Content uploaded by Gang-Jin Wang
Author content
All content in this area was uploaded by Gang-Jin Wang on Jan 26, 2021
Content may be subject to copyright.
When Bitcoin meets economic policy uncertainty (EPU):
Measuring risk spillover effect from EPU to Bitcoin
Gang-Jin Wanga,∗
, Chi Xiea, Danyan Wenb, Longfeng Zhaoc,∗
aBusiness School and Center for Finance and Investment Management,
Hunan University, Changsha 410082, China
bSchool of Economics and Management, Nanjing University of Science and Technology,
Nanjing 210094, China
cSchool of Management, Xi’an Polytechnic University, Xi’an 710048, China
Abstract
Bitcoin was launched to solve the distrust and uncertainty in the existing financial
system. Here we investigate risk spillover effect from economic policy uncertainty
(EPU) to Bitcoin using a multivariate quantile model and the Granger causality
risk test. We use the US EPU index, equity market uncertainty index, and VIX
as proxies for EPU. We find that risk spillover effect from EPU to Bitcoin is
negligible in most conditions. Our work provides useful information on building
asset portfolios for investors who have investment strategies in Bitcoin, because
Bitcoin can be acted as a safe-haven or a diversifier under EPU shocks.
Keywords: Bitcoin, Economic policy uncertainty, Risk spillover effect,
Multivariate quantile model, Granger causality risk test
JEL: G15, C32, D81
1. Introduction
In the context of distrusting the existing international financial system and
facing the extreme economic uncertainty, on 31 October 2008, Nakamoto (2008)
released a nine-page white paper describing a new monetary system, i.e., Bitcoin,
a fully decentralized cryptocurrency based on the blockchain technology. On 16
December 2008, Satoshi Nakamoto released an early version of Bitcoin network
∗Corresponding authors.
Email addresses: wanggangjin@hnu.edu.cn (Gang-Jin Wang), xiechi@hnu.edu.cn
(Chi Xie), wendy2018@njust.edu.cn (Danyan Wen), zlfccnu@mails.ccnu.edu.cn
(Longfeng Zhao)
Preprint submitted to FRL November 8, 2018
and the first units of the bitcoin cryptocurrency (called bitcoins). Since then,
Bitcoin has attracted much attention from practitioners, academics and the media
due to its unique decentralized payment or trust system that does not rely on
the third parties (e.g., financial institutions). Especaillay during the 2010–2013
European sovereign debt crisis and the 2012–2013 Cypriot banking crisis, many
people resorted to Bitcoin as a safe-heaven or hedging asset to avoid risk and
market uncertainty (Bouri et al.,2017b).
The current world is full of economic policy uncertainty (EPU), e.g., financial
crisis and trade war. In this background, the price of Bitcoin increased sharply
from $0.09 on 18 July 2010 to $7487.19 on 31 May 2018, and its price and market
capitalization reached their record high of $19343.04 and $326 billion on 16–17
December 2017, respectively.1Thus a natural question is raised: when Bitcoin
meets EPU, does EPU affect the behavior of Bitcoin? In the existing literature,
few studies have focused on this topic. To the best of our knowledge, Bouri et al.
(2017b) and Demir et al. (2018) are the only two works to investigate the impact
of EPU on Bitcoin using the ordinary least squares (OLS) and (wavelet-based)
quantile-on-quantile regressions. Another closely related work on this topic is that
of Bouri et al. (2018b), who use copula-based techniques to examine the quan-
tile conditional dependence and causality between Bitcoin and the global financial
stress index (GFSI), showing that Bitcoin can be a safe-haven against global fi-
nancial stress. However, their studies are limited to analyzing the relationship
between EPU/GFSI and Bitcoin return and ignore the risk spillover effect from
EPU to Bitcoin. Here we aim to investigate whether EPU affects the behavior
of Bitcoin from a risk spillover perspective. Namely, we study the risk spillover
effect from EPU to Bitcoin or the impact of EPU shocks on Bitcoin. Meanwhile,
a hypothesis is proposed: if Bitcoin is really independent of the existing economic
and financial system, it will be little affected by EPU, i.e., the risk spillover effect
from EPU to Bitcoin should be weak or negligible. For this propose, we use two
different approaches, i.e., the multivariate quantile model (MVQM) of White et al.
(2015) and the Granger causality risk test of Hong et al. (2009), to examine the
risk spillover effect from EPU to Bitcoin. The US EPU index and the US equity
market uncertainty (EMU) index developed by Baker et al. (2016) are used to
represent EPU,2and the CBOE volatility index (VIX) is also considered for ro-
bustness test. We employ daily and weekly data of Bitcoin price and the US EPU,
EMU and VIX indeices during the period from 18 July 2010 to 31 May 2018.
1Source: http://www.coindesk.com/price/ and https://coinmarketcap.com.
2Note that in this paper the term “EPU” denotes economic policy uncertainty, and
such terms as “the EPU index” and “the US EPU index” refer to the US economic policy
uncertainty index developed by Baker et al. (2016).
2
Our work has the following three contributions. First, our study is the first at-
tempt to examine the impact of EPU on Bitcoin from a risk spillover perspective,
which complements to the previous research of Bouri et al. (2017b) and Demir
et al. (2018).3Unlike their research that only uses daily data of either the US
EPU index or the VIX index for empirical analysis, we use daily and weekly data
of the US EPU, EMU and VIX indices, making our results more comprehensive
and robust. Second, our work is the first one using two different approaches. i.e.,
the MVQM and the Granger causality risk test, to examine the risk spillover effect
from EPU to Bitcoin at different quantiles (i.e., risk levels). Based on a condi-
tional autoregressive value-at-risk (CAViaR) model of Engle & Manganelli (2004)
using regression quantiles, both of the MVQM and the Granger causality risk test
are inherently nonlinear models, which allow us to quantify the risk spillover ef-
fect varying across the quantiles (e.g., downside and upside risks). Especially, the
Granger causality risk test also allows us to investigate the risk spillover effect
at different time-lags, and this is important because the information transmission
usually has a time-lag effect. Third, our results confirm the proposed hypothesis
that the risk spillover effect from EPU to Bitcoin is negligible. Using the MVQM
approach we find that VaRs of Bitcoin at different quantiles are only related to its
own lagged volatilities and VaRs, and are independent of both the lagged volatil-
ities and lagged VaR of EPU, meaning that the impact of EPU on Bitcoin is
insignificant from the risk spillover view. The results by the Granger causality risk
test show that the risk spillover effect from EPU to Bitcoin is negligible or weak
varying across quantiles and lags.
In the existing literature, much empirical research (see, e.g., Beckmann & Czu-
daj,2017;Chuli´a et al.,2017a,b) finds that traditional financial asset markets are
somewhat affected by EPU. However, this conclusion does not hold true for cur-
rent the largest digital currency — Bitcoin, based on our investigation on risk
spillover effect from EPU to Bitcoin using the MVQM-CAViaR and the Granger
causality risk test. Unlike the research conducted by Bouri et al. (2017b) and
Demir et al. (2018) who find Bitcoin returns are negatively correlated with EPU
and conclude that Bitcoin can act as a hedge against uncertainty, we find negligible
or weak risk spillover effect from EPU to Bitcoin, indicating that Bitcoin differs
from traditional financial asset markets and is somewhat isolated from the current
3According to Corbet et al. (2018a) who provide an excellent and systemic review on
the topics of Bitcoin and other cryptocurrencies as a financial asset, most of the current
literature focuses on diversification benefits by Bitcoin or relationships between Bitcoin
and other traditional assets (see, e.g., Bri`ere et al.,2015;Dyhrberg,2016a,b;Bouri et al.,
2017c,d;Baur et al.,2018a,b;Giudici & Abu-Hashish,2018;Ji et al.,2018;Bouri et al.,
2018a,b;Symitsi & Chalvatzis,2018;Feng et al.,2018;Corbet et al.,2018b).
3
economic and financial environment. Our findings provide important information
for investors who have an interest in using Bitcoin as an investment or hedging
asset in their diversified portfolios, because the negligible risk spillover effect from
EPU to Bitcoin means that when investors are in an extremely uncertain econom-
ic policy situation Bitcoin can be considered as a safe-haven or a diversifier. For
example, according to recent media reports (see, e.g., Masters.,2018;Maurya.,
2018) and CoinMarketCap, since the Turkish currency and debet crisis sparked
and the Lira started plunging, many Turkish investors have shifted their holdings
to Bitcoin and other cryptocurrencies, leading to a surge in the cryptocurrency
trading volume on Turkey’s cryptocurrency exchanges (e.g., Paribu and BtcTurk).
Note that regulatory polices on Bitcoin or cryptocurrencies can heavily influence
the decision-making behavior of market participants. Currently, different countries
or regulatory agencies show different attitudes and views on cryptocurrencies. For
example, some counties such as Germany, the UK and the USA are still careful-
ly examining and weighing the regulatory and legal framework for cryptocurren-
cies. Thus our study on risk spillover effect from EPU to Bitcoin also supplies
valuable information for regulatory agencies when they define the positioning of
cryptocurrencies in their financial system and set relevant regulatory policies on
cryptocurrencies.
2. Methodology
2.1. Multivariate quantile model (MVQM)
The MVQM is a VAR extension to quantile models, also called as VAR for
VaR. White et al. (2015) develop a reduced and structural form of the MVQM,
i.e., a bivariate MVQM(1,1), which is a multivariate extension of CAViaR models
proposed by Engle & Manganelli (2004). Thus MVQM is also called as MVQM-
CAViaR model. The MVQM is straightforward, i.e., quantiles (i.e., VaR) of a
time series (e.g., Bitcoin returns r1,t) distribution depend on the lags of exogenous
and endogenous variables, e.g., its own lags and the lags of other related variables
(e.g., EPU changes r2,t). Mathematically, the MVQM-CAViaR(1,1) we used for
examining impact of EPU on Bitcoin is defined as follows:
q1,t =c1+a11|r1,t−1|+a12 |r2,t−1|+b11q1,t−1+b12 q2,t−1,(1)
q2,t =c2+a21|r1,t−1|+a22 |r2,t−1|+b21q1,t−1+b22 q2,t−1,(2)
where |r1,t−1|and |r2,t−1|represent absolute values of returns of Bitcoin and
changes of EPU respectively, which can be considered as volatility of Bitcoin and
EPU, and q1,t and q2,t are conditional quantiles of the distributions of Bitcoin re-
turns and EPU changes. The definition of q1,t and q2,t shows that they actually
4
represent VaRs of the two variables (Shen,2018), which are defined as
qi,t = VaRi,t =−Qθ(ri,t |It−1) = −inf
q{q∈R|Pr(ri,t ≤q|It−1)≥θ}, i = 1,2,(3)
where Qθis a quantile function at the confidence level θ∈(0,1) and It−1is the
available information set at time t−1.
In Eq. (1), quantiles of Bitcoin returns r1t(q1t), at the confidence level θ,
depend on itself with a lag (q1,t−1) by b11, on its volatility with a lag (|r1,t−1|)
through a11, on EPU’s volatility with a lag (|r2,t−1|) via a12 , and importantly
on the uncertainly-quantiles with a lag (q2,t−1) by b22. Eq. (2) holds a similar
interpretation for EPU. Since we mainly consider the risk spillover effect from
EPU to Bitcoin or the influence of EPU on Bitcoin, here we focus our attention
on Eq. (1) by analyzing the four coefficients a11,a12,b11 and b12 , especially b12
that represents the degree of risk spillover effect from EPU to Bitcoin. Thus in our
empirical analysis, we follow Chuli´a et al. (2017b) and only report the estimated
coefficients associated with Eq. (1).
2.2. The Granger causality risk test
The idea behind the Granger causality risk test of Hong et al. (2009) is that
given two variables, e.g., Bitcoin and EPU, EPU can be considered as Granger
causes risk to Bitcoin if the capacity to predict the future risk information of
Bitcoin is improved via adding the past risk information of EPU (Wang et al.,
2017). In what follows we briefly introduce the Granger causality risk test.
We first obtain VaR estimation for Bitcoin and EPU using the asymmetric
slope model in the CAViaR framework of Engle & Manganelli (2004), and then
transform the VaR into a risk indicator, i.e.,
Zi,t =1(ri,t <−VaRi,t ), i = 1,2,(4)
where 1(·) is an indicator function.
The statistic for the Granger causality risk test builds on the cross-correlation
function (CCF) between two risk indicators ˆ
Zi,t (i=1,2), and the sample CCF is
defined as
ˆρ(j) = ˆ
C(j)
ˆ
S1ˆ
S2
,(5)
where ˆ
C(j) is the sample cross-covariance function between two risk indicators at
positive lag j, which is defined as
ˆ
C(j) = T−1
T
t=1+j
(ˆ
Z1,t −ˆα1)( ˆ
Z2,t−j−ˆα2),1≤j≤T−1,(6)
5
where the sample mean ˆαi=T−1T
t=1 Zi,t,Tis the sample length, and the sample
variance ˆ
S2
i= ˆαi(1 −ˆαi).
Based on the sample CCF, a kernel-based statistic proposed by Hong et al.
(2009) for examining the one-way Granger causality in risk from one variable (e.g.,
EPU) to another variable (e.g., Bitcoin) is defined as
Q(M) =
T
T
j=1
k2(j/M)ˆρ2(j)−CT(M)
[DT(M)]1/2,(7)
where the centering and standardization constants are defined as
CT(M) = T−1
j=1 (1 −j/T )k2(j/M),(8)
DT(M) = 2 T−1
j=1 (1 −j/T )(1 −(j+ 1)/T )k4(j/M),(9)
k(·) is a kernel function assigning weights to various lags, and Mis the bandwidth,
i.e., the lag order. Common kernel functions include the truncated kernel, the
Daniel kernel, the Bartlett kernel, the Parzen kernel and the Quadratic-Spectral
kernel. Following Wang et al. (2016,2017) and Shen (2018), here we use the Daniel
kernel, defined as k(x)=sin(πx)/(πx). Under the null hypothesis that one variable
(e.g., EPU) does not Granger-cause risk to another variable (e.g., Bitcoin), the
kernel-based statistic Q(M) obeys an asymptotically standard normal distribution
N(0,1) (Hong et al.,2009). If the null hypothesis is rejected when the value of
Q(M) is larger than the right-tailed critical value of N(0,1) at the significance
level β, there is one-way Granger causality in risk from EPU to Bitcoin, namely,
there is a risk spillover effect from EPU to Bitcoin.
3. Data
Following Chuli´a et al. (2017b) who investigate spillover effects from the US
EPU and EMU indices to emerging and developed stock markets using the MVQM-
CAViaR and Demir et al. (2018) who study whether the US EPU index can predict
the Bitcoin returns using the three approaches (e.g., the quantile regression), we
use the US EPU and EMU indices developed by Baker et al. (2016) as two proxies
for EPU. We obtain daily data of the US EPU and EMU indices from the website
of Economic Policy Uncertainty Index (http://www.policyuncertainty.com) devel-
oped by Baker et al. (2016). We also follow Bouri et al. (2017b), who examine
whether Bitcoin can hedge EPU, and use the VIX index as an alternative proxy
for EPU. We collect daily prices of the VIX index from Yahoo Finance (http-
s://finance.yahoo.com). Like Bouri et al. (2017b) and Demir et al. (2018), we
6
collect daily price data of Bitcoin in USD from the website of CoinDesk (http-
s://www.coindesk.com/price/). The reason why we use Bitcoin price data from
CoinDesk is that its Bitcoin price index is an average of Bitcoin prices across major
Bitcoin exchanges in the world and thus it can well reflect the overall trend of the
Bitcoin market. The sample period for the above data is from 18 July 2010 to 31
May 2018. Following Demir et al. (2018), we consider the logarithmic returns of
Bitcoin or changes of the EPU, EMU, and VIX indices, i.e., ri,t = ln(Pi,t/Pi,t−1),
where Pi,t is Bitcoin price or the value of the three indices at time t.
About data frequency, we mainly use daily data of the above variables for
empirical analysis and also consider weekly data for robustness test. Note that
data of Bitcoin and the US EPU and EMU indices are based on calendar days
(including weekdays, weekends and holidays), thus their daily returns or changes
have 2874 observations during the entire period from 19 July 2010 to 31 May
2018. Because the VIX data are based on business days (with 1981 observations),
in order to study the risk spillover effect from the VIX to Bitcoin, we match
Bitcoin data with the VIX data by removing weekend and holiday data of Bitcoin.
Thus the data set can be divided two groups, group I (Bitcoin, EPU and EMU)
with calendar-day observations and group II (Bitcoin and VIX) with business-day
observations. We transform the daily prices into weekly prices by averaging prices
of the whole week. Similar to daily data, weekly data of Bitcoin in group I and
group II are different (see Table 1).
Previous research has different conclusions on market efficiency of Bitcoin and
relations between Bitcoin and other assets before and after the December 2013
Bitcoin price crash. For example, Urquhart (2016) obtains evidence of the inef-
ficiency of Bitcoin during the entire sample period, but he finds that the Bitcoin
market has moved toward an efficient market since the latter half of 2013. Bouri
et al. (2017c) find that the hedge and safe-haven properties of Bitcoin against com-
modities during the pre-crash period and the post-crash period are significantly
different. Bouri et al. (2017a) show that there exist differences in the return-
volatility relationship of Bitcoin before and after the crash. Thus, to figure out
whether structural breakpoints (e.g., the December 2013 Bitcoin price crash) in
Bitcoin returns affect the risk spillover effect from EPU to Bitcoin, we check the
robustness over the sample period by using multiple structural change models of
Bai & Perron (2003) to detect breakpoints in Bitcoin returns. The detection re-
sults point toward a structural breakpoint on 5 December 2013 in Bitcoin returns,4
which approaches to the December 2013 Bitcoin price crash. This is also consistent
4Detailed results on the structural breakpoint testing are not reported here but available
from the authors.
7
with the results on structural breakpoint testing by Cheah & Fry (2015) and Bouri
et al. (2017a), even though our sample length differs from (i.e., is larger than) their
sample lengths. Thus we divide the entire period into (i) subperiod I from 19 July
2010 to 4 December 2013 and (ii) subperiod II from 5 December 2013 to 31 May
2018.
In Table 1we show descriptive statistics on daily and weekly Bitcoin returns
and changes of the US EPU, EMU and VIX indices during the entire period. We
also present descriptive statistics for daily data during two subperiods. Note that
for all periods including the entire periods (with daily observations and weekly
observations) and two subperiods, there are two groups for the data set, meaning
that there are two data sets for Bitcoin during each period in Table 1. For all
periods, Bitcoin has the positive average return and large standard deviation (SD),
showing an attractive investment opportunity in Bitcoin but with high risk. The
changes of three proxies for EPU (i.e., the US EPU, EMU and VIX indices) show
negative mean values and very high SDs during all periods (except for the EMU
index in subperiod II with positive mean value), suggesting that economic policy
has extreme uncertainty and large fluctuation. Except for the EMU index during
the entire period (weekly observations) and subperiod I, all returns or changes are
skewed and have a kurtoisis value in excess of the critical value (i.e., three) of a
normal distribution, and their Jarque-Bera statistics reject the null hypothesis at
1% significance level, suggesting that all returns or changes disobey the normal
distribution and are fat-tailed. This finding supports our decision on using CAViaR
and its extension to estimate VaRs or conditional quantiles, which do not need any
assumption on the distributions of Bitcoin returns and EPU changes. The ADF
statistic show that all returns or changes are a stationary series without unit root
and are suitable for the further modeling.
4. Empirical results
4.1. Results from MVQM-CAViaR(1,1)
Here we investigate Bitcoin’s reaction to EPU, i.e., risk spillover effects from
EPU to Bitcoin using the reduced and structural form VAR-quantile, i.e., MVQM-
CAViaR(1,1). We consider four quantiles (or confidence levels), including θ=0.01,
0.05, 0.95, and 0.99. Especially, VaRs at the former two quantiles and the latter
two quantiles represent downside risk and upside risk, respectively.
In Table 2we report estimated coefficients associated with Eq. (1) at four
quantiles during the entire period, where Panels A and B show the results based
on daily and weekly data, respectively. Both Panels A and B show that the
estimated coefficients b12 at different quantiles, representing the impact level of
the US EPU and EMU indices on Bitcoin in VaR, are very small (close to zero)
8
and statistically insignificant. This finding suggests that downside and upside risk
spillover effects from EPU to Bitcoin are statistically negligible during the entire
period. Almost all estimated coefficients a12, measuring the influence level of the
volatility of the US EPU index or EMU index on Bitcoin’s VaR, are also small
and statistically insignificant, except for the two cases of the weekly EPU index
at the 99th percentile (θ=0.99) and the weekly EMU index at the 5th percentile
(θ=0.05). This means that the influence of the volatility of EPU on Bitcoin’s VaR
is negligible. Almost all estimated autoregressive coefficients b11 are statistically
significant, except for the weekly EMU index at 1st percentile (θ=0.01), suggesting
that the VaR of Bitcoin is autocorrelated. The estimated coefficients a11 are
statistically significant except for two cases of the weekly EMU index at the 1st
and 5th percentiles, which indicates that the VaR of Bitcoin is affected by its lagged
volatility.5In summary, we have the following findings: (i) the risk spillover effect
from EPU to Bitcoin is insignificant, and (ii) the VaR of Bitcoin is affected by its
lagged VaR and volatility. These findings suggest that the risk of Bitcoin price is
independent of the changes of EPU and is related to its previous risk and volatility
information.
For robustness test, we also estimate coefficients associated with Eq. (1) at
four quantiles during two subperiods (subperiods I and II) to check whether the
December 2013 Bitcoin price crash affects our results. In Table 3we report the
estimated results for the two subperiods. Both Panels A and B of Table 3show that
the estimated coefficients b12 at different quantiles are statistically insignificant,
suggesting that the December 2013 Bitcoin price crash does not change our finding
on the negligible risk spillover effect from EPU to Bitcoin.
We further use the VIX index instead of the US EPU and EMU indices for
examining the risk spillover effect from EPU to Bitcoin. In Table 4we present
estimated coefficients associated with Eq. (1) at four quantiles during the entire
period and two subperiods. Based on the daily and weekly data of Bitcoin and
the VIX index during the entire period and the daily data of Bitcoin and the VIX
index during two subperiods (i.e., before and after the December 2013 Bitcoin
price crash), the estimated coefficients b12 at different quantiles confirm that there
is no risk spillover effect from EPU to Bitcoin, or the impact of EPU on Bitcoin is
negligible. The empirical results based on the VIX index also confirm (i) that the
effect of volatility of EPU on Bitcoin is insignificant and (ii) that the risk of Bitcoin
(in terms of VaR) is observably affected by its own lagged risk and volatility.
5It is not difficult to understand that the VaR is related to the volatility. For exam-
ple, we usually use variance-covariance approaches (e.g., RiskMetrics and GARCH-type)
to estimate the VaR of financial asset, but this type of approaches needs to know the
distribution of returns.
9
4.2. Results from the Granger causality risk test
In this section, we estimate statistics Q(M) associated with Eq. (7) and the
corresponding p-values of one-way Granger causality in risk for measuring risk
spillover effect from EPU to Bitcoin at different lags. We follow Hong et al. (2009)
and Wang et al. (2016) and examine risk spillover effects with the lag orders M=5,
10 and 20 days/weeks. Like the MVQM-CAViaR analysis in Section 4.1, we also
consider four quantiles (or confidence levels), i.e., θ=0.01, 0.05, 0.95, and 0.99, and
three periods including the entire period and two subperiods.6
In Table 5we show estimated statistics Q(M) for measuring risk spillover effect
from the US EPU and EMU indices to Bitcoin during the entire period. Panel
A based on daily data shows that almost all estimated statistics across different
quantiles and lags are insignificant, except for a case of the EPU index at 10%
significant level when the confidence level θ=0.01 and lag order M=20, suggesting
the absence of risk spillover effect from EPU to Bitcoin. Panel B shows that most
of estimated statistics based on weekly data support the above finding. But there
are some exceptions, including that estimated statistics are significant from the
EPU index to Bitcoin when θ=0.95 and M=5 and 10 and from the EMU index
to Bitcoin when θ=0.99 and M=5, 10 and 20. When the US EPU index or EMU
index shows extreme upside risk representing that the uncertainty shock increases
sharply, EPU has an upside risk spillover effect on Bitcoin, leading to Bitcoin price
increases. But note that this result only holds up in certain cases for weekly data.
For robustness check, in Table 6we report estimated statistics Q(M) for mea-
suring risk spillover effect from the US EPU and EMU indices to Bitcoin during
two subperiods. Panel A for subperiod I shows that most of estimated statistics
support the finding on negligible risk spillover effect from EPU to Bitcoin. There
are three exceptions, including the existence of risk spillover effect from the US
EPU index to Bitcoin (i) when θ=0.99 and M=10 and 20, and (ii) when θ=0.01
and M=10 and 20, and (iii) from the US EMU index to Bitcoin when θ=0.01 and
M=20. The latter two exceptions can be interpreted as follows: when the US
EPU index or EMU index shows extreme downside risk meaning that the uncer-
tainty shock decreases steeply, EPU has a downside risk spillover effect on Bitcoin,
6Note that both Bouri et al. (2017b) and Demir et al. (2018) find negative relations
between Bitcoin and EPU, which are built on contemporaneous or instantaneous correla-
tions between Bitcoin and the EPU or VIX index. To figure out whether contemporaneous
or instantaneous correlations affect our finding, we follow Wang et al. (2016) and also use
a modified statistic of one-way Granger causality in risk by adding a CCF with a lag order
of zero. We find that contemporaneous or instantaneous correlations do not change our
central finding on the insignificant risk spillover effect from EPU to Bitcoin. The detailed
results can be obtained from the authors.
10
and this results in a decrease of Bitcoin price. However, after the December 2013
Bitcoin price crash, the above exceptions all disappear in subperiod II, confirming
again the inexistence of risk spillover risk effect from EPU to Bitcoin.
In Table 7we present estimated statistics Q(M) for measuring risk spillover
effect from the VIX index to Bitcoin during the entire period and two subperiods.
Panel A based on daily data during the entire period shows evidence of downside
risk spillover effect from the VIX index to Bitcoin, because estimated statistics
Q(M) are significant when θ=0.01 and M=5, 10, and 20 and when θ=0.05 and
M=5 and 10. But this evidence disappears when examining weekly data during
the entire period (see Panel B), meaning the inexistence of risk spillover effect from
the VIX index to Bitcoin. For subperiods I and II in Panels C and D, only when
θ=0.01 estimated statistics Q(M) are significant, and the insignificant risk spillover
effect from the VIX to Bitcoin holds in other conditions. On the whole, Table 7
shows that estimated statistics Q(M) for the VIX index in most conditions are
statistically insignificant, confirming that risk spillover effect from EPU to Bitcoin
is insignificant.
5. Conclusions
We have investigated the risk spillover effect from the US EPU, EMU and VIX
indices to Bitcoin using two different approaches, i.e., the MVQM-CAViaR and
the Granger causality risk test. We have examined risk spillover effect using daily
and weekly data during the entire period and also checked whether the December
2013 Bitcoin price crash influences the spillover results by splitting the sample
period into two subperiods. We have further studied whether contemporaneous
or instantaneous correlations impact risk spillover effect. The empirical results
based on the MVQM-CAViaR approach show that the impact of the US EPU,
EMU and VIX shocks on Bitcoin’s risk is negligible, while the Granger causality
risk test shows that risk spillover effect from the US EPU, EMU and VIX indices
to Bitcoin is insignificant in most conditions (i.e., different quantiles and time-
lags). Our results are robust to data frequency, the influence of the December
2013 Bitcoin price crash, and contemporaneous or instantaneous correlations. Our
finding on the inexistence risk spillover effect from EPU to Bitcoin provides new
information for investors when they construct asset portfolios, e.g., Bitcoin can
be used as a safe-haven or a diversifier in condition of extreme EPU shocks. In
the future, our work can be extended to examine the impact of economic policy
uncertainty on more cryptocurrencies (e.g., Ethereum, Ripple and Litecoin) for
checking whether the cryptocurrency market is immune from EPU shocks.
11
References
Bai, J., & Perron, P. (2003). Computation and analysis of multiple structural change
models. Journal of Applied Econometrics,18 , 1–22.
Baker, S. R., Bloom, N., & Davis, S. J. (2016). Measuring economic policy uncertainty.
Quarterly Journal of Economics,131 , 1593–1636.
Baur, D. G., Dimpfl, T., & Kuck, K. (2018a). Bitcoin, gold and the US dollar – A
replication and extension. Finance Research Letters,25 , 103–110.
Baur, D. G., Hong, K., & Lee, A. D. (2018b). Bitcoin: Medium of exchange or speculative
assets? Journal of International Financial Markets, Institutions and Money,54 , 177–
189.
Beckmann, J., & Czudaj, R. (2017). Exchange rate expectations and economic policy
uncertainty. European Journal of Political Economy,47 , 148–162.
Bouri, E., Azzi, G., & Haubo Dyhrberg, A. (2017a). On the return-volatility relationship
in the Bitcoin market around the price crash of 2013. Economics: The Open-Access,
Open-Assessment E-Journal,11 , 1–16.
Bouri, E., Gupta, R., Tiwari, A. K., & Roubaud, D. (2017b). Does bitcoin hedge global
uncertainty? evidence from wavelet-based quantile-in-quantile regressions. Finance
Research Letters,23 , 87–95.
Bouri, E., Jalkh, N., Moln´ar, P., & Roubaud, D. (2017c). Bitcoin for energy commodities
before and after the December 2013 crash: diversifier, hedge or safe haven. Applied
Economics,49 , 5063–5073.
Bouri, E., Moln´ar, P., Azzi, G., Roubaud, D., & Hagfors, L. I. (2017d). On the hedge and
safe haven properties of Bitcoin: Is it really more than a diversifier? Finance Research
Letters,20 , 192–198.
Bouri, E., Das, M., Gupta, R., & Roubaud, D. (2018a). Spillovers between Bitcoin and
other assets during bear and bull markets. Applied Economics,50 , 5935–5949.
Bouri, E., Gupta, R., Lau, C. K. M., Roubaud, D., & Wang, S. (2018b). Bitcoin and
global financial stress: A copula-based approach to dependence and causality in the
quantiles. The Quarterly Review of Economics and Finance ,69 , 297–307.
Bouri, E., Shahzad, S. J. H., Raza, N., & Roubaud, D. (2018c). Oil volatility and sovereign
risk of BRICS. Energy Economics,70 , 258–269.
Bri`ere, M., Oosterlinck, K., & Szafarz, A. (2015). Virtual currency, tangible return:
Portfolio diversification with bitcoin. Journal of Asset Management,16 , 365–373.
Cheah, E.-T., & Fry, J. (2015). Speculative bubbles in Bitcoin markets? An empirical
investigation into the fundamental value of Bitcoin. Economics Letters,130 , 32–36.
Chuli´a, H., Guill´en, M., & Uribe, J. M. (2017a). Spillovers from the United States to Latin
American and G7 stock markets: A VAR quantile analysis. Emerging Markets Review,
31 , 32–46.
Chuli´a, H., Gupta, R., Uribe, J. M., & Wohar, M. E. (2017b). Impact of US uncertain-
ties on emerging and mature markets: Evidence from a quantile-vector autoregressive
approach. Journal of International Financial Markets, Institutions and Money,48 ,
178–191.
Corbet, S., Lucey, B. M., Urquhart, A., & Yarovaya, L. (2018a). Cryptocurrencies as a
financial asset: A systematic analysis. International Review of Financial Analysis,
doi:10.1016/j.irfa.2018.09.003.
12
Corbet, S., Meegan, A., Larkin, C., Lucey, B., & Yarovaya, L. (2018b). Exploring the
dynamic relationships between cryptocurrencies and other financial assets. Economics
Letters,165 , 28–34.
Demir, E., Gozgor, G., Lau, C. K. M., & Vigne, S. A. (2018). Does economic policy
uncertainty predict the bitcoin returns? An empirical investigation. Finance Research
Letters,26 , 145–149.
Dyhrberg, A. H. (2016a). Bitcoin, gold and the dollar — A GARCH volatility analysis.
Finance Research Letters ,16 , 85–92.
Dyhrberg, A. H. (2016b). Hedging capabilities of bitcoin. Is it the virtual gold? Finance
Research Letters,16 , 139–144.
Engle, R. F., & Manganelli, S. (2004). CAViaR: Conditional autoregressive value at risk
by regression quantiles. Journal of Business & Economic Statistics,22 , 367–381.
Feng, W., Wang, Y., & Zhang, Z. (2018). Can cryptocurrencies be a safe haven: A tail
risk perspective analysis. Applied Economics ,50 , 4745–4762.
Giudici, P., & Abu-Hashish, I. (2018). What determines bitcoin exchange prices? A
network VAR approach. Finance Research Letters, doi:10.1016/j.frl.2018.05.013.
Hong, Y., Liu, Y., & Wang, S. (2009). Granger causality in risk and detection of extreme
risk spillover between financial markets. Journal of Econometrics,150 , 271–287.
Ji, Q., Bouri, E., Gupta, R., & Roubaud, D. (2018). Network causality structures among
bitcoin and other financial assets: A directed acyclic graph approach. The Quarterly
Review of Economics and Finance, doi:10.1016/j.qref.2018.05.016.
Masters, C. Bitcoin purchases grow popular in Turkey as Lira falls. (2018).
https://cryptovest.com/news/bitcoin-purchases-grow-popular-in-turkey-as-lira-falls/
Accessed 13 August 2018.
Maurya, N. Crypto trading volume hikes as Turkey citizens interest shifts from plung-
ing Lira to Bitcoin. (2018). https://coingape.com/crypto-trading-volume-hikes-turkey-
shifts-plunging-lira/ Accessed 13 August 2018.
Nakamoto, S. (2008). Bitcoin: A peer-to-peer electronic cash system. http-
s://bitcoin.org/bitcoin.pdf .
Shen, Y. (2018). International risk transmission of stock market movements. Economic
Modelling,69 , 220–236.
Symitsi, E., & Chalvatzis, K. J. (2018). Return, volatility and shock spillovers of bitcoin
with energy and technology companies. Economics Letters,170 , 127–130.
Urquhart, A. (2016). The inefficiency of bitcoin. Economics Letters ,148 , 80–82.
Wang, G.-J., Xie, C., He, K., & Stanley, H. E. (2017). Extreme risk spillover network:
Application to financial institutions. Quantitative Finance,12 , 1417–1433.
Wang, G.-J., Xie, C., Jiang, Z.-Q., & Stanley, H. E. (2016). Extreme risk spillover effects
in world gold markets and the global financial crisis. International Review of Economics
& Finance,46 , 55–77.
White, H., Kim, T.-H., & Manganelli, S. (2015). VAR for VaR: Measuring tail dependence
using multivariate regression quantiles. Journal of Econometrics,187 , 169–188.
Yi, S., Xu, Z., & Wang, G.-J. (2018). Volatility connectedness in the cryptocurrency
market: Is Bitcoin a dominant cryptocurrency?. International Review of Financial
Analysis,60 , 98–114.
13
Table 1: Descriptive statistics on returns of Bitcoin and changes of the US EPU, EMU and VIX indices during the entire
period and two subperiods.
Mean Median Maximum Minimum Std. Dev. Skewness Kurtosis Jarque–Bera ADF Observations
Panel A: The entire period (Daily observations)
BTCa0.0039 0.0014 0.4246 −0.4915 0.0585 −0.3486 14.6854 16409.74∗∗∗ −52.25∗∗∗ 2874
EPU −0.0003 −0.0070 2.6025 −3.1483 0.4961 −0.0479 4.6511 327.57∗∗∗ −25.11∗∗∗ 2874
EMU −0.0004 −0.0492 4.2257 −4.1866 1.0237 0.0408 3.4428 24.28∗∗∗ −23.92∗∗∗ 2874
BTCb0.0058 0.0020 0.4997 −0.4700 0.0664 −0.0319 11.7345 6297.63∗∗∗ −18.96∗∗∗ 1981
VIX −0.0003 −0.0045 0.7682 −0.3141 0.0766 1.1791 11.1477 5938.53∗∗∗ −47.46∗∗∗ 1981
Panel B: The entire period (Weekly observations)
BTCa0.0283 0.0120 0.6782 −0.4508 0.1291 1.0522 7.2292 381.22∗∗∗ −12.50∗∗∗ 410
EPU −0.0023 −0.0102 0.9581 −0.9417 0.2645 −0.0499 4.2020 24.85∗∗∗ −16.61∗∗∗ 410
EMU −0.0024 −0.0305 2.0781 −2.1360 0.6341 0.1627 3.0601 1.87 −14.60∗∗∗ 410
BTCb0.0057 0.0027 0.1561 −0.1645 0.0323 0.5155 8.6041 556.04∗∗∗ −17.44∗∗∗ 410
VIX −0.0003 −0.0015 0.2281 −0.1198 0.0333 1.3725 11.6181 1400.94∗∗∗ −15.44∗∗∗ 410
Panel C: Subperiod I
BTCa0.0077 0.0006 0.4246 −0.4915 0.0744 −0.3918 11.6987 3925.30∗∗∗ −33.76∗∗∗ 1235
EPU −0.0004 −0.0114 2.2326 −3.1483 0.4402 −0.2592 6.6475 698.46∗∗∗ −16.27∗∗∗ 1235
EMU −0.0009 0.0288 3.1326 −3.7556 1.0647 0.0070 2.9236 0.31 −27.10∗∗∗ 1235
BTCb0.0112 0.0021 0.4997 −0.4700 0.0834 −0.0881 9.4757 1489.78∗∗∗ −14.62∗∗∗ 852
VIX −0.0007 −0.0042 0.4055 −0.3141 0.0697 0.7517 6.9187 625.38∗∗∗ −20.48∗∗∗ 852
Panel D: Subperiod II
BTCa0.0011 0.0016 0.2908 −0.2696 0.0426 −0.4158 9.9130 3310.86∗∗∗ −41.06∗∗∗ 1639
EPU −0.0002 −0.0031 2.6025 −1.9826 0.5345 0.0417 3.7521 39.11∗∗∗ −18.90∗∗∗ 1639
EMU 0.0001 −0.0849 4.2257 −4.1866 0.9920 0.0723 3.9276 60.18∗∗∗ −28.73∗∗∗ 1639
BTCb0.0058 0.0020 0.4997 −0.4700 0.0664 −0.0319 11.7345 6297.63∗∗∗ −18.96∗∗∗ 1129
VIX −0.0003 −0.0045 0.7682 −0.3141 0.0766 1.1791 11.1477 5938.53∗∗∗ −47.46∗∗∗ 1129
Notes: The entire period is from 19 July 2010 to 31 May 2018, subperiod I is from 19 July 2010 to 4 December 2013, and subperiod
II is from 5 December 2013 to 31 May 2018. aand bindicate Bitcoin data with calendar-day observations and with business-
day observations, respectively. Jarque–Bera statistic tests for the null hypothesis of normal distribution. The ADF (Augmented
DickeyCFuller) statistic tests for a unit root. ∗∗∗,∗∗ , and ∗indicate statistical significance at 1%, 5%, and 10% level, respectively.
14
Table 2: Estimated coefficients associated with Eq. (1) of MVQM-CAViaR(1,1) using the US EPU and EMU indices during
the entire period.
θEPU EMU
c1a11 a12 b11 b12 c1a11 a12 b11 b12
Panel A: The entire period (Daily observations)
0.01 −0.049 −0.548∗∗ −0.005 0.790∗∗∗ −0.032 −0.012 −0.464∗∗∗ −0.007 0.847∗∗∗ −0.005
(0.045) (0.223) (0.013) (0.073) (0.038) (0.060) (0.087) (0.008) (0.031) (0.028)
0.05 0.000 −0.153∗∗∗ 0.004 0.924∗∗∗ 0.002 −0.003 −0.147∗∗∗ −0.001 0.931∗∗∗ −0.002
(0.005) (0.026) (0.005) (0.017) (0.007) (0.007) (0.022) (0.005) (0.012) (0.006)
0.95 0.005 0.406∗∗∗ −0.004 0.767∗∗∗ 0.001 0.000 0.399∗∗ 0.005 0.768∗∗∗ 0.001
(0.008) (0.042) (0.004) (0.027) (0.011) (0.179) (0.176) (0.038) (0.135) (0.122)
0.99 0.000 0.427∗∗∗ −0.014 0.848∗∗∗ 0.009 0.002 0.417∗∗∗ 0.008 0.841∗∗∗ −0.001
(0.026) (0.100) (0.010) (0.037) (0.022) (0.405) (0.147) (0.006) (0.129) (0.169)
Panel B: The entire period (Weekly observations)
0.01 0.000 −0.436∗∗∗ −0.011 0.811∗∗∗ 0.005 −0.121 −0.624 0.024 0.369 0.000
(0.014) (0.061) (0.020) (0.033) (0.022) (0.211) (0.672) (0.101) (0.527) (0.154)
0.05 0.010 −0.146∗−0.004 0.917∗∗∗ 0.016 −0.534∗∗∗ 0.008 0.060∗∗ −1.141∗∗∗ −0.183
(0.011) (0.081) (0.034) (0.053) (0.037) (0.148) (0.117) (0.027) (0.269) (0.112)
0.95 0.179∗1.073∗∗∗ −0.056 0.285∗∗∗ −0.241 0.075 1.140∗∗∗ 0.016 0.286∗∗ −0.032
(0.096) (0.170) (0.040) (0.096) (0.180) (0.085) (0.198) (0.018) (0.126) (0.071)
0.99 0.109∗∗∗ 1.111∗∗∗ −0.128∗∗∗ 0.452∗∗∗ −0.040 0.155 1.190∗∗∗ 0.048 0.394 −0.072
(0.033) (0.040) (0.014) (0.045) (0.044) (0.113) (0.137) (0.042) (0.281) (0.030)
Notes: Here we only report estimated coefficients in Eq. (1), mainly representing the impact of the US EPU and EMU
indices on Bitcoin, but the estimated results of Eq. (2) can be available upon request. The entire period is from 19 July
2010 to 31 May 2018. Numbers in parentheses are standard errors of the corresponding estimated coefficients. ∗∗∗ ,∗∗ , and
∗indicate statistical significance at 1%, 5%, and 10% level, respectively.
15
Table 3: Estimated coefficients associated with Eq. (1) of MVQM-CAViaR(1,1) using the US EPU and EMU indices during
two subperiods.
θEPU EMU
c1a11 a12 b11 b12 c1a11 a12 b11 b12
Panel A: Subperiod I
0.01 −0.055 −0.981 0.010 0.611 −0.011 0.333 −0.684 0.007 0.751 0.146
(1.761) (13.295) (1.443) (6.021) (0.355) (1.427) (0.637) (0.013) (0.316) (0.601)
0.05 −0.006 −0.318∗∗∗ 0.011 0.833∗∗∗ −0.001 0.007 −0.321∗∗∗ 0.002 0.828∗∗∗ 0.005
(0.004) (0.059) (0.007) (0.027) (0.008) (0.017) (0.059) (0.004) (0.028) (0.010)
0.95 0.000 0.575∗∗∗ −0.009 0.654∗∗∗ 0.022 0.001 0.343∗∗∗ 0.002 0.832∗∗∗ 0.000
(0.017) (0.085) (0.010) (0.056) (0.025) (0.044) (0.072) (0.009) (0.051) (0.026)
0.99 0.048 0.951 −0.030 0.503∗∗∗ 0.009 0.024 0.842 0.043 0.442 0.001
(0.179) (0.829) (0.038) (0.163) (0.202) (329.100) (48.504) (8.431) (872.430) (65.519)
Panel B: Subperiod II
0.01 0.002 −0.355 −0.013 0.885∗∗∗ 0.000 0.002 −0.216 0.000 0.910 0.003
(0.088) (0.547) (0.053) (0.215) (0.073) (47.301) (29.765) (2.376) (451.730) (33.869)
0.05 0.003 −0.181∗∗∗ 0.014 0.905∗∗∗ 0.012 −0.019 −0.461 0.000 0.679∗−0.007
(0.007) (0.060) (0.005) (0.030) (0.010) (0.027) (0.342) (0.008) (0.365) (0.011)
0.95 −0.001 0.370∗∗∗ 0.006 0.739∗∗∗ 0.005 0.046 0.289 0.014 0.420 −0.017
(0.015) (0.064) (0.009) (0.045) (0.021) (0.675) (0.940) (0.075) (0.271) (0.418)
0.99 0.010 0.482∗∗∗ −0.001 0.755∗∗∗ 0.001 0.039 0.726 0.004 0.183 0.010
(0.016) (0.098) (0.013) (0.072) (0.015) (0.168) (0.216) (0.060) (0.337) (0.061)
Notes: Here we only report estimated coefficients in Eq. (1), mainly representing the impact of the US EPU and EMU
indices on Bitcoin, but the estimated results of Eq. (2) can be available upon request. Subperiod I is from 19 July
2010 to 4 December 2013 and subperiod II is from 5 December 2013 to 31 May 2018. Numbers in parentheses are
standard errors of the corresponding estimated coefficients. ∗∗∗ and ∗indicate statistical significance at 1% and 10%
level, respectively.
16
Table 4: Estimated coefficients associated with Eq. (1) of MVQM-CAViaR(1,1) using the VIX index during the entire period
and two subperiods.
θ c1a11 a12 b11 b12 c1a11 a12 b11 b12
Panel A: The entire period (Daily observations) Panel B: The entire period (Weekly observations)
0.01 0.005 −0.166∗∗∗ −0.110 0.946∗∗∗ 0.005 −0.001 −0.444∗∗ −0.103 0.856∗∗∗ −0.034
(0.008) (0.044) (0.131) (0.028) (0.079) (0.059) (0.223) (0.188) (0.099) (1.057)
0.05 0.000 −0.187∗∗∗ −0.001 0.916∗∗∗ 0.006 −0.072∗∗∗ 0.033 −0.102 −0.968∗∗∗ 0.107
(0.003) (0.030) (0.054) (0.021) (0.044) (0.026) (0.039) (0.092) (0.082) (0.474)
0.95 0.007 0.389∗∗∗ −0.061 0.788∗∗∗ 0.009 0.052 1.107∗∗∗ −0.184∗0.304 −0.686
(0.022) (0.088) (0.097) (0.055) (0.225) (0.073) (0.408) (0.111) (0.496) (1.285)
0.99 0.018∗∗∗ 0.347∗∗∗ −0.106 0.879∗∗∗ −0.023 0.021 0.781∗∗∗ −0.315∗∗∗ 0.591∗∗ 0.026
(0.006) (0.062) (0.057) (0.042) (0.022) (0.035) (0.266) (0.049) (0.065) (0.243)
Panel C: Subperiod I Panel D: Subperiod II
0.01 −0.013 −0.429∗−0.578 0.694∗∗∗ 0.023 −0.002 −0.483∗−0.164 0.810∗∗∗ −0.003
(0.072) (0.250) (1.147) (0.145) (0.754) (0.092) (0.250) (0.863) (0.200) (0.970)
0.05 0.004 −0.198∗∗∗ −0.027 0.897∗∗∗ 0.039 −0.003 −0.207∗∗ −0.004 0.878∗∗∗ −0.010
(0.006) (0.059) (0.093) (0.042) (0.088) (0.005) (0.097) (0.130) (0.053) (0.099)
0.95 0.015 0.458∗∗∗ 0.021 0.697∗∗∗ −0.010 0.007 0.269∗∗∗ −0.054 0.839∗∗∗ −0.003
(0.031) (0.071) (0.155) (0.060) (0.363) (0.006) (0.050) (0.022) (0.035) (0.068)
0.99 0.052 0.618∗−0.137 0.787∗∗∗ −0.138∗0.007 0.291∗∗∗ −0.103∗0.902∗∗∗ 0.007
(0.021) (0.329) (0.150) (0.125) (0.073) (0.016) (0.062) (0.054) (0.021) (0.076)
Notes: Here we only report estimated coefficients in Eq. (1), mainly representing the impact of the US EPU and EMU
indices on Bitcoin, but the estimated results of Eq. (2) can be available upon request. The entire period is from 19 July
2010 to 31 May 2018, subperiod I is from 19 July 2010 to 4 December 2013, and subperiod II is from 5 December 2013
to 31 May 2018. Numbers in parentheses are standard errors of the corresponding estimated coefficients. ∗∗∗,∗∗ , and ∗
indicate statistical significance at 1%, 5%, and 10% level, respectively.
17
Table 5: Estimated statistics Q(M) associated with Eq. (7) for measuring extreme risk
spillover effects from the US EPU and EMU indices to Bitcoin using the Granger causality
risk test during the entire period.
θEPU⇒BTC EMU⇒BTC
M= 5 M= 10 M= 20 M= 5 M= 10 M= 20
Panel A: The entire period (Daily observations)
0.01 −0.653 0.289 1.522∗−0.700 −0.157 0.542
[0.743] [0.386] [0.064] [0.758] [0.562] [0.294]
0.05 −0.765 −0.714 −0.059 −0.810 −1.203 −1.230
[0.778] [0.762] [0.523] [0.791] [0.886] [0.891]
0.95 0.598 0.756 1.229 −0.202 −0.037 0.546
[0.275] [0.225] [0.110] [0.580] [0.515] [0.293]
0.99 −0.891 −0.942 −1.103 −0.905 −1.254 −1.752
[0.813] [0.827] [0.865] [0.817] [0.895] [0.960]
Panel B: The entire period (Weekly observations)
0.01 −0.044 0.058 0.906 −0.294 −0.623 −0.862
[0.518] [0.477] [0.183] [0.616] [0.733] [0.806]
0.05 −0.786 −1.056 −1.247 0.113 0.635 0.949
[0.784] [0.855] [0.894] [0.455] [0.263] [0.171]
0.95 2.452∗∗∗ 1.362∗0.797 −0.106 −0.232 −0.718
[0.007] [0.087] [0.213] [0.542] [0.592] [0.764]
0.99 −1.182 −1.537 −2.159 4.364∗∗∗ 8.670∗∗∗ 6.440∗∗∗
[0.881] [0.938] [0.985] [0.000] [0.000] [0.000]
Notes: “EPU⇒BTC” and “EMU⇒BTC” represent one-way Granger
causality in risk from the US EPU index to Bitcoin and from the US
EMU index to Bitcoin, respectively. The entire period is from 19 July
2010 to 31 May 2018. Numbers in brackets are p-values of the corre-
sponding estimated coefficients. ∗∗∗ and ∗indicate statistical signifi-
cance at 1% and 10% level, respectively.
18
Table 6: Estimated statistics Q(M) associated with Eq. (7) for measuring extreme risk
spillover effects from the US EPU and EMU indices to Bitcoin using the Granger causality
risk test during two subperiods.
θEPU⇒BTC EMU⇒BTC
M= 5 M= 10 M= 20 M= 5 M= 10 M= 20
Panel A: Subperiod I
0.01 −0.349 2.615∗∗∗ 6.687∗∗∗ −0.696 1.128 3.164∗∗∗
[0.636] [0.005] [0.000] [0.757] [0.130] [0.000]
0.05 −0.477 −0.932 0.256 0.152 −0.196 −0.764
[0.683] [0.824] [0.399] [0.440] [0.578] [0.778]
0.95 −0.277 −0.457 1.209 0.340 0.141 −0.025
[0.609] [0.676] [0.113] [0.367] [0.444] [0.510]
0.99 −1.256 2.588∗∗∗ 6.382∗∗∗ −1.077 −1.488 −2.047
[0.895] [0.005] [0.000] [0.859] [0.932] [0.980]
Panel B: Subperiod II
0.01 −0.683 −0.846 −1.174 −0.734 −0.636 −0.641
[0.753] [0.801] [0.880] [0.769] [0.738] [0.739]
0.05 −0.745 −0.787 −0.805 −0.730 −0.894 −0.380
[0.772] [0.784] [0.790] [0.767] [0.814] [0.648]
0.95 −0.262 0.265 0.164 −0.536 −0.834 −0.502
[0.604] [0.396] [0.435] [0.704] [0.798] [0.692]
0.99 −0.469 0.035 0.244 −0.846 −1.129 −1.076
[0.680] [0.486] [0.404] [0.801] [0.871] [0.859]
Notes: “EPU⇒BTC” and “EMU⇒BTC” represent one-way Granger
causality in risk from the US EPU index to Bitcoin and from the US
EMU index to Bitcoin, respectively. Subperiod I is from 19 July 2010 to
4 December 2013 and subperiod II is from 5 December 2013 to 31 May
2018. Numbers in brackets are p-values of the corresponding estimated
coefficients. ∗∗∗ indicates statistical significance at 1% level.
19
Table 7: Estimated statistics Q(M) associated with Eq. (7) for measuring risk spillover
effect from the VIX index to Bitcoin using the Granger causality risk test during the entire
period and two subperiods.
θ M = 5 M= 10 M= 20 M= 5 M= 10 M= 20
Panel A: The entire period (Daily) Panel B: The entire period (Weekly)
0.01 1.369∗2.296∗3.329∗∗∗ 0.176 −0.312 −0.419
[0.086] [0.011] [0.000] [0.430] [0.622] [0.662]
0.05 2.037∗∗ 1.302∗1.074 −0.954 −0.519 −0.686
[0.021] [0.096] [0.141] [0.830] [0.698] [0.754]
0.95 0.857 0.104 0.163 −0.319 −0.646 −0.979
[0.196] [0.459] [0.435] [0.625] [0.741] [0.836]
0.99 −0.872 −1.063 −0.545 −1.074 −1.538 −0.461
[0.809] [0.856] [0.707] [0.859] [0.938] [0.678]
Panel C: Subperiod I Panel D: Subperiod II
0.01 2.306∗∗ 3.789∗∗∗ 3.737∗∗∗ 0.905 1.829∗∗ 3.419∗∗∗
[0.011] [0.000] [0.000] [0.183] [0.034] [0.000]
0.05 0.777 0.190 −0.240 0.186 0.002 −0.280
[0.219] [0.425] [0.595] [0.426] [0.499] [0.610]
0.95 0.396 −0.113 −0.193 −0.422 −0.999 −1.465
[0.346] [0.545] [0.576] [0.664] [0.841] [0.929]
0.99 0.184 0.820 0.893 −0.757 0.431 1.063
[0.427] [0.206] [0.186] [0.775] [0.333] [0.144]
Notes: The entire period is from 19 July 2010 to 31 May 2018, subperiod I is from
19 July 2010 to 4 December 2013, and subperiod II is from 5 December 2013 to
31 May 2018. Numbers in brackets are p-values of the corresponding estimated
coefficients. ∗∗∗,∗∗ , and ∗indicate statistical significance at 1%, 5%, and 10% level,
respectively.
20
