higher levels of com
unique nature of its transaction cost, which monality in CC markets to the following demand
increases with an increase in volumes beyond cer and supply related factors. First, the process of
tain threshold levels. mining that governs the supply of cryptocurrencies
is similar across all these currencies. Second, even
though these currencies are traded across a number
Quantile approach to commonality
of markets, the transaction costs essentially reflect
The dynamic transaction costs of CC markets make the computational resources spent on mining these
the commonality in these markets time-varying, currencies and may exhibit uniformity due to similar
and dependent on market liquidity conditions. blockchain management protocols. This leads to an
Hence, the methodologies based on conditional artificial similarity in the microstructure design of all
10
Results from Equation (2) are qualitatively similar, and not reported for brevity.
11
The goodness-of-fit measure in quantile regression framework is provided by the ‘goodfit’ function in the package ‘WRTDStidal’. In nature, this function is
similar to the conventional R2 measure from regression. The statistic is calculated as one minus the ratio of the appropriately weighted sum of absolute
deviations from the conditional model to that from the unconditional model.
1734 A. TRIPATHI ET AL.
Table 6. Liquidity commonality across CCs, using CRS (2000) market model regressions.
Panel A: Results for market model regressions without control variables
Market Liquidity
Number of observations Coefficient t-statistic adjusted:R2 (%) F-value
Volume 49,958 1.001*** 25.33 9.94% 219.94***
Amihud 49,958 0.993*** 27.65 19.85% 299.99***
CS 49,958 1.000*** 32.60 5.82% 361.56***
RHL 49,958 0.996*** 36.63 24.78% 454.33***
HLR 49,958 1.012*** 37.33 28.35% 469.91***
VOV 49,958 1.011*** 36.44 28.59% 444.85***
Panel B: Results for market model regressions with control variables
Market Liquidity DGSVI
Number of observations Coefficient t-statistic Coefficient t-statistic adjusted:R2 (%) F-value
Volume 49,958 0.987*** 25.20 0.125*** 7.81 20.54% 179.42***
Amihud 49,958 0.998*** 26.73 0.093*** 4.54 20.84% 180.35***
CS 49,958 1.001*** 32.59 −0.013 −0.69 5.82% 194.86***
RHL 49,958 0.996*** 37.81 0.088*** 7.47 31.65% 316.13***
HLR 49,958 0.996*** 37.05 0.084*** 7.49 32.08% 353.43***
VOV 49,958 1.002*** 35.74 0.082*** 7.38 32.06% 340.00***
Panel C: Robustness tests (Chordia, Roll, and Subrahmanyam 2000)for the sign and significance of market liquidity measures
Without control [Equation (1)] With control [Equation (2)]
Positive and Positive and Positive and Positive and
significant insignificant significant insignificant
Volume 50 2 50 3
Amihud 51 2 51 2
CS 52 1 52 1
RHL 52 1 52 1
HLR 52 1 52 1
VOV 52 1 52 1
*, ** and *** indicate significance at the 10%, 5% and 1% levels, respectively.
Panels A and B summarize the results of market model regressions in the panel fixed-effect framework [Equations (1) and (2)]. Results include the regression
2
coefficients and their t-statistics for the market liquidity (and DGSVI measure),�adjusted:
� R , F-statistics, and the number of observations. For brevity, the table
GSVIt
does not report the statistics pertaining to control variables. DGSVI [ ¼ ln GSVIt 1 � measure reported in Panel B is google search volume index (GSVI)
corresponding to the ‘full name’ of currency. The other variant of the GSVI measure corresponding to the symbol of the currency yields similar results (not
reported for brevity). Following Chordia, Roll, and Subrahmanyam (2000), the regression models [Equations (1) & (2)] are also estimated separately on the
individual CCs (53). In Panel C, we report the number of CCs for which the t-statistics for market liquidity measures (volume, Amihud, CS, RHL, HLR, and VOV)
are positive and significant, and positive but not significant.
the CCs. Moreover, the majority of market partici that the behaviour of commonality is significantly dif
pants in these cryptocurrencies share common traits; ferent around extreme quantiles than that observed
these are tech-savvy committed retail investors that during normal conditions. We contend that the unique
broadly fall under the definition of noise-traders transaction cost structure integral to the CC market –
(influenced by behavioural biases such as overconfi increasing volumes beyond certain thresholds driving
dence and the representative heuristics; Barber, up the transaction costs – attenuates the escalating
Odean, and Zhu 2008). We argue that a set of similar commonality around extreme liquidity quantiles.
factors (e.g. news reports, attention-grabbing public Moreover, investor attention – measured using
events) attract the attention of these retail investors, GSVI – appears to play an important role in driving
and in turn, contribute to liquidity commonality. the liquidity commonality in CCs. The results suggest
that the rising investor attention relates positively to the
spread measures (i.e. illiquidity). The observed relation
V. Conclusion ship between the liquidity and investor attention is in
This paper investigates the phenomenon of liquidity contrast to that observed for other asset classes, and
commonality in the cryptocurrency (CC) market, may be ascribed to the unique transaction cost struc
using 53 CCs. The results show that CCs exhibit higher ture of the CC market. Portfolio managers are particu
commonality compared to other asset classes. The larly concerned about the liquidity commonality
quantile regression approach to commonality suggests behaviour around extreme quantiles, as the rising
APPLIED ECONOMICS 1735
Figure 1. Quantile regression results [Equation (1)] across the conditional quantiles of liquidity. Note: The figure summarizes
the results from the model [Equation (1)] in a quantile regression framework, across the conditional quantiles of liquidity. The time
series of the daily changes in the aggregate (market) liquidity of CCs (contemporaneous, lead, and lag) act as the independent
variables. The corresponding values of the daily changes in the liquidity of individual CCs act as the dependent variable. Panels A &
B show the t-statistics and goodness-of-fit measure (GoF), respectively. The results are shown for the volume, Amihud, CS, RHL, HLR,
and VOV measures.The horizontal axis represents 20 quantiles with increments of 0.05. The goodness-of-fit measure is calculated as
one minus the ratio of the appropriately weighted sum of absolute errors (deviations) from the conditional model and that from the
unconditional model. The goodness-of-fit measure in the quantile regression framework is provided by the ‘goodfit’ function in the
package ‘WRTDStidal’. In nature, this function is similar to the conventional R2 measure for regression. Detailed results corresponding
to this figure are provided in Table B1 of Appendix B.
transaction costs can offset the diversification benefits. manuscript has been read and approved by all named authors
Therefore, the results from this study are useful to and that there are no other persons who satisfied the criteria
investors as well as academics interested in alternative for authorship but are not listed. I further confirm that the
order of authors listed in the manuscript has been approved
asset classes such as CCs. by all of us. I understand that the Corresponding Author is the
sole contact for the Editorial process and is responsible for
communicating with the other authors about progress, sub
Confirmation of corresponding author missions of revisions and final approval of proofs.
Fry, J. 2018. “Booms, Busts and Heavy-tails: The Story of Koenker, R., S. Portnoy, P. T. Ng, A. Zeileis, P. Grosjean, and
Bitcoin and Cryptoc urrency Markets?” Ec onomics B. D. Ripley. 2018. “Package ‘Quantreg’.” Cran R-Project. Org.
Letters 171: 225–229. doi:10.1016/j.econlet.2018.08.008. Koenker, R., and Z. Xiao. 2006. “Quantile Autoregression.”
Galariotis, E. C., and E. Giouvris. 2007. “Liquidity Journal of the American Statistical Association 101 (475):
Commonality in the London Stock Exchange.” Journal of 980–990. doi:10.1198/016214506000000672.
Business Finance & Accounting 34 (1–2): 374–388. Korajczyk, R. A., and R. Sadka. 2008. “Pricing the common
doi:10.1111/j.1468-5957.2006.00664.x. ality across alternative measures of liquidity.” Journal of
Hameed, A., W. Kang, and S. Viswanathan. 2010. “Stock market Financial Economics, 87(1): 45–72.
declines and liquidity.” The Journal of Finance, 65(1): 257–293. Koutmos, D. 2018. “Liquidity Uncertainty and Bitcoin’s
Hasbrouck, J., and D. J. Seppi. 2001. “Common Factors in Prices, Market Microstructure.” Economics Letters 172: 97–101.
Order Flows, and Liquidity.” Journal of Financial Economics doi:10.1016/j.econlet.2018.08.041.
59 (3): 383–411. doi:10.1016/S0304-405X(00)00091-X. Mancini, L., A. Ranaldo, and J. Wrampelmeyer. 2013.
Hirshleifer, D., S. S. Lim, and S. H. Teoh. 2011. “Limited “Liquidity in the foreign exchange market: Measurement,
Investor Attention and Stock Market Misreactions to commonality, and risk premiums.” The Journal of Finance,
Accounting Information.” The Review of Asset Pricing 68(5): 1805–1841.
Studies 1 (1): 35–73. doi:10.1093/rapstu/rar002. Marshall, B. R., N. H. Nguyen, and N. Visaltanachoti. 2013.
Ho, T. S., and H. R. Stoll. 1983. “The Dynamics of Dealer “Liquidity Commonality in Commodities.” Journal of
Markets under Competition.” The Journal of Finance Banking & Finance 37 (1): 11–20. doi:10.1016/j.
38 (4): 1053–1074. doi:10.1111/j.1540-6261.1983.tb02282.x. jbankfin.2012.08.013.
Howell, S. T., M. Niessner, and D. Yermack. 2018. “Initial Marshall, B. R., N. N. H. Nguyen, and N. Visaltanachoti. 2019.
Coin Offerings: Financing Growth with Cryptocurrency “Bitcoin Liquidity.” SSRN Working Paper No. ID 3194869.
Token Sales.” The Review of Financial Studies. 33(9): 10.2139/ssrn.3194869
3925–3974. https://doi.org/10.1093/rfs/hhz131 Moshirian, F., X. Qian, C. K. G. Wee, and B. Zhang. 2017.
Kamara, A., X. Lou, and R. Sadka. 2008. “The divergence of “The Determinants and Pricing of Liquidity Commonality
liquidity commonality in the cross-section of stocks.” around the World.” Journal of Financial Markets 33: 22–41.
Journal of Financial Economics, 89(3): 444–466. doi:10.1016/j.finmar.2017.02.004.
Kang, W., and H. Zhang. 2013.“ Limit order book and com Nakamoto, S. 2008. “Bitcoin: A Peer-to-peer Electronic Cash
monality in liquidity. Financial Review, 48(1): 97–122. System.” https://bitcoin.org/bitcoin
Karolyi, G. A., K.-H. Lee, and M. A. Van Dijk. 2012. Pástor, L., and R. F. Stambaugh. 2003.“ Liquidity risk and expected
“Understanding Commonality in Liquidity around the stock returns.” Journal of Political Economy, 111(3): 642–685.
World.” Journal of Financial Economics 105 (1): 82–112. Rösch, C. G., and C. Kaserer. 2013.“ Market liquidity in the
doi:10.1016/j.jfineco.2011.12.008. financial crisis: The role of liquidity commonality and flight-
Kempf, A., and D. Mayston. 2008. “Liquidity Commonality to-quality. ”Journal of Banking & Finance, 37(7): 2284–2302.
beyond Best Prices.” Journal of Financial Research 31 (1): Smith, C., and A. Kumar. 2018. “Crypto-Currencies–An
25–40. doi:10.1111/j.1475-6803.2008.00230.x. Introduction to Not-So-Funny Moneys.” Journal of
Koch, A., S. Ruenzi, and L. Starks. 2016. “Commonality in Economic Surveys 32 (5): 1531–1559. doi:10.1111/
Liquidity: A Demand-side Explanation.” The Review of joes.12289.
Financial Studies 29 (8): 1943–1974. doi:10.1093/rfs/hhw026. Stoll, H. R. 1978. “The Supply of Dealer Services in
Koenker, R. 2012. “Quantile Regression in R: A Vignette.” Securities Markets.” The Journal of Finance 33 (4):
Retrieved November. 1133–1151. doi:10.1111/j.1540-6261.1978.tb02053.x.
Koenker, R., and G. Bassett Jr. 1978. “Regression Quantiles.” Tasca, P. 2019. “Token-based Business Models.” In Disrupting
Econometrica: Journal of the Econometric Society 46 (1): Finance, edited by Theo Lynn, John G. Mooney, Pierangelo
33–50. doi:10.2307/1913643. Rosati, Mark Cummins,135–148. Cham: Palgrave Pivot. http://
Koenker, R., and J. A. Machado. 1999. “Goodness of Fit and library.oapen.org/handle/20.500.12657/23126 .
Related Inference Processes for Quantile Regression.” Tripathi, A., Dixit, A., and Vipul. 2019. “Liquidity of Financial
Journal of the American Statistical Association 94 (448): Markets: A Review.” Studies in Economics and Finance 37
1296–1310. doi:10.1080/01621459.1999.10473882. (2): 201–227. https://doi.org/10.1108/SEF-10-2018-0319
Koenker, R., and K. F. Hallock. 2001. “Quantile Regression.” Wei, W. C. 2018. “Liquidity and Market Efficiency in
Journal of Economic Perspectives 15 (4): 143–156. Cryptocurrencies.” Economics Letters 168: 21–24.
doi:10.1257/jep.15.4.143. doi:10.1016/j.econlet.2018.04.003.
1738 A. TRIPATHI ET AL.
Appendix A
Appendix B
Table B1: Quantile regression results [Equation (1)] across the conditional quantiles of liquidity.
Panel A: Volume measure
Quantiles (τ = .05 to 0.50)
Variables 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50
Number of observations 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958
t-stat (Market) 25.53 39.77 52.47 63.60 69.05 72.65 75.50 75.28 75.38 75.44
GoF 3.94% 4.86% 5.61% 6.16% 6.44% 6.61% 6.70% 6.76% 6.83% 6.91%
Quantiles (τ = .55 to 0.95)
Variables 0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95
Number of observations 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958
t-stat (Market) 75.62 75.77 76.48 73.13 69.25 65.89 58.64 41.31 29.79
GoF 7.01% 7.13% 7.26% 7.29% 7.23% 7.01% 6.59% 5.96% 4.95%
Panle B: Amihud measure
Quantiles (τ = .05 to 0.50)
Variables 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50
Number of observations 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958
t-stat (Market) 29.36 41.17 52.80 59.70 67.96 75.38 77.30 79.04 80.56 81.01
GoF 8.97% 10.06% 10.80% 11.38% 11.85% 12.15% 12.38% 12.53% 12.60% 12.56%
Quantiles (τ = .55 to 0.95)
Variables 0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95
Number of observations 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958
t-stat (Market) 81.05 79.35 76.64 74.30 69.79 61.11 53.39 43.03 29.11
GoF 12.66% 12.69% 12.60% 12.40% 12.09% 11.64% 11.02% 10.22% 8.97%
Panle C: CS measure
Quantiles (τ = .05 to 0.50)
Variables 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50
Number of observations 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958
t-stat (Market) 15.64 28.14 34.56 43.31 47.23 47.85 46.02 40.62 35.06 31.70
GoF 2.29% 3.19% 3.82% 4.06% 4.18% 4.08% 3.76% 3.13% 2.10% 1.54%
Quantiles (τ = .55 to 0.95)
Variables 0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95
Number of observations 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958
t-stat (Market) 32.10 35.34 40.06 43.71 46.85 42.91 39.93 30.80 20.07
GoF 1.87% 2.83% 3.46% 3.87% 4.08% 4.13% 3.94% 3.53% 2.87%
Panle D: RHL measure
Quantiles (τ = .05 to 0.50)
Variables 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50
Number of observations 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958
t-stat (Market) 40.42 57.10 69.79 81.93 95.83 105.89 109.34 118.42 123.46 124.11
GoF 8.82% 11.03% 12.70% 13.94% 14.93% 15.74% 16.39% 16.94% 17.33% 17.59%
Quantiles (τ = .55 to 0.95)
Variables 0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95
Number of observations 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958
t-stat (Market) 121.05 118.33 116.21 114.50 104.98 95.31 84.88 67.85 42.22
GoF 17.79% 17.99% 18.05% 17.95% 17.65% 17.19% 16.42% 14.96% 12.03%
Panel E: HLR measure
Quantiles (τ = .05 to 0.50)
Variables 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50
Number of observations 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958
t-stat (Market) 46.08 64.64 74.68 87.42 102.15 112.50 115.67 124.09 129.23 129.30
GoF 11.13% 13.26% 14.84% 15.96% 16.84% 17.62% 18.23% 18.75% 19.13% 19.38%
Quantiles (τ = .55 to 0.95)
Variables 0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95
Number of observations 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958 49,958
t-stat (Market) 127.56 125.72 124.72 122.73 112.20 102.17 91.76 71.15 45.88
GoF 19.62% 19.86% 19.99% 19.96% 19.77% 19.40% 18.74% 17.46% 14.85%
Panel F: VOV measure
Quantiles (τ = .05 to 0.50)
(Continued)
APPLIED ECONOMICS 1741